A neural network model for predicting postures during non-repetitive manual materials handling tasks

Modeling the Impact of Ergonomic Interventions and Occupational Factors on Work-Related Musculoskeletal Disorders in the Neck of Office Workers with Machine Learning Methods

BACKGROUND

Modeling with methods based on machine learning (ML) and artificial intelligence can help understand the complex relationships between ergonomic risk factors and employee health. The aim of this study was to use ML methods to estimate the effect of individual factors, ergonomic interventions, quality of work life (QWL), and productivity on work-related musculoskeletal disorders (WMSDs) in the neck area of office workers. Study Design: A quasi-randomized control trial.

METHODS

To measure the impact of interventions, modeling with the ML method was performed on the data of a quasi-randomized control trial. The data included the information of 311 office workers (aged 32.04±5.34). Method neighborhood component analysis (NCA) was used to measure the effect of factors affecting WMSDs, and then support vector machines (SVMs) and decision tree algorithms were utilized to classify the decrease or increase of disorders.

RESULTS

Three classified models were designed according to the follow-up times of the field study, with accuracies of 86.5%, 80.3%, and 69%, respectively. These models could estimate most influencer factors with acceptable sensitivity. The main factors included age, body mass index, interventions, QWL, some subscales, and several psychological factors. Models predicted that relative absenteeism and presenteeism were not related to the outputs.

CONCLUSION

In this study, the focus was on disorders in the neck, and the obtained models revealed that individual and management interventions can be the main factors in reducing WMSDs in the neck. Modeling with ML methods can create a new understanding of the relationships between variables affecting WMSDs.

An artificial neural network for full-body posture prediction in dynamic lifting activities and effects of its prediction errors on model-estimated spinal loads

Musculoskeletal models have indispensable applications in occupational risk assessment/management and clinical treatment/rehabilitation programs. To estimate muscle forces and joint loads, these models require body posture during the activity under consideration. Posture is usually measured via video-camera motion tracking approaches that are time-consuming, costly, and/or limited to laboratories. Alternatively, posture-prediction tools based on artificial intelligence can be trained using measured postures of several subjects performing many activities. We aimed to use our previous posture-prediction artificial neural network (ANN), developed based on many measured static postures, to predict posture during dynamic lifting activities. Moreover, effects of the ANN posture-prediction errors on dynamic spinal loads were investigated using subject-specific musculoskeletal models. Seven individuals each performed twenty-five lifting tasks while their full-body three-dimensional posture was measured by a 10-camera Vicon system and also predicted by the ANN as functions of the hand-load positions during the lifting activities. The measured and predicted postures (i.e., coordinates of 39 skin markers) and their model-estimated L5-S1 loads were compared. The overall root-mean-squared-error (RMSE) and normalized (by the range of measured values) RMSE (nRMSE) between the predicted and measured postures for all markers/tasks/subjects was equal to 7.4 cm and 4.1 %, respectively (R2 = 0.98 and p < 0.05). The model-estimated L5-S1 loads based on the predicted and measured postures were generally in close agreements as also confirmed by the Bland-Altman analyses; the nRMSE for all subjects/tasks was < 10 % (R2 > 0.7 and p > 0.05). In conclusion, the easy-to-use ANN can accurately predict posture in dynamic lifting activities and its predicted posture can drive musculoskeletal models.

Multi-task artificial neural networks and their extrapolation capabilities to predict full-body 3D human posture during one- and two-handed load-handling activities

Machine-learning based human posture-prediction tools can potentially be robust alternatives to motion capture measurements. Existing posture-prediction approaches are confined to two-handed load-handling activities performed at heights below 120 cm from the floor and to predicting a limited number of body-joint coordinates/angles. Moreover, the extrapolating power of these tools beyond the range of the input dataset they were trained for (e.g., for underweight, overweight, or left-handed individuals) has not been investigated. In this study, we trained/validated/tested two posture-prediction (for full-body joint coordinates and angles) artificial neural networks (ANNs) using both 70%/15%/15% random-hold-out and leave-one-subject-out methods, based on a comprehensive kinematic dataset of forty-one full-body skin markers collected from twenty right-handed normal-weight (BMI = 18-26 kg/m2) subjects. Subjects performed 204 one- and two-handed unloaded activities at different vertical (0 to 180 cm from the floor) and horizontal (up to 60 cm lateral and/or anterior) destinations. Subsequently, the extrapolation capability of the trained/validated/tested ANNs was evaluated using data collected from fifteen additional subjects (unseen by the ANNs); three individuals in five groups: underweight, overweight, obese, left-handed individuals, and subjects with a hand-load. Results indicated that the ANNs predicted body joint coordinates and angles during various activities with errors of ∼ 25 mm and ∼ 10°, respectively; considerable improvements when compared to previous posture-prediction ANNs. Extrapolation errors of our ANNs generally remained within the error range of existing ANNs with obesity and being left-handed having, respectively, the most and least compromising effects on their accuracy. These easy-to-use ANNs appear, therefore, to be robust alternatives to common posture-measurement approaches.

The role of machine learning in the primary prevention of work-related musculoskeletal disorders: A scoping review

To determine the applications of machine learning (ML) techniques used for the primary prevention of work-related musculoskeletal disorders (WMSDs), a scoping review was conducted using seven literature databases. Of the 4,639 initial results, 130 primary research studies were deemed relevant for inclusion. Studies were reviewed and classified as a contribution to one of six steps within the primary WMSD prevention research framework by van der Beek et al. (2017). ML techniques provided the greatest contributions to the development of interventions (48 studies), followed by risk factor identification (33 studies), underlying mechanisms (29 studies), incidence of WMSDs (14 studies), evaluation of interventions (6 studies), and implementation of effective interventions (0 studies). Nearly a quarter (23.8%) of all included studies were published in 2020. These findings provide insight into the breadth of ML techniques used for primary WMSD prevention and can help identify areas for future research and development.

Improved artificial neural networks for 3D body posture and lumbosacral moment predictions during manual material handling activities

Body posture measurement approaches, required in biomechanical models to assess risk of musculoskeletal injuries, are usually costly and/or impractical for use in real workplaces. Therefore, we recently developed three artificial neural networks (ANNs), based on measured posture data on several individuals, to predict whole body 3D posture (coordinates of 15 markers located on body's main joints), segmental orientations (Euler angles of 14 body segments), and lumbosacral (L5-S1) moments during static manual material handling (MMH) activities (ANN_Posture, ANN_Angle, and ANN_Moment, respectively). These ANNs require worker's body height, body weight (only for ANN_Moment), hand-load 3D position, and its mass as inputs to accurately predict 3D marker coordinates (RMSE = 7.0 cm), segmental orientations (RMSE = 29.9°) and L5-S1 moments (RMSE = 16.5 N.m) for various static MMH activities. The current work aims to further improve the accuracy of these ANNs by performing outlier elimination and data normalization (as effective tools to improve the accuracy of ANNs) as well as by introducing participant's knee flexion angle (i.e., lifting technique: stoop, semi-squat, and full-squat) and body weight as new inputs into these ANNs. Results indicate that the RMSE of the new ANN_Posture, ANN_Angle, and ANN_Moment reduced by, respectively, ∼43%, 10%, and 29% (from 7.0 cm, 29.9°, and 16.5 Nm in the original ANNs to, respectively, 4.0 cm, 27.0°, and 11.8 Nm). Such significant improvements in the predictive power of our ANNs further confirm their effectiveness as alternative posture-prediction approaches requiring minimal in vivo data collection in real workplaces.

Coupled artificial neural networks to estimate 3D whole-body posture, lumbosacral moments, and spinal loads during load-handling activities

Biomechanical modeling approaches require body posture to evaluate the risk of spine injury during manual material handling. The procedure to measure body posture via motion-analysis techniques as well as the subsequent calculations of lumbosacral moments and spine loads by, respectively, inverse-dynamic and musculoskeletal models are complex and time-consuming. We aim to develop easy-to-use yet accurate artificial neural networks (ANNs) that predict 3D whole-body posture (ANN_posture), segmental orientations (ANN_angle), and lumbosacral moments (ANN_moment) based on our measurements during load-handling activities. Fifteen individuals each performed 135 load-handling activities by reaching (0 kg) or handling (5 and 10 kg) weights located at nine different horizontal and five vertical (0, 30, 60, 90, and 120 cm from the floor) locations. Whole-body posture was measured via a motion capture system and lumbosacral moments were calculated via a 3D top-down eight link-segment inverse-dynamic model. ANN_posture, ANN_angle, and ANN_moment were trained (RMSEs = 6.7 cm, 29.8°, and 16.2 Nm, respectively) and their generalization capability was tested (RMSE = 7.0 cm and R² = 0.97, RMSE = 29.9° and R² = 0.85, and RMSE = 16.5 Nm and R² = 0.97, respectively). These ANNs were subsequently coupled to our previously-developed/validated ANN_load, which predicts spinal loads during 3D load-handling activities. The results showed outputs of the coupled ANNs for L4-L5 intradiscal pressure (IDPs) during a number of activities were in agreement with measured IDPs (RMSE = 0.37 MPa and R² = 0.89). Hence, coupled ANNs were found to be robust tools to evaluate posture, lumbosacral moments, spinal loads, and thus risk of injury during load-handling activities.

Lifting motion simulation using a hybrid approach

In this study, a hybrid dynamic model for lifting motion simulation is presented. The human body is represented by a two-dimensional (2D) five-segment model. The lifting motions are predicted by solving a nonlinear optimisation problem, the objective function of which is defined based on a minimal-effort performance criterion. In the optimisation procedure, the joint angular velocities are bounded by time-functional constraints that are determined by actual motions. Symmetric lifting motions performed by younger and older adults under varied task conditions were simulated. Comparisons between the simulation results and actual motion data were made for model evaluation. The results showed that the mean and median joint angle errors were less than 10°, which suggests the proposed model is able to accurately simulate 2D lifting motions. The proposed model is also comparable with the existing motion simulation models in terms of the prediction accuracy. Strengths and limitations of this hybrid model are discussed. Practitioner Summary: Human motion simulation is a useful tool in assessing the risks of occupational injuries. Lifting motions are associated with low-back pain. A hybrid model for lifting motion simulation was constructed. The model was able to accurately simulate 2D lifting motions in varied task scenarios for younger and older subjects.