Abstract
Accurate, reliable, and cost-effective quantification of real-time biomechanical exposures in occupational settings remains an enduring pursuit in ergonomics. Miniaturized, wireless, body-worn inertial sensors offer opportunities to directly measure vast and personalized kinematics data in both laboratory and applied settings. This review investigated the contemporary and emerging uses of wearable inertial sensing technology in occupational ergonomics research related to biomechanical exposure assessment in physical work. A review and narrative synthesis of 78 peer-reviewed studies was conducted. A conceptual framework was used for scoping and synthesizing the reviewed scientific literature. Review findings help to contextualize contributions of this emerging technology to the broader goals of reducing work-relevant musculoskeletal trauma disorders. The review made evident that despite the growing interest in wearable inertial sensing technologies for ergonomics research, its use in applied settings still lags. The review also identified differences in sensor attachment locations and methods and measures for calibration and validation, and inconsistent criteria for reporting and assessing biomechanical exposures even across studies with similar objectives. Emerging applications include combining inertial sensing with predictive modeling for obtaining cumulative exposure data, and providing real-time feedback about biomechanical work demands. The manuscript concludes with research directions for enabling inertial sensing technologies as a tool for online biomechanical exposure assessment and feedback, which has particular appeal in non-repetitive work settings.
Keywords: ergonomics, exposure assessment, inertial sensors, wearable sensors, accelerometers
1. Introduction
Wearable inertial sensors have garnered much attention for human motion measurement in both laboratory and naturalistic settings (Bergmann et al., 2009). Much of this interest has stemmed from improvements in the miniaturization and reduced weight of the sensor hardware, decreased manufacturing cost (Barbour and Schmidt, 2001), advancement in wireless connectivity and data processing software, and increased battery life and data bandwidth which allows for continuous monitoring over long durations (e.g., multiple days or work shifts). The resulting hardware portability and less interference with a worker’s natural work postures (Scott and Browning, 2016) have made body-worn inertial sensors a suitable alternative for direct measurement of postures and movements in ergonomics studies over camera-based motion tracking systems.
Despite the growing interest in wearable inertial sensing technologies for ergonomics research, its adoption in applied occupational settings still lags. A recent survey of ergonomics practitioners in the US attributed this slow uptake to concerns about employee data privacy and compliance, sensor durability, information validity and efficacy (Schall Jr. et al., 2018). Efficacy in this context includes the usefulness and practicality of the exposure-related information conveyed to the wearer (i.e., instrumented worker) or to the ergonomist/practitioner, for instance about cumulative workload, work-relevant Musculoskeletal Disorder (MSD) risk, or about priorities for job modification based on the type and severity of the exposure level (David, 2005). Beyond generating accurate and reliable data, wearable sensor-based exposure assessment tools would need to convey actionable but straightforward information about cumulative exposures and/or risk of musculoskeletal trauma to effectively inform intervention designs (David, 2005). Furthermore, wireless inertial sensing technologies connected to web services (i.e., cloud-based) have the potential to compute and communicate such information in near real-time. In order to better understand these contemporary and emerging uses of wearable inertial sensing technology in occupational ergonomics, this review focused on synthesizing research literature related to body-worn inertial sensing for assessing biomechanical exposures and MSD risk resulting from physical work.
1.1. Why Biomechanical Exposure Assessment?
Prior epidemiological studies provide strong evidence of a biomechanical pathway between physical work exposures and the increased risk of work-relevant MSDs (Marras et al., 1995; Putz-Anderson et al., 1997; Punnett et al., 1991; Keyserling et al., 1992). The cumulative trauma model provides one explanation for the occupational relevance of musculoskeletal disorders specifically due to repeated and long durations of external load handling and/or forceful exertions performed throughout the workday (Radwin et al., 2001). This model posits that injury results from the accumulated effect of transient external loads that in isolation may be not exceed internal tissue tolerances. Exposures from repetitive and/or prolonged duration cause cumulative microdamage such that the internal tolerances of tissues are eventually exceeded.
Biomechanical exposures during physical work are often characterized by three main dimensions: intensity (e.g., load magnitudes, extent of non-neutral postures), repetition (frequency or number of force exertions and motions), and duration (the time the physical activity is performed, sustained non-neutral postures) (Winkel and Mathiassen, 1994). Traditional exposure assessment techniques have multiple limitations when applied to field evaluations of biomechanical exposures. Observational methods such as the Owako Working Posture System (OWAS) (Buchholz et al., 1996; Karhu et al., 1977), Rapid Upper Limb Assessment (RULA) (McAtamney and Corlett, 1993), and Rapid Entire Body Assessment (REBA) (Hignett and McAtamney, 2000) can be time and labor intensive (Takala et al., 2010). Direct measurement methods allow for long duration of data collection, and are considered advantageous in terms of performance (e.g., accuracy, precision) and cost (Winkel and Mathiassen, 1994). However, portability and wearability of direct instrumentation is vital to minimize potential interference with worker movements and work performance. Examples of direct instrumentation used for field measurement of work postures include electronic goniometers (Radwin and Lin, 1993), the lumbar motion monitor (Marras et al., 1992), and inclinometers (Hansson et al., 2001). Examples of force measurement instrumentation in situ are pressure mapping insoles (Cordero et al., 2004), instrumented force shoes (Faber et al., 2010), instrumented gloves (Castro and Cliquet, 1997), and electromyography (EMG) for estimating the magnitude of force exertion from muscle activity (Theado et al., 2007).
When studying routine or cyclical work, biomechanical exposures are typically measured by sampling work to estimate either all or a subset of the three main exposure dimensions (i.e., intensity, repetition, and duration). These estimates are usually extrapolated to a longer time period (e.g., workday or shift) to quantify the cumulative biomechanical exposure. On the other hand, non-repetitive jobs defined as jobs that display variation and diversity in terms of work element frequency, duration, or content (Gold et al., 2006; Mathiassen, 2006) present unique challenges for biomechanical exposure assessment. Non-repetitive jobs may contain repetitive elements or fundamental cycles (Silverstein et al., 1986), but the intensity, repetition, and duration of the task varies over time and between workers. Examples include patient handling (Jensen, 1988; Videman et al., 1984; Garg and Owen, 1992; Knibbe and Knibbe, 1996), material handling at construction sites (Liira et al., 1996), distribution centers and ware-housing operations (Waters et al., 1998), and team lifting (Sharp et al., 1997) where the magnitude of hand loads vary during the workday and/or are difficult to measure directly. In such cases, exposure assessment performed on a small sample of workers or using discrete-interval work sampling may not capture the work-relevant exposures and MSD risk (Paquet et al., 2005). This implies the work assessment may need to be performed across multiple workers over long work periods to capture a representative profile of biomechanical exposures. Accurate, reliable and cost-effective quantification of real-time biomechanical exposures in such work settings remains an enduring pursuit in ergonomics, and provides the underlying motivation for this review.
1.2. Study objectives
The objectives of this review were to: (1) examine the current state of the science of wearable inertial sensing technologies applied to biomechanical exposure assessment, and (2) identify research needs and opportunities for leveraging wearable inertial sensing for real-time biomechanical exposure assessment in applied settings. Specifically, a narrative review was conducted on inertial sensing based approaches for quantifying worker’s prolonged exposures to awkward postures, repetitive movements, and forceful exertions for the purpose of measuring and mitigating risk of work-relevant MSDs. This review begins with a brief introduction to inertial sensing technologies (1.3), a conceptual ergonomics-based framework to scope the review (1.4), followed by the review methodology, findings, and implications for research.
1.3. Inertial Sensing Technologies
Inertial sensing refers to sensing technologies that use the property of inertia. This review uses the term inertial sensing to include accelerometers, inclinometers, gyroscopes, and inertial measurement units (IMUs) which are a combination of accelerometers, gyroscopes, and/or magnetometers. A brief description of each type of sensor is provided below.
1.3.1. Accelerometers or Inclinometers
A single-axis accelerometer measures the linear acceleration of the sensor’s reference frame relative to the earth’s gravitational field vector (Pedley, 2013). Accelerometers, ranging from uniaxial to triaxial, also function as inclinometers when used for measuring inclination or tilt angles of the sensor relative to the earth’s gravitational field vector if the acceleration is small compared to the gravity (Luinge, 2002). In ergonomics applications, gravity-based inclinometers are typically used to measure orientations (e.g., flexion/extension or elevation angle) of a specific body segment. However, the orientation estimates can be unstable when one of the rotation axes is aligned with gravity since accelerometers are insensitive to rotation about the earth’s gravitational field vector (Pedley, 2013).
1.3.2. Gyroscopes
A gyroscope measures angular velocity in response to rotation of the sensor with respect to the inertial frame, i.e., the Coriolis force (Abyarjoo et al., 2015). Integrating the gyroscope sensor data provides angular displacement about the sensor frame, so it can be used to estimate the rotation angle of the sensor axis. However, the integrated sensor results drift over time due to the accumulation of noise and offsets in the raw signal. If the integration exceeds a short duration (e.g., < 5s) then filtering techniques are typically applied to compensate for drift. Gyroscope and accelerometer data are often combined for calculating sensor rotation angles.
1.3.3. Inertial Measurement Units (IMUs)
IMUs consist of a triaxial accelerometer and a gyroscope to measure 3D acceleration and angular velocity of the sensor with respect to gravity (Luinge, 2002). Calculating 3D rotations of the sensor using both accelerometer and gyroscope signals has advantages over using any one individual signal since each sensor data can compensate for the limitations of the other (Foxlin, 1996) with the use of sensor data fusion algorithms such as a Kalman filter or complimentary filter (Madgwick et al., 2011). For instance, the effects of drift error introduced by integrating gyroscope data can be corrected by the accelerometer-based orientation estimates. The orientation error which is affected by earth’s gravity in the accelerometer can be corrected by gyroscope-based rotation estimates. Often, IMUs also include a triaxial magnetometer to measure the geomagnetic field to estimate the heading of the sensor relative to the earth’s magnetic polarity. Certain algorithms combine information about the direction and magnitude of the earth’s magnetic field from magnetometers with measurements from accelerometers and gyroscopes to obtain a single estimate of three-dimensional orientation (e.g., Madgwick et al. (2011)). However, magnetometers are susceptible to local magnetic field disturbances from local ferrous objects and electrical appliances, which affects the accuracy of 3D orientation estimates (Bachmann et al., 2007) and could limit the use of IMUs in some applied settings. Creating non-proprietary methods for measuring 3D orientations using IMUs that mitigate effects of magnetic disturbances or altogether forego magnetometer data is an area of active research (Chen et al., 2018; Ligorio and Sabatini, 2016; Robert-Lachaine et al., 2017a). Sensor fusion algorithms for calculating 3D rotations in most commercial IMUs are proprietary and thus not inspectable.
1.4. Modeling-Sensing-Analysis-Assessment-Intervention (MSAAI) Framework
To scope this review and adequately synthesize the contemporary uses of inertial sensing in ergonomics research, we propose the Modeling-Sensing-Analysis-Assessment-Intervention (MSAAI) framework (Figure 1). This conceptual framework draws upon the reviewed studies to depict the typical role of wearable inertial sensing and sensor data for biomechanical exposure assessment. Each component of the MSAAI framework is briefly described.
Figure 1:
Conceptual framework for biomechanical exposure assessment using wearable inertial sensors spanning Modeling, Sensing, Analysis, Assessment, and Intervention (MSAAI).
The use of inertial sensing is typically preceded by some form of biomechanical modeling in order to represent the relationships between critical elements of the person-equipment-task system and to inform the sensing phase. The present review makes a distinction between studies that use a very explicit form of modeling such as modeling the musculoskeletal system or finite element modeling, and studies that use implicit kinematics assumptions, i.e., using inertial sensors to obtain kinematic information based on the relative orientation between pairs of sensors or with respect to the gravity axis. Sensing includes decisions on the types of sensors and measurement variables appropriate for a particular study, the locations for sensor attachment, and the process of measuring worker postures and movements. The analysis component refers to the myriad approaches to analyze sensor data spanning descriptive and inferential statistical analyses, and more recently using predictive modeling (i.e., machine learning) techniques to obtain measures of specific biomechanical exposures including information about work-relevant activities and postures, movements, and potentially joint loads. Descriptive statistics are typically used to summarize key dimensions of biomechanical exposures, namely, intensity (e.g., extent of non-neutral postures, magnitude of motions, joint loads), duration (e.g., time spent in a particular posture or on a task, time to complete a movement), repetition (i.e., frequency of motions), and exposure variation or the change in the exposure over time (Winkel and Mathiassen, 1994; Mathiassen and Winkel, 1991; Mathiassen, 2006). Inferential statistics include hypothesis testing using techniques such as ANOVA to compare exposures between conditions (e.g., between task types or equipment) or to quantify associations between inertial sensor-based biomechanical exposure measures and ergonomics outcomes (e.g., injury prevalence, productivity). Certain exposure characteristics such as force exertions and loads handled are often not directly available from inertial sensor data, hence additional methods (e.g., direct observation, indirect video-based observations, direct measurement, or work diaries) get used in conjunction with body-worn inertial sensing.
Assessment, very broadly, implies that the quantified exposures are either compared to previously established ” absolute” limits and thresholds to assess physical workload and/or risk of MSDs (e.g., NIOSH Lifting Index, RULA or OWAS scores) or compared in ”relative” terms. Examples of the latter include comparisons between different occupations, task conditions, types of equipment design, or pre- vs. post-intervention (e.g., workplace modification) to assess relative differences in biomechanical exposures and/or the risk of work-relevant MSDs.
The assessment of exposures (either absolute and relative) and MSD risk help to inform and prioritize interventions, i.e., actions aimed at modifying work content in order to augment task performance, reduce the risk of work-relevant MSDs, and/or improve worker well-being. Interventions in ergonomics can take many forms, including lowering of task demands by changing work equipment/tool or the environment, by augmenting worker capacity using assistive technologies (e.g., exoskeleton), or by means of worker training and worker selection. Since this review is focused on new opportunities afforded by wireless wearable inertial sensing in real-time biomechanical exposure assessment (i.e., study aim), synthesis and discussion about interventions are limited to direct feedback strategies in the workplace that are near real-time (e.g., using prompts or alerts from wearable devices) aimed at either altering or reinforcing certain actions/behaviors in contrast to the aforementioned interventions that are asynchronous and post priori. The closed feedback loop in the conceptual framework indicates that the biomechanical exposures are assessed across time (i.e., longitudinally, or pre- and post-intervention), or compared between conditions (i.e., cross-sectional A-B designs) towards some salutary goal. The modeling component can be updated based on comparisons between the expected (i.e., modeled or hypothesized) behavior and actual (i.e., measured or observed) behavior of the system (Figure 1, indicated by a dotted line pointing to the biomechanical modeling in a closed feedback loop).
2. Material and Methods
2.1. Search Strategy
This narrative review follows the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) methodology for conducting and reporting systematic reviews in health sciences (Liberty Mutual Insurance, 2017; Moher et al., 2009). For the review, three databases (i.e., Web of Science, PubMed, and ProQuest) were searched for articles printed through June, 2019. The search keywords used to identify studies on inertial sensing in biomechanical exposure assessment included [‘work*’ or ‘Ergonomic*’ or ‘Biomechanic*’ or ‘Occupation*’] and [‘Exposure’ or ‘Assessment’ or ‘Evaluat*’ or ‘Posture’] and [‘inertia*’ or ‘acceler*’ or ‘inclino*’ or ‘gyro*’ or ‘wearable’]. Any peer-reviewed journal articles or conference proceedings including the search keywords in the document title, abstract, or keywords were selected. Only studies written in English were searched. The search process and exclusion criteria are summarized in Figure 2.
Figure 2:
Flowchart summarizing the phases of the document search and selection process.
2.2. Exclusion Criteria
The initial search identified a total of 737 studies based on title, abstract, and keywords, and after discarding duplicate articles. Review articles were excluded unless the document also contained an empirical case-study originating from the authors’ own research. The selected studies (n = 737) were reviewed in their entirety based on the title and abstract (Step 1 in Figure 2) and full text (Step 2 in Figure 2) and were excluded from further analysis if the article described a study that:
did not include human subjects; or
used inertial sensing for applications other than measuring some aspect of body posture or postural kinematics. For example, studies using accelerometers to measure movement or vibration of an inanimate object or equipment were excluded; or
did not focus on occupational (work) activities. For example, studies focusing activities of daily living, fitness, sports, athletics, and clinical rehabilitation were excluded.
2.3. Data Extraction and Analyses
A total of 78 studies met all criteria after the full-text review (Step 2 in Figure 2). These were analyzed and coded according to the PRISMA methodology and MSAAI framework (Figure 1). Extracted data entries included the following:
General information: Author(s), year of publication, title, journal, study objective, study sample size and composition (i.e., participant characteristics), study setting (e.g., lab vs. field), and type of study site (i.e., job type) if it were a field study.
Modeling: Formal recognition of using biomechanical modeling [yes/no], biomechanical model details, main purpose or objective for the model, method for model validation, and key finding(s).
Sensing: Sensor attachment location(s), sensor pairs if used for measuring inter-segment or joint kinematics, and additional methods used to record the type of work tasks performed.
Analysis: Biomechanical exposure characteristics (intensity, duration, repetition), type of statistical technique used for analysis (i.e., descriptive statistics, inferential statistics, or predictive modeling), specific technique used, main purpose/objective of the analysis, model assessment and findings if predictive modeling was used.
Assessment: Relative vs. absolute assessment, and specific criteria used for the assessment if an absolute assessment was performed.
Intervention: Type of intervention used, content and modality if real-time feedback was used, and whether biomechanical exposures were re-examined post-intervention either in a cross-sectional or longitudinal manner.
3. Results
Appendix A (Table A1) summarizes general information on the 78 studies reviewed including each study’s sample composition and size, study design (e.g., whether the study was performed in a lab or field setting), types of occupational activities examined, number of sensors attached on the subject, and the specific analysis and assessment methods used. Review results indicated that the number of research studies that employed inertial sensors for biomechanical exposure assessment has steadily increased in the past decade. Excluding the current year, over 50% of the studies reviewed (n = 40 of 78) were published in the previous 4 years (2015 – 2018; Figure 3a).
Figure 3:
Summary of the review results in terms of the number of studies by publication year between Jan-1998 and Jun-2019 (a), by field study site where the data collection was performed (b), and by component of the Modeling-Sensing-Analysis-Assessment-Intervention (MSAAI) framework addressed (c).
Five of the 78 studies performed data collection in both applied and laboratory settings, 44 (56.7%) were conducted in applied field-based settings, while 29 (37.2%) were performed in a laboratory environment (Figure 3b). Health-care facilities (e.g., hospitals, dentistry clinics, operating rooms, and long-term care facilities) were the most frequently studied work sites (14 of the 44 + 5 = 49 field-based studies, 28.6%). Ten of the 49 field-based studies (20.4%) investigated multiple site locations mostly to compare biomechanical exposures between job types. For example, Teschke et al. (2009) used measurements of angular displacement and velocity from a triaxial inclinometer attached at the thoracic vertebra (T6) to compare the magnitude of non-neutral torso postures (i.e., flexion/extension and lateral bending) over a work shift in workers across forestry, wood and paper products, transportation, warehousing, and construction jobs. Offices comprising white-collar work were the primary study site in 9 of the 49 (18.4%) field-based studies.
With reference to the proposed framework, only six studies spanned all five components of the framework with the intervention including real-time feedback (M-S-A-A-I in Figure 3c). Over half of the reviewed studies (n = 42 out of 78, 53.8%) included the four components of modeling, sensing, analysis, and assessment of biomechanical exposures (M-S-A-A in Figure 3c). Thirty-seven of these studies were performed in applied settings. Some of the 37 studies reported on portions of a large cohort while using the same inertial sensing methodology (refer Appendix A – Table A1 under Study Sample). Examples include studies by Lagersted-Olsen et al. (2016), Villumsen et al. (2016, 2017) and Locks et al. (2018) that were all part of the “Danish Physical ACTivity cohort with Objective measurements” (DPhacto; Jørgensen et al. (2019)), and the study by Palm et al. (2018) that relied on the “New method for Objective Measurements of physical Activity in Daily living” (NOMAD; Gupta et al. (2015)) cohort. Hence it is possible for any potential bias in methodology and findings to influence all studies derived from the same cohort. About one-third of the reviewed studies (n = 24 of 78) included only modeling and sensing as the sole focus to validate the accuracy or reliability of the inertial sensor-based kinematic measures, i.e., the study did not compare measured exposures to any known limits and thresholds or between measured tasks, occupations, or interventions.
The subsequent sections summarize prominent themes organized by each framework component, i.e., modeling (M), sensing (S), analysis (A), assessment (A) and feedback intervention (I), identified from the analysis and synthesis of the 78 studies reviewed.
3.1. Modeling
The review results on modeling approaches are limited to 9 of the 78 studies (11.5%) that explicitly used biomechanical models to estimate joint loads and moments. Table 1 summarizes the type of biomechanical models used across the studies, along with the modeling purpose, approaches to validation, and key findings. Examples of modeling approaches used spanned inverse dynamics models (Faber et al., 2012, 2016, 2018; Holmes et al., 2010; Kim and Nussbaum, 2013; Shojaei et al., 2016), and musculoskeletal finite element-based models (Gholipour and Arjmand, 2016). Across the 9 studies, inertial sensor-derived posture angles were used as an input to the biomechanical model along with information from other instrumentation such as ground reaction forces from force plates (Kim and Nussbaum, 2013) or in-shoe pressure sensors (Kim and Nussbaum, 2013; Faber et al., 2018), or estimated locations of body or segmental Center of Mass (CoM) (Faber et al., 2016). The locations for inertial sensor attachment in these studies were guided by the specific segment postures or joint angles needed for calculating joint loads and moments in the biomechanical model (refer subsequent Section 3.2.1 for details). The accuracy of estimated joint loads and moments were validated by comparing these inertial sensor derived estimates to corresponding measures computed using data from optoelectronic motion capture and force plates. Valero et al. (2016) proposed an additional step by using a deterministic postural angle-based threshold model to distinguish different lifting strategies (i.e., stoop, squat, or combined) from inertial sensor-based postural angles (Valero et al., 2016). Trunk and leg inclination angles were used as criteria to discriminate between lifting strategies, for example, a trunk inclination angle ≥ 20° and shank inclination angle between 0° and 30° was used to characterize a stoop lift (Valero et al., 2016).
Table 1:
Summary of the biomechanical modeling techniques that were used in 9 of the 78 reviewed studies.
| Study | Purpose | Validation | Key findings |
|---|---|---|---|
| Inverse dynamics model | |||
| Faber et al. (2012) | To estimate the knee, hip, and L5/S1 joint loadings and moments | Compared with optical motion capture and force plate-derived joint moments | RMSE% < 4% (knee moments), RMSE% < 14 % (L5/S1 moment) |
| Faber et al. (2016) | To estimate the L5/S1 moment from the estimated CoM location | Compared with optical motion capture and force plate-derived L5/S1 moments calculation | RMSE < 10 Nm (L5/S1 moment) |
| Faber et al. (2018) | To estimate the hand force from the full-body postural angles measured from IMUs and instrumented force shoes | Compared with optical motion capture and force plate-derived hand forces | RMSE < 17–21 N |
| Holmes et al. (2010) | To estimate the L4/L5 joint loadings and moments | ||
| Kim and Nussbaum (2013) | To estimate the joint moments at the knees, hips, L5/S1, and shoulders | Compared with optical motion capture and force plate-derived joint moments | Peak Absolute Errors (PAE) < 14.4 Nm (L5/S1 moment), PAE < 11.2 Nm (hip moments) |
| Shojaei et al. (2016) | To estimate the L5/S1 compressive and shear loads and moments | ||
| Musculoskeletal finite element model | |||
| Gholipour and Arjmand (2016) | To estimate the L5/S1 compressive and shear loads from the estimated 3D rotation of trunk postures, given load, and link length | Compared with optical motion capture-derived L5/S1 loadings | RMSE = 173 N (L5/S1 compression load), RMSE = 35 N (L5/S1 shear load) |
| Single-segment inverted pendulum model | |||
| Nocerino et al. (2011) | To estimate the CoM location | ||
| Deterministic postural angle-based threshold model | |||
| Valero et al. (2016) | To discriminate different lifting strategies (stoop, squat, or a combination of both) | Compared with video-based posture analysis | |
3.2. Sensing
3.2.1. Sensor Attachment Locations
The number of inertial sensors used per study ranged from 1 to 17, with a median of 3 sensors per study. Figure 4 depicts the total number of times an inertial sensor was attached at a particular body location (denoted by circles) yielding 175 total sensor placements across the 78 studies reviewed. Sensors were placed bilaterally in 146 out of the 202 sensor placement locations (72.3%). When a sensor was placed unilaterally, either the right side or the dominant side was chosen more often (46 out of 202, 22.8%) than either the left or non-dominant side in the remaining 10 out of 202 (5.0%) studies. The upper arm was the most frequently used location for attaching an inertial sensor (37 out of 78 studies, 47.4%), followed by the thoracic vertebra (between T1 to T12) and pelvis in 23 of the 78 studies (29.5%) each, and thigh in 21 of the 78 studies (26.9%).
Figure 4:
Body locations of the inertial sensors across the reviewed studies (n=78).
Lines connecting two circles in Figure 4 denote pairs of sensors used in conjunction such as for computing the joint angle between a proximal and distal body segment, or for computing the movement coordination between two segments (e.g., ratio of lumbar flexion to pelvic rotation; Shojaei et al. (2016)). The shoulder and knee were the most frequently investigated joints, namely, in 10 out of 78 studies (12.8%). Notably, the studies used different pairs of sensor locations to derive joint angle kinematics for similar joints such as at the neck, shoulder, trunk, and hip, suggesting a lack of standardization in joint description. For example, rotations at the shoulder (glenohumeral) joint (i.e., flexion/extension, abduction/adduction, internal rotation) were calculated by combining data from a sensor attached on the upper arm and either the scapula, chest/sternum, or thoracic spine, with the scapula being the most frequent choice (6 out of 10, 60.0%). For measuring neck posture, pairs of sensors were attached either on the head and thoracic vertebra (Battini et al., 2014; Robert-Lachaine et al., 2017b), or forehead and cervical vertebra (Jonker et al., 2009; Moriguchi et al., 2013; Jun et al., 2019), or head and chest or sternum (Vignais et al., 2013; Yu et al., 2017).
3.2.2. Evaluating Sensor Performance
Studies validating inertial sensor measurements for accuracy and precision in ergonomics applications were all performed in lab settings. Accuracy was evaluated by statistical comparison of kinematic measures obtained using inertial sensing to corresponding measures obtained from a reference instrumentation system. Examples of the latter included optoelectronic motion capture (Bonnet and Heliot, 2007; Faber et al., 2016; Kim and Nussbaum, 2013; Lebel et al., 2017; Robert-Lachaine et al., 2017b; Schall Jr. et al., 2016b), magnetic motion tracking (Amasay et al., 2010; Favre et al., 2008), ultrasound-based motion capture (Dejnabadi et al., 2005), lumbar motion monitor (Schall Jr. et al., 2015), and previously validated accelerometers (e.g., Logger Teknologi HB; Dahlqvist et al. (2016)). Kinematic measures compared for validation purposes included segment or joint angular displacement, velocity, acceleration, and range of motion (ROM), and kinetic measures such as joint moments. Statistical measures typically used for assessing the accuracy of inertial sensor-derived measures compared to reference instrumentation measures included correlation coefficients, paired and independent sample t-tests, and summary statistics on root mean squared errors (RMSE), mean absolute errors, peak absolute errors (PAE), mean peak errors, and Bland-Altman bias and limits of agreement.
Two studies validated kinetic measures derived using inertial sensor data. Faber et al. (2016) used IMU-derived joint position estimates to compute CoM positions which were subsequently used for calculating 3D L5/S1 moments. Kim and Nussbaum (2013) computed joint moments at the L5/S1, shoulders, hips, and knees during lifting and lowering, carrying, and pushing and pulling tasks using a combination of IMUs for posture and in-shoe pressure measurement system for measuring vertical force component at the ground (Kim and Nussbaum, 2013). In both studies, the difference in L5/S1 joint moments calculated using optoelectronic motion capture and force plates to inertial sensor derived estimates were found to be comparable with a RMSE < 10 Nm (Faber et al., 2016) and PAE < 14.4 Nm (Kim and Nussbaum, 2013), respectively.
Precision of sensor measurements was evaluated by assessing inter-rater reliability and intra-rater repeatability, respectively. To examine inter-rater reliability, multiple examiners took turns attaching sensors to the same participant and with the participant repeating the same movement each time (Schiefer et al., 2015; Stenlund et al., 2014). Intra-rater repeatability was typically examined by having the same examiner attach sensors to the same participant multiple times and the resulting measurements compared (Amasay et al., 2010; Ambusam et al., 2015; Favre et al., 2008; Schall Jr. et al., 2015).
A direct comparison of inertial sensor accuracy and precision across the different studies reviewed was hindered by differences in experimental design that influence estimated errors. Examples include differences in the location of sensor attachment, type of kinematic and kinetic measures compared, the reference instrumentation systems used, type of reference tasks performed, duration of the measurement, signal processing techniques used, type of error estimates reported, and the type of statistical test performed on the error estimates. However, these studies provide some generalizable findings, namely, that: comparisons of segment orientations had smaller error magnitudes than joint angle estimates (Lebel et al., 2017), long complex task produced larger errors than short duration movements (e.g., manual material handling task performed over 32 minutes vs. short functional movements such as flexion/extension at head, trunk, upper arms, and upper legs (Robert-Lachaine et al., 2017b)), movement velocity in dynamic tasks influenced sensor accuracy (e.g., motions performed between 50°/s and 75°/s showed better accuracy than motions above these angular velocities) (Lebel et al., 2017), and differences in the biomechanical models used contributed more to the error in outcome measures than the inherent instrumentation error of the sensor itself (Robert-Lachaine et al., 2017b).
3.2.3. Relationships between Inertial-based Direct vs. Other Subjective Measurements
Seven studies (10.4%) investigated associations between inertial-sensor derived direct postural measures and subjective measures of workload such as self-reported questionnaires (e.g., Pedersen et al., 2016b). A subset of these studies used commercial software to classify different postures from inertial sensor data across time (e.g., activePAL, Acti4, and ActiLife 6). The time spent in the different postures were compared to durations obtained from self-reported questionnaires such as the Occupational Sitting and Physical Activity (Pedersen et al., 2016b; Wick et al., 2016) or the Multicontext Sitting Time Questionnaire (Whitfield et al., 2013). Results showed that the total duration of walking obtained from self-reported methods were overestimated by 4.6%-12.4% (Pedersen et al., 2016b; Whitfield et al., 2013) compared to durations obtained from an inertial sensor-based posture classification algorithm. Standing was either not significantly different (Pedersen et al., 2016b) or underestimated by 4.6% (Whitfield et al., 2013) in self-reported methods compared to the inertial sensor-based classification. Teschke et al. (2009) compared inertial sensor derived measurements of torso flexion-extension and lateral bending angles to direct observations using the Back Exposure Sampling Tool (Back-EST) (Village et al., 2009) and also with a structured questionnaire on the level of biomechanical exposures similar to questions in the Back-EST. In that study, observations and self-reported questionnaire responses were found to account for 30%-61% and 33%-40% of the variance in inertial sensor measurement, respectively. Overall, these studies demonstrated advantages of using direct inertial sensing measurements over observation- and questionnaire-based methods which are time and labor intensive and more susceptible to measurement bias. Continuous inertial sensing data were also amenable to leveraging statistically-based posture classification algorithms to identify postures and durations in specific work activities such as material handling (e.g., lifting, lowering, carrying), standing, walking, and sitting.
3.3. Analysis
3.3.1. Type of Analysis Methods
All of the studies reviewed provided descriptive statistics of the measured or processed sensor data in some form (e.g., mean, median, standard deviation, range, percentiles) calculated either for the entire data collection duration or stratified by condition for comparison (e.g., by job category, task types, equipment used, or before vs. after intervention; refer Appendix for study details). A few studies were limited to reporting descriptive statistics (e.g., Arvidsson et al., 2012; Bertrand et al., 2016; Douphrate et al., 2012). Over half the reviewed studies with an analysis component (n = 28 out of 54 studies, 51.9%) used inferential statistics such as ANOVA or mixed effects analyses with repeated measures either to contrast biomechanical exposure characteristics between conditions, or to examine associations between biomechanical exposures and work-relevant MSDs (Lagersted-Olsen et al., 2016; Villumsen et al., 2016).
Eleven of the reviewed studies (20.4%) used predictive modeling (i.e., machine learning) techniques either to classify postures and type of work task performed from the inertial sensor data, or to predict inertial-sensor derived postures from different incomplete sources of information (e.g., hand position, observation). Table 2 summarizes these 11 studies in terms of the type of predictive model used, modeling purpose/objective, model validation and key findings. Eight studies used inertial sensor data as predictors for classifying tasks and/or postures. Three of the 8 studies relied on commercial posture classification software, e.g., SenseWear Professional 6.1 (Breidahl et al., 2015), ADL Monitor software (Busser et al., 1998), and ActiLife 6 (Wick et al., 2016) to identify work-related activities such as walking, sitting, standing, and reclined resting. These three studies did not perform any validation of the classification algorithm of their own in the study, neither was the algorithm inspectable. However, if the validations were available elsewhere, the results and references were also included in Table 2.
Table 2:
Summary of predictive modeling techniques used in the reviewed studies (n = 11).
| Study | Model | Purpose | Validation | Result |
|---|---|---|---|---|
| Task/posture classification | ||||
| Anderson et al. (2019) | Decision tree | To classify activities: sitting, standing, weight-shifting, shuffling, and walking | Leave-one-subject-out cross-validation | 93.3%-96.8% accuracy |
| Brandt et al. (2018) | Linear Discriminant Analysis | To classify the low and high-risk lifting based on the guidelines of the Danish Working Environment Authority | Monte-Carlo cross validation (random sampling of training and test set repeated 100 times) | 65.1%-65.5% accuracy |
| Breidahl et al. (2015) | Proprietary, unspecified (used SenseWear Professional 6.1 software) | To classify postures: sleep, supine position; and steps taken | ||
| Busser et al. (1998) | Proprietary, unspecified (used ADL Monitor software) | To classify postures: locomotion, standing, sitting, and lying | Compared the classified physical activities with the video-based task analysis | Accuracy of 89%-96.5% were reported in (Uiterwaal et al., 1998) |
| Hosseinian et al. (2019) | Random-forest | To classify manual material handling activities: four static and seven dynamic activities | 93%-98.2% accuracy | |
| Kim and Nussbaum (2014) | Linear Discriminant Analysis, k-nearest neighbor, Multilayer feedforward neural network | To classify task types: walking, carrying, lifting, lowering, pushing and pulling | Three fold cross-validation | Precision of > 90% and recall of 80% |
| Peppoloni et al. (2016) | State machine | To classify task types: neutral pose, reach, grasp, and move | Strain Index and RULA score calculated with the measured postures within the identified task duration were compared with human evaluator’s ratings | Accuracy of 94.8% for RULA action level, 44.8% for Strain Index |
| Wick et al. (2016) | Decision tree (used ActiLife 6 software) | To classify postures: sitting, standing, and walking | Sensitivity of 99% and specificity of 100% were reported in (Skotte et al., 2014) using the same algorithm | |
| Posture prediction | ||||
| Gholipour and Arjmand (2016) | Artificial Neural Networks | To predict the 3D spinal posture (x, y, z rotation angle at T1, T12, S1) by using hand position (x, y, z) and body height as inputs | 70%-15%-15% (training, validation, testing data set) | RMSE < 8° between the measured angles from the inertial sensors and the predicted angles validation |
| Heiden et al. (2017) | Linear mixed model | To estimate the trunk and arm inclination postures from the other sources (administrative data, worker ratings, observation) | Resampling validation using bootstrapping (500 virtual datasets drawn with replacement from the original dataset) | Model fit (R2) < 15% (administrative data), < 36% (worker ratings), and < 56% (observation) |
| Nath et al. (2017) | Linear model | To extract the feature vector from the raw acceleration data that reliably represents the observed postural angles | Absolute error in trunk and shoulder flexion < 7°. | |
Five studies proposed their own predictive model for classifying work tasks or postures with inertial-derived kinematic data as predictors. Brandt et al. (2018) used accelerometer-based measurements of torso flexion and lateral bending angles recorded while participants performed “high-risk lifting” (i.e., 16, 18 and 20 kg loads) vs. “low-risk lifting” (i.e., 22, 24 kg at short distance, 8, 12 and 16 kg asymmetric, and 16 kg with long reaching distance lifting) as input to a statistical classification algorithm. On average, the measured postural angles correctly separated 65% of the high vs. low-risk lifting tasks. Classification accuracy increased to 76% when the accelerometer-derived predictor variables were supplemented with RMS electromyography activity of the upper trapezius and erector spinae muscles (Brandt et al., 2018). In a laboratory study of simulated MMH, Kim and Nussbaum (2014) compared the performance of three different predictive modeling techniques to classify work tasks such as walking, carrying, lifting, lowering, pushing, and pulling using continuous data from 10 body-worn IMUs. The precision of classification exceeded 90% in all of the work tasks, while the duration of work tasks were underestimated by 14% (Kim and Nussbaum, 2014). Hosseinian et al. (2019) developed an algorithm based on a Random forest model to classify four static activities (i.e., standing, trunk flexion, trunk lateral bending to the left and right side) and seven dynamic activities (i.e., bidirectional trunk twisting, lateral bending, flexion/extension, squatting, slow walking, fast walking, and running) using data features from a single tri-axial accelerometer on the chest. The algorithm classified data from simulated activities with a prediction accuracy of 93%−98.2% (Hosseinian et al., 2019). Peppoloni et al. (2016) developed a machine learning-based segmentation algorithm to identify the start and end of different hand activities including neutral posture, reaching, grasping, and moving in a study simulating the repetitive hand motions of supermarket cashiers. Deviations from neutral posture of the upper limb, neck, trunk, and leg during the segmented activities were used to calculate the RULA score for each activity. The study reported an accuracy of 94.8% within each activity cycle for RULA scores obtained from the algorithm compared to a human coder (Peppoloni et al., 2016). Anderson et al. (2019) developed an algorithm using decision trees to classify sitting, standing, weight-shifting, shuffling, and walking using data from a single accelerometer attached on the thigh. The prediction accuracy of the algorithm from leave-one-subject-out cross-validation tests were between 93.3% to 96.8% (Anderson et al., 2019).
Three other studies implemented predictive modeling techniques to predict postures and movements obtained from inertial sensors using task variables and/or person attributes as predictors. Gholipour and Arjmand (2016) successfully demonstrated the use of Artificial Neural Networks with individual stature and position of the hand load as inputs to predict 3D spinal posture measured by IMUs attached at S1, T12 and T1, with an overall RMSE < 11° and R2 < 0.95. In a study of workers at a paper mill, Heiden et al. (2017) used a combination of administrative data (e.g., worker’s demographics, work shifts such as the shift type, number of staff on shift, number of paper rolls produced), worker’s subjective ratings before and after each shift using a Borg CR-10 scale (Borg, 1982) and video-based observational data to predict the trunk and arm inclination angles measured by accelerometers. Results indicated that the coefficient of determination for predicting trunk and arm inclination angles was higher in the case of direct observation (R2 < 56%) compared to worker’s subjective ratings (R2 < 36%) and administrative data (R2 < 15%). Nath et al. (2017) demonstrated the use of raw acceleration data obtained from a smart phone worn on the upper arm as feature vectors in a linear predictive model to estimate the range of trunk and shoulder flexion angles during a simulated construction task.
3.3.2. Biomechanical Exposure Variables Measured
Inertial sensors were exclusively used for quantifying the intensity, repetition, and/or duration of extreme motions and postures. Table 3 provides a summary for the 54 studies that reported on inertial sensor-derived exposure characteristics in terms of these three dimensions. Exposure characteristics measured using methods other than inertial sensors (e.g., observer-based) were excluded from the table. Intensity of extreme motions and postures was expressed in three ways. First, by ordinal categories based on threshold. For example, thresholds for arm elevations of 30°, 60°, and 90° were typically used to calculate either the duration and/or repetition of movement in the range of 0° to 30°, 30° to 60°, 60° to 90°, and ≥ 90° of arm elevation, with larger deviations from neutral (0° ) implying greater postural load or intensity. Threshold levels were decided either based on previously validated absolute limits or study-specific thresholds (see details in Section 3.4). Second, the intensity of the exposure was expressed in terms of absolute magnitude such as by calculating an average or cumulative summary statistic, for instance, average arm accelerations (m/s2), cumulative joint loading, and cumulative activity counts. Multiple studies used physical activity level on the job measured as cumulative activity counts to convey intensity (Arias et al., 2012, 2017; Straker et al., 2014; Waters et al., 2016; Whitfield et al., 2013), with an activity being identified as any movement with accelerations above a pre-defined threshold (0.016 m/s2) (Arias et al., 2017). Third, exposure intensity was conveyed using percentiles (e.g., 10th, 90th, 95th) calculated from the cumulative distribution of posture angles, angular velocities and accelerations over the measurement duration. Most studies used a threshold-based approach (31 out of 54, 57.4%) compared to percentiles (12 out of 54, 22.2%) and average/cumulative intensity measures (9 out of 54, 16.7%).
Table 3:
Summary of the 54 studies reviewed that reported exposure characteristics of intensity, duration, and/or repetition calculated using inertial sensor-derived data. The fifth column (Task-based) represents whether the exposure characteristics were summarized either for specific tasks (T) or aggregated over the entire work duration irrespective of task (A).
| Reference | Biomechanical exposure characteristics |
Task-based | Additional Methods | ||
|---|---|---|---|---|---|
| Intensity | Duration | Repetition | |||
| Acuna and Karduna (2012) | Threshold-based (arm elevation > 30°, > 60°, > 90°), Percentile (50th, 90th, 99th of arm elevation angle and velocity) | %time | A | ||
| Ailneni et al. (2019) | Threshold-based (neck flexion > 15° for more than 30 sec) | Time | Count | A | |
| Åkesson et al. (2012) | Percentile (1st, 50th, and 90th of head flexion and lateral flexion, upper arm elevation angle and velocity) | T | Diary | ||
| Alvarez et al. (2016) | Threshold-based | %time | Frequency analysis | A | |
| Anderson et al. (2019) | N/A | %time | Count (number of events/hour) | T | |
| Arias et al. (2012) | Threshold-based (trunk flexion < −10°, −10° to 20°, 20° to 45°, > 45°, Activity count <100 counts/min, 101-759 counts/min, 760-1952 counts/min, 1953-5724 counts/min, > 5725 counts/min) | Time | Count (bends/hour) | A | |
| Arias et al. (2017) | Threshold-based (trunk flexion < −10°, −10° to 20°, 20° to 45°, > 45°, Activity count <100 counts/min, 101-759 counts/min, 760-1952 counts/min, 1953-5724 counts/min, > 5725 counts/min) | Time | Count (bends/hour) | A | |
| Arvidsson et al. (2012) | Percentile (1st, 10th, 50th, 90th, and 99th of wrist flexion, head flexion, upper arm elevation angle and velocity) | A | |||
| Battini et al. (2014) | Threshold-based (varies depending on the selected method) | Time | Count | T | Direct observation |
| Bergsten et al. (2017) | Threshold-based (arm elevation > 60°, or < 20° and movement velocity < 5°/s) | %time | A | Diary, video-based task analysis | |
| Bertrand et al. (2016) | Magnitude (average elbow/shoulder flexion angle, and shoulder abduction/adduction angle) | A | |||
| Bootsman et al. (2019) | Threshold-based (trunk flexion > 20° for more than 1.5 sec) | Time | Count | T | Work-logs |
| Brandt et al. (2015) | N/A | T | Video-based task analysis | ||
| Brandt et al. (2018) | Threshold-based (high vs. low risk lifting based on the guidelines of the Danish Working Environment Authority) | T | |||
| Breidahl et al. (2015) | Magnitude (energy expenditure/hour) | T | Self-reported questionnaire | ||
| Busser et al. (1998) | Threshold-based (trunk flexion > 50°) | %time | Count (bends/hour) | T | Video-based task analysis |
| Douphrate et al. (2012) | Threshold-based (shoulder elevation < 20°, > 45°, > 60°), Percentile (10th, 50th, 90th of shoulder elevation) | Time | Frequency analysis | A | |
| Estill et al. (2000) | Magnitude (average arm acceleration) | T | Work-logs | ||
| Fisher et al. (2018) | N/A | Time | Count (steps/hour) | T | Self-reported questionnaire |
| Gholipour and Arjmand (2016) | Magnitude (average flexion angle and L5-S1 compression and shear loadings) | A | |||
| Hansson et al. (2010) | Percentile (10th, 50th, 90th, and 99th of head flexion, upper arm elevation) | A | |||
| Headley et al. (2018) | N/A | Time | Count (steps/hour) | T | Self-reported questionnaire |
| Heiden et al. (2017) | Threshold-based (trunk flexion 0° - 20°, upper arm inclination < 20°) | Time | Count (uninterrupted neutral posture/min) | A | Video-based task analysis |
| Hodder et al. (2010) | Percentile (1st, 10th, 50th, 90th, and 99th of trunk flexion angle, angular velocity, and angular acceleration) | T | Video-based task analysis | ||
| Holmes et al. (2010) | Magnitude (peak and cumulative joint loading, lateral and anterior-posterior shear loading at lumbar) | T | Direct observation | ||
| Horn et al. (2015) | Threshold-based and Magnitude (hip, ankle, wrist acceleration counts) | A | |||
| Jonker et al. (2009) | Percentile (10th, 50th, and 90th of head flexion and lateral flexion, neck flexion and lateral flexion, arm elevation angle and velocity) | T | Direct observation | ||
| Jonker et al. (2013) | Percentile (50th of head flexion, trunk flexion, upper arm elevation angle and velocity) | T | Video-based task analysis | ||
| Jun et al. (2019) | Threshold-based (torso flexion angle < −10°, −10° - 10°, 10° - 20°, > 20°, head flexion angle < −5°, −5° - 10°, 10° - 30°, > 30° , neck flexion angle < −10°, −10° - 10°, 10° - 30°, > 30°, arm angle < −20°, −20° - 20°, 20° - 45°, > 45°) | %time | A | ||
| Kim and Nussbaum (2014) | Time | T | |||
| Lagersted-Olsen et al. (2016) | Threshold-based (trunk flexion > 30°, > 60°, > 90°) | %time | T | Diary | |
| Locks et al. (2018) | N/A | Time | T | Diary | |
| Möller et al. (2004) | Threshold-based (head flexion and arm elevation > 30°) | %time | A | Video-based task analysis | |
| Moriguchi et al. (2013) | Threshold-based (arm elevation > 60°, > 90°), percentile (1st, 10th, 50th, 90th, and 99th of arm elevation, head flexion, neck flexion angle) | %time | T | Direct observation | |
| Moser et al. (2014) | Magnitude (average relative acceleration between the equipment and aperson) | A | |||
| Nath et al. (2017) | Threshold-based (trunk flexion/elbow flexion 0° to 30°, 30° to 60°, 60° to 90°, >90°, trunk lateral bend 0° to 15°, 15° to 30°, 30° to 45°, shoulder flexion/abduction 0° - 30°, 30° to 60°, 60° to 90°, 90° to 120°, > 120°) | T | Direct observation | ||
| Palm et al. (2018) | Percentile (50th, 90th, and 99th of arm elevation angle), Threshold-based (arm elevation angle > 60 ° , > 90 ° ) | %time | T | ||
| Peppoloni et al. (2016) | Threshold-based (wrist extension < 10°, 11° - 25°, 2° - 4°, 41° - 55°, > 60°, wrist flexion < 5°, 6° - 15°, 16° - 30°, 31° - 50°, > 50°, ulnar deviation < 10°, 11° - 15°, 16° - 20°, 21° - 25°, > 25°) | %time | T | Self-reported questionnaire | |
| Ribeiro et al. (2011) | Threshold-based (trunk flexion > 30°, > 45°, > 60°) | %time | Count (> 5s over the threshold) | A | |
| Ribeiro et al. (2017) | N/A | A | |||
| Samani et al. (2012) | Percentile (10th, 50th, and 90th of trunk flexion and lateral bending) | A | |||
| Schall Jr. et al. (2016a) | Percentile (10th, 50th, and 90th of trunk flexion, arm elevation), threshold-based (trunk flexion and upper arm elevation < 20°, > 45°, > 60°, velocities < 5°/s, > 90°/s) | %time | Count (3s period in neutral trunk or arm posture with velocity < 5°/s per min) | A | |
| Shafti et al. (2016) | Threshold-based (same as RULA) | A | Video-based task analysis | ||
| Singh et al. (2016) | Threshold-based (neck flexion 0° to 10°, 10° to 20°, >20°, neck extension > 0°, trunk flexion at 0°, 0° to 20°, 20° to 60°, > 60°, shoulder elevation 0° to 20°, 20° to 45°, 45° to 90°, > 90°) | %time | A | ||
| Straker et al. (2014) | Threshold-based (activity count <100 counts/min, 101-1952 counts/min, 1953-5724 counts/min, > 5725 counts/min) | Time | T | Diary | |
| Valero et al. (2016) | Threshold-based (trunk flexion < 20° and shank inclination 0° to 30°, >30°) | Time | Count (bends/hour) | T | |
| Vignais et al. (2013) | Threshold-based (same as RULA) | %time | A | ||
| Villumsen et al. (2016) | Threshold-based (trunk flexion > 30°) | Time | T | Diary | |
| Villumsen et al. (2017) | Threshold-based (trunk flexion levels expressed as a percentage with a 10% increment) | Time | T | Diary | |
| Vinstrup et al. (2017) | N/A | T | Direct observation | ||
| Waters et al. (2016) | Threshold-based (activity count <150 counts/min, 150-2019 counts/min, 2020-5999 counts/min, > 5999 counts/min) | %time | T | Self-reported questionnaire | |
| Whitfield et al. (2013) | Threshold-based (activity count < 100 counts/min) | %time | T | Self-reported questionnaire | |
| Wick et al. (2016) | Magnitude (percentage time spend on each posture) | %time | T | Self-reported questionnaire | |
| Yu et al. (2017) | Threshold-based (neck flexion > 10°, trunk flexion > 20°, shoulder elevation > 45°), magnitude (mean and ROM of neck, torso flexion and shoulder elevation) | %time | A | ||
Duration characterized sustained postures and motions represented as an absolute time spent in certain postures or activities or as percent time normalized to total measurement time. Measures of percent time were used more often (19 out of 34, 55.9%) than absolute time (15 out of 34, 44.1%). Repetition was quantified either as the number of posture changes across a pre-defined threshold per time duration (counts/min) (Arias et al., 2012; Yu et al., 2017) or by using frequency analysis such as a Fourier transformation to estimate motion repeatability in repetitive tasks (Alvarez et al., 2016; Douphrate et al., 2012).
Exposure characteristics were computed either over the entire duration of data collection (24 out of 54, 44.4%) or stratified by specific tasks or activities (30 out of 54, 55.6%) (Table 3). For the latter, studies (25 out of 30, 83.3%) used information from additional methods such as direct or video-based observations to annotate and segment sensor data with the start and end of tasks and task type (14 out of 25 studies, 56.0%), diaries or work-logs (9 out of 25 studies, 36.0%) to record the time and duration of work vs. non-work (e.g., sleep, leisure) activities especially for data collection sessions spanning multiple days (Lagersted-Olsen et al., 2016; Villumsen et al., 2016, 2017), and self-reported questionnaires for estimating the approximate duration of work and non-work activities (7 out of 25 studies, 28.0%). In 3 studies, the exposure characteristics were stratified by tasks by first identifying the duration and type of specific work tasks using more objective techniques such a deterministic posture thresholds for classifying lifting strategy (i.e., stoop, squat, or combined) (Valero et al., 2016) and predictive modeling such as Linear Discriminant Analysis (Brandt et al., 2018; Kim and Nussbaum, 2014).
3.4. Assessment
Measured exposure characteristics were assessed for work-relevant MSD risk using either “relative” vs. “absolute” assessment criterion. Relative assessments involved comparing measured biomechanical exposure characteristics (i.e., intensity, duration, and repetition of tasks or motions) within each study to evaluate differences in risk of work-relevant MSDs between type of task performed (Åkesson et al., 2012; Breidahl et al., 2015; Hodder et al., 2010; Holmes et al., 2010; Jonker et al., 2009, 2013; Moriguchi et al., 2013; Vinstrup et al., 2017; Yu et al., 2017; Locks et al., 2018; Palm et al., 2018), job category (Busser et al., 1998; Straker et al., 2014; Headley et al., 2018), equipment used (Bertrand et al., 2016; Moser et al., 2014; Singh et al., 2016; Vinstrup et al., 2017), environmental conditions (e.g., load location, workstation set-up) (Arvidsson et al., 2012; Horn et al., 2015; Möller et al., 2004; Samani et al., 2012; Fisher et al., 2018), level of social support (Villumsen et al., 2016), and effects of ergonomics training and feedback (Samani et al., 2012; Ailneni et al., 2019; Bootsman et al., 2019). Alternatively, inertial sensor-based biomechanical exposures were compared with previously established or validated “absolute” thresholds to create an ordinal scale of MSD risk. For example, Arias et al. (2012) and Arias et al. (2017) used thresholds derived from prior epidemiological studies on low-back disorders (Fathallah et al., 1998; Punnett et al., 1991; Keyserling et al., 1992) to categorize low-back disorder risk from torso flexion exposures into three ordinal levels: 1) low risk: trunk flexion < 15° during 86% of the work shift, 2) high risk: trunk flexion > 20° during 33% of the shift, and 3) medium risk: all other exposure values. Bergsten et al. (2017) used thresholds for exposures describing ‘time in extreme’ and ‘time in neutral’ developed using findings reported by Wahlström et al. (2016). In a study of mill workers, Heiden et al. (2017) categorized trunk and arm postural exposures obtained from inclinometry using limits for neutral vs. non-neutral working postures specified in the European Standard ‘Evaluation of working postures and movements in relation to machinery’ (Standard, 2005). Others have assessed inertial sensor-derived postural exposures using ordinal risk scores analogous to observation-based coding methods such as RULA and OWAS (e.g., Battini et al. (2014); Peppoloni et al. (2016); Shafti et al. (2016); Singh et al. (2016); Vignais et al. (2013)).
3.5. Intervention
Only six studies of the reviewed studies used real-time direct feedback as a method of intervention (Ailneni et al., 2019; Battini et al., 2014; Bootsman et al., 2019; Peppoloni et al., 2016; Valero et al., 2016; Village et al., 2009). Visual display was the most common feedback modality used, typically depicting graphical representations of highlighted or color-coded “high risk” motions and postures identify from the exposure assessment (Battini et al., 2014; Peppoloni et al., 2016; Valero et al., 2016; Village et al., 2009). In a study of manufacturing workers, Vignais et al. (2013) used a head-mounted display to highlight joints and segments exceeding a threshold level from an inertial-based RULA assessment. Auditory feedback such as a warning beep when the postural angle exceeds a threshold level (Bootsman et al., 2019; Valero et al., 2016; Vignais et al., 2013) and vibrotactile feedback (Ailneni et al., 2019; Bootsman et al., 2019; Valero et al., 2016) were also used in combination with visual displays. When real-time feedback was provided directly to the worker instrumented with inertial sensors to instantly modify their work postures, auditory and vibrotactile feedback were found to be more effective (e.g., less obtrusive) since workers could attend to it in conjunction with a visually-dominant work task compared to a visual display that potentially interferes with work performance (Valero et al., 2016).
4. Discussion
Body-worn inertial sensor technology provides a number of opportunities to advance the safety and health of workers engaged in physical work. This research reviewed 78 references in the ergonomics literature on inertial sensing-based biomechanical exposure assessment to examine the current state of the science and potential gaps in different areas of sensor use based on a proposed MSAAI framework. The subsequent sections discuss key themes and gaps identified with an eye towards informing future research on wearable inertial sensing-based tools for biomechanical exposure assessment.
4.1. Modeling, Sensing and Sensor Placement
Choosing where to place sensors on the body is a critical decision when considering the use of body-worn inertial sensors for novel ergonomics applications. Sensor locations in most studies were primarily informed by either implicit or explicit biomechanical modeling criteria related to that study’s focal research question, i.e., the relevant segments or joints that were being modelled. However, the number and anatomical locations of the sensors differed across the reviewed studies even when investigating the same type of segment or joint kinematics. Only a handful of reviewed articles cited the International Society of Biomechanics guidelines for constructing 3D joint coordinate systems when calculating 3D joint motions Wu and Cavanagh (1995); Wu et al. (2002, 2005). In addition, consistency in the reporting of joint angular motions in the anatomical or segment reference frame such as flexion/extension, abduction/adduction, or internal/external rotation would aid the interpretation of results compared to using the sensor frame of reference (e.g., roll, pitch, and yaw). Importantly, the lack of standardized procedures for constructing joint coordinate systems and calculating joint motions across studies undermined comparisons, for instance when comparing sensor accuracy (Figure 4). Overall, this review identified a need for validated procedures for selecting and evaluating sensor attachment locations for use in applied settings that consider technical aspects such as the underlying biomechanical modeling, sensor performance issues, and wearability (e.g., user comfort, acceptability, obtrusiveness).
Some studies relied on up to 17 body-worn inertial sensors, which can be obtrusive and time-consuming, and affect wearability. Clearly, minimizing the number of sensors would be advantageous. Trade-offs between the number of sensors and user comfort are rarely documented in the ergonomics literature. A study by Mokhlespour Esfahani and Nussbaum (2018) reported that respondents rated the ankle, wrist, waist, thigh, and foot as the most preferred location for inertial sensor placement compared to the neck, torso, elbow, fingers. Beeler et al. (2018) reported that sensors attached on the upper arm, hip, and back were rated relatively higher in comfort and acceptability compared to the chest location which was rated the lowest.
4.1.1. Sensor Performance
Studies validating inertial sensor performance (e.g., accuracy, reliability, and repeatability) largely focused on determining whether the inertial sensor-derived kinematic or kinetic information were accurate and precise compared to traditional lab-based methods (e.g., optical motion capture). While metrics of sensor performance were not directly comparable across different studies due to the differences in study design and factors affecting estimated errors, the overall trend showed consensus that commercial research-grade body-worn inertial sensors are accurate and reliable for measuring human movement in practical applications relevant to ergonomics. However, several issues pertaining to inertial sensor performance remain unanswered or under-explored despite the many validation studies. For example, few validation studies have examined the effects of different biomechanical modeling methods on the performance of inertial sensor based exposure metrics (Kim and Nussbaum, 2013; Robert-Lachaine et al., 2017b; Schall Jr. et al., 2016b). The focus of most validation studies to date was on investigating the inherent instrumentation error rather than the comparison between different biomechanical modeling methods. However, two studies in particular demonstrated that differences in biomechanical models contributed more to the accuracy of the kinematic data compared to the inherent instrumentation error (Kim and Nussbaum, 2013; Robert-Lachaine et al., 2017b). Also, sensor durability, reliability, and repeatability can become a critical issue when data collection spans a long duration such as across multiple days wherein users themselves or different researchers need to reattach the sensors on the user each time. Furthermore, if sensors are not attached firmly on the body, for instance using adhesive tape or straps (e.g., Velcro™ bands), then subtle changes in sensor location over time are unavoidable.
Multiple studies required that one of the sensor axes was aligned with the anatomical axis of the body, so that linear and angular movement of the sensor can be directly mapped to segment kinematics. However, it can be difficult to perfectly align a sensor axis to the anatomical axis of the body segment, especially when the user (e.g., worker) has to attach the sensor themselves. Data processing algorithms that are robust to account for small misalignment would be ideal. Alternately, improved dynamic calibration methods may provide a more practical solution to minimize effects of small deviations in sensor placement (i.e., using a calibration pose or reference movements) than assuming that sensor locations remain unchanged over time. Some studies employed such static or dynamic calibration methods to correct for misalignment between the sensor axis and the anatomical axis (Acuna and Karduna, 2012; Favre et al., 2008). In a study measuring the upper arm elevation angle, Acuna and Karduna (2012) performed a static calibration by asking participants to align their arm perpendicular to the floor by laterally bending their trunk while maintaining a seated upright posture and holding a 1-kg weight. Favre et al. (2008) developed a dynamic calibration method involving hip abduction and adduction movement to align two IMUs attached on thigh and shank segments. Wearable sensors incorporated into clothing fabric (Mokhlespour Esfahani et al., 2019) and body-worn equipment such as exoskeletons and personal protective devices (e.g., back support belts) that have specific location and orientation could altogether alleviate some of these misalignment concerns.
4.2. Analysis: Quantifying Biomechanical Exposures and the Emergence of Predictive Modeling
Quantifying biomechanical exposures and identifying high-risk work conditions require knowing the specific work context (Burdorf et al., 1997; Chung and Shorrock, 2011; Dempsey and Mathiassen, 2006). Most studies that examined biomechanical exposures by task (25 out of 30, 83.3%) incorporated additional methods (e.g., direct observations, self-reported measures) to identify task type and duration (i.e., start and end of specific tasks) since such information is not readily apparent from inertial sensor data alone. However, direct observation and video-based task analysis are time and labor intensive, and not feasible in some applied settings for reasons of data privacy and logistical barriers. Self-reported questionnaires, diaries, and work logs can be subjective and susceptible to systematic bias (Pedersen et al., 2016b; Whitfield et al., 2013) and often cannot provide the specific time-frame of work activities unless workers are asked to log their activities over very short time interval (< 1 min).
As an alternative, predictive modeling techniques have shown strong potential for extracting rich contextual information from inertial-sensor based kinematic data in non-occupational domains. For example, activity recognition in daily activities (Oshima et al., 2010; Ravi et al., 2005) and detecting falls (Bagala et al., 2012; Schwickert et al., 2013; Wu and Xue, 2008) have gained much research attention in healthcare, smart homes and assisted living, and sports/kinesiology applications. In contrast the application of predictive modeling techniques in biomechanical exposure assessment remains under-explored. A few of the studies reviewed (Breidahl et al., 2015; Busser et al., 1998; Wick et al., 2016) used commercial software to detect different work postures. However biomechanical exposure assessment requires information beyond just postures to identifying the specific tasks performed, durations (i.e., start and end times), repetition, and intensity (e.g., magnitude of loads carried). In this regard, studies on classifying work-relevant tasks (Kim and Nussbaum, 2014; Peppoloni et al., 2016) and task intensity (Brandt et al., 2018; Lim and D’Souza, 2017, 2018, 2019b) from inertial sensor data though limited in task conditions were comparable in performance to traditional methods for exposure assessments (e.g., direct observations) with the benefit of less time and cost. Collectively these studies show potential for combining inertial sensor data with predictive modeling techniques in developing work activity and intensity classification algorithms for use in biomechanical exposure assessment. However, as predictive modeling techniques gain prominence in ergonomics, it is important that the research community also attend to topics of algorithmic bias, classification parity, and fairness in order to build trust with users and realize the potential of novel predictive modeling techniques for improving worker health and safety (Lim and D’Souza, 2019a).
4.3. Assessment of Work-relevant MSD Risk
This review revealed gaps in our ability to assess the level of ergonomic risk from the obtained biomechanical exposure characteristics. Considering that the use of inertial sensors in biomechanical exposure assessment is still nascent, understandably many of the studies reviewed were focused on validating the accuracy and precision of sensor data. Reviewed studies that were focused on exposure assessment followed traditional observation-based methods by substituting observed categorical posture codes with inertial sensor-based posture angles into assessment tools (e.g., RULA and OWAS scores, Strain Index). Since these traditional methods were developed assuming that human observers visually code posture angles categorically each time, the coding scheme and the resulting rating scales of MSD risk are very basic. Research is needed to develop more valid and sensitive measures of work-relevant MSD risk relying on directly measured exposure data.
4.4. Intervention: Real-time Feedback
Most commercial wearable inertial sensors have the capability to stream data to a web server (cloud) or remote computer in real-time (e.g., APDM Inc.). This allows for recording, processing and reviewing sensor data online and in near real-time affording new opportunities for quick feedback about work postures to workers (e.g., to prompt changes in work postures) or to supervisors and managers (e.g., about duration and repetition of worker postures over time). However, few studies in this review utilized such real-time feedback functionality. Leveraging this capability will require developing software algorithms that receive, process, and calculate biomechanical exposure metrics or risk levels of over-exertion in real-time.
From the hardware standpoint, combining feedback functions in the sensor unit itself to provide online feedback about potentially harmful work postures or activities may be more effective than reviewing exposure data post hoc. At present, not many commercial inertial sensors incorporate functionality to communicate auditory, tactile, or visual feedback coupled with the sensor unit. Vignais et al. (2013) combined a head-mounted display to provide immediate visual and auditory feedback to the wearer, but this required wearing additional instrumentation that is potentially cumbersome and restrictive. Implementing vibro-tactile feedback in the sensor units is a promising option to provide real-time prompts to the wearer about posture corrections or specific body locations at harm (Valero et al., 2016; Ailneni et al., 2019). Based on these few studies, however, it is early and the evidence insufficient to infer the general effectiveness of real-time direct feedback approaches that use wearable inertial sensing. As such, studies that assess the effectiveness of direct feedback approaches combining inertial sensing and real-time feedback modalities in applied settings are needed.
4.5. Study Limitations
The current review was limited to the use of inertial sensing technologies for continuous monitoring of biomechanical exposures. The study intentionally focused on aspects of modeling, sensing, analysis, assessment and intervention. Aspects related to data privacy and user adoption though relevant to the uptake of body-worn sensor technologies in occupational settings (Schall Jr. et al., 2018) were outside the scope of this study. The narrow set of search keywords used could have resulted in some related studies on inertial sensing being missed in this review. For example, this review did not include the application of inertial sensing-based activity classification and gait analysis to estimate energy expenditure and fatigue onset (e.g., from walking and carrying; Zhang et al. (2014)), which are quite relevant to the assessment of biomechanical exposures. As a narrative review, the present study also did not formally evaluate or rate the quality of the individual studies reviewed. Applications combining inertial sensing with other wearable technologies for collecting data on human performance and physiology (e.g., wireless surface electromyography, heart rate, galvanic skin response) and ambient work conditions (e.g., temperature, air particulates, etc.) are also fast developing although were not explored in this research review. However, these wearable technologies present some similar concerns as those raised in this study related to modeling, sensing, analysis, assessment, intervention, and wearability. Wearable inertial sensing in ergonomics is a rapidly evolving domain and as such, this review is only a small part of a larger conversation about realizing the full potential of wearable sensors in ergonomics research and practice.
5. Conclusion
This study reviewed the current state of the science on wearable inertial sensors for biomechanical exposure assessment. The study used a conceptual framework comprising modeling, sensing, analysis, assessment, and intervention to scope and organize the review. Findings suggest a growing body of knowledge pertaining to wearable inertial sensing in ergonomics. At present, the main contribution of this emerging technology to ergonomics has been its ability to increase the quality and quantity of postural information, and thus appealing as a research tool. The majority of studies reviewed pertained to laboratory evaluations to validate inertial sensor data compared to reference motion capture derived measurements, or in applied settings for direct measurement of motions and postures in place of observational methods. Translating inertial sensing technologies as a viable tool for online biomechanical exposure assessment and real-time feedback towards proactively identifying and mitigating the risk of work-relevant MSDs disorders in applied settings still lags. To this effect, a few key conclusions and directions for future research are provided:
Modeling and sensing: Since misalignment in sensor placement are unavoidable, posture analysis techniques using body-worn inertial sensing need to be robust to minor misalignment, or involve calibration postures or movements that are simple enough for practitioners to administer.
Modeling: Quantifying inertial sensor “measurement error” is important but has been overemphasized in previous research. For example, when quantifying low-back disorder risk, the effect of modeling assumptions used in calculating internal low-back biomechanical loads (e.g., a single-vector back extensor muscle when calculating low-back compressive forces) exceeds the influence of inertial sensor “error” in measuring torso flexion angle (Kim and Nussbaum, 2013; Robert-Lachaine et al., 2017b). It is important that research involving new emerging sensing technologies not lose perspective of the end goal, namely, improving our ability to assess and reduce the risk of work-relevant MSDs.
Assessment: Multiple studies were found to rely on ordinal risk indices developed for observational methods such as RULA and OWAS to convey the risk of work-relevant MSDs from inertial sensor derived exposure measures (e.g., Peppoloni et al. (2016)). New algorithmic frameworks are needed that relate or map continuous biomechanical exposure measures from body-worn inertial sensing to the risk of work-relevant MSDs.
- Analysis: Analysis of inertial sensor data was largely limited to characterizing the intensity, repetition, and duration of non-neutral postures. The ability to extract information about work content from kinematics data is underutilized. Two approaches from this review appear promising:
- Deterministic threshold-based methods for activity classification, particularly when clear criteria for separating tasks using posture information (e.g., standing, vs. walking vs. sitting; Valero et al. (2016); Arias et al. (2012, 2017)) can be identified, and
- Statistical prediction techniques to identify tasks or behaviors (e.g., lifting strategy, load carrying mode; Brandt et al. (2018); Kim and Nussbaum (2014); Hosseinian et al. (2019)), particularly when subtle changes in multiple sensor data features are involved.
Sensing: Alongside the modeling requirements, wearability issues associated with the number and location of body-worn sensors needs to be considered. With some studies requiring up to 17 body-worn inertial sensors it is clear that the trade-offs between the number/location of sensors and wearability are not often prioritized (Battini et al., 2014).
Intervention: Wearable sensing technologies (inertial as well as other physiological sensing technologies) present unprecedented opportunity for feedback about biomechanical exposures. Early studies indicate that auditory and vibrotactile feedback modalities are more effective when real-time feedback derived from inertial-based kinematic data is provided directly to the worker (e.g., to instantly modify their work postures (Vignais et al., 2013)). Visual displays may be more effective when the viewer is not the worker, but rather a work-supervisor, manager, or ergonomics practitioner that is reviewing a workers’ cumulative biomechanical exposures across time (e.g., spinal compression loads at the end of a work-shift). Research is needed to understand the content and modality of information being presented, and identify appropriate use-cases for such real-time feedback capability.
Supplementary Material
Relevance to industry:
Despite the growing interest in wearable inertial sensing technologies for ergonomics research, its use in applied settings still lags. This manuscript explores contemporary and emerging uses of body-worn inertial sensing for assessing biomechanical exposures and reducing the risk of work-relevant musculoskeletal disorders.
Acknowledgements
The contents of this manuscript were developed under a grant from the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR 90IF0094-01-00). NIDILRR is a Center within the Administration for Community Living (ACL), Department of Health and Human Services (HHS). The contents of this manuscript do not necessarily reflect the policy of NIDILRR, ACL, HHS, and you should not assume endorsement by the Federal Government.
Contributor Information
Sol Lim, Department of Systems and Industrial Engineering, University of Arizona, Tucson, Arizona.
Clive D’Souza, Center for Ergonomics, Department of Industrial and Operations Engineering, University of Michigan, Ann Arbor, Michigan.
References
- Abyarjoo F, Barreto A, Cofino J, and Ortega FR (2015). Implementing a sensor fusion algorithm for 3D orientation detection with inertial/magnetic sensors, pages 305–310. Springer. [Google Scholar]
- Acuna M and Karduna AR (2012). Wrist activity monitor counts are correlated with dynamic but not static assessments of arm elevation exposure made with a triaxial accelerometer. Ergonomics, 55(8):963–970. [DOI] [PubMed] [Google Scholar]
- Ailneni RC, Syamala KR, Kim I-S, and Hwang J (2019). Influence of the wearable posture correction sensor on head and neck posture: Sitting and standing workstations. Work, 62(1):27–35. [DOI] [PubMed] [Google Scholar]
- Åkesson I, Balogh I, and Hansson G-Å (2012). Physical workload in neck, shoulders and wrists/hands in dental hygienists during a work-day. Applied Ergonomics, 43(4):803–811. [DOI] [PubMed] [Google Scholar]
- Alvarez D, Alvarez JC, Gonzalez RC, and Lopez AM (2016). Upper limb joint angle measurement in occupational health. Comput Methods Biomech Biomed Engin, 19(2):159–70. [DOI] [PubMed] [Google Scholar]
- Amasay T, Latteri M, and Karduna AR (2010). In vivo measurement of humeral elevation angles and exposure using a triaxial accelerometer. Human Factors, 52(6):616. [DOI] [PubMed] [Google Scholar]
- Ambusam S, Omar B, Joseph L, Meng SP, and Padzil FAM (2015). Is a triaxial accelerometer a reliable device to measure head excursion? Technology and Health Care, 23(5):691–697. [DOI] [PubMed] [Google Scholar]
- Anderson J, Granat MH, Williams AE, and Nester C (2019). Exploring occupational standing activities using accelerometer-based activity monitoring. Ergonomics, pages 1–11. [DOI] [PubMed] [Google Scholar]
- Arias OE, Umukoro PE, Stofell S, Dennerlein JT, and Sorensen G (2012). Association between trunk flexion and physical activity in patient care unit workers. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, volume 56, pages 1188–1191. SAGE Publications Sage CA: Los Angeles, CA. [Google Scholar]
- Arias OE, Umukoro PE, Stoffel SD, Hopcia K, Sorensen G, and Dennerlein JT (2017). Associations between trunk flexion and physical activity of patient care workers for a single shift: A pilot study. Work, 56(2):247–255. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Arvidsson I, Balogh I, Hansson G-, Ohlsson K, Åkesson I, and Nordander C (2012). Rationalization in meat cutting - consequences on physical workload. Applied Ergonomics, 43(6):1. [DOI] [PubMed] [Google Scholar]
- Bachmann ER, Yun X, and Brumfield A (2007). Investigating the effects of magnetic variations on inertial/magnetic orientation sensors. Technical report, NAVAL POSTGRADUATE SCHOOL; MONTEREY CA. [Google Scholar]
- Bagala F, Becker C, Cappello A, Chiari L, Aminian K, Hausdorff JM, Zijlstra W, and Klenk J (2012). Evaluation of accelerometer-based fall detection algorithms on real-world falls. PloS one, 7(5):e37062. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Barbour N and Schmidt G (2001). Inertial sensor technology trends. IEEE sensors journal, 1(4):332–339. [Google Scholar]
- Battini D, Persona A, and Sgarbossa F (2014). Innovative real-time system to integrate ergonomic evaluations into warehouse design and management. Computers & Industrial Engineering, 77:1–10. [Google Scholar]
- Beeler N, Roos L, Delves SK, Veenstra BJ, Friedl K, Buller MJ, and Wyss T (2018). The wearing comfort and acceptability of ambulatory physical activity monitoring devices in soldiers. IISE Transactions on Occupational Ergonomics and Human Factors, 6(1):1–10. [Google Scholar]
- Bergmann JH, Mayagoitia RE, and Smith IC (2009). A portable system for collecting anatomical joint angles during stair ascent: a comparison with an optical tracking device. Dyn Med, 8:3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bergsten EL, Mathiassen SE, Kwak L, and Vingard E (2017). Daily shoulder pain among flight baggage handlers and its association with work tasks and upper arm postures on the same day. Annals of Work Exposures and Health, 61(9):1145–1153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bertrand P, Reyes S, and Newman D (2016). Pressure and kinematic in-suit sensors: Assessing human-suit interaction for injury risk mitigation. The Institute of Electrical and Electronics Engineers, Inc. (IEEE) Conference Proceedings., pages 1–10. [Google Scholar]
- Bonnet S and Heliot R (2007). Magnetometer-based approach for studying human movements. Ieee Transactions on Biomedical Engineering, 54(7):1353–1355. [DOI] [PubMed] [Google Scholar]
- Bootsman R, Markopoulos P, Qi Q, Wang Q, and Timmermans AA (2019). Wearable technology for posture monitoring at the workplace. International Journal of Human-Computer Studies, 132:99–111. [Google Scholar]
- Borg GA (1982). Psychophysical bases of perceived exertion. Medicine and Science in Sports and Exercise, 14(5):377–381. [PubMed] [Google Scholar]
- Brandt M, Madeleine P, Ajslev JZ, Jakobsen MD, Samani A, Sundstrup E, Kines P, and Andersen LL (2015). Participatory intervention with objectively measured physical risk factors for musculoskeletal disorders in the construction industry: study protocol for a cluster randomized controlled trial. BMC Musculoskelet Disord, 16:302. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Brandt M, Madeleine P, Samani A, Jakobsen MD, Skals S, Vinstrup J, and Andersen LL (2018). Accuracy of identification of low or high risk lifting during standardised lifting situations. Ergonomics, 61(5):710–719. [DOI] [PubMed] [Google Scholar]
- Breidahl T, Christensen M, Jepsen JR, and Omland O (2015). The influence of vessel movements on the energy expenditure of fishermen in relation to activities and occupational tasks on board. International Maritime Health, 66(4):196–202. [DOI] [PubMed] [Google Scholar]
- Buchholz B, Paquet V, Punnett L, Lee D, and Moir S (1996). Path: a work sampling-based approach to ergonomic job analysis for construction and other non-repetitive work. Applied ergonomics, 27(3):177–187. [DOI] [PubMed] [Google Scholar]
- Burdorf A, Rossignol M, Fathallah FA, Snook SH, and Herrick RF (1997). Challenges in assessing risk factors in epidemiologic studies on back disorders. American journal of industrial medicine, 32(2):142–152. [DOI] [PubMed] [Google Scholar]
- Busser HJ, de Korte WG, Glerum EB, and van Lummel RC (1998). Method for objective assessment of physical work load at the workplace. Ergonomics, 41(10):1519–1526. [DOI] [PubMed] [Google Scholar]
- Castro MCF and Cliquet A (1997). A low-cost instrumented glove for monitoring forces during object manipulation. IEEE Transactions on Rehabilitation Engineering, 5(2):140–147. [DOI] [PubMed] [Google Scholar]
- Chen H, Schall MC Jr, and Fethke N (2018). Accuracy of angular displacements and velocities from inertial-based inclinometers. Applied ergonomics, 67:151–161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Chung AZ and Shorrock ST (2011). The research-practice relationship in ergonomics and human factors–surveying and bridging the gap. Ergonomics, 54(5):413–429. [DOI] [PubMed] [Google Scholar]
- Cordero AF, Koopman H, and Van Der Helm F (2004). Use of pressure insoles to calculate the complete ground reaction forces. Journal of biomechanics, 37(9):1427–1432. [DOI] [PubMed] [Google Scholar]
- Dahlqvist C, Hansson G-, and Forsman M (2016). Validity of a small low-cost triaxial accelerometer with integrated logger for uncomplicated measurements of postures and movements of head, upper back and upper arms. Applied Ergonomics, 55:108. [DOI] [PubMed] [Google Scholar]
- David G (2005). Ergonomic methods for assessing exposure to risk factors for work-related musculoskeletal disorders. Occupational medicine, 55(3):190–199. [DOI] [PubMed] [Google Scholar]
- Dejnabadi H, Jolles BM, and Aminian K (2005). A new approach to accurate measurement of uniaxial joint angles based on a combination of accelerometers and gyroscopes. Ieee Transactions on Biomedical Engineering, 52(8):1478–1484. [DOI] [PubMed] [Google Scholar]
- Dempsey PG and Mathiassen SE (2006). On the evolution of task-based analysis of manual materials handling, and its applicability in contemporary ergonomics. Applied Ergonomics, 37(1):33–43. [DOI] [PubMed] [Google Scholar]
- Douphrate DI, Fethke NB, Nonnenmann MW, Rosecrance JC, and Reynolds SJ (2012). Full shift arm inclinometry among dairy parlor workers: A feasibility study in a challenging work environment. Applied Ergonomics, 43(3):604. [DOI] [PubMed] [Google Scholar]
- Estill CF, MacDonald LA, Wenzl TB, and Petersen MR (2000). Use of accelerometers as an ergonomic assessment method for arm acceleration-a large-scale field trial. Ergonomics, 43(9):1430–45. [DOI] [PubMed] [Google Scholar]
- Faber G, Chang CC, Kingma I, and Dennerlein JT (2012). A novel wearable measurement system for ambulatory assessment of joint loading in the occupational setting. Work-a Journal of Prevention Assessment & Rehabilitation, 41:5527–5528. [DOI] [PubMed] [Google Scholar]
- Faber GS, Chang CC, Kingma I, Dennerlein JT, and van Dieen JH (2016). Estimating 3d l5/s1 moments and ground reaction forces during trunk bending using a full-body ambulatory inertial motion capture system. J Biomech, 49(6):904–912. [DOI] [PubMed] [Google Scholar]
- Faber GS, Kingma I, Bruijn SM, and van Dieën JH (2009). Optimal inertial sensor location for ambulatory measurement of trunk inclination. Journal of biomechanics, 42(14):2406–2409. [DOI] [PubMed] [Google Scholar]
- Faber GS, Kingma I, and van Dieën JH (2010). Bottom-up estimation of joint moments during manual lifting using orientation sensors instead of position sensors. Journal of biomechanics, 43(7):1432–1436. [DOI] [PubMed] [Google Scholar]
- Faber GS, Koopman A, Kingma I, Chang C, Dennerlein J, and van Dieën J (2018). Continuous ambulatory hand force monitoring during manual materials handling using instrumented force shoes and an inertial motion capture suit. Journal of biomechanics, 70:235–241. [DOI] [PubMed] [Google Scholar]
- Fathallah FA, Marras WS, and Parnianpour M (1998). An assessment of complex spinal loads during dynamic lifting tasks. Spine, 23(6):706–716. [DOI] [PubMed] [Google Scholar]
- Favre J, Jolles BM, Aissaoui R, and Aminian K (2008). Ambulatory measurement of 3d knee joint angle. Journal of Biomechanics, 41(5):1029–1035. [DOI] [PubMed] [Google Scholar]
- Fisher A, Ucci M, Smith L, Sawyer A, Spinney R, Konstantatou M, and Marmot A (2018). Associations between the objectively measured office environment and workplace step count and sitting time: cross-sectional analyses from the active buildings study. International journal of environmental research and public health, 15(6):1135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Foxlin E (1996). Inertial head-tracker sensor fusion by a complementary separate-bias kalman filter. In Virtual Reality Annual International Symposium, 1996., Proceedings of the IEEE 1996, pages 185–194. IEEE. [Google Scholar]
- Garg A and Owen B (1992). Reducing back stress to nursing personnel: an ergonomic intervention in a nursing home. Ergonomics, 35(11):1353–1375. [DOI] [PubMed] [Google Scholar]
- Gholipour A and Arjmand N (2016). Artificial neural networks to predict 3d spinal posture in reaching and lifting activities; applications in biomechanical models. Journal of Biomechanics, 49(13):2946–2952. [DOI] [PubMed] [Google Scholar]
- Gold JE, Park J-S, and Punnett L (2006). Work routinization and implications for ergonomic exposure assessment. Ergonomics, 49(1):12–27. [DOI] [PubMed] [Google Scholar]
- Gupta N, Christiansen CS, Hallman DM, Korshøj M, Carneiro IG, and Holtermann A (2015). Is objectively measured sitting time associated with low back pain? a cross-sectional investigation in the nomad study. PLoS One, 10(3):e0121159. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hansson G, Asterland P, Holmer N. g., and Skerfving S (2001). Validity and reliability of triaxial accelerometers for inclinometry in posture analysis. Medical and Biological Engineering and Computing, 39(4):405–13. [DOI] [PubMed] [Google Scholar]
- Hansson G-A, Balogh I, Ohlsson K, Granqvist L, Nordander C, Arvidsson I, Aakesson I, Unge J, Rittner R, Stroemberg U, and Skerfving S (2010). Physical workload in various types of work: Part ii. neck, shoulder and upper arm. International Journal of Industrial Ergonomics, 40(3):267–281. [Google Scholar]
- Headley S, Hutchinson J, Wooley S, Dempsey K, Phan K, Spicer G, Janssen X, Laguilles J, and Matthews T (2018). Subjective and objective assessment of sedentary behavior among college employees. BMC public health, 18(1):768. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Heiden M, Garza J, Trask C, and Mathiassen SE (2017). Predicting directly measured trunk and upper arm postures in paper mill work from administrative data, workers’ ratings and posture observations. Annals of Work Exposures and Health, 61(2):207–217. [DOI] [PubMed] [Google Scholar]
- Hignett S and McAtamney L (2000). Rapid entire body assessment (reba). Applied ergonomics, 31(2):201–205. [DOI] [PubMed] [Google Scholar]
- Hodder JN, Holmes MWR, and Keir PJ (2010). Continuous assessment of work activities and posture in long-term care nurses. Ergonomics, 53(9):1097–1107. [DOI] [PubMed] [Google Scholar]
- Holmes MWR, Hodder JN, and Keir PJ (2010). Continuous assessment of low back loads in long-term care nurses. Ergonomics, 53(9):1108–1116. [DOI] [PubMed] [Google Scholar]
- Horn GP, Kesler RM, Motl RW, Hsiao-Wecksler ET, Klaren RE, Ensari I, Petrucci MN, Fernhall B, and Rosengren KS (2015). Physiological responses to simulated firefighter exercise protocols in varying environments. Ergonomics, 58(6):1012–1021. [DOI] [PubMed] [Google Scholar]
- Hosseinian SM, Zhu Y, Mehta RK, Erraguntla M, and Lawley MA (2019). Static and dynamic work activity classification from a single accelerometer: Implications for ergonomic assessment of manual handling tasks. IISE Transactions on Occupational Ergonomics and Human Factors, 7(1):59–68. [Google Scholar]
- Jensen RC (1988). Epidemiology of work-related back pain. Top Acute Care Trauma Rehabil, 2(3):1–15. [Google Scholar]
- Jonker D, Rolander B, and Balogh I (2009). Relation between perceived and measured workload obtained by long-term inclinometry among dentists. Applied ergonomics, 40(3):309–315. [DOI] [PubMed] [Google Scholar]
- Jonker D, Rolander B, Balogh I, Sandsjo L, Ekberg K, and Winkel J (2013). Rationalisation in public dental care - impact on clinical work tasks and mechanical exposure for dentists - a prospective study. Ergonomics, 56(2):303–313. [DOI] [PubMed] [Google Scholar]
- Jørgensen MB, Gupta N, Korshøj M, Lagersted-Olsen J, Villumsen M, Mortensen OS, Skotte J, Søgaard K, Madeleine P, Samani A, et al. (2019). The dphacto cohort: an overview of technically measured physical activity at work and leisure in blue-collar sectors for practitioners and researchers. Applied ergonomics, 77:29–39. [DOI] [PubMed] [Google Scholar]
- Jun D, Johnston V, McPhail SM, and O’Leary S (2019). Are measures of postural behavior using motion sensors in seated office workers reliable? Human factors, page 0018720818821273. [DOI] [PubMed] [Google Scholar]
- Karhu O, Kansi P, and Kuorinka I (1977). Correcting working postures in industry: a practical method for analysis. Applied ergonomics, 8(4):199–201. [DOI] [PubMed] [Google Scholar]
- Keyserling WM, Brouwer M, and Silverstein BA (1992). A checklist for evaluating ergonomic risk factors resulting from awkward postures of the legs, trunk and neck. International Journal of Industrial Ergonomics, 9(4):283–301. [Google Scholar]
- Kim S and Nussbaum MA (2013). Performance evaluation of a wearable inertial motion capture system for capturing physical exposures during manual material handling tasks. Ergonomics, 56(2):314–326. [DOI] [PubMed] [Google Scholar]
- Kim S and Nussbaum MA (2014). An evaluation of classification algorithms for manual material handling tasks based on data obtained using wearable technologies. Ergonomics, 57(7):1040–1051. [DOI] [PubMed] [Google Scholar]
- Knibbe NE and Knibbe HJ (1996). Postural load of nurses during bathing and showering of patients: Results of a laboratory study. Professional Safety, 41(11):37. [Google Scholar]
- Lagersted-Olsen J, Thomsen BL, Holtermann A, Sogaard K, and Jorgensen MB (2016). Does objectively measured daily duration of forward bending predict development and aggravation of low-back pain? a prospective study. Scandinavian Journal of Work Environment & Health, 42(6):528–537. [DOI] [PubMed] [Google Scholar]
- Lebel K, Boissy P, Nguyen H, and Duval C (2017). Inertial measurement systems for segments and joints kinematics assessment: towards an understanding of the variations in sensors accuracy. Biomedical Engineering Online, 16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Liberty Mutual Insurance (2017). https://www.libertymutualgroup.com/about-liberty-mutual-site/research-institute-site/Pages/Workplace-Safety-Index.aspx. [Online; accessed 04-09-2019].
- Ligorio G and Sabatini A (2016). Dealing with magnetic disturbances in human motion capture: A survey of techniques. Micromachines, 7(3):43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Liira JP, Shannon HS, Chambers LW, and Haines TA (1996). Long-term back problems and physical work exposures in the 1990 ontario health survey. American Journal of Public Health, 86(3):382–387. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lim S and D’Souza C (2017). Statistical prediction of hand force exertion levels in a simulated push task using posture kinematics. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, volume 61, pages 1031–1035. SAGE Publications Sage CA: Los Angeles, CA. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lim S and D’Souza C (2018). Inertial sensor-based measurement of thoracic-pelvic coordination predicts hand-load levels in two-handed anterior carry. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, volume 62, pages 798–799. SAGE Publications Sage CA: Los Angeles, CA. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lim S and D’Souza C (2019a). Gender and parity in statistical prediction of anterior carry hand-loads from inertial sensor data. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting, volume 63. SAGE Publications Sage CA: Los Angeles, CA. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lim S and D’Souza C (2019b). Statistical prediction of load carriage mode and magnitude from inertial sensor derived gait kinematics. Applied Ergonomics, 76:1–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Locks F, Gupta N, Hallman D, Birk Jørgensen M, Oliveira AB, and Holtermann A (2018). Association between objectively measured static standing and low back pain-a cross-sectional study among blue-collar workers. Ergonomics, 61(9):1196–1207. [DOI] [PubMed] [Google Scholar]
- Luinge HJ (2002). Inertial sensing of human movement. Twente University Press; Enschede. [Google Scholar]
- Madgwick SOH, Harrison AJL, and Vaidyanathan R (2011). Estimation of imu and marg orientation using a gradient descent algorithm. In 2011 IEEE International Conference on Rehabilitation Robotics, pages 1–7. [DOI] [PubMed] [Google Scholar]
- Marras W, Fathallah F, Miller R, Davis S, and Mirka G (1992). Accuracy of a three-dimensional lumbar motion monitor for recording dynamic trunk motion characteristics. International Journal of Industrial Ergonomics, 9(1):75–87. [Google Scholar]
- Marras WS, Lavender SA, Leurgans SE, Fathallah FA, Ferguson SA, Gary Allread W, and Rajulu SL (1995). Biomechanical risk factors for occupationally related low back disorders. Ergonomics, 38(2):377–410. [DOI] [PubMed] [Google Scholar]
- Mathiassen SE (2006). Diversity and variation in biomechanical exposure: what is it, and why would we like to know? Applied ergonomics, 37(4):419–427. [DOI] [PubMed] [Google Scholar]
- Mathiassen SE and Winkel J (1991). Quantifying variation in physical load using exposure-vs-time data. Ergonomics, 34(12):1455–1468. [DOI] [PubMed] [Google Scholar]
- McAtamney L and Corlett EN (1993). Rula: a survey method for the investigation of work-related upper limb disorders. Applied ergonomics, 24(2):91–99. [DOI] [PubMed] [Google Scholar]
- Moher D, Liberati A, Tetzlaff J, Altman DG, and Group P (2009). Reprint-preferred reporting items for systematic reviews and meta-analyses: the prisma statement. Physical therapy, 89(9):873–880. [PubMed] [Google Scholar]
- Mokhlespour Esfahani MI and Nussbaum MA (2018). A“smart”undershirt for tracking upper body motions: Task classification and angle estimation. IEEE Sensors Journal, 18:7650–7658. [Google Scholar]
- Mokhlespour Esfahani MI, Nussbaum MA, and Kong Z (2019). Using a smart textile system for classifying occupational manual material handling tasks: Evidence from lab-based simulations. Ergonomics, (just-accepted):1–25. [DOI] [PubMed] [Google Scholar]
- Möller T, Mathiassen SE, Franzon H, and Kihlberg S (2004). Job enlargement and mechanical exposure variability in cyclic assembly work. Ergonomics, 47(1):19–40. [DOI] [PubMed] [Google Scholar]
- Moriguchi CS, Carnaz L, Veiersted KB, Hanvold TN, Haeg LB, Hansson G-, and Coury HJCG (2013). Occupational posture exposure among construction electricians. Applied Ergonomics, 44(1):86. [DOI] [PubMed] [Google Scholar]
- Moser DJ, Graham RB, Stevenson JM, and Costigan PA (2014). Subjective and objective analysis of three water pump systems carried by forest firefighters. Work-a Journal of Prevention Assessment & Rehabilitation, 47(1):45–53. [DOI] [PubMed] [Google Scholar]
- Nath ND, Akhavian R, and Behzadan AH (2017). Ergonomic analysis of construction worker’s body postures using wearable mobile sensors. Appl Ergon, 62:107–117. [DOI] [PubMed] [Google Scholar]
- Nocerino E, Ackermann S, Del Pizzo S, Menna F, and Troisi S (2011). Low-cost human motion capture system for postural analysis onboard ships. In Videometrics, Range Imaging, and Applications XI, volume 8085, page 80850L. International Society for Optics and Photonics. [Google Scholar]
- Oshima Y, Kawaguchi K, Tanaka S, Ohkawara K, Hikihara Y, Ishikawa-Takata K, and Tabata I (2010). Classifying household and locomotive activities using a triaxial accelerometer. Gait Posture, 31(3):370–4. [DOI] [PubMed] [Google Scholar]
- Palm P, Gupta N, Forsman M, Skotte J, Nordquist T, and Holtermann A (2018). Exposure to upper arm elevation during work compared to leisure among 12 different occupations measured with triaxial accelerometers. Annals of work exposures and health, 62(6):689–698. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Paquet V, Punnett L, Woskie S, and Buchholz B (2005). Reliable exposure assessment strategies for physical ergonomics stressors in construction and other non-routinized work. Ergonomics, 48(9):1200–1219. [DOI] [PubMed] [Google Scholar]
- Pedersen ESL, Danquah I, Petersen C, and Tolstrup J (2016a). Intra-individual variability in day-to-day and month-to-month measurements of physical activity and sedentary behaviour at work and in leisure-time among danish adults. BMC public health, 16(1):1222. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pedersen SJ, Kitic CM, Bird M-L, Mainsbridge CP, and Cooley PD (2016b). Is self-reporting workplace activity worthwhile? validity and reliability of occupational sitting and physical activity questionnaire in desk-based workers. BMC Public Health, 16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Pedley M (2013). Tilt sensing using a three-axis accelerometer. Freescale semiconductor application note, 1:2012–2013. [Google Scholar]
- Peppoloni L, Filippeschi A, Ruffaldi E, and Avizzano CA (2016). (wmsds issue) a novel wearable system for the online assessment of risk for biomechanical load in repetitive efforts. International Journal of Industrial Ergonomics, 52:1–11. [Google Scholar]
- Punnett L, Fine LJ, Keyserling WM, Herrin GD, and Chaffin DB (1991). Back disorders and nonneutral trunk postures of automobile assembly workers. Scandinavian journal of work, environment & health, pages 337–346. [DOI] [PubMed] [Google Scholar]
- Putz-Anderson V, Bernard BP, Burt SE, Cole LL, Fairfield-Estill C, Fine LJ, Grant KA, Gjessing CC, Jenkins L, and Hurrell JJ Jr (1997). Musculoskeletal disorders and workplace factors. National Institute for Occupational Safety and Health (NIOSH), 104. [Google Scholar]
- Radwin RG and Lin ML (1993). An analytical method for characterizing repetitive motion and postural stress using spectral analysis. Ergonomics, 36(4):379–389. [DOI] [PubMed] [Google Scholar]
- Radwin RG, Marras WS, and Lavender SA (2001). Biomechanical aspects of work-related musculoskeletal disorders. Theoretical Issues in Ergonomics Science, 2(2):153–217. [Google Scholar]
- Ravi N, Dandekar N, Mysore P, and Littman ML (2005). Activity recognition from accelerometer data. AAAI, 3:1541–1546. [Google Scholar]
- Ribeiro DC, Milosavljevic S, and Abbott JH (2017). Effectiveness of a lumbopelvic monitor and feedback device to change postural behaviour: a protocol for the elf cluster randomised controlled trial. BMJ Open, 7(1):e015568. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Ribeiro DC, Sole G, Abbott JH, and Milosavljevic S (2011). Cumulative postural exposure measured by a novel device: a preliminary study. Ergonomics, 54(9):858–865. [DOI] [PubMed] [Google Scholar]
- Robert-Lachaine X, Larue C, Denis D, Delisle A, Mecheri H, Corbeil P, and Plamondon A (2019). Feasibility of quantifying the physical exposure of materials handlers in the workplace with magnetic and inertial measurement units. Ergonomics, pages 1–10. [DOI] [PubMed] [Google Scholar]
- Robert-Lachaine X, Mecheri H, Larue C, and Plamondon A (2017a). Effect of local magnetic field disturbances on inertial measurement units accuracy. Applied ergonomics, 63:123–132. [DOI] [PubMed] [Google Scholar]
- Robert-Lachaine X, Mecheri H, Larue C, and Plamondon A (2017b). Validation of inertial measurement units with an optoelectronic system for whole-body motion analysis. Medical & Biological Engineering & Computing, 55(4):609–619. [DOI] [PubMed] [Google Scholar]
- Samani A, Holtermann A, Spgaard K, Holtermann A, and Madeleine P (2012). Following ergonomics guidelines decreases physical and cardiovascular workload during cleaning tasks. Ergonomics, 55(3):295. [DOI] [PubMed] [Google Scholar]
- Schall MC Jr., Fethke NB, and Chen H (2016a). Working postures and physical activity among registered nurses. Applied Ergonomics, 54:243. [DOI] [PubMed] [Google Scholar]
- Schall MC Jr., Fethke NB, Chen H, and Gerr F (2015). A comparison of instrumentation methods to estimate thoracolumbar motion in field-based occupational studies. Appl Ergon, 48:224–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schall MC Jr., Fethke NB, Chen H, Oyama S, and Douphrate DI (2016b). Accuracy and repeatability of an inertial measurement unit system for field-based occupational studies. Ergonomics, 59(4):591–602. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schall MC Jr., Sesek RF, and Cavuoto LA (2018). Barriers to the adoption of wearable sensors in the workplace: A survey of occupational safety and health professionals. Human factors, page 0018720817753907. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schiefer C, Kraus T, Ellegast RP, and Ochsmann E (2015). A technical support tool for joint range of motion determination in functional diagnostics - an inter-rater study. Journal of Occupational Medicine and Toxicology, 10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Schwickert L, Becker C, Lindemann U, Marechal C, Bourke A, Chiari L, Helbostad J, Zijlstra W, Aminian K, and Todd C (2013). Fall detection with body-worn sensors. Zeitschrift fur Gerontologie und Geriatrie, 46(8):706–719. [DOI] [PubMed] [Google Scholar]
- Scott KA and Browning RC (2016). Occupational physical activity assessment for chronic disease prevention and management: A review of methods for both occupational health practitioners and researchers. J Occup Environ Hyg, 13(6):451–63. [DOI] [PubMed] [Google Scholar]
- Shafti A, Lazpita BU, Elhage O, Wurdemann HA, and Althoefer K (2016). Analysis of comfort and ergonomics for clinical work environments. Conf Proc IEEE Eng Med Biol Soc, 2016:1894–1897. [DOI] [PubMed] [Google Scholar]
- Sharp MA, Rice VJ, Nindl BC, and Williamson TL (1997). Effects of team size on the maximum weight bar lifting strength of military personnel. Human factors, 39(3):481–488. [DOI] [PubMed] [Google Scholar]
- Shojaei I, Vazirian M, Croft E, Nussbaum MA, and Bazrgari B (2016). Age related differences in mechanical demands imposed on the lower back by manual material handling tasks. Journal of Biomechanics, 49(6):896–903. [DOI] [PubMed] [Google Scholar]
- Silverstein BA, Fine LJ, and Armstrong TJ (1986). Hand wrist cumulative trauma disorders in industry. British journal of industrial medicine, 43(11):779–784. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Singh R, Carranza Leon DA, Morrow MM, Vos-Draper TL, Mc Gree ME, Weaver AL, Woolley SM, Hallbeck S, and Gebhart JB (2016). Effect of chair types on work-related musculoskeletal discomfort during vaginal surgery. Am J Obstet Gynecol, 215(5):648.e1–648.e9. [DOI] [PubMed] [Google Scholar]
- Skotte J, Korshøj M, Kristiansen J, Hanisch C, and Holtermann A (2014). Detection of physical activity types using triaxial accelerometers. Journal of physical activity and health, 11(1):76–84. [DOI] [PubMed] [Google Scholar]
- Standard B (2005). Safety of machinery: Human physical performance. BS EN, pages 1005–4. [Google Scholar]
- Stemland I, Ingebrigtsen J, Christiansen CS, Jensen BR, Hanisch C, Skotte J, and Holtermann A (2015). Validity of the acti4 method for detection of physical activity types in free-living settings: comparison with video analysis. Ergonomics, 58(6):953–965. [DOI] [PubMed] [Google Scholar]
- Stenlund TC, Ohberg F, Lundstrom R, Lindroos O, Hager C, and Rehn B (2014). Inter- and intra-tester reliability when measuring seated spinal postures with inertial sensors. International Journal of Industrial Ergonomics, 44(5):732. [Google Scholar]
- Straker L, Campbell A, Mathiassen SE, Abbott RA, Parry S, and Davey P (2014). Capturing the pattern of physical activity and sedentary behavior: Exposure variation analysis of accelerometer data. Journal of Physical Activity & Health, 11(3):614–625. [DOI] [PubMed] [Google Scholar]
- Takala E-P, Pehkonen I, Forsman M, Hansson G-A, Mathiassen SE, Neumann WP, Sjøgaard G, Veiersted KB, Westgaard RH, and Winkel J (2010). Systematic evaluation of observational methods assessing biomechanical exposures at work. Scandinavian journal of work, environment & health, pages 3–24. [DOI] [PubMed] [Google Scholar]
- Teschke K, Trask C, Johnson P, Chow Y, Village J, and Koehoorn M (2009). Measuring posture for epidemiology: Comparing inclinometry, observations and self-reports. Ergonomics, 52(9):1067–1078. [DOI] [PubMed] [Google Scholar]
- Theado E, Knapik G, and Marras W (2007). Modification of an emg-assisted biomechanical model for pushing and pulling. International Journal of Industrial Ergonomics, 37(11):825–831. [Google Scholar]
- Uiterwaal M, Glerum E, Busser H, and Van Lummel R (1998). Ambulatory monitoring of physical activity in working situations, a validation study. Journal of medical engineering & technology, 22(4):168–172. [DOI] [PubMed] [Google Scholar]
- Valero E, Sivanathan A, Bosche F, and Abdel-Wahab M (2016). Musculoskeletal disorders in construction: A review and a novel system for activity tracking with body area network. Applied Ergonomics, 54:120. [DOI] [PubMed] [Google Scholar]
- Videman T, Nurminen T, Tola S, Kuorinka I, Vanharanta H, and Troup J (1984). Low-back pain in nurses and some loading factors of work. Spine, 9(4):400–404. [DOI] [PubMed] [Google Scholar]
- Vignais N, Miezal M, Bleser G, Mura K, Gorecky D, and Marin F (2013). Innovative system for real-time ergonomic feedback in industrial manufacturing. Applied Ergonomics, 44(4):566. [DOI] [PubMed] [Google Scholar]
- Village J, Trask C, Luong N, Chow Y, Johnson P, Koehoorn M, and Teschke K (2009). Development and evaluation of an observational back-exposure sampling tool (back-est) for work-related back injury risk factors. Applied ergonomics, 40(3):538–544. [DOI] [PubMed] [Google Scholar]
- Villumsen M, Madeleine P, JØrgensen MB, Holtermann A, and Samani A (2016). Social support modifies association between forward bending of the trunk and low-back pain: Cross-sectional field study of blue-collar workers. Scandinavian Journal of Work Environment & Health, 42(2):125–134. [DOI] [PubMed] [Google Scholar]
- Villumsen M, Madeleine P, Jørgensen MB, Holtermann A, and Samani A (2017). The variability of the trunk forward bending in standing activities during work vs. leisure time. Applied Ergonomics, 58:273. [DOI] [PubMed] [Google Scholar]
- Vinstrup J, Madeleine P, Jakobsen MD, Jay K, and Andersen LL (2017). Patient transfers and risk of back injury: Protocol for a prospective cohort study with technical measurements of exposure. Jmir Research Protocols, 6(11). [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wahlström J, Bergsten E, Trask C, Mathiassen SE, Jackson J, and Forsman M (2016). Full-shift trunk and upper arm postures and movements among aircraft baggage handlers. Annals of Occupational Hygiene, 60(8):977–990. [DOI] [PubMed] [Google Scholar]
- Waters CN, Er PL, Chu AHY, Ng SHX, Chia A, Yee Wei L, and Muller-Riemenschneider F (2016). Assessing and understanding sedentary behaviour in office-based working adults: a mixed-method approach. BMC Public Health, 16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Waters TR, Putz-Anderson V, and Baron S (1998). Methods for assessing the physical demands of manual lifting: a review and case study from warehousing. American Industrial Hygiene Association Journal, 59(12):871–881. [DOI] [PubMed] [Google Scholar]
- Whitfield GP, Pettee Gabriel KK, and Kohl HW (2013). Assessing sitting across contexts: Development of the multicontext sitting time questionnaire. Research Quarterly for Exercise and Sport, 84(3):323–328. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wick K, Faude O, Schwager S, Zahner L, and Donath L (2016). Deviation between self-reported and measured occupational physical activity levels in office employees: effects of age and body composition. International Archives of Occupational and Environmental Health, 89(4):575–582. [DOI] [PubMed] [Google Scholar]
- Winkel J and Mathiassen SE (1994). Assessment of physical work load in epidemiologic studies: concepts, issues and operational considerations. Ergonomics, 37(6):979–988. [DOI] [PubMed] [Google Scholar]
- Wu G and Cavanagh PR (1995). Isb recommendations for standardization in the reporting of kinematic data. J Biomech, 28(10):1257–61. [DOI] [PubMed] [Google Scholar]
- Wu G, Siegler S, Allard P, Kirtley C, Leardini A, Rosenbaum D, Whittle M, D’Lima D, Cristofolini L, Witte H, Schmid O, and Stokes I (2002). Isb recommendation on definitions of joint coordinate system of various joints for the reporting of human joint motion—part i: ankle, hip, and spine. Journal of Biomechanics, 35(4):543–548. [DOI] [PubMed] [Google Scholar]
- Wu G, Van der Helm FC, Veeger HD, Makhsous M, Van Roy P, Anglin C, Nagels J, Karduna AR, McQuade K, and Wang X (2005). Isb recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion—part ii: shoulder, elbow, wrist and hand. Journal of biomechanics, 38(5):981–992. [DOI] [PubMed] [Google Scholar]
- Wu G and Xue S (2008). Portable preimpact fall detector with inertial sensors. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 16(2):178–183. [DOI] [PubMed] [Google Scholar]
- Yu D, Dural C, Morrow MM, Yang L, Collins JW, Hallbeck S, Kjellman M, and Forsman M (2017). Intraoperative workload in robotic surgery assessed by wearable motion tracking sensors and questionnaires. Surg Endosc, 31(2):877–886. [DOI] [PubMed] [Google Scholar]
- Zhang J, Lockhart TE, and Soangra R (2014). Classifying lower extremity muscle fatigue during walking using machine learning and inertial sensors. Annals of biomedical engineering, 42(3):600–612. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.