Comparison of Fourier and wavelet analysis for fatigue assessment during repetitive dynamic exertion
Introduction
Repetitive arm and neck movements constitute a major facet of workplace tasks in several industries. A causal relationship between such exertions and neck/shoulder musculoskeletal disorders has been proposed in the literature (Côté et al., 2008). Repetitive exertions performed over a sustained period of time lead to fatigue, which is believed to be the precursor of musculoskeletal disorders (MSDs). Therefore, in order to eliminate the risk of MSDs due to repetitive exertions, it is essential that accurate fatigue assessment methods are available.
Among the methods that are used to estimate muscle fatigue, surface electromyography (SEMG) is a preferred method due to its high precision, non-invasiveness, and unobtrusiveness. Previous studies have used SEMG based measures to investigate the muscle fatigue associated with various occupational tasks (Bosch et al., 2007, Chowdhury et al., 2013). Three measures – mean frequency, median frequency, and the power of the spectrum estimated using the fast Fourier transform (FFT), have been commonly used to report neuromuscular fatigue. A shift in the mean or median frequencies toward lower values or an increase in the power of the low frequency components and a decrease in the power of high frequency components of the SEMG signal have been identified as indicators of muscle fatigue (Eberstein and Beattie, 1985, Georgakis et al., 2003).
The FFT algorithm is based upon the hypothesis that any signal can be represented by a sum of appropriately weighted sine and cosine functions (Bracewell and Bracewell, 1986). The algorithm assumes that the signal under investigation is stationary (Bilodeau et al., 1997). A signal is said to be stationary if the joint distribution of any set of samples within the signal are time independent and randomly distributed, i.e., the average, variance, and frequency contents of the signal do not alter over time. In case of non-stationary signals, the set of samples are statistically dependent and the mean, variance, and frequency content of the signals change over time (Bendat and Piersol, 2000). The SEMG signals are stationary during isometric exertion due to a stable and constant force exertion and body posture (Basmajian and De Luca, 1985). However, during dynamic exertion, the magnitude and direction of force application as well as body posture changes continuously. The SEMG signal recorded under such a condition may not satisfy the stationarity assumption (Merletti et al., 2004). Alternatively, short time Fourier transforms (STFT) are used to analyze nonstationary dynamic signals (MacIsaac et al., 2001). In STFT, the window sizes are altered (mostly reduced) to satisfy the stationarity assumption of the Fourier transform. However, the window size approximation used in STFT may introduce time and frequency resolution issues; i.e., a short window size may provide better time resolution but poor frequency resolution while a relatively longer window provides better frequency resolution but poor time resolution (Bonato et al., 1996).
Another viable choice to FFT based spectral analysis for fatigue estimation is the use of discrete wavelet transform (DWT) analysis. Unlike FFT, which is restricted to one feature morphology, the sinusoid, DWT provides a range of orthogonal wavelet functions for conducting spectral analysis (Polikar, 2006). There is no assumption of signal stationarity under this analysis and the SEMG signal can be evaluated to elucidate spectral and temporal information simultaneously. However, the computation method for DWT analysis is more tedious than FFT. The signal decomposition takes place at several levels based on the duration and/or selection of the window size. The DWT has been used in analyzing non-stationary SEMG signals for studying the muscle fatigue in a variety of exertion tasks (Beck et al., 2005, Hostens et al., 2004, Kumar et al., 2003, Nimbarte et al., 2013). The root mean square (RMS), the power, and the center frequency of the wavelet coefficients at different levels (frequency bands) have been used in previous DWT studies to report the onset and development of fatigue. A drop in the center frequency (Beck et al., 2005, Kumar et al., 2003) and an increase in the power (Chowdhury et al., 2013, Kumar et al., 2003) and the RMS values (Sparto et al., 2000) of the wavelet coefficients in the lower frequency bands were associated with the development of muscle fatigue.
Previous studies comparing the fatigue prediction ability of DWT with FFT have reported contradictory findings. Canal (2010) reported better resolution and visualization performance for DWT compared to STFT during fatiguing isokinetic exertions. Karlsson et al. (1999) estimated the SEMG power spectrum and the mean frequency using both wavelet and the Fourier transform methods. The wavelet based methods (DWT and wavelet packet transform) were more accurate than the Fourier transform for the nonstationary dynamic signal. In another study, Camata et al. (2010) compared STFT and continuous wavelet transform (CWT) in assessing muscle fatigue during a constant load dynamic task. The results indicated higher variability in the median frequency estimated by the STFT compared to the CWT method. In contrast to the aforementioned studies, Beck et al. (2005) reported a similar pattern of center frequency when comparing FFT and DWT in analyzing fatiguing isokinetic exertions. In another study, Hostens et al. (2004) compared Fourier and wavelet transform methods when examining the spectral pattern of the SEMG signal during fatiguing submaximal isometric and dynamic exertions. Similarities in the wavelet and Fourier based center frequency patterns were also observed in this study.
In summary, spectral analysis performed using FFT and DWT based algorithms were used in the previous studies to perform fatigue estimation. Each algorithm has its strengths and weaknesses and previous comparison studies have reported conflicting findings. Moreover, the dynamic exertions investigated in previous studies were mostly isokinetic in nature whereas most occupational dynamic exertions are rarely isokinetic. The comparative ability of the FFT and DWT algorithms in estimating spectral changes during fatiguing real world dynamic exertions has never been studied. The present study attempted to compare the spectral trends computed by the FFT algorithm with the time–frequency approaches of the DWT algorithm in assessing muscle fatigue during sub-maximal repetitive dynamic exertions. The sub-maximal repetitive exertions were defined as the exertions that a worker/ human participant is capable of performing for a longer period of time.
Section snippets
Approach
A lab-based study was performed in order to compare the abilities of the FFT and DWT algorithms in assessing fatigue generated by dynamic exertions. Human participants completed a task consisting of 40 min of repetitive upper extremity exertions. Activity of the muscle responsible for these exertions was recorded using SEMG. Spectral analysis was performed using FFT and DWT algorithms to estimate the power of different frequency bands.
Participants
Ten healthy male participants (age = 27 ± 4.8 years; weight = 71.1 ±
Results
The mean duration of loading and unloading exertions ranged from 2.02 s to 1.57 s. The mean duration of exertion decreased from 2.02 s to 1.67 s and from 1.87 s to 1.57 s during sessions 1 and 2, respectively.
The mean discomfort scores recorded using Borg’s CR-10 scale showed an increasing discomfort trend towards the end of sessions 1 and 2. A baseline discomfort of 1 (= nothing at all) was recorded at the beginning of the first session. After twenty minutes of loading and unloading exertions, a
Discussion
In this study the power trend of different frequency bands was used to evaluate the fatigue prediction ability of the FFT and DWT algorithms. The muscle fatigue was also assessed using a standardized and widely used subjective tool (Borg’s CR-10 scale). A significant correlation between CR-10 ratings and the muscle fatigue was reported in the past studies (Dedering et al., 1999, Dedering et al., 2002). A 30% reduction in the median and mean frequencies, and endurance time were observed at a
Conclusions
The comparative ability of the FFT and DWT algorithms to predict fatigue generated during repetitive dynamic exertions was evaluated in the current study. The results showed less variability in the power values computed by the DWT algorithm compared to the FFT algorithm. Frequency band (lower vs higher) and amount of fatigue (less vs more) were found to influence the trend in the power values.
Conflict of interest
The authors declare that they have no conflict of interest
Suman Kanti Chowdhury is a PhD student at the faculty of Industrial and Management System Engineering of West Virginia University. He received his M.S. degree in Industrial Engineering from West Virginia University in 2012, and received his B.Sc. degree in Industrial and Production Engineering from Bangladesh University of Engineering and Technology, Bangladesh in 2006. His research interest is in the general areas of industrial ergonomics and human factors engineering. His recent research
References (36)
- et al.
Comparison of Fourier and wavelet transform procedures for examining the mechanomyographic and electromyographic frequency domain responses during fatiguing isokinetic muscle actions of the biceps brachii
J Electromyogr Kinesiol
(2005) - et al.
Normality and stationarity of EMG signals of elbow flexor muscles during ramp and step isometric contractions
J Electromyogr Kinesiol
(1997) - et al.
Discrete wavelet transform analysis of surface electromyography for the fatigue assessment of neck and shoulder muscles
J Electromyogr Kinesiol
(2013) - et al.
Correlation between electromyographic spectral changes and subjective assessment of lumbar muscle fatigue in subjects without pain from the lower back
Clin Biomech
(1999) - et al.
Electromyography and ratings of lumbar muscle fatigue using a four-level staircase protocol
Clin Biomech
(2002) - et al.
Validation of the wavelet spectral estimation technique in biceps brachii and brachioradialis fatigue assessment during prolonged low-level static and dynamic contractions
J Electromyogr Kinesiol
(2004) - et al.
The short-time Fourier transform and muscle fatigue assessment in dynamic contractions
J Electromyogr Kinesiol
(2001) - et al.
Empirical evaluation of neck muscle fatigue generated by healthcare related exertions
Safety Science
(2013) - et al.
Wavelet analysis of neuroelectric waveforms: a conceptual tutorial
Brain Lang
(1999) - et al.
Changes in the myoelectric signal (MES) power spectra during dynamic contractions
Electroencephalogr Clin Neurophysiol
(1989)
Fatigue-induced changes in muscle fiber action potentials estimated by wavelet analysis
J Electromyogr Kinesiol
Muscle Alive
Random data analysis and measurement procedures
Meas Sci Technol
Analysis of myoelectric signals recorded during dynamic contractions
Eng Med Biol Mag IEEE
Development of fatigue and discomfort in the upper trapezius muscle during light manual work
Ergonomics
The Fourier transform and its applications
Myoelectric signal conduction velocity and spectral parameters: influence of force and time
J Appl Physiol
Fourier and wavelet spectral analysis of EMG signals in supramaximal constant load dynamic exercise
Cited by (30)
Electromyographic traces of motor unit synchronization of fatigued lower limb muscles during gait
2021, Human Movement ScienceCitation Excerpt :It, however, does not offer enough time resolution to properly resolve the changes in the distribution of energy over the power spectrum during dynamic tasks. Wavelet transformations give a better resolution of the power spectrum, both in time and frequency, than the Fourier Transform for the study of spectral characteristic of the EMG (Chowdhury & Nimbarte, 2015; Chowdhury, Nimbarte, Jaridi, & Creese, 2013). Wavelet transformations are dependent on mother wavelets (Chowdhury et al., 2013; Chowdhury & Nimbarte, 2015; Sharma & Veer, 2016a, 2016b) and some authors suggest that using non-linearly scaled wavelets gives the best results.
Evaluating photosynthetic pigment contents of maize using UVE-PLS based on continuous wavelet transform
2020, Computers and Electronics in AgricultureStrength, endurance and fatigue response of rotator cuff muscles during isometric exertions
2019, International Journal of Industrial ErgonomicsCitation Excerpt :The 10-point data was fitted with a linear regression line and the slope of the line was computed. A decrease in the median frequency, indicated by negative slope, was considered as a sign of muscle fatigue (De Luca, 1984; Chowdhury and Nimbarte, 2015; Niemeyer et al., 2004; Minning et al., 2007; Kai et al., 2012). The statistical analysis was performed to understand if the strength at MVC is different between the muscles, i.e., to test if the muscle has any effect on the strength at MVC.
Muscle fatigue response of rotator cuff muscles in different postures
2023, Archives of Orthopaedic and Trauma SurgeryWavelet analysis of the EMG signal to assess muscle fatigue in the lower extremities during symmetric movement on a rowing ergometer
2023, Acta of Bioengineering and Biomechanics
Suman Kanti Chowdhury is a PhD student at the faculty of Industrial and Management System Engineering of West Virginia University. He received his M.S. degree in Industrial Engineering from West Virginia University in 2012, and received his B.Sc. degree in Industrial and Production Engineering from Bangladesh University of Engineering and Technology, Bangladesh in 2006. His research interest is in the general areas of industrial ergonomics and human factors engineering. His recent research focuses on the estimation of muscle fatigue using advanced digital signal processing techniques.
Dr. Ashish D. Nimbarte is an Assistant Professor and the Director of Occupational Safety and Health Engineering program in the Industrial and Management Systems Engineering Department at West Virginia University. His training and expertise is in the general areas of ergonomics and occupational biomechanics. Dr. Nimbarte’s research interest is motivated by the need to better characterize occupational risk factors so that effective control strategies can be implemented in the workplace. Specifically, he focuses on improving the understanding of the pathogenesis of work-related musculoskeletal disorders of the upper extremities and cervical spine.