Friction and vibration generated during sliding over bare skin correlate with perceived smoothness and moistness

Abstract

Objective

Tactile perception of the skin is a key factor for consumers when evaluating skincare products, yet conventional sensory assessments are subjective, costly and often lack clear physical interpretations. This study aims to bridge this gap by developing a tactile sensor that quantifies physical parameters relevant to skin feel and combines these measurements with sensory evaluation.

Methods

A tactile sensor with a stainless‐steel probe replicating a human fingerprint was developed to measure physical parameters—specifically, vibration and friction—during controlled rubbing on the cheek. Simultaneously, blinded sensory evaluations were performed by assessors rating six attributes: smooth–rough, moisturized–dry, soft–hard (surface), soft–hard (base), elastic–not elastic and oily–not oily. Multiple regression analyses were conducted using vibration features and friction coefficients as independent variables and sensory evaluation scores as dependent variables.

Results

Regression analysis revealed that perceived smoothness (smooth–rough) was predicted by amplitudes of high‐frequency vibration (R ² = 0.708; adjusted R ² = 0.676). Moistness (moisturized–dry) was best predicted by a combination of high‐frequency vibration amplitude and friction coefficients (R ² = 0.584; adjusted R ² = 0.538). No significant regression models were obtained for the other sensory attributes.

Conclusion

These findings highlight the potential of objective, instrument‐based tactile measurements to elucidate the physical basis of sensory evaluation terms in skincare. This approach may enhance the training of sensory evaluation panels and improve the consistency and reliability of product assessment, ultimately supporting more effective product development.

A tactile sensor with a probe replicating a fingerprint quantified skin texture by measuring vibration and friction. Regression analysis linked smoothness and moistness perceptions with high‐frequency vibration amplitude and friction. These findings support the objective assessment of skin texture, enhancing product evaluation consistency and bridging physical measures with subjective sensory impressions.

INTRODUCTION

Skin texture is a critical factor for consumers when assessing their skin condition, making accurate evaluation methods essential in skincare product development. Beyond subjective consumer impressions, trained evaluators perform sensory assessments of various skin attributes using different tactile motions [1]. While sensory evaluation remains the standard for skin assessment, objective instrumental measurements have also been introduced [2, 3]. Most commercially available devices assess skin condition using non‐contact methods or by applying pressure or suction [4, 5, 6]. However, both consumers and evaluators naturally assess skin texture not only by pressing but also by sliding their fingers across the surface [7, 8], a motion that many devices do not accommodate.

Material texture is generally defined along five dimensions: microroughness, macroroughness, hardness/softness, friction and thermal properties, with sliding motions being particularly relevant to microroughness and friction [9, 10]. Microroughness is detected as vibrations generated during rubbing [10] and linked with dynamic friction [11, 12], and thus, measuring both vibrations and friction is important for characterizing tactile perception of materials and fabrics [13, 14, 15, 16]. For this reason, we developed a handheld tactile sensor capable of simultaneously measuring vibration and friction during rubbing on human skin [17]. This automated, reproducible device enables consistent measurements across various skin areas. Although previous sensors have been developed to capture contact information during skin rubbing—including handheld devices measuring friction [18, 19] and a sensor by Nakatani et al. that simultaneously measured contact force and vibration [20]—these systems have not been designed to measure both friction and vibration simultaneously.

Sensory evaluation has limitations, including the need for extensive training and multiple evaluators, which are both time‐consuming and costly. Automated skin assessment using tactile sensors has the potential to reduce these costs and improve objectivity. Furthermore, integrating sensory evaluations with instrumental data can help validate sensory approaches and clarify the physical properties underlying perceived texture.

Friction generated during skin rubbing has been extensively studied by previous researchers. Pailler‐Mattei et al. [21] demonstrated that the removal of the stratum corneum increases both the frictional force and the amplitude of stick–slip vibrations. Hendriks et al. [19] reported that higher humidity and temperature nearly doubled the average friction coefficient. Gerhardt et al. [22] found friction coefficients ranging from 0.3 to 0.43 between forearm skin and textiles, with no age‐related differences. Elkhyat et al. [23] showed that greater skin hydrophobicity results in lower friction coefficients. However, these studies did not examine the relationship between friction and sensory evaluation.

Horiuchi et al. [24] investigated frictional signals and sensory comfort during cosmetic foundation application but found that static and kinetic friction were not always dominant determinants of tactile comfort. Akiyama et al. [25] reported that friction coefficients when applying lotion to the arm were related to perceived friction. Egawa et al. [26] found correlations between average friction coefficients and sensory attributes such as moistness and stickiness, although they did not include vibration measurements or cheek assessments. Tanaka et al. [27] developed a polyvinylidene‐fluoride‐based tactile sensor and found that vibration measurements correlated with clinical findings for diseased skin. Nevertheless, no previous studies have comprehensively examined how friction and vibration during rubbing relate to sensory parameters such as smoothness and moistness.

The present study investigates the relationship between vibrations and friction, measured during probe‐based rubbing on bare skin, and sensory evaluation scores, with the aim of improving texture evaluation methods for skincare development. Our previous work [17] focused on sensor development and validation using skin models and assessed only smoothness. In this study, we apply the sensor to human skin, evaluating a broader range of sensory attributes to estimate sensory evaluation scores from measured physical parameters.

METHODS

Tactile sensor design

Figure 1 shows the tactile sensor developed in our previous work [17] and utilized in the present study. The sensor housing was designed for easy grasping (Figure 1a) to measure both vibrations and contact forces generated as the probe slides across the skin.

FIGURE 1.

Tactile sensor used in this study. (a) Photograph of the handheld sensor. (b) Stainless steel probe designed to mimic the surface shape of a fingerprint; the cross‐section shows an array of semicylindrical ridges with a radius of 250 μm. (c) Internal structure of the sensor: Right, overview of the device; left, detailed view of the sensor tip. (d) Example photographs of the sensor measuring the cheek area. Adapted from [17].

To accommodate the limited space of areas such as the cheek (Figure 1d), the probe was rotated along the x‐axis with a radius of 7.35 mm, driven by a stepping motor to ensure a sufficient sliding distance. Two probes were symmetrically mounted about the rotational centre to stabilize the motion. The probe surface featured an array of semi‐cylindrical ridges with a radius of 250 μm, referencing designs commonly used in fabric texture evaluation (Figure 1b) [14]. The probe dimensions were 6.5 × 9.5 mm. A depth adjustment mechanism at the tip of the housing maintained an approximately constant normal contact force (0.1–0.4 N) between the probe and the skin, independent of the operator.

As depicted in Figure 1c, the sensor's internal structure included a force sensor (S10S1‐WM155‐K1‐P5A, Touchence Inc.) and a three‐axis accelerometer (LIS344ALH, STMicroelectronics), both positioned directly beneath one of the probes. The outputs from these sensors were synchronously recorded at a sampling frequency of 2000 Hz, enabling the capture of high‐frequency vibrations within the perceptual range of human touch. Details regarding the sensor's resolution and linearity have been reported previously [17].

Participants

This study was approved by the Human Research Ethics Committee of the Shiseido Global Innovation Center (Study Nos. C02205, C02206), and written informed consent was obtained from all participants and evaluators. Data from our previous study [1] were partially included in this analysis.

The study involved four sensory evaluators: two trained professional assessors and two untrained researchers. The professional evaluators had 5 and 16 years of experience, respectively, and conducted tactile evaluations of skin and cosmetics for hundreds of individuals as part of their professional duties. The untrained evaluators were researchers with no prior experience in the tactile assessment of bare skin.

The participants evaluated comprised 22 women, aged 25–57 years.

Sensory evaluation

The evaluation site was a 40 mm‐diameter area on the left cheek adjacent to the nose, covered with a 0.2‐mm‐thick polypropylene sheet (Figure 2a). Although professional evaluators typically do not blindfold themselves when assessing bare skin, all evaluators in this study wore blindfolds to prevent identification of the participants (Figure 2b). Because the evaluators could not visually locate the evaluation site, participants guided the evaluators' hands to the correct area.

FIGURE 2.

Images of the evaluator and the participant during sensory evaluation. (a) Evaluation area, (b) experimental scene. Left: Participant. Right: Evaluator.

The experiment was conducted over 3 days, with six to eight participants assessed each day. Before evaluation, participants removed their makeup and washed their faces; sensory assessments were conducted 5–10 min after washing. Each participant was evaluated twice in a blinded manner. Evaluators were not informed of the number of participants or how many times each participant was evaluated. Additionally, no participant was evaluated consecutively by the same evaluator. To minimize changes in skin condition, the second evaluation for each participant was completed within 30 min of the first.

Six attributes were assessed: ‘smooth–rough’, ‘moisturized–dry’, ‘soft–hard (surface)’, ‘soft–hard (base)’, ‘elastic–not elastic’ and ‘oily–not oily’. Human skin has a layered structure, and the physical properties of its superficial and deeper layers differ [28, 29, 30]. For example, Isogai et al. [30] reported that the Young's moduli of subcutaneous tissue and muscle in the buttocks were not correlated, suggesting the importance of evaluating the superficial and deep layers independently. Accordingly, the evaluations for surface softness and underlying softness were separated in our design. To assess surface softness, the evaluator gently touched the cheek, whereas to assess underlying softness, the evaluator applied a pressing motion. The applied force or depth of pressing was not explicitly measured. Professional evaluators were trained to differentiate between these two types of softness, and non‐professional evaluators received prior instruction in the procedure. Definitions of each attribute, shown in Table 1, were provided in both Japanese and English. These attributes align with those commonly used in skin sensory evaluation [7, 8, 31, 32, 33].

TABLE 1.

Definitions of sensory evaluation items.

Attribute	Definition
Smooth (0)–rough (5)	Overall assessment of unevenness and roughness perceived when sliding a finger across the skin surface. Evaluation performed using reference artificial skin models
Elastic (0)–not elastic (5)	The force with which the skin pushes back against the finger after pressing. Evaluation performed using reference artificial skin models
Soft (0)–hard (surface) (5)	Resistance encountered when pressing to a depth of approximately 2 mm. Evaluation performed using reference artificial skin models
Soft (0)–hard (base) (5)	Resistance encountered when pressing to a depth of approximately 5 mm. Evaluation performed using reference artificial skin models
Moisturized (0)–dry (5)	Perceived stickiness felt on the finger when gently pressing and sliding on the skin. If some areas feel moisturized and others feel dry, provide an average evaluation
Oily (0)‐not oily (5)	Assessment of the degree to which oiliness is perceived on the skin surface

For four of the six attributes, reference artificial skin models were used; the remaining two attributes were evaluated without references. Two sets of four polyurethane models served as standards: one set (diameter: 45 mm, height: 6 mm) for smoothness and another set (diameter: 50 mm, height: 15 mm) for soft–hard (surface), soft–hard (base) and elasticity. All reference models were created under the supervision of experienced professional evaluators to ensure even score distribution across four levels. Due to confidentiality agreements with the manufacturers, details of the artificial skin fabrication process cannot be disclosed.

Evaluations were performed on an 11‐point scale (0–5, in 0.5‐point increments). Hand movements during evaluation were not restricted, allowing each evaluator to use their preferred assessment technique.

Measurements of the skin using sensors

Following the sensory evaluation, skin measurements were conducted using the developed tactile sensor as well as five additional skin measurement instruments, including commercially available devices, for comparative purposes. All measurements were performed on the same area of the left cheek as used in the sensory evaluation.

For the developed tactile sensor, the probe was rubbed across the skin at a speed of 25.7 mm/s for 4 s, a motion consistent with natural human rubbing and designed to avoid resonance within the sensor system. This corresponded to a probe rotation speed of 200°/s, resulting in approximately 2.2 rotations over the 4‐s period. Three measurements were obtained for each participant.

The commercial devices used for comparison included a video microscope, VisioScan VC98 (Courage & Khazaka, Cologne, Germany), Corneometer CM825 (Courage & Khazaka), Skicon‐200EX (Yayoi Co., Ltd., Tokyo, Japan) and Cutometer CT580 (Courage & Khazaka). For the video microscope, VisioScan and Cutometer, three measurements were taken for each participant; for the Corneometer CM825, five measurements were taken; and for the Skicon‐200EX, seven measurements were taken. Cutometer measurements were performed using a suction diameter of 2 mm and a suction pressure of 400 mbar.

Data analysis

Outlier removal

Among the 22 participants, one was excluded from the analysis as an outlier. This exclusion was due to sweating during the instrumental skin measurements, which led to a noticeably different tactile sensation during the sensory evaluation.

Analysis of sensory evaluation scores

Scores for the ‘oily–not oily’ attribute were excluded from analysis, as values were nearly zero across all participants. This likely resulted from the evaluations being conducted immediately after face washing, when sebum levels were minimal. Each participant was evaluated twice by the same evaluator; any participant showing large discrepancies between the first and second scores for a given attribute was considered an outlier. The outlier criterion was set as twice the standard deviation for all evaluators. Based on this criterion, five trials or participants were excluded as outliers for ‘smooth–rough’ and ‘moisturized–dry’, and seven trials were excluded for ‘elastic–not elastic’, ‘soft–hard (surface)’ and ‘soft‐hard (base)’.

No significant differences were found in the intra‐class correlation coefficients between professional evaluators and researchers. Therefore, the arithmetic mean of the scores from all four evaluators was used for subsequent analyses, consistent with previous methods [1].

Data analysis of the developed sensor

Vibration features were calculated from the acceleration in the x‐direction, corresponding to the direction of probe rotation. The analysis interval was set to one complete rotation of the probe, specifically from 2.2 to 4 s after measurement initiation, yielding 3600 data points. A Fast Fourier Transform (FFT) was applied to the acceleration data, and the amplitude of each frequency component $y\left(\mathrm{\omega }\right)$ was converted to decibel values $y_{\mathrm{dB}}\left(\mathrm{\omega }\right)$ using Equation (1), where the reference acceleration $y_0\left(\omega \right)$ was set to 1 G:

$$y_{\mathit{dB}}\left(\omega \right)=20{\log }_{10}\frac{y\left(\omega \right)}{y_0\left(\omega \right)}$$ (1)

Given the high dimensionality of the amplitude spectrum, partial least squares (PLS) regression was employed to reduce dimensionality while preserving features most relevant to the sensory evaluation scores. Such approaches are commonly employed in statistical analyses when dealing with a large number of explanatory variables [34, 35]. Unlike traditional dimensionality reduction methods such as principal component analysis, PLS aligns the extracted components with the dependent variables, making it particularly suitable for this study. Sensory evaluation scores served as dependent variables, whereas the decibel values of the averaged vibration amplitudes at 10 Hz intervals were used as independent variables. All variables were standardized to a mean of 0 and a standard deviation of 1 prior to analysis.

The optimal number of principal components (PCs) was determined by two‐fold cross‐validation, repeated 100 times, selecting the number of components that minimized the mean squared error (MSE). PC scores were then computed for all samples based on the selected component number. For the ‘smooth–rough’ attribute, two vibration components minimized MSE; for the ‘moisturized–dry’ attribute, one component was optimal. Vibration features were not used for the other sensory attributes, as they did not contribute to the prediction of sensory evaluation scores.

In addition, three features were derived from the normal force ($F_z$) and the friction force ($F_x$) measured by the force sensor: the maximum initial friction coefficient (${\mathrm{\mu }}_{\text{init}}$), the kinetic friction coefficient (${\mathrm{\mu }}_{\mathrm{k}}$), and the mean deviation of the friction coefficient (MMD). Both $F_z$ and $F_x$ were first processed using a low‐pass filter (cut‐off frequency: 35 Hz). The friction coefficient at each time point $t$ was then calculated as follows:

$$\begin{array}{c}\mathrm{\mu }\left(t\right)=\frac{F_x\left(t\right)}{F_z\left(t\right)}.\end{array}$$ (2)

${\mathrm{\mu }}_{\text{init}}$ was defined as the maximum value of $\mathrm{\mu }$ from rubbing initiation to 0.3 s. ${\mathrm{\mu }}_k$ was the average value of $\mathrm{\mu }$ during the same interval as the vibration features, from 2.2 to 4 s. To calculate the MMD, the raw data for $F_z$ and $F_x$ were used without applying a low‐pass filter, and the mean deviation of the friction coefficient was calculated for the interval between 2.2 and 4 s.

By first applying PLS regression to reduce the dimensionality of the vibration features, we ensured that only the most relevant components were retained for subsequent analysis. This approach allowed us to focus on the contributions of vibration and friction to tactile sensation while avoiding overfitting and maintaining interpretability. The analysis was conducted using MATLAB software (2022b).

Data analysis of commercial sensors

A video microscope was used to capture images of a 5.0 mm × 6.7 mm area on the cheek. Image analysis was performed to extract features characterizing skin texture amount and unevenness. The amount of texture was quantified as the number of black pixels in a binarized image [36]. Texture unevenness was defined as the coefficient of variation in the number of black pixels across 12 × 12 pixel subregions of the image, with higher values indicating greater unevenness and lower values indicating more uniform texture. Three images were acquired from the evaluated cheek area, and the mean values for texture amount and unevenness were used as representative measures.

Surface condition of the skin was further evaluated using the VisioScan VC98 [6]. Four parameters were analysed: SEw (wrinkles), SEsm (smoothness), SEr (roughness) and SEsc (scaling). The average values of three measurements were calculated for each parameter.

For the Corneometer CM825, the average value of five measurements was used. For the Skicon‐200EX, the average value of seven measurements was used. For the Cutometer CT580, the average value of three measurements was calculated for each of the 12 defined features [37].

Multiple regression analysis

Multiple regression analyses were performed for each sensory evaluation item, with sensory scores as dependent variables. An overview of the data‐analysis procedure is illustrated in Figure 3. Features extracted from the developed tactile sensor (Figure 3a) and those obtained from commercial sensors (Figure 3b) were used as independent variables. Variable selection began by including all candidate features, followed by model construction using a stepwise variable‐reduction approach. Variables with a variance inflation factor >5 were excluded to prevent multicollinearity. Among the remaining variables, combinations were compared, and the model yielding the highest adjusted R ² was selected as the final model. All analyses were conducted using SPSS software (IBM Corp., Armonk, NY, USA).

FIGURE 3.

Overall data analysis flow. (a) Multiple regression analysis using data acquired from the developed sensor. (b) Multiple regression analysis using data acquired from commercial sensors.

RESULTS

Figure 4 presents representative time‐series graphs of acceleration in the direction of probe movement, normal force, friction force and friction coefficient, measured on the cheek using the developed sensor for two participants (Panel Nos 3 and 19). The data shown correspond to approximately one rotation of the probe during the first 2 s after measurement initiation. Panel Nos 3 and 19 were selected as examples because their sensory evaluation results differed notably.

FIGURE 4.

Representative measurement results by the developed sensor from Panel Nos 3 and 19. From the top, the raw waveform of acceleration in the direction of the probe, the waveform with the gravity acceleration removed from the raw acceleration, the waveforms of normal force and friction force, and the waveform of the friction coefficient are shown. The data for the first 2 s of the measurement is shown.

The mean sensory evaluation scores for these two panels were as follows: for ‘smooth–rough’ (0–5 scale), Panel No. 3 scored 2.00, and Panel No. 19 scored 0.50; for ‘moisturized–dry’, Panel No. 3 scored 3.06, and Panel No. 19 scored 2.08. Thus, Panel No. 3 was rated as less smooth and drier, whereas Panel No. 19 was perceived as smoother and more moisturized. Comparing the acceleration data, Panel No. 3 exhibited larger amplitude vibrations than Panel No. 19. Although the cheek surface is gently curved rather than flat, stable measurements were obtained without significant fluctuations.

Figure 5 presents representative vibration spectra for two participants (Panel Nos. 3 and 19). Both participants exhibited vibration peaks around 50 Hz, corresponding to the probe's surface geometry, with amplitudes gradually attenuating towards higher frequencies. Differences between the two spectra were observed primarily in the 200–1000 Hz frequency range, where the spectrum for Panel No. 3 was generally larger than that for Panel No. 19.

FIGURE 5.

Vibration spectra for two participants (Panel Nos 3 and 19).

Figure 6 shows the loadings of the PC for the vibration features obtained through partial least squares regression for ‘smooth–rough’ (Figure 6a) and ‘moisturized–dry’ (Figure 6b). When ‘smooth–rough’ was used as the dependent variable, two vibration components were extracted. The first PC exhibited neutral (near‐zero) or positive loadings across the entire frequency range, with consistently positive values above 200 Hz. The second PC exhibited a positive loading for amplitudes below 10 Hz and a negative loading for amplitudes between 10 and 200 Hz. For ‘moisturized–dry’, only one vibration component was extracted, with its loading pattern similar to that of the first PC for ‘smooth–rough’—that is, amplitudes above 200 Hz had a positive influence.

FIGURE 6.

Loadings from the partial least square regression: (a) smooth–rough as the dependent variable and the averaged vibration amplitudes (10 Hz intervals) as the independent variables and (b) moisturized–dry as the dependent variable and with the same independent variables.

Table 2 presents the results of the multiple regression analysis for smoothness. Furthermore, each regression equation is shown in Equations (3) and (4). The model based on features from the developed sensor showed a high degree of fit, with an R ² value of 0.708 and an adjusted R ² of 0.676. In comparison, the model combining features from commercial sensors yielded an R ² of 0.520 and an adjusted R ² of 0.360, indicating a lower predictive fit than the developed sensor model.

TABLE 2.

Multiple regression analysis results for the smooth–rough score.

Developed tactile sensor		Combination of commercial sensors
R 2	0.708	R 2	0.524
Adjusted R 2	0.676	Adjusted R 2	0.366

Valuable	Standardized coefficient	p‐value	Valuable	Standardized coefficient	p‐value
PC1 for smooth (vibration component)	0.93	<0.001	Corneometer score (a.u.)	−0.408	0.041
PC2 for smooth (vibration component)	0.395	0.012	Amount of texture	−0.487	0.031
			SEsc	−0.492	0.064
			SEsm	0.584	0.098
			SEw	−1.039	0.017

Examining the variable impacts in the developed sensor model, the standardized coefficient for PC1 for smooth (vibration component) was 0.93 (p < 0.001), and for PC2 for smooth (vibration component), it was 0.395 (p = 0.012), both indicating significant influence. The resulting regression equation is expressed as follows:

$$\begin{array}{c}\begin{array}{c}\text{Smooth}–\text{rough score}=0.93\mathrm{PC}1\text{for smooth}\\ +0.395\mathrm{PC}2\text{for smooth}.\hfill \end{array}\hfill \end{array}$$ (3)

These results suggest that vibration features play an important role in estimating sensory evaluation scores for smoothness and roughness.

In the model using commercial sensors, the standardized coefficients for the Corneometer score (a.u.), amount of texture, SEsc, SEsm and SEw were −0.408 (p = 0.041), −0.487 (p = 0.031), $-$0.492 (p = 0.064), 0.584 (p = 0.098) and −1.039 (p = 0.017), respectively. The regression equation is expressed as follows:

$$\begin{array}{c}\begin{array}{c}\text{Smooth}-\text{rough score}=-0.408\text{Corneometer score}-0.487\text{Amount of texture}\\ -0.492\text{SEsc}+0.584\text{SEsm}-1.039\mathrm{SEw}.\end{array}\hfill \end{array}$$ (4)

These results indicate that variables related to skin capacitance and the surface characteristics of the stratum corneum contribute to the estimation of sensory smooth–rough evaluation scores.

Table 3 presents the results of the multiple regression analysis for ‘moisturized–dry’. Furthermore, each regression equation is shown in Equations (5) and (6). The model using features from the developed sensor showed a good fit, with an R ² value of 0.584 and an adjusted R ² of 0.538. In comparison, the model incorporating commercial sensor features achieved a higher fit, with an R ² of 0.700 and an adjusted R ² of 0.625.

TABLE 3.

Multiple regression analysis results for the moisturized–dry score.

Developed tactile sensor		Combination of commercial sensors
R 2	0.584	R 2	0.700
Adjusted R 2	0.538	Adjusted R 2	0.625

Variable	Standardized coefficient	p‐value	Variable	Standardized coefficient	p‐value
${\mathrm{\mu }}_{\text{init}}$	−0.493	0.012	Skicon200Ex score ($\mathrm{\mu }$ S)	−0.363	0.029
PC 1 for moisture (vibration component)	0.384	0.043	Unevenness of the texture	0.626	<0.001
			SEr	−0.306	0.059
			SEsc	−0.278	0.087

Regarding variable contributions in the developed sensor model, the standardized coefficient for μ_init − 0.493 (p = 0.012), indicating a significant negative association with the moisturized–dry evaluation. The standardized coefficient for PC1 for moisture was 0.384 (p = 0.043), demonstrating a significant positive effect. The regression equation for the score is expressed as follows:

$$\begin{array}{c}\begin{array}{c}\text{Moisturized}–\mathrm{dry}\text{score}=-0.493{\mu }_{\text{init}}\\ +0.384\mathrm{PC}1\text{for moisture}.\hfill \end{array}\hfill \end{array}$$ (5)

Both friction and vibration features were suggested to be important for predicting the sensory evaluation scores of moisturized–dry.

In the model with commercial sensor features, the standardized coefficients for the Skicon‐200EX score (μS), Unevenness of the texture, SEr and SEsc were −0.363 (p = 0.029), 0.626 (p < 0.001), −0.306 (p = 0.059) and −0.278 (p = 0.087), respectively.

This indicates that variables related to skin conductance and the surface morphology of the stratum corneum are important for estimating the moisturized–dry state. The regression equation for the score is as follows:

$$\begin{array}{c}\text{Moisturized}–\mathrm{dry}\text{score}=-0.363\text{Skicon}200\mathrm{EX}\text{score}+0.626\text{Unevenness of texture}\\ -0.306\mathrm{SEr}-0.278\text{SEsc}.\end{array}$$ (6)

No valid regression models were obtained for the attributes ‘elastic–not elastic’, ‘soft–hard (surface)’ and ‘soft–hard (base)’.

DISCUSSION

This study demonstrated that vibration and friction data obtained by rubbing the skin with a probe can be used to estimate sensory evaluation scores for smoothness and moistness, with adjusted R ² values of 0.676 and 0.538, respectively. Skin perceived as smooth in sensory evaluation exhibited smaller amplitudes of high‐frequency vibrations. These findings are consistent with those of Tanaka et al. [27], who reported that vibration features measured by a sensor correlated with physicians' evaluations of skin roughness and hardness, supporting the use of vibration as an indicator of roughness. In the present study, however, skin hardness could not be estimated, whereas Tanaka et al. successfully predicted this attribute. Notably, their study included 14 out of 30 participants with skin disorders such as xerosis (n = 5), atopic dermatitis (n = 7) and psoriasis (n = 2), whereas our study evaluated only healthy skin. This difference in participant backgrounds likely contributed to the discrepancies in results between the two studies.

The results of this study indicate that the perceived moistness of bare skin is associated with both a smaller amplitude of high‐frequency vibrations and a higher friction coefficient. Hendriks and Franklin [19] reported that increased skin moisture under humid weather conditions leads to higher friction coefficients. Egawa et al. [26] found a positive correlation between the friction coefficient measured after lotion application and sensory evaluation scores for moisture obtained the following morning. Similarly, Sakata et al. [38] observed an increase in the friction coefficient following skin moisturization. Veijgen et al. [39] reported that the kinetic friction coefficient rises with increasing skin moisture, peaking at a certain point. Klaassen et al. [40] demonstrated that friction coefficients increase in warm and humid environments by independently varying temperature and humidity. Gerhardt et al. [41] also showed that higher skin moisture content increases the friction coefficient between skin and fabrics. Thus, our finding of high friction when the skin was moisturized aligned with previous reports.

Several studies have reported that the MMD or fluctuations in the coefficient of friction influence the sensory evaluation of moistness [26, 38, 42, 43, 44]. For instance, Kawasegi et al. [42] created plastic samples with microscale textures and reported a correlation between perceived moistness and MMD. Egawa et al. [26] reported a positive correlation between the MMD measured after applying a lotion for 7 h and the sensory evaluation score for moisture obtained the following morning. The MMD represents the degree of fluctuation in friction. Large fluctuations in friction can cause vibrations in both the skin and the probe. However, herein, the MMD did not influence the evaluation of moisture. The reason for this may be that during sensor measurements, no obvious stick–slip phenomena were visually confirmed, making it difficult for the MMD to serve as a criterion for evaluating moisture. Kawazoe and Norihisa [43] suggested that although MMD is effective for friction perception, it may not be effective for assessing moistness, depending on the definition of the sensory evaluation attributes.

One possible explanation for the correlation between vibrations and moisture evaluation is that fluctuations in friction generate skin vibrations. However, to our knowledge, no previous studies have specifically examined the relationship between vibrations and the perceived moisture of bare skin, highlighting the importance of evaluating both friction and vibration.

Sensory evaluation items related to softness and elasticity could not be estimated using vibration and friction data. Arakawa et al. [45] suggested that hardness assessments based on pressing actions differ from those based on rubbing actions. In this study, softness and elasticity were evaluated through pressing, which likely limited the relevance of friction and vibration information. Incorporating both rubbing‐ and pressing‐type sensors, as used in this study, may enable the estimation of a broader range of sensory evaluation attributes.

In this study, the rubbing action of the assessors was not strictly controlled for constant speed, reflecting the natural variability in human exploratory movements when rubbing the skin. Additionally, the developed sensor used a stainless‐steel probe with a fingerprint‐mimicking surface, whereas previous studies often employed soft, finger‐like materials as contact elements [46, 47, 48, 49]. Optimizing both the rubbing action and the material and surface structure of the probe remains an important direction for future research.

The results of sensory evaluation can be estimated using combinations of commercial sensors; however, the direct causal relationships between independent and dependent variables often remain unclear in such methods, making them less suitable for understanding the principles underlying a phenomenon. Understanding these phenomena would allow for targeted adjustments to sensory evaluation items. Furthermore, translating sensory evaluation terms can be challenging when developing products for consumers across different linguistic regions. In such cases, the ability to interpret the meaning of sensory evaluation terms through physical quantities, as demonstrated in this study, would be beneficial.

CONCLUSION

This study investigated the relationship between vibrations and friction information obtained via rubbing of the human skin with a newly developed tactile sensor and the sensory evaluation scores for smoothness and moistness. The results indicate that vibrations and friction are important indicators for the evaluation of the smoothness and moisture content of the bare skin. Specifically, a smaller high‐frequency vibration amplitude was associated with smoother evaluations, whereas higher moisture content was associated with lower high‐frequency vibrations and higher friction.

These findings enable the physical interpretation of sensory evaluation terms used in skin care product development, thereby allowing clearer design guidelines. In particular, when developing products for consumers in different linguistic regions, translating sensory evaluation terms can be challenging. Our results are expected to contribute to addressing such issues.

Although the sensory evaluation items for softness and elasticity could not be estimated using vibration and friction information, this was believed to be due to differences in the evaluation methods. Future advances in sensor development, such as the inclusion of pressing actions and probe optimization, are expected to enable the estimation of a wider variety of sensory evaluation items.

CONFLICT OF INTEREST STATEMENT

One of the professional assessors is one of the authors. However, she was not informed of the details of the test procedure. Furthermore, a blind design was adopted in the experiments.

DATA AVAILABILITY STATEMENT

Research data are not shared.