Behavioral analysis of assembly work processes using temporal convolutional networks with feature selection

Abstract

Recently, it has been growing interest on so-called action segmentation, dividing a time series of behavioral data into fundamental action units.Though many deep learning models have been proposed, it still remains as a challenging issue to interpret how the model prioritizes the features for its decision making.This paper presents a novel action segmentation model that incorporates a feature selection technique into the existing general model, called MS-TCN, based on the temporal convolutional network.By employing the proposed model to assembly processing data, it becomes feasible to visualize the significant features that highlight each worker and process.The paper shows how the model switches the features in accordance with the prior work process by analyzing multi-dimensional behavior data generated from cameras, acceleration, angular rate sensors, and voice recordings.