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

Area of Science:

• Biomechanics
• Ergonomics
• Artificial Intelligence

Background:

• Manual materials handling (MMH) tasks require accurate posture prediction for design and evaluation.
• Existing computational methods for posture prediction have limitations.

Purpose of the Study:

• To evaluate artificial neural network (ANN) models for predicting initial and final lifting postures in 2-D and 3-D scenarios.
• To assess the generalization ability of ANNs to novel manual materials handling conditions.

Main Methods:

• ANN models were trained using subsets of a posture database.
• Participant and task descriptors were used as model inputs, with joint angles as outputs.
• Model predictions were evaluated on data unseen during training, including novel conditions.

Main Results:

• ANN models demonstrated consistent prediction errors across data subsets, indicating good generalization.
• Per-joint errors ranged from 5 to 20 degrees, with higher errors in 3-D conditions.
• The models achieved reasonably accurate predictions, outperforming some previous computational approaches.

Conclusions:

• ANN models offer a viable method for predicting postures in common manual materials handling tasks.
• The models provide insights into potential lifting strategies based on individual characteristics.
• Further refinement of ANN models can enhance prediction accuracy for ergonomic assessments.