Predicting Ergonomic Risk and Worker Performance Using Neural Networks with EMG and Kinematic Data

Open Access
Article
Conference Proceedings
Authors: Sohyung Cho
Abstract

Ergonomic performance is fundamentally governed by the interaction between postural configuration and muscle activation; however, frontier based methodologies such as Data Envelopment Analysis (DEA) yield relative efficiency scores rather than predictive estimates. This study investigates the extent to which supervised neural networks (NNs) can generate predictive estimates of DEA derived ergonomic efficiency using electromyography (EMG) and kinematic inputs. In this study, ten participants completed five sweeping task trials with bilateral EMG on biceps brachii and flexor carpi radialis, and elbow angles captured via motion tracking. Trial features were aggregated into six inputs per participant to compact feedforward NNs trained for regression of left/right DEA efficiencies and classification of efficient frontier participants. Leave one out cross validation and probability calibration were applied. As results, left side regression achieved R² = 0.57 and right side performance was weaker with R² = 0.11. The classifier reached 80% accuracy with high distinguishing power of efficient frontiers about 75% (F1 accuracy) and high ability to rank-order efficient vs. inefficient frontiers about 92% (AUC accuracy). Key predictors included left elbow angle and bilateral biceps activity. Two NN–DEA disagreements reflected differences between normative and predictive modeling. As a conclusion, neural networks accurately predicted ergonomic efficiency and frontier membership from the same EMG/kinematic data used in DEA, demonstrating their value for prospective, real time screening and future extension to sequence-based models.

Keywords: Ergonomics, Electromyography (EMG), Upper Limb Kinematics, Data Envelopment Analysis (DEA), Neural Networks

DOI: 10.54941/ahfe1008476

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Experimental investigation of damping and mass parameters of an adjustable physical hand-arm model for reproducing human mechanical impedanceThe Process of Information Integration in Children's Spatial Cognition: An Analysis of Newly Acquired Knowledge and Mapping Rates via Fieldwork →
View all articles in Human Factors in Design, Engineering, and Computing →