Inertial Sensor-based Postural Demand Profiles for Cumulative Physical Workload Estimation

Research Trainee: Sol Lim, MS, PhD Student, Department of Industrial and Operations Engineering at the University of Michigan

Faculty Sponsor: Clive D’Souza, PhD, Assistant Professor of Industrial and Operations Engineering at the University of Michigan 


Sol Lim, PhD Student


Professor D’Souza

Project Abstract

Prolonged manual work in awkward constrained postures and requiring high force exertions are known risk factors for musculoskeletal injuries. Reliable, accurate, and unobtrusive instrumentation methods are needed to identify, measure and manage such physical exposures in real work environments. The current project aimed to validate use of low-cost body-worn inertial sensors for measuring body posture, and to develop an algorithm for estimating physical workload and injury risk using inertial sensor-derived body posture measurements.

Fifteen able-bodied male participants performed three physical tasks simulated in the laboratory that are relevant to patient transferring and manual material handling, namely, one- and two-handed pushing and pulling, two-handed material lifting and lower, and carrying a weighted object. Task intensity levels were manipulated to induce changes in body posture.

Algorithms were developed to measure body posture using accelerometry, gyroscope and magnetometer data from body-worn inertial sensors. For accuracy we compared inertial sensor derived measures of torso posture to conventional motion capture derived measures. Error differences ranged between 1.80-3.69 degrees for the static push-pull task and 2.59-16.9 degrees for the dynamic lifting-lowering task. Overall, these values were similar to those reported in research literature.

Participant pulling with one hand

Participant pushing with one hand

Statistical analysis was conducted for each of the three tasks separately to quantify the effects of task intensity levels on body posture. Multinomial logistic regression was used to predict the task intensity (ex. force exerted or external load carried) as a function of the inertial sensor derived posture variables. Lastly, we present an algorithmic model that combines information about the task type, inertial sensor derived body posture and predicted task intensity level to compute the cumulative low back compressive force, i.e., a measure of physical workload and musculoskeletal injury risk.

Collectively, our results have broad applicability for the assessment of ergonomic workload in field settings using unobtrusive, low-cost instrumentation. The study provides the methodology and metrics to quantify posture demand and workload in manual work conditions. The algorithms we present to classify worker postures and generate more precise exposure profiles improves our ability to assess injury risk, identify prevention strategies and evaluate effectiveness of ergonomic interventions over prolonged periods of time.

Participant performing an isometric one-handed push (left) vs. pull (right) task using their dominant hand. Multiple optical motion tracking markers and inertial sensors are used to quantify 3D body segment postures during the task for subsequent statistical comparison.