PDMS-embedded wearable FBG sensors for gesture recognition and communication assistance

Abstract

This study introduces fiber Bragg grating (FBG) sensors embedded in polydimethylsiloxane (PDMS) silicone elastomer specifically engineered for recognizing intricate gestures like wrist pitch, finger bending, and mouth movement. Sensors with different PDMS patch thicknesses underwent evaluation including thermal, tensile strain, and bending deformation characterization, demonstrating a stability of at least four months. Experiments revealed the FBG sensors’ accurate wrist pitch recognition across participants after calibration, confirmed by statistical metrics and Bland-Altman plots. Utilizing finger and mouth movements, the developed system shows promise in assisting post-stroke patients and individuals with disabilities, enhancing their interaction capabilities with the external surroundings.

1. Introduction

Stroke, recognized as a leading cause of death related to cerebrovascular diseases in China, exhibits alarmingly high morbidity and mortality rates [1]. The sequelae of stroke, including hemiparesis, facial paralysis, and speech or swallowing disorders, significantly impair the daily activities of affected individuals, leading to considerable declines in quality of life and imposing substantial burdens on families and society [2]. Despite enhanced survival rates attributed to medical advancements, the growing prevalence of stroke, especially in an aging population, poses challenges in alleviating post-stroke symptoms and improving survivors’ functional capabilities [3]. This underscores the urgent need for comprehensive stroke rehabilitation services, with a particular focus on motor recovery.

The integration of innovative assistive technologies and assessment methods into post-stroke care is essential for enhancing the efficacy and accuracy of rehabilitation strategies [4], including robot-assisted devices [5], gait analysis technologies [6], brain imaging techniques [7], and wearable technologies [8–10]. However, challenges persist in rehabilitating fine motor skills, such as wrist, finger, and mouth movements, due to the extensive monitoring required. Existing motion sensors often face issues such as bulkiness, data accuracy problems, and the need for substantial data processing [6,7,10]. There are ongoing efforts to develop cost-effective, portable, flexible, and comfortable devices for long-term use in home settings, capable of providing repeatable and reliable signal changes for applications in personalized healthcare and rehabilitation monitoring [8,10,11].

There is an increasing trend in optical fiber sensor usage in biomedical sciences, ranging from cardiorespiratory and hemodynamics [12–14] to blood pressure monitoring [15,16]. Among many proposed systems, the fiber Bragg gratings (FBGs) are particularly interesting due to their unique geometrical, metrological, and mechanical properties for bodily integration [17]. FBG-based sensors are also employed in posture recognition and movement monitoring, accurately measuring strain, pressure, angle, and torque [18–20]. The growing interest in this technology in biomechanics and rehabilitation engineering highlights its potential in medical diagnostics and patient care [21,22]. However, conventional silica-based FBG sensors face limitations in sensitive human motion monitoring due to their low temperature and strain sensitivity [23], and their rigidity [24]. Efforts have been made to improve the robustness, adaptability, and sensitivity of FBG sensors by embedding them in elastic materials for kin-mountable applications and enhanced sensitivity. Researchers like Tavares et al. [13], Zaltieri et al. [20], and Lo Presti et al. [12] have demonstrated the efficacy of FBG sensors in respiratory and posture monitoring, as well as multi-point measurements, when integrated with elastic substrates. Kim et al. [25] and Guo et al. [26,27] innovated FBG sensor arrays within a polymer matrix for intricate applications like finger joint movement monitoring and gesture recognition. Among the elastic materials, Polydimethylsiloxane (PDMS) stands out for its superior flexibility in fabricating complex patterned structures [28] and seamless skin adherence, enabling the detection of subtle biological signals [21,22], accurate perception of both thermal and mechanical stimuli [29], and ensuring mechanical compliance with human skin [14]. The curvature sensitivity of FBGs, significantly influenced by the strain transfer rate proportional to the distance from the fiber axis to the sensor surface [30], can be enhanced by embedding the FBGs in polymer patches, thus altering this distance and amplifying the FBG sensor's wavelength shift [31]. Innovative contributions by Guo et al. [19] and Rao et al. [32] in stretchable FBG-based strain sensors highlight the versatility of these systems. Despite these advancements, the development of standardized FBG sensors with varied dimensions for precise calibration and detection of fine joint movements remains unexplored.

In this study, we investigated the real-time monitoring capabilities of wearable FBG sensors embedded within PDMS, specifically targeting the recognition of fine gestures including wrist pitch, finger bending, and mouth movement. The impact of varying PDMS patch thicknesses on sensor response was assessed. Through calibration processes, we successfully demonstrated the accuracy of these sensors in wrist pitch recognition. A personal communication assistance system was developed to aid post-stroke patients and individuals with disabilities, offering a potential enhancement in their ability to interact and communicate effectively.

2. Methodology

In this section, we introduce the working principle, the design method, and the fabrication process of the developed FBG-based sensor. Three sensing devices of varying thicknesses were manufactured to compare the performance.

2.1. Working principle

A fiber Bragg grating is characterized as a narrow-band device in which a periodic perturbation of the refractive index along the fiber length is induced to monitor the shift in wavelength of the returned “Bragg” signal. The Bragg wavelength, ${\lambda }_B$ , manifests at the point of strongest mode coupling between the incident optical field and the index variations of the core [23]:

$${\lambda }_B=2n_{\mathit{eff}}\mathrm{\Lambda }.$$ (1)

Here, $n_{\mathit{eff}}$ represents the effective modal index and $\mathrm{\Lambda }$ denotes the periodicity; both parameters are functions of strain $\epsilon$ and temperature T. The variation in the Bragg wavelength is dependent on these parameters and can be expressed as follows [31]:

$$\frac{\mathrm{\Delta }{\lambda }_B}{{\lambda }_B}=S_{\epsilon }\mathrm{\Delta }ϵ+S_T\mathrm{\Delta }T.$$ (2)

Here, $S_{\epsilon }$ and $S_T$ represent the longitudinal strain sensitivity and the thermal sensitivity, corresponding to the photo-elastic parameter, and the thermal expansion coefficient combined with the thermal-optic coefficient of the fiber, respectively. The PDMS silicone elastomer was chosen as the embedded material owing to its flexibility in fabricating intricate patterned structures [28] and its mechanical compatibility with human skin [14]. It remains operational over a wide temperature range of -45°C to 200°C for long periods and can undergo either room temperature or heat curing. The elastomer is based on a two-part liquid component. Upon thorough mixing of these liquid components, the resultant substance solidifies into a flexible elastomer, which is particularly effective for protecting the delicate silica fiber. The external force applied to the PDMS elastomer results in elastic deformation, thereby enhancing the sensitivity of the FBG sensors embedded within it.

2.2. Biomedical sensor fabrication

This section details the fabrication process of the sensing device in which an FBG is embedded in a PDMS patch, as outlined in Fig. 1(a). The optical fiber (Corning SMF-28e) utilized in our experiments has a bare region for inscribing the FBG of 10 mm in length. The FBG was inscribed into the SMF-28e fibers using a photomask technique with ultraviolet lasers. After the inscription process, the bare region was coated with acrylate to enhance the mechanical strength of the FBG region. The Bragg wavelengths, ${\lambda }_B$ , for the utilized fibers approximate 1550.2 nm, and the reflectivity at ${\lambda }_B$ exceeds 92%. To analyze the influence of the thickness of the FBG patch on deformation sensing applications, three types of 3D printed resin molds with varying thicknesses (t) were developed, while maintaining consistent dimensions in length (l) and width (w) for all patches, measuring 40 mm and 20 mm, respectively. Two square cavities, each featuring a guide channel, were engineered at both ends of the mold. Their dimensions were calculated to ensure the fiber is clamped straight in the middle of the mold by using an adhesive (Blu-Tack) in the cavities, thus ensuring the fiber's central positioning in the PDMS patch. The PDMS-coated FBG patch was subsequently fabricated by pouring the PDMS precursor (Dow Corning Sylgard 184, with a 10:1 weight mixing ratio between the main component and the curing agent) into the mold, then undergoing a curing process for 72 hours at room temperature. Finally, the polymerized FBG patch was extricated from the mold, featuring a smooth surface and excellent transparency, facilitating easy inspection of the embedded FBG. Utilizing this fabrication method, three sensitive FBG patches were developed, each with different thicknesses: $t=$ 1 mm, 2 mm, and 3 mm for FBG-1, FBG-2, and FBG-3 sensors, respectively. A photograph of the fabricated FBG-3 sensor and its mold is presented in Fig. 1(b), where the blue arrow indicates the FBG region.

Fig. 1.

(a) Fabrication steps of the sensor, and (b) photograph of a fabricated FBG-3 sensor and its mold.

A commercial optical interrogation unit (FS22SI Industrial BraggMETER SI) was utilized in tandem with the proposed FBG sensor throughout our study. The unit offers a broadband measurement range between 1500 nm and 1600 nm, featuring a resolution of < 0.5 pm and a stability of 1 pm. The sampling rate was 1 Hz. To evaluate the FBG sensor, we analyzed the output spectra of the manufactured sensors both before and after embedding in the PDMS patch, as shown in Fig. 2. The results demonstrate that the embedded FBGs exhibit a Bragg wavelength with a slight blue shift of approximately 0.01 nm, maintaining the same 3 dB bandwidth of 0.2 nm as the bare FBGs in all three cases, while the optical power of the reflected peak remains unchanged. This can be attributed to the uniform mechanical stress on the gratings imparted by the curing process at room temperature, thereby protecting the FBG fibers without altering their spectra and ensuring accurate interrogation in peak tracking [31].

Fig. 2.

Output spectra of the three FBG sensors before and after embedding in the PDMS patch.

2.3. Thermal characterization

The PDMS-coated FBG sensors underwent temperature characteristic testing to investigate their thermal sensitivity, due to their critical role in a human movement sensing system. A heating station (DLAB, MS7-H550-Pro) was utilized to study the thermal response. The Bragg wavelength shifts of the PDMS-coated FBGs and a bare FBG were monitored in a temperature range of 30°C to 70°C, employing a stepwise methodology that encompassed both heating and cooling processes. Data collection was facilitated using the interrogation unit, with each step sustained for 10 minutes. Here, upward- and downward-pointing triangles denote temperature increase and decrease, respectively. The results, displayed in Fig. 3, reveal linear behavior and excellent reversibility for each FBG, with R-squared values exceeding 0.997. The standard deviation of each data point did not exceed 0.5 pm, indicating robust stability of the FBG sensors across various environmental temperatures.

Fig. 3.

Thermal characterization results of three PDMS-coated FBG sensors and a bare FBG.

A thermal responsivity of 12.2 pm/°C was observed for the bare FBG, which serves as a reference and closely aligns with the values reported in the literature [23]. All FBGs embedded in the PDMS patches exhibited thermal responsivity values higher than the reference sensor. Specifically, the thermal responsivities were measured at 16.5 pm/°C, 23.5 pm/°C, and 28.9 pm/°C for FBG-1, FBG-2, and FBG-3, respectively. The thermal responsivity increased with the thickness of the PDMS patch, potentially due to the increasing strain contribution originating from the patch's thermal expansion, with the PDMS's thermal expansion coefficient significantly exceeding that of silica ( $280\times {10}^{-6}\circ C^{-1}$ vs. $6.67\times {10}^{-6}\circ C^{-1}$ ) [23,33]. The observed thermal responsivity values are comparable to the bare FBG sensitivity, a phenomenon that can be ascribed to the high elasticity of the PDMS rubber, evidenced by its much lower Young’s modulus compared to that of single mode fiber (0.75 MPa vs. 13.75 GPa) [34,35], and to the low adhesion strength between the glass fibers and PDMS (the average interfacial shear strength is 0.31 MPa) [36]. In applications involving human movement detection, exhibiting greater dynamics compared to body temperature variations, the Bragg wavelength variation induced by body temperature can be neglected, even with the maximum sensitivity of 28.9 pm/°C observed in the FBG-3 case.

3. Deformation experiments and discussion

The PDMS-coated FBG sensors underwent strain characteristic testing to evaluate their potential application as biomedical sensors. To evaluate the strain response of the FBG sensors, experiments were focused on tensile and bending deformation separately. Each experiment resulted in a functioning fiber sensor, operating at a steady temperature of 25°C.

3.1. Tensile strain characterization

The experimental setup for the tensile strain characterization is presented in Fig. 4(a), featuring the FBG patch mounted on two translation stages (BOCIC, PTS306 M) using base plates. The distance between the two stages when no tensile strain was applied to the fiber patch was equal to 32 mm. Each FBG sensor was subjected to horizontal displacement ranging from 0 mm to 0.2 mm, with consistent increments of 0.02 mm. Data collection was conducted using the interrogation unit, with each step sustained for 2 minutes. The tensile strain responses of the FBG sensors were obtained through repeated calibration experiments conducted over four months, as illustrated in Fig. 4(b)-(d), with upward- and downward-pointing triangles indicating the stretching and releasing processes, respectively. Bragg wavelength shifts, resulting from axial displacement, were measured, and the average tensile strain sensitivity of each FBG sensor was calculated using linear fit curves. This analysis yielded sensitivities of 3.69 nm/mm, 3.40 nm/mm, and 2.96 nm/mm (equivalent to 0.118 pm/µµ, 0.109 pm/µε, and 0.095 pm/µε) for the FBG-3, FBG-2, and FBG-1 sensors, respectively, with all R-squared values exceeding 0.997. The results demonstrate commendable repeatability in tensile strain sensitivities for all FBG sensors, with a maximum observed deviation of 5.1% among the sensors during the period from June 2023 to October 2023. Low hysteresis behavior in the strain response, due to viscoelasticity, was observed, with hysteresis errors remaining below 6.0% in all cases [37,38], a result that can be attributed to the high elasticity of the PDMS rubber [39]. These findings align with previous experimental studies [40], noting that the tensile strain sensitivity marginally increases with the patch's thickness. This is attributed to the increased force required to elongate a patch with a larger cross-sectional area, leading to greater axial deformation of the fiber and a subsequent increase in Bragg wavelength shift, though the difference is minimal. Tensile strain characterization was not conducted on bare FBG due to its limited capacity to endure substantial tensile strain.

Fig. 4.

Tensile strain characterization results for three PDMS-coated FBG sensors. (a) Experimental setup for sensor calibration. (b)-(d) Comparative analysis of output Bragg wavelength shifts in FBG sensors, measured from June 2023 to October 2023, relative to varying external tensile strains.

An assessment of the mechanical durability of the sensors was performed through repetitive cycle tests in December 2023. During these tests, the sensors underwent over 1000 repeated cycles of stretching and releasing, with a peak strain reaching 90% of the tensile range used in the previous tensile strain characterization experiments. Given the interrogation unit's sampling rate of 1 Hz, the testing frequency for the sensors was set at 0.1 Hz. As shown in Fig. 5, the results indicate a stable response during sensor utilization, with the central wavelength in each cycle exhibiting variations of less than 0.02 nm, even after 1000 cycles. All sensors demonstrated high operational stability with consistent outputs during the cycling test, exhibiting changes of no more than 0.10 nm in the amplitude of wavelength shifts, thus confirming their excellent durability.

Fig. 5.

Results of cycle tests involving 1000 repetitions of stretching and releasing with a peak strain of 90%, demonstrating long-term durability.

3.2. Bending deformation characterization

Beyond tension sensing, the PDMS-embedded FBGs were also employed to detect strains resulting from bending deformations. The schematic of the experimental setup for strain testing of the devices is depicted in Fig. 6(a). A thin steel plate, measuring 30 cm in length and 3 cm in width with a thickness of approximately 1 mm, is clamped onto two translation stages. Sensor patches were attached to the steel plate by adhesive tapes, and were positioned 10 cm from the center and parallel to the plate, with various weights centrally loaded onto it. Weights were selected based on the stiffness and curvature performance of the thin steel plate. The test involved loads up to 600 g, with an increment of 200 g. Each test involved bending the plate both into a concave shape and a convex shape, using the same sequence of weights, achieved by flipping the steel plate with the attached sensors. In addition to the three FBG patches under test, a bare FBG was also affixed to the plate for reference purposes.

Fig. 6.

Bending strain response of three PDMS-coated FBG sensors and a bare FBG. (a) Schematic representation of the experimental setup for strain testing. (b) Observed strain trends recorded by the FBG sensors in two configurations: concave and convex bending of the plate.

Data collection was conducted using the interrogation unit, with each measurement taken over 2 minutes following the application of each load to the plate. The experimental results, depicted in Fig. 6(b), reveal that all strain curves stabilize over time, indicating a gradual equilibration of strain within the patches [41]. The wavelength shifts progressively increased/decreased with little hysteresis, returning to the baseline upon unloading [19]. Noticeable reciprocal behavior was observed when the plate was flipped for each FBG sensor; specifically, the Bragg wavelength is red-shifted under convex bending, and in the opposite direction under concave bending. The magnitudes of the wavelength shifts in concave and convex scenarios were nearly identical, as anticipated. For each FBG sensor, the wavelength shift ( $\Delta {\lambda }_B$ ) increased proportionally with the load applied to the plate, consistent with the theory that $\Delta {\lambda }_B$ is proportional to the plate's curvature when other parameters remain constant [42]. In comparison, the bare FBG demonstrated the least bending strain responsivity relative to the PDMS-embedded FBG sensors.

It is noteworthy, as indicated by the experimental results, that the greater the thickness of the PDMS patch, the more pronounced the wavelength shift, suggesting increased sensitivity. Specifically, when a 600 g weight was applied to the plate, the stabilized $\Delta {\lambda }_B$ values were measured as 0.26 nm, 0.16 nm, 0.07 nm, and 0.01 nm for FBG-3, FBG-2, FBG-1, and the bare FBG, respectively. This phenomenon can be attributed to a higher strain transfer rate over the longer distance from the fiber axis to the plate surface, approximately half the thickness of the PDMS patch in our experiments [30]. Specifically, when the SMF-28e fiber was securely clamped in the center of the mold, the heights of the fiber (i.e., the distances from the fiber axis to one surface of the PDMS patches) were 0.473 mm, 0.923 mm, and 1.423 mm for FBG-1, FBG-2, and FBG-3, respectively. Therefore, it can be inferred that a thicker PDMS patch enhances bending strain responsivity, particularly when the patches are securely attached to the plate. Additionally, it is important to note the more prominent distortion of the response curve and wavelength drift observed with thicker PDMS patches each time the load is altered. This phenomenon, attributed to the slippage between the silica fiber and the PDMS patch, could be ascribed to the more substantial deformation transmitted to the FBG in thicker sensors.

4. Wrist pitch recognition experiments and discussion

4.1. FBG sensors calibration

Wrist pitch recognition using PDMS-embedded FBG sensors was explored through a series of experimental trials aimed at assessing the capability of the proposed device in monitoring wrist pitch movements. During these trials, FBG-2 and FBG-3 were selected for the experiments due to their significantly higher bending strain responsivities. Furthermore, FBG-2 and FBG-3 sensors maintained structural integrity in all the experiments detailed in this study, attributed to the effective protection provided by the thicker PDMS patches for the silica FBGs. In the initial wrist pitch experiment trials, FBG-1 experienced a fiber break at one end of its FBG region after the wrist was bent close to 90°. This was ascribed to the insufficient protection afforded by the 1-mm PDMS patch.

Calibration of the wrist pitch sensors was conducted in a laboratory, with one sensor attached at a time to the wrist, where the center of the FBGs was positioned between the carpal and long bones (radius and ulna) of the forearm. The long edge of the FBG sensors aligned with the forearm direction to investigate wrist pitch, as shown in Fig. 7(g). Medical tapes were used to ensure firm attachment of the sensor to the skin. The sensor's position was established by flexing the wrist and identifying the bending area for placing the center of the FBGs. Given that muscle and skin conditions vary among individuals, three participants with diverse BMIs (Calibr 1-3) were enlisted to calibrate the sensors. Participant ages and anthropometric data, such as height and body mass, were listed in Table 1. Throughout the calibration process, each participant was seated with their forearm secured to the chair arm, consistently applying wrist pitch to the FBG sensor through varied hand positions in the vertical plane. An inertial measurement unit (IMU) sensor (WitMotion BWT901BLECL5.0) was attached to the back of the hand using a sports elastic band (not visible in the photo) to provide a movement reference. As shown in Fig. 7(g), real-time angles from the IMU sensor's rotation axis (x) correspond to the actual wrist pitch in degrees with an accuracy of 0.05°, registering as positive for the downward pitch and negative for the upward one. The data output frequency was standardized to 1 Hz to match the FBG sensor's interrogation unit. Each trial began with the participants’ wrists pitched near 0°, necessitating the recalibration of FBG sensors to establish a baseline for wavelength shifts originating from this initial posture. Participants were instructed to sweep their wrist pitch through the widest possible angle range, ensuring the collection of detailed and sufficient datasets for effective calibration.

Fig. 7.

Calibration of FBG sensors for wrist pitch recognition. (a)-(c) Calibration results for FBG-2 sensor with three participants. (d)-(f) Calibration results for FBG-3 sensor with the same participants. (g) FBG sensor setup for wrist pitch recognition alongside an IMU reference. (h)-(i) Relationship between wrist pitch and wavelength shift for both FBG sensors.

Table 1. Physiological information for participants.

Index	Gender	Age	Height (cm)	Weight (kg)	BMI
Calibr 1	Female	36	170	55	19.0
Calibr 2	Male	22	172	60	20.3
Calibr 3	Male	36	181	86	26.3
Vol 1	Male	25	178	74	23.4
Vol 2	Male	23	186	85	24.6
Vol 3	Male	22	174	55	18.2
Vol 4	Male	24	178	75	23.7
Vol 5	Male	26	177	70	22.3

Calibration results of FBG-2 and FBG-3 involving three participants are illustrated in Fig. 7(a)-(f), with black lines in each figure representing IMU sensor data. Overall, wavelength shifts showed a strong correlation with wrist pitch angles. For both FBG-2 and FBG-3, downward wrist pitch resulted in positive wavelength shifts, whereas upward pitch yielded negative shifts. The thicker PDMS patch of FBG-3 resulted in larger wavelength shifts, consistent with observations in Fig. 6. All data sets were compiled and plotted to illustrate the relationship between wrist pitch and wavelength shift for each sensor, as displayed in Fig. 7(h)-(i). Had a linear fit been applied, it would have yielded R-squared values of only 0.913 for FBG-2 and 0.861 for FBG-3, respectively. This behavior could be attributed to the wrist anatomy, as the movement of the bones and muscles around the wrist was not linear. As a result, a 6th-degree polynomial fit was applied, yielding R-squared values of approximately 0.98. Subsequently, this polynomial fit was utilized to convert wavelength shifts into estimated wrist pitch values in subsequent experiments. Figure 7(h)-(i) illustrates that the wrist pitch range was limited when employing FBG-3, a constraint attributed to the stiffness of its thicker PDMS patch, which consequently reduced wrist flexibility. In contrast, FBG-2 exhibited a broader tuning range for wrist pitches, approaching ±90°. Despite the reduced pitch range of FBG-3, it exhibited greater sensitivity compared to FBG-2. For instance, at a wrist pitch of 45°, the sensitivities of FBG-2 and FBG-3 were measured at 5.92 pm/° and 7.47 pm/°, respectively. Similarly, at a wrist pitch of -45°, the sensitivities of FBG-2 and FBG-3 were found to be 12.85 pm/° and 24.90 pm/°, respectively. This discrepancy in sensitivity arises from the intricate anatomy of the wrist, as previously mentioned, resulting in varying bending performance of the sensors for positive and negative pitches. The achieved sensitivity represents an improvement compared to the literature [18], where a sensitivity of 5.6 pm/° was reported. This enhancement can be attributed to the specific dimensions of the PDMS patches that we designed. For both sensors, the data set began to form clusters when wrist pitches fell below 0°, particularly at larger angles. This clustering may be attributed to the participants’ muscle and skin conditions, such as slippage between skin and muscle during upward wrist pitching. Variations in participants’ BMIs might also amplify this effect.

4.2. Different position test

Participant Calibr 1 was engaged in a further experimental trial to assess the performance of FBG sensors in wrist pitch recognition, considering different sensor placements. A ruler was aligned along the arm and marked on the wrist for reference. The zero position (0 mm) was established on the median line between the carpal and long bones (radius and ulna) of the forearm. Previously, during calibration trials, participants Calibr 1-3 had aligned the FBG sensor's center with this zero position. In this specific trial, Calibr 1 methodically adjusted the FBG sensor's position from -10 mm to 10 mm, following the ruler's positive direction from hand to arm, to evaluate its correlation, stability, and agreement with the IMU sensor. For each trial, the participant initiated with a wrist pitch near 0°, and the FBG sensors were recalibrated to ensure wavelength shifts were referenced from this initial posture. Figure 8(a)-(e) display correlation plots comparing estimated wrist pitch values with reference values. Estimated values derived from the wavelength shift of the FBG-2 sensor, calculated via the polynomial fit established in earlier calibration experiments, with an average of approximately 500 data points collected at each position. Reference values were directly sourced from the IMU sensor. These figures demonstrate a strong correlation between the estimated and reference wrist pitch values, with a Pearson correlation coefficient (r) consistently exceeding 0.988. The Root Mean Square Error (RMSE) metric was employed to evaluate the stability of the estimated wrist pitch, indicated in each figure, and is calculated as follows:

$$RMSE=\sqrt{\frac{1}{n}\sum _{i=1}^n{(X_i-X_{i0})}^2},$$ (3)

where X refers to wrist pitch, the subscripts i and i₀ denote the estimated and reference values respectively, and n indicates the number of measurements. Overall, RMSE values across all cases were determined to be no greater than 9.14°, with smaller values achieved when the FBG patch was positioned closer to the wrist's median line.

Fig. 8.

Evaluation of FBG-2 sensor at varied positions. (a)-(e) Correlation plots for the FBG-2 sensor placed at different positions. (f)-(j) Bland-Altman diagrams corresponding to each position.

Bland-Altman diagrams for each position are presented in Fig. 8(f)-(j). This method analyzed the agreement between estimated and reference values, highlighting mean error (µ) and the corresponding limits of agreement (LoAs), denoted by horizontal dash-dotted lines in each figure, with

$$\mu =\frac{1}{n}\sum _{i=1}^n(X_i-X_{i0}),$$ (4)

$$LoA=\mu \pm 1.96\sigma ,$$ (5)

where $\sigma$ is the standard deviation (SD) and is calculated as

$$\sigma =\sqrt{\frac{{\sum }_{i=1}^n{(X_i-X_{i0}-\mu )}^2}{n-1}}.$$ (6)

Mean errors across all positions were comparable, falling within ±5.5°. Standard deviation varied across positions, decreasing as the FBG patch was placed closer to the wrist's median line, mirroring the RMSE trend. Notably, with the FBG sensor's center positioned at -5 mm and 0 mm, LoA intervals ranged from -9.41° to 2.23°, and from -9.60° to 3.96°, respectively. For each position, more than 93.8% of measurements fell within the LoA range, indicating good agreement between wrist pitch estimates from the FBG sensor and the IMU sensor.

The results were consistent and analogous behavior was demonstrated with the FBG-3 sensor in the experimental trials, with an average of 500 data points at each position. Correlation plots, shown in Fig. 9(a)-(e), demonstrate a strong correlation between the estimated and reference wrist pitch values, with a Pearson correlation coefficient (r) consistently exceeding 0.988 in each case. An overall RMSE of up to 8.66° was recorded for each position, reducing to 3.7° when the FBG patch was positioned near the wrist's median line. Bland-Altman plots in Fig. 9(f)-(j) reveal mean errors for the estimated and reference values using the FBG-3 sensor, within a range of ±3.9°, which is slightly lower than those obtained with the FBG-2 sensor. With the FBG sensor's center positioned at 0 mm and 5 mm, LoA intervals ranged from -8.05° to 5.34°, and from -7.17° to 7.17°, respectively. In each position, more than 96.0% of measurements fell within the LoA range, signifying enhanced agreement between the FBG-3 sensor and the IMU sensor.

Fig. 9.

Evaluation of FBG-3 sensor at varied positions. (a)-(e) Correlation plots for the FBG-3 sensor placed at different positions. (f)-(j) Bland-Altman diagrams corresponding to each position.

The results demonstrated the effectiveness of both sensors across a broad range of positions on the wrist for accurate pitch recognition based on previous calibration. Specifically, positioning the FBG sensors within a range of -5 mm to 5 mm around the median line yielded precise wrist pitch measurements. This was evidenced by an average correlation coefficient of 0.994 for both FBG-2 and FBG-3 sensors, with an average RMSE of 5.42° for FBG-2 and 4.82° for FBG-3, and an average standard deviation of 4.07° for FBG-2 and 4.70° for FBG-3.

4.3. Wrist pitch monitoring

Subsequent experiments involved five volunteers, whose physiological details are listed in Table 1. Each volunteer was instructed in a sitting posture with their forearm secured on the chair arm, mirroring the setup in the previous section. Volunteers were invited to place the FBG sensor around the middle area (approximately 0 mm position) of their wrist, as self-determined, for each trial. The IMU sensor was attached to the back of the hand to provide movement reference. The duration of each trial with the FBG sensor was approximately 10 minutes on average. Experimental results were processed and analyzed similarly to the previous section, with findings presented in Fig. 10 and Fig. 11. The results demonstrated that our FBG sensors are capable of accurately predicting wrist pitch across individuals with varying BMIs. Figure 10(a)-(e) display correlation plots for the FBG-2 sensor, demonstrating good performance in comparing estimated and reference wrist pitch, with a Pearson correlation coefficient (r) above 0.991 in each case. An RMSE up to 7.08° was recorded for each volunteer. Figure 10(f)-(j) present Bland-Altman plots, showing mean errors for the estimated and reference values using the FBG-2 sensor, within a range of ±4.4°, and an average standard deviation of 4.66°. For each volunteer, 93.1%, 99.1%, 95.9%, 92.1%, and 96.5% of measurements respectively fell within the LoA range, indicating good agreement between the FBG-2 sensor and the IMU sensor.

Fig. 10.

Performance analysis of FBG-2 sensor with different volunteers. (a)-(e) Correlation plots for the FBG-2 sensor as worn by individual volunteers. (f)-(j) Bland-Altman diagrams for each volunteer.

Fig. 11.

Performance analysis of FBG-3 sensor with different volunteers. (a)-(e) Correlation plots for the FBG-3 sensor as worn by individual volunteers. (f)-(j) Bland-Altman diagrams for each volunteer.

Compared to the FBG-2 sensor, results with the FBG-3 sensor exhibited a marginally lower Pearson correlation coefficient (r), as shown in Fig. 11(a)-(e), with a value of 0.983 for Vol 3. However, the results still indicated a strong correlation between the estimated and reference wrist pitches, with an RMSE of up to 8.49°. Figure 11(f)-(j) show mean errors for the estimated and reference values using the FBG-3 sensor, within a range of ±3.9°, and an average standard deviation of 5.94°. For each volunteer, 94.6%, 97.1%, 99.8%, 98.4%, and 95.3% of measurements fell within the LoA range, indicating strong agreement between the FBG-3 sensor and the IMU sensor.

These results, complemented by the outcomes from the previous sections, demonstrate the potential of PDMS-embedded FBG-based sensors as an effective method for wrist pitch monitoring, exhibiting commendable correlation, stability, and agreement with commercial IMU sensors. The correlation coefficients obtained across our participants are higher than those reported in recent research on FBG-based wearable systems for sitting posture recognition and respiratory rate evaluation [20] the effectiveness of our calibration methodology in wrist pitch recognition. Compared to previous studies that utilized FBG arrays within polymer matrices for applications such as finger joint movement monitoring and gesture recognition [25,26,32], our research takes a comprehensive approach by engaging a varied participant group. We calibrated the FBG sensors with three participants and subsequently applied this calibration curve to an additional five individuals. While variations in bone shape, muscle, and skin conditions exist among individuals, our experimental results confirm the adaptability and potential of our fabricated FBG sensors for broader applications among diverse individuals with enhanced accuracy. With additional advancements in the interrogation unit software, real-time estimation of wrist pitch using FBG sensors becomes feasible using the calibration curve, enabling direct display on a computer screen.

5. Gesture recognition and communication assistance

A volunteer was engaged to conduct experiments on finger and mouth gesture recognition using FBG sensors. Figure 12(a)-(d) illustrate the use of FBG sensors in detecting finger joint movements, with the center of the FBG sensors attached to the middle phalanges of the index finger. This setup translated the movement of the distal and proximal phalanges into sensitive wavelength shifts detected by the interrogation unit, resulting in a negative direction of wavelength shift due to the concave bending of the FBG sensors. Medical tape was used to ensure firm attachment of the sensor to the finger. When the knuckles were bent stepwise from 0° to 90° to 0° with 30° increments, the FBG sensors exhibited corresponding wavelength shifts in response to the bending deformation. The angles were determined by aligning a transparent protractor against the finger. As depicted in Fig. 12(a)-(b), the magnitude of the wavelength shift was larger for FBG-3, indicating higher sensitivity compared to the FBG-2 sensor, consistent with results from the previous sections. The average standard deviation for the FBG-3 and FBG-2 sensors was 0.03 nm and 0.02 nm, respectively, indicating a stable response and consistent angle maintenance. From the participant’s perspective, more effort was required to bend the finger and maintain the gesture with FBG-3 due to its thicker structure. A steady series of wavelength shifts was observed when the finger was continuously bent, as depicted in Fig. 12(c)-(d). The bending gestures were typically maintained for approximately 1 second, allowing the interrogation unit to record the wavelength shifts and display them. The slight variation in the magnitude of wavelength shifts was attributed to the volunteer's difficulty in precisely controlling the finger to the designated angle during testing.

Fig. 12.

Wavelength shift responses to finger bending angles. (a), (c): Responses of the FBG-2 sensor. (b), (d): Responses of the FBG-3 sensor.

The experimental outcomes suggest the potential for developing a personal communication assistance system using FBG sensors, designed to aid post-stroke individuals with disabilities in interacting with the external environment. In this context, international Morse code was adopted as the communication protocol, utilizing binary signals derived from finger joint movements to convey information. As explained in Fig. 13(a), Morse code employs dots (·) and dashes (−) to transmit messages. Various combinations of these elements represent individual letters, enabling the formation of words and sentences. To align with the interrogation unit's sampling rate of 1 Hz, the duration of each Morse code unit was set to 1 second. Participants managed the signal transmission effectively with the aid of a metronome.

Fig. 13.

Application of a personal communication assistance system with FBG sensors. (a) Morse code rules for message transmission. (b) FBG-2 and FBG-3 sensors attached to fingers for simultaneous bending. (c), (d): Fingers bending demonstrating the interpretation of the words “BIO” and “SENSOR”. (e), (f): Volunteer opens mouth to silently articulate “a,” spelling the word “BNU”. (g), (h): Volunteer pouts to silently articulate “u,” forming the word “FIBER”.

A volunteer attached the FBG-2 and FBG-3 sensors to the middle phalanges of their respective index fingers, as depicted in Fig. 13(b). Both index fingers were simultaneously bent following the metronome's beat for comparative analysis. The volunteer was instructed to bend the FBG sensors with approximately equal force. Figure 13(c)-(d) demonstrate that through appropriate management of knuckle motion, the words “BIO” and “SENSOR” were interpretable according to Morse code conventions. The FBG sensors were alternated between the left and right index fingers in Fig. 13(c) and (d). Results indicated that the magnitude of wavelength shifts for the FBG-3 sensor was smaller than that of the FBG-2 sensor when the volunteer bent both fingers with approximately the same force. Referencing results in Fig. 12, the bending angle for the FBG-3 sensor here was estimated to be between 30° and 60°, reflecting its stiffness. This trend remained consistent upon switching the two sensors.

An investigation was conducted into the sensing of minor strain around the risorius muscle when the volunteer simulated mouth movements to silently pronounce “a” and “u” sounds. When the FBG sensors were vertically attached adjacent to the mouth, as shown in Fig. 13(e), elongation of the sensor occurred as the volunteer opened his mouth to silently articulate “a”. Similarly, with horizontal placement adjacent to the mouth, as shown in Fig. 13(g), sensor elongation was observed when the volunteer pouted to silently articulate “u”. An FBG-2 and an FBG-3 were symmetrically attached to the volunteer’s face for comparison. Medical tapes were used to ensure firm attachment of the sensors to the face. Results shown in Fig. 13(f) and (h) exhibit a series of wavelength shifts, enabling the interpretation of the words “BNU” and “FIBER” according to Morse code conventions. As facial muscles are typically controlled simultaneously on both sides, the wavelength shift behaviors of FBG-2 and FBG-3 were quite similar. However, due to the greater force required to elongate the FBG-3 sensor, it tended to resist elongation more strongly and was prone to slipping when the mouth was opened or pouted, even with the aid of medical tape. Consequently, the magnitude of the wavelength shifts for the FBG-3 sensor was smaller than that of the FBG-2 sensor. Considering these findings, FBG-2 emerges as an effective communication assistance tool, offering individuals with disabilities a means to communicate externally, particularly in post-stroke settings, due to its sufficient sensitivity and accessibility.

6. Conclusion

In this study, the capacity of PDMS-embedded FBG sensors to recognize fine gestures, including wrist pitch, finger bending, and mouth movement was evaluated. By investigating three different thicknesses of PDMS patches, we assessed the sensors’ thermal, tensile strain, and bending deformation responses, demonstrating at least a 4-month stability of the sensors. Findings reveal that these FBG sensors are capable of accurately recognizing wrist pitch recognition across various positions following meticulous calibration. Both FBG-2 and FBG-3 sensors demonstrated consistent performance in estimating wrist pitch with reference from an Inertial Measurement Unit, as evidenced by mean errors within ±4.4° and an average standard deviation not exceeding 5.94°. Bland-Altman plots reinforced the strong agreement between the FBG sensors and the IMU sensor. Utilizing finger bending or mouth movement, this system shows promise in aiding post-stroke patients and individuals with disabilities, enhancing their ability to interact effectively with their surroundings. This is particularly relevant in post-stroke rehabilitation, where the restoration of fine motor skills is crucial. Future research could focus on the development of a portable and cost-effective interrogation system for these sensors to further enhance their practicality and versatility in a variety of settings, further integrating them into personalized healthcare and rehabilitation technologies.

Funding

Basic and Applied Basic Research Foundation of Guangdong Province10.13039/501100021171 (2021A1515011997, 2021A1515110310); Natural Science Foundation of Guangxi Province10.13039/501100004607 (2023GXNSFDA026040); National Natural Science Foundation of China10.13039/501100001809 (62003046, 62111530238).

Disclosures

The authors declare no conflict of interest.

Data availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.