Introduction

Repetitive manual tasks, particularly those involving high force and high repetition (HIF.HIR), significantly increase the risk of debilitating hand and wrist cumulative trauma disorders (CTDs)1. Similarly, moderate-intensity tasks in occupational settings like assembly lines and keyboard use are a primary cause of work-related musculoskeletal disorders (WMSDs) of the upper extremity, including carpal tunnel syndrome2. These activities can lead to chronic muscle fatigue and injury through cumulative effects over time3. Given that muscle fatigue is a key precursor to muscle damage, its early and accurate monitoring is essential for WMSD prevention4.

Traditional fatigue assessment methods, including surface electromyography (sEMG), maximum voluntary contraction (MVC) measurements, and subjective scales (e.g., Borg scale), face significant limitations in dynamic occupational contexts5,6. For instance, sEMG signals are highly susceptible to skin impedance, perspiration, and electromagnetic interference during dynamic movements7,8. MVC measurements disrupt ongoing tasks, making them unsuitable for continuous monitoring, while subjective scales can be influenced by individual psychological factors. In contrast, mechanomyography (MMG) offers a compelling non-invasive technique that monitors muscle function by recording transverse mechanical vibrations9,10. MMG signals are inherently less affected by skin surface changes and external electrical noise compared to sEMG8, making them more robust for dynamic tasks11,12. Recent studies further highlight MMG sensors’ superior stability and fatigue sensitivity during fatigue testing13.

The quality of MMG signals, however, depends on sensor characteristics like weight and size, particularly in dynamic motion where heavy or large sensors can introduce artifacts14. Common MMG acquisition methods rely on accelerometers (prone to motion artifacts15) or microphones (with debatable signal-to-noise performance in dynamic contractions16). This study employs polyvinylidene fluoride (PVDF)-based piezoelectric sensors, which offer distinct advantages for dynamic, high-frequency tasks: their flexibility enables thin-film, lightweight designs that conform to skin contours, enhancing comfort and stable mechanical coupling17; miniaturization minimizes interference with natural movements18; and they are unaffected by skin impedance changes19. This strategic choice enhances MMG reliability for prolonged dynamic monitoring, representing a methodological innovation.

While existing research confirms MMG’s correlation with muscle fatigue, these studies have predominantly focused on static or extreme (high or low) load conditions20,21,22,23,24,25,26,27. A significant research gap persists in understanding MMG signal behavior during dynamic gripping tasks at moderate loads (30% MVC) and high frequencies (1 Hz), which are highly representative of industrial production3.In particular, the Flexor Digitorum Superficialis (FDS)-as the core muscles of fine hand manipulation28, whose tendons are located superficially and can be easily palpated in the distal forearm-is suitable for acquiring MMG signals for research29,30, remains underexplored with MMG in these specific dynamic, moderate-intensity contexts.

Beyond task characteristics, the effectiveness of intermittent rest strategies (e.g., 30-second micro-rests) in alleviating fatigue during such tasks—and how MMG signals reflect fatigue progression or rest-induced recovery in the FDS—remains unclear. Previous research has shown some benefits of micro-breaks: Balci and Aghazadeh31 found that 30-second micro-breaks every 15 min reduced upper extremity discomfort compared to longer rest intervals, and Sundelin et al.32 demonstrated that pausing activity relieved shoulder and neck muscle fatigue. However, the applicability of these findings to the FDS with its distinct functional and loading characteristics is unclear. The FDS is a key muscle for fine motor control, often operating under low-force, high-repetition patterns distinct from the global muscles previously studied. Furthermore, while EMG studies like Iridiastadi and Nussbaum33 have explored how contraction level, duty cycle, and load cycle affect endurance and fatigue, evidence on traditional EMG metrics’ sensitivity to work/rest patterns is inconsistent. Christensen et al.34 found no differences in EMG spectral metrics between forearm tasks with distinct work/rest patterns in moderate-intensity work. In contrast, Marina et al.35 showed EMG sensitivity depended on recovery duration, with changes in short-rest tasks but not long-rest ones. This inconsistency highlights the need for more sensitive methods like MMG to fully understand such fatigue. Critically, current literature lacks investigations into how MMG signals reflect fatigue progression or respond to 30-second micro-rests in the FDS during high-frequency (1 Hz), moderate-intensity (30% MVC) dynamic grasping. This represents a significant gap, especially given that such work patterns are common in industries prone to cumulative trauma disorders.

Equally important is understanding how individual factors—gender and hand dominance—influence fatigue responses in these tasks. These factors shape neuromuscular control and fatigue resistance through distinct mechanisms, yet their impact on the FDS remains underexamined. For gender differences, Hicks36 and Hunter37 demonstrated females exhibit greater fatigue resistance in submaximal contractions, linked to higher proportions of fatigue-resistant type I fibers and enhanced perfusion. However, Hunter38 noted dynamic fatigability differences depend on contraction velocity and muscle group, while Hill39 found faster recovery in females post-fatigue. Clarifying how these sex-specific adaptations affect the FDS is critical, as its functional role in grasping differs from globally studied muscles. For hand dominance, Adam40 showed dominant hands have lower motor unit recruitment thresholds and reduced force variability due to neural adaptations, while Madarshahian and Latash28 found non-dominant hands use stronger force-stabilizing synergies. How these differences translate to FDS fatigue in dynamic tasks, as measured by MMG, is largely unknown—despite their relevance to occupations prone to CTDs, where dominant hands often bear greater workloads.

To address these research gaps, this study had the following primary objectives: (1) To characterize the dynamic evolution of multidimensional MMG signal features (i.e., time-domain, frequency-domain, and nonlinear) in the FDS muscle during a fatiguing, moderate-intensity, high-frequency dynamic grasping task. (2) To determine if periodic 30-second micro-rests are sufficient for preventing fatigue accumulation, as measured by MMG features, MVC, and perceived exertion. (3) To investigate the influence of individual factors, specifically gender and hand dominance, on MMG-based fatigue responses and subjective reports of exertion. By achieving these objectives, this research aims to provide a scientific basis for developing more effective ergonomic interventions and personalized, real-time fatigue monitoring systems for high-frequency manual occupations.

Experiments and methods

Subjects

Twenty-four healthy volunteers were recruited for this study. After full explanation of the experimental purpose and protocol, a total of 12 males (height: 179.2 ± 5.5 cm, weight: 82.8 ± 8.3 kg, age: 24.3 ± 1.7 years) and 12 females (height: 163.3 ± 3.7 cm, weight: 53.2 ± 3.3 kg, age: 22.2 ± 0.4 years) met the inclusion criteria and were enrolled. All subjects had no history of upper limb musculoskeletal disorders and were right-handed. Twelve subjects were randomly assigned to have their dominant hand (right hand) tested, and the remaining twelve to have their non-dominant hand (left hand) tested, with gender and hand side balanced across groups.

Subjects were informed of potential risks and benefits and provided written informed consent prior to participation. The study adhered to the ethical principles outlined in the Helsinki Declaration adopted in June 1964 (Helsinki, Finland) and revised in October 2000 (Edinburgh, Scotland). The experimental procedures were approved by the Ethics Committee of Nanjing Qixia District Hospital (Ethics Review Number: 2024-QX029).

Experimental protocol

Subjects were seated comfortably on a standardized testing chair, maintaining an upright torso posture. For the tested limb: the upper arm hung naturally parallel to the torso, with the elbow joint flexed at 90°; the forearm was positioned in semi-pronation (thumb oriented upward); the wrist remained neutrally suspended and immobile during movement. The non-tested limb was relaxed at the subject’s side.

First, a handheld specialized grip dynamometer was used to measure three maximal voluntary contractions (MVCs) in the seated position. Each contraction lasted 3–5 s, with 5-minute inter-trial rests to prevent fatigue. The highest value from these three trials was designated as the baseline MVC. 30% of this MVC (30% MVC) was calculated as the target force for the experimental task, and the dynamometer was calibrated to this target value.

Prior to formal testing, a force sensor was attached to the dynamometer. Subjects completed five practice grip trials at 30% MVC to ensure task comprehension and consistent execution. Simultaneous force sensor data were recorded to verify the dynamometer’s calibration accuracy. During this familiarization phase, the region of maximal forearm muscle excursion (typically at the mid-forearm, ~ 1/3 forearm length from the olecranon process) was identified. The corresponding skin area was cleansed, and a PVDF sensor was securely affixed over the prominence of the FDS muscle. Subsequent references to this muscle use the abbreviation FDS for brevity.

The experimental procedure is shown schematically and the evaluation criteria for subjective muscle fatigue are shown in Fig. 1. During formal testing, subjects performed cyclic grips at 30% MVC intensity, paced by a 1 Hz metronome (one grip per second). Subjective fatigue was assessed every 30 s using the Borg CR-10 scale41. Each testing block lasted 5 min, followed by a 30-second rest. The total duration of ~ 16 min (three consecutive blocks) was determined based on pilot tests, where participants performed the task until voluntary grip failure. Pilot data revealed that within 16 min, all participants could complete the task without excessive physical strain, while still reaching measurable muscle fatigue. Beyond this duration, some participants began to fail to fully close the dynamometer for three consecutive trials. Thus, 16 min was identified as the optimal duration: it ensures all participants can complete the experiment without experiencing such failure or excessive fatigue, while still inducing sufficient fatigue for reliable measurement. In line with this, the formal experiment included a predefined termination criterion as a safeguard: the task would end early if subjects failed to fully close the dynamometer for three consecutive trials, further ensuring no participant was forced to continue beyond their capacity to complete the protocol.

At the end of the trial, subjects tested the MVC in the original position and recorded it.

Fig. 1
figure 1

Schematic illustration of the experimental protocol and criteria for subjective muscle fatigue assessment.

Full size image

Data collection and processing

MMG signals were collected by a flexible PVDF piezoelectric film sensor (SDT1-028 K, size 28.6 mm×11.2 mm×0.13 mm, sensitivity 20 mV/g). In this experiment, instead of using straps for fixation, we chose to use medical double-sided adhesive tape (3 M TP77) and medical tape to affix to the test muscle, to reduce the influence of external pressure on the test, and to ensure that the sensor and the muscle can be in good contact with each other.

The sensor was amplified and read by a VK102 charge amplifier (input range 0-± 5,000 Pc, sensitivity Ac = 100pC/100m V) and an Arduino-compatible board (e.g., DFRduino UNO R3) for data acquisition. MMG signals were then transferred to a host computer and synchronously stored using custom scripts in Python (via PyCharm) and the Arduino IDE environment, at a sampling rate of 800 Hz. The sensor and acquisition system are shown in Fig. 2. and acquisition system is shown in Fig. 2. Subjective fatigue perception was assessed using the Borg CR-10 scale, with ratings recorded manually every 30 s.

Fig. 2
figure 2

Sensors and Acquisition systems.

Full size image

The raw MMG signals were processed using a 5–100 Hz fourth-order Butterworth bandpass filter followed by Kalman filtering for noise reduction. Signal segmentation was performed in 30-second windows, aligned with the 30-second interval of Borg fatigue ratings and validated to yield comparable results with 10-second or single-movement segmentation protocols.

MMG signal features across multiple domains reveal muscle activity physiology from distinct mechanistic perspectives42. Time-domain metrics—including RMS, integrated MMG (IMMG), contraction rise time, and relaxation decay slope—quantify amplitude, energy accumulation, and contraction-relaxation dynamics. Frequency-domain analyses, such as MPF, median frequency (MDF), and zero-crossing rate, characterize myofiber recruitment strategies via spectral component distribution; zero-crossing rate specifically reflects signal periodicity and frequency modulation, widely used in fatigue-associated spectral drift research. Elevated high-frequency energy ratios (> 40 Hz) indicate increased fast-twitch fiber engagement. Nonlinear indices—sample entropy (SampEn), approximate entropy (ApEn), and largest Lyapunov exponent (LLE)—capture signal complexity and chaotic dynamics43. Reductions in entropy or shifts in LLE may signify enhanced motor unit recruitment synchronization, potentially linked to fatigue-induced neural control adaptations44.

Given the limited understanding of fatigue biomarkers in FDS, this study systematically extracted multi-domain features from MMG signals, including time-domain metrics such as RMS, contraction rise time, relaxation decay slope, IMMG, and zero-crossing rate; frequency-domain parameters like MPF, MDF, high-frequency energy ratio (> 40 Hz), and instantaneous frequency; nonlinear-domain indices such as SampEn, ApEn, and LLE; and additional metrics including waveform factor (WF), compensatory tremor index (CTI), and instantaneous energy fluctuation rate. These features were analyzed to identify sensitive indicators capable of characterizing muscle fatigue signal changes in the FDS by integrating temporal dynamics, spectral adaptations, and nonlinear complexity.

After obtaining the feature values, Z-Score standardization was applied to each subject’s MMG features to eliminate inter-individual baseline variability and enable cross-sample comparisons. The MVC decline rate was calculated to quantify reductions in maximal voluntary contraction capacity, defined as:

$$\:\frac{{MVC}_{pre}-{MVC}_{post}}{{MVC}_{pre}}\times\:100\%$$

Statistical analysis of data

Figure 3 illustrates the complete analytical pipeline, from data collection to statistical testing, for integrating MMG signals, MVC measurements, and Borg ratings. To first confirm the induction of fatigue, paired t-tests were performed to compare MVC values and Borg subjective fatigue ratings at task onset and completion, assessing objective strength decrements and shifts in subjective perception induced by the task. Cohen’s d effect sizes were calculated to quantify the practical significance of these pre-to-post differences.

Fig. 3
figure 3

Analytical workflow for MMG, MVC, and Borg Data.

Full size image

Beyond initial fatigue confirmation, the workflow further integrated advanced statistical analyses to characterize fatigue progression, relationships between measures, and subgroup differences. To analyze the effects of fatigue progression over time and across trial sets, a series of Two-Way Repeated-Measures ANOVAs were conducted for each dependent variable (i.e., all extracted MMG features, MVC, and Borg ratings). For each ANOVA, the within-subjects factors were Time (10 levels, representing 30-second intervals within each trial) and Trial (3 levels, representing the three 5-minute work blocks), enabling assessment of main effects of Time and Trial, as well as their critical interaction, on each outcome. Results were reported as F-values, p-values, and partial η², with significant main effects further probed using Tukey HSD post-hoc tests to identify specific time-point or trial-based differences.

Spearman’s rank correlation analyses were then used to explore associations between total changes in MMG features, cumulative Borg score increases, and percentage MVC declines, with nonparametric ρ coefficients quantifying the strength and direction of linear relationships between subjective fatigue perception and objective signal alterations. For gender-difference analyses, percentage MVC declines were first computed; MVC decline magnitude and Borg ratings were tested for normality via Shapiro-Wilk tests, with normally distributed data compared using independent samples t-tests (Welch’s correction) and non-normal data analyzed via Mann-Whitney U tests. A mixed linear model (gender as a between-subjects factor, trial as a within-subjects factor, and subjects as random effects) further characterized gender-specific fatigue responses by examining main effects and interactions.

Given the 1:1 ratio of dominant (right) vs. non-dominant (left) hand testing, additional analyses investigated hand-side fatigue differences: Mann-Whitney U tests compared MMG features between hands, followed by mixed ANOVA to assess hand-side × trial × time interactions. All statistical tests utilized a significance threshold of \(\:p\:<\:0.05\), with Tukey HSD corrections applied for multiple comparisons to control Type I error rates, ensuring result validity and reproducibility.

Results

MVC and Borg fatigue confirmation

Figures 4 and 5 depict pre- to post-experiment mean changes in MVC and Borg ratings across 24 subjects. Subjects’ MVC decreased by 19.7%, from 36.38 ± 9.76 kg to 29.21 ± 10.47 kg (t = 12.457, p < 0.0001, Cohen’s d = 0.734). Borg ratings increased ~ 15-fold, from 0.35 ± 0.15 to 5.46 ± 1.22 (t = -20.722, p < 0.0001, Cohen’s d = -33.324). All participants completed three trial sets without early termination.

Repeated-measures ANOVA of Borg ratings revealed significant main effects of Time (\(\:F\:=\:33.24,\:p\:<\:0.0001\)), Trial (\(\:F\:=\:60.23,\:p\:<\:0.0001\)), and a Time×Trial interaction (\(\:F\:=\:63.81,\:p\:<\:0.0001\)), indicating cumulative subjective fatigue with increasing trial duration and set number.

Fig. 4
figure 4

Before and after MVC tasks.

Full size image
Fig. 5
figure 5

Borg score start vs. end.

Full size image

Impact of time and trial sets on MMG features and Borg ratings

Two-Way Repeated-Measures ANOVA (Two-Way RM ANOVA) on MMG features and Borg ratings identified significant effects as shown in Table 1.

Table 1 Significant Two-Way RM ANOVA results for Features.
Full size table

MPF, high-frequency energy ratio, IF, and LLE exhibited significant trial set main effects (\(\:p\:<\:0.05\)). Tukey HSD post-hoc tests revealed that mean MPF (\(\:p\:=\:0.024\)), high-frequency energy ratio (\(\:p\:=\:0.040\)), and IF (p = 0.049) in trial set 3 were significantly lower than those in trial set 1.

The within-time main effect of IMMG was significant, with post-hoc tests showing that IMMG values at late time points (e.g., 240 s, 270 s, 300 s) were significantly lower than those in early task phases (e.g., 30 s, 60 s).

WF and SampEn displayed significant Trial Set×Time interaction effects. Main effect analyses indicated divergent WF temporal patterns across trial sets: WF values in trial set 3 were significantly higher than those in trial sets 1 and 2 at late time points (e.g., 270 s). Temporal changes in entropy were significant in trial sets 2 (p < 0.001) and 3 (p < 0.01) but not in trial set 1 (\(\:p\:>\:0.05\)). Between-set differences emerged at late experimental stages (240s, 270 s; \(\:p\:<\:0.05\)).

Other metrics from the methodology—including time-domain (relaxation decay slope, RMS), frequency-domain (MDF), nonlinear (ApEn) metrics, and CTI (muscle co-activation time index)—showed no significant Time, Trial Set, or interaction effects: relaxation decay slope (Trial Set: \(\:F=1.67,\:p=0.20\); Time:\(\:\:F=1.19,\:p=0.32\); Interaction: \(\:F=0.84,\:p=0.55\)), RMS (Trial Set: \(\:F=1.43,\:p=0.25\); Time: \(\:F=0.98,\:p=0.45\); Interaction:\(\:\:F=0.72,\:p=0.68\)), MDF (Trial Set:\(\:\:F=2.11,\:p=0.13\); Time: \(\:F=1.35,\:p=0.26\); Interaction: \(\:F=0.91,\:p=0.51\)), ApEn (Trial Set: \(\:F=1.58,\:p=0.21\); Time: \(\:F=1.05,\:p=0.40\); Interaction: \(\:F=0.65,\:p=0.76\)), and CTI (Trial Set: \(\:F=2.05,\:p=0.14\); Time: \(\:F=1.53,\:p=0.20\); Interaction:\(\:\:F=0.76,\:p=0.61\)) all exhibited no systematic changes across conditions with minimal mean fluctuations.

Correlation analysis

A significant positive correlation between subjective fatigue perception and objective functional decline was revealed through Spearman’s correlation analysis. As illustrated in Fig. 6, a significant positive association was observed between the total increase in Borg ratings and the percentage decline in MVC (ρ = 0.575, p = 0.003).

Fig. 6
figure 6

Borg score increase vs. percentage MVC decline.

Full size image

Regarding the MMG features, a significant positive correlation was observed between the total change in contraction rise time and both the percentage MVC decline (ρ = 0.512, p = 0.011) and the total Borg score increase (\(\:\rho\:=0.429,\:p=0.036\)), as illustrated in Fig. 7.

Fig. 7
figure 7

Total change in contraction rise time vs. percentage MVC decline and total Borg score increase.

Full size image

In contrast, total changes in other MMG features across domains showed no significant correlations with Borg ratings or MVC decline (all \(\:p\:>\:0.05\)): time-domain metrics including RMS (\(\:\rho\:=0.19,\:p=0.37\)), integrated MMG (IMMG; \(\:\rho\:=0.23,\:p=0.30\)), and relaxation decay slope (\(\:\rho\:=0.15,\:p=0.48\)); frequency-domain metrics such as mean power frequency (MPF; \(\:\rho\:=0.12,\:p=0.56\)), median frequency (MDF; \(\:\rho\:=0.21,\:p=0.33\)), zero-crossing rate (ZCR; \(\:\rho\:=0.17,\:p=0.42\)), and high-frequency energy ratio (> 40 Hz; \(\:\rho\:=0.20,\:p=0.3\)5); and nonlinear indices including sample entropy (SampEn; \(\:\rho\:=0.18,\:p=0.39\)), approximate entropy (ApEn;\(\:\:\rho\:=0.14,\:p=0.50\)), and largest Lyapunov exponent (LLE; \(\:\rho\:=0.22,\:p=0.31\)).

Results of gender difference analysis

Gender-specific analyses indicated that the percentage decline in MVC was significantly more pronounced in females (28.06%) than in males (14.77%, \(\:p<0.0001\)). Correspondingly, females reported significantly elevated Borg subjective fatigue ratings across all trial sets compared to males (Trial 1: 1.36 vs. 0.61; Trial 2: 3.88 vs. 1.97; Trial 3: 5.13 vs. 3.80; all p < 0.0001).

Fig. 8
figure 8

Gender analysis of sample entropy.

Full size image

A significant main effect of gender on sample entropy (p = 0.0271) was identified through linear mixed-effects modeling, as illustrated in Fig. 8.

In contrast, no statistically significant gender main effects were observed for other MMG features across domains: time-domain features including contraction rise time (\(\:p=0.3381\)), relaxation decay slope (\(\:p=0.8655\)), integrated MMG (IMMG; \(\:p=0.4779\)), and RMS (\(\:p=0.5831\)); frequency-domain features such as median frequency (MDF;\(\:\:p=0.1764\)), mean power frequency (MPF;\(\:\:p=0.0691\)), high-frequency energy ratio (> 40 Hz; \(\:p=0.1464\)), zero-crossing rate (ZCR; \(\:p=0.2761\)), and instantaneous frequency (IF; \(\:p=0.2723\)); and nonlinear and other features including largest Lyapunov exponent (LLE;\(\:\:p=0.4269\)), waveform factor (WF; \(\:p=0.0914\)), compensation tremor index (CTI;\(\:\:p=0.9582\)), and instantaneous energy mutation rate (\(\:p=0.1307\)), all with \(\:p\:>0.05\) indicating no significant gender-related differences in their overall magnitudes.

Results of Left-Right hand difference analysis

Differences between dominant and non-dominant hands were also examined. As presented in Table 2, Mann-Whitney U test results indicated significant overall mean differences (p < 0.05) between the left and right hands across 10 metrics. These included time-domain features (IMMG, ZCR, WF), frequency-domain features (MDF, MPF, high-frequency energy ratio, IF), and nonlinear features (SampEn, LLE).

Table 2 Mann-Whitney U test results for Left-Right hand differences in MMG Metrics.
Full size table

All other features showed no significant differences, including contraction rise time (\(\:p=0.3742\)), relaxation decay slope (\(\:p=0.0732\)), RMS (\(\:p=0.7570\)), compensation tremor index (CTI;\(\:\:p=0.0624\)), and instantaneous energy mutation rate (\(\:p=0.0812\)).

Further analysis using linear mixed-effects modeling, detailed in Table 3, identified significant Tested Hand×Trial interactions \(\:(p<0.05\)) for seven features. Specifically, frequency-domain metrics (MDF, MPF) exhibited amplified left-right differences during Trial 2. Time-domain and energy metrics (ZCR, IMMG, RMS) showed divergent hand-side trends across trials, while nonlinear features (LLE, IF) presented significant hand-side differences in late fatigue stages.

Table 3 Significant Tested_Hand×Trial interactions and trial main effects (p < 0.05).
Full size table

Independent samples t-tests, depicted in Fig. 9, revealed significant differences in Borg rating slopes between hands (left: 2.01 ± 0.53 vs. right: 1.47 ± 0.35; t = 2.948, p = 0.0082). This finding suggested a faster progression of subjective fatigue perception in the left hand across trials.

Fig. 9
figure 9

Comparison of Borg Rating Slopes between Left and Right Hands (Box Plots).

Full size image

Discussion

The evolution of MMG signals during muscle fatigue, the intervention effects of periodic micro-breaks, and differences in fatigue responses between genders and dominant/non-dominant hands were investigated in this study. The research focused on the FDS as the target muscle during a moderate-intensity (30% MVC), high-frequency (1 Hz) dynamic gripping task.

Successful induction of fatigue and dynamic evolution of MMG signals

Following the completion of three sets of repetitive gripping tasks, a decrease of approximately 19.7% in subjects’ MVC was observed, alongside an increase in subjective fatigue perception of around 15-fold. Both of these changes were statistically significant (\(\:p<0.0001\)). This decline in MVC and increase in perceived exertion confirm the successful induction of fatigue within the FDS under the experimental conditions. Further validation of the consistency between subjective and objective fatigue metrics was provided by a significant positive Spearman correlation (\(\:\rho\:=0.575,\:p=0.003\)) between the total increase in Borg ratings and the percentage MVC decline. This finding aligns with previous research, such as that by Enoka and Duchateau, and reaffirms the validity of the Borg scale for assessing muscle fatigue in the context of these specific task conditions2.

Analysis of MMG features revealed that the time-domain metric IMMG significantly decreased during late task phases. This reduction in IMMG is interpreted as reflecting a diminished mechanical output capacity of the muscle as fatigue progressed. Frequency-domain features, including MPF, high-frequency energy ratio, IF, and ZCR, exhibited significant between-trial set effects. The observed decline in MPF, a widely recognized marker of fatigue, is hypothesized to stem from reduced motor unit discharge rates and a slowing of muscle fiber conduction velocity due to accumulated fatigue. These physiological changes would consequently alter the frequency characteristics of motor unit activity7. The presence of a between-trial set effect for the nonlinear feature LLE, coupled with significant Trial×Time interactions for SampEn and WF, suggests that fatigue involves adaptive neuromuscular control adjustments rather than being solely characterized by a unidimensional decline in strength.

Notably, changes in contraction rise time were found to be significantly positively correlated with both the percentage decline in MVC and the increases in Borg scores. This suggests that the deceleration of muscle contraction velocity can serve as a functional indicator of fatigue severity for this specific dynamic gripping task, a finding consistent with Bigland-Ritchie et al.’s research on prolonged muscle reaction times during fatigue45.

In contrast, other MMG metrics did not exhibit significant correlations with overall fatigue severity or showed limited dynamic changes across trials. Time-domain metrics such as RMS \(\:(\rho\:=0.19,\:p=0.37)\) and relaxation decay slope (\(\:\rho\:=0.15,\:p=0.48\)) showed no significant associations with MVC decline or Borg increases. This may be attributed to the stable submaximal force output (30% MVC) of the task, which minimized large amplitude fluctuations in RMS, while relaxation decay slope—sensitive to passive muscle stiffness—may not have captured active fatigue-related changes in dynamic contractions. Frequency-domain metrics like MDF (\(\:\rho\:=0.21,\:p=0.33\)) also lacked strong correlations, potentially due to the FDS’s reliance on fine motor control rather than global shifts in motor unit firing frequency. Nonlinear features including ApEn (\(\:\rho\:=0.14,\:p=0.50\)) and CTI (\(\:p>0.05\) for time/trial effects) showed no significant associations, likely because the task’s repetitive nature constrained variability in neural control complexity (for ApEn) and muscle co-activation timing (for CTI) despite accumulating fatigue.

Other MMG metrics also exhibited significant variations across trial sets (as detailed in Sect. “Impact of time and trial sets on MMG features and borg ratings”, Table 1), thus indicating their sensitivity to the progression of the task. However, direct linear correlations between the magnitude of their total change and the overall fatigue severity (as measured by total MVC decline or Borg increase) were not as prominent as that of contraction rise time under the current experimental conditions. It is acknowledged that some prior studies have noted that MPF, for instance, may remain unchanged during low-load fatigue despite marked subjective fatigue20, further supporting the task and metric specificity of fatigue-related MMG changes. These findings collectively highlight that MMG metrics vary in their sensitivity to fatigue in dynamic fine-motor tasks, with contraction rise time emerging as a particularly robust marker for functional decline.

Evaluation of periodic Micro-Breaks on fatigue accumulation

Despite the implementation of a periodic micro-break pattern—consisting of 5 min of work followed by 30 s of rest—significant cumulative effects across trial sets were observed in both Borg ratings and multiple MMG features (e.g., MPF, high-frequency energy ratio, LLE). This finding suggests that a 10% rest ratio was insufficient to prevent continuous fatigue accumulation during the high-frequency (1 Hz) dynamic repetitive gripping tasks. Both the muscle’s mechanical output capacity (as indicated by declining IMMG in late phases) and neuromuscular control efficiency (evidenced by Trial×Time interactions in SampEn and WF) appeared to gradually decline over the course of the trials. This observation aligns with research by Balci and Aghazadeh on high-frequency hand manipulation tasks, where EMG monitoring of wrist flexors indicated that even with frequent short breaks, physiological load metrics and subjective discomfort ratings significantly increased with work duration31. By analyzing objective MMG signals, the current study provides additional evidence regarding the limited recovery efficacy of traditional micro-break strategies under high-frequency dynamic loads, thereby revealing commonalities across high-frequency tasks involving small muscle groups.

The limited effectiveness of these micro-breaks may be attributed to the sustained high metabolic activation of the FDS during 1 Hz contractions. Small muscle groups like the FDS have higher per-unit-volume metabolic demands than larger muscles (e.g., trapezius), yet their smaller cross-sectional area may restrict capillary blood flow, hindering efficient clearance of metabolites (e.g., lactate) during 30-second passive rest periods29. This could result in a net accumulation of fatigue-inducing substances, even with periodic rests. While MMG cannot directly measure lactate, changes in muscle mechanical properties (e.g., declining IMMG) may indirectly reflect the impact of such metabolic stress.

Notably, certain MMG metrics showed no significant recovery trends during rest intervals, further supporting the inadequacy of short breaks. Time-domain metrics like RMS (\(\:p=0.45\) for Time effect) exhibited minimal fluctuations across rest periods, likely because residual muscle tension persisted in the FDS—critical for maintaining grip stability—preventing full amplitude normalization even during rest. Similarly, the compensation tremor index (CTI) showed no significant Time or Trial Set effects (\(\:p>0.05\)), indicating that 30-second rests failed to reset the timing of muscle co-activation patterns. This aligns with findings by Madarshahian and Latash28, who demonstrated that the FDS relies on stable motor unit synergies to maintain fine motor control during high-frequency tasks; such synergies may persist beyond brief rest periods, limiting CTI’s responsiveness to short breaks.

Additionally, fixed-proportion passive rest periods may not align with the neural control characteristics of small muscles. The FDS’s reliance on precise, repetitive motor unit recruitment patterns may require more active recovery strategies to disrupt sustained neural adaptations (e.g., motor unit synchronization), which short passive rests do not address. This is reflected in the lack of recovery in nonlinear features like ApEn (p = 0.40 for Time effect), as neural control complexity remained suppressed despite rest intervals.

These findings suggest that rest strategies for high-frequency hand tasks should be tailored to muscle group-specific characteristics. Adjustments might involve extending rest duration to enhance metabolic clearance, incorporating active recovery (e.g., light stretching) to reset neuromuscular synergies, or implementing task-specific rest ratios that account for the FDS’s high metabolic demands. Such modifications could better address fatigue recovery needs, enhance work efficiency, and reduce risks of work-related musculoskeletal disorders (WMSDs) in occupations involving repetitive fine motor tasks.

Differences in fatigue responses by gender and Dominant/Non-Dominant hands

In terms of gender differences, female subjects exhibited more pronounced subjective and objective fatigue, with significantly greater MVC decline and higher Borg ratings than males, aligning with task-specific gender effects on fatigue37,38. While females show stronger fatigue resistance in low-to-moderate isometric contractions46, the FDS—a small muscle for fine motor control—differs: Tantipoon29 noted lower resting stiffness in female FDS but similar late-fatigue stiffness increases, potentially reflecting greater metabolic sensitivity accelerating fatigue.

Gender differences in MMG features were primarily observed in SampEn: males showed SampEn decreases consistent with simplified neural control (“muscular wisdom hypothesis”44, while females exhibited late-fatigue SampEn rebound, indicating dynamic motor unit recruitment to resist force decline47, possibly explaining their greater MVC loss. Notably, no significant gender effects were found in time-frequency features (e.g., MPF). This aligns with Hill et al.39 but contrasts with large muscle studies27, as FDS—reliant on fine control—may exhibit gender differences in neural flexibility (via SampEn) rather than global frequency shifts, given its slow-twitch dominance and stable amplitude output45. Non-significant CTI (\(\:p=0.9582\)) further suggests gender similarities in basic muscle co-activation timing, with differences emerging in dynamic tasks.

For dominant/non-dominant hands, although contraction rise time (\(\:p=0.3742\)) and RMS (\(\:p=0.7570\)) showed no differences, non-dominant hands exhibited faster increases in Borg rating slopes (left: 2.01±0.53 vs. right: 1.47±0.35,\(\:\:p=0.0082\)), with 10 MMG metrics demonstrating baseline differences and 7 exhibiting hand×trial interactions. Consistent with Madarshahian and Latash28, dominant hands may maintain coordination through stable motor unit synergies: their preserved zero-crossing rate (ZCR) stability in late trials and higher sample entropy (SampEn) suggest optimized vibration encoding, which delays subjective fatigue perception40. The non-significant difference in MVC decline rate alongside faster perceived fatigue indicates that non-dominant hands struggle with “signal stability maintenance” (e.g., differences in largest Lyapunov exponent LLE in Trial 3) rather than deficits in force output, highlighting the need to tailor ergonomic interventions to hand dominance, leveraging the sensitivity of PVDF sensors to such nuanced differences.

Limitations

This study is subject to several limitations that may influence the generalizability of its findings. The relatively small sample size (n = 24) and the exclusive inclusion of right-handed participants inherently restrict the direct applicability of the results to left-handed individuals or more diverse populations. Furthermore, the experiment was conducted within controlled laboratory settings, which, by its nature, cannot fully replicate the inherent complexity and variability of real-world work scenarios. Future research endeavors should aim to validate the applicability of MMG in larger, more diverse cohorts. Additionally, incorporating complementary physiological measurement methods (e.g., electromyography (EMG) or near-infrared spectroscopy)48 in future studies could provide a more comprehensive assessment of fatigue in authentic work environments.

Conclusion

This research utilizes PVDF sensors to collect MMG signals from the FDS during simulated moderate-load repetitive dynamic gripping tasks. By analyzing multi-dimensional features including time-domain (IMMG), frequency-domain (MPF, IF, ZCR), and nonlinear (SampEn, LLE, WF) characteristics, the research explores the muscle’s immediate, cumulative, and complex fatigue responses during such tasks. Findings validate that MMG signals sensitively capture the dynamic processes of muscle fatigue, with short rest periods (10% duty cycle) demonstrating limited efficacy in alleviating fatigue. Significant differences in fatigue progression and physiological indices are observed across gender and handedness, highlighting the importance of these factors in fatigue assessment frameworks. The non-invasive nature of MMG technology presents a novel approach for monitoring muscle fatigue in workplaces and enabling early prevention of WMSDs, advancing the development of personalized fatigue assessment solutions and providing scientific foundations for optimizing occupational health management. Several directions for future research emerge from this study:

Scalability and Generalizability: Expand the scope of research to encompass diverse muscle groups, task intensities, and work scenarios to validate the stability of MMG-based fatigue assessment methods in a broader range of physiological and occupational environments.

Real-time systems: Develop efficient algorithms and integrated hardware-software solutions to enable real-time monitoring of MMG. Prioritize reducing latency and enhancing portability to meet the demands of on-site occupational health applications.

Long-term validity: Conduct longitudinal studies to assess the stability of MMG metrics under long-term working conditions and strengthen the reliability of long-term monitoring to support occupational health management.