Prediction-Based Assembly Assistance System

This paper presents the design of a prediction-based assembly assistance system for manual operations and the results obtained on the data collected from experiments of assembling a customizable product. We integrated into the proposed system a Markov predictor improved with a padding mechanism whose role is to recommend the next assembly step and to detect the worker’s errors. The predictor is trained with correct assembly patterns and tested with real assembly/manufacturing data. The proposed predictor improves the coverage and, thus, there is a significantly higher number of assembly steps which are correctly correlated with the real intentions of the workers.