Functional ANOVA for Upper Extremity Fatigue Analysis During Dynamic Order Picking.
Setareh Kazemi Kheiri,Zahra Vahedi,Hongyue Sun,Fadel M. Megahed,Lora A. Cavuoto
DOI: https://doi.org/10.1080/24725838.2024.2331182
2023-01-01
IISE Transactions on Occupational Ergonomics and Human Factors
Abstract:Musculoskeletal disorders are prevalent among warehouse workers who engage in repetitive and dynamic tasks. To prevent such injuries, it is vital to identify the factors that influence fatigue in the upper extremities during these repetitive activities. Our study reveals that task factors, namely the bottle mass and picking rate, significantly influence upper extremity fatigue. In most cases, the fatigue indicator is a functional variable, meaning that the fatigue score or measurement is a curve captured over time, which could be modeled as a function. In this study, we demonstrate that functional data analysis tools, such as functional analysis of variance (FANOVA), prove more effective than traditional methods in specifying how task factors contribute to the development of fatigue in the upper extremities. Furthermore, since there are inherent differences among workers that could affect their fatigue development process, the data heterogeneity could be tackled by employing clustering methods. TECHNICAL ABSTRACT Background: Preventing musculoskeletal disorders is a paramount safety concern for industries, with order pickers in warehouses being particularly vulnerable due to their repetitive and dynamic tasks. Understanding the factors contributing to upper-extremity fatigue in such settings is crucial. Purpose: This paper investigates the impact of task-related factors on two upper-extremity fatigue indicators: ratings of perceived fatigue and relative muscle strength. Several statistical approaches were used and compared in terms of their capability in eliciting these effects. Methods: Simulated over-shoulder, order-picking lab experiments were conducted under different combinations of two bottle loads and three picking paces. Fourteen participants, evenly distributed between genders, completed the experiment. A FANOVA was executed as the principal analytical approach, considering the functional nature of the two fatigue indicators measured over the work period. To underscore the benefits of considering the whole functional curve instead of discrete variables, we also conducted repeated-measures and two-way ANOVA as benchmark analyses. Results: FANOVA outcomes affirmed that both task factors (load and pace) significantly influenced both fatigue indicators. The FANOVA method identified larger effect sizes (0.11< eta(2)(p) < 0.19) for both task factors compared to the conventional methods (0< eta(2)(p) < 0.11), supporting the efficacy of FANOVA in identifying the importance of these factors. Conclusions: The FANOVA approach proved effective in detecting the impact of task factors on fatigue indicators, yielding superior results compared to conventional benchmark methods. To address participant heterogeneity, functional clustering and gender-based clustering were introduced into the FANOVA framework, both effectively mitigating this challenge. Notably, FANOVA with functional clusters had superior performance compared to the one with gender clustering, suggesting functional clustering as a more suitable method in overcoming participant heterogeneity.