Hand Motion Recognition of Shipyard Welder Using 9-DOF Inertial Measurement Unit and Multi Layer Perceptron Approach

Abstract

A viable system that can monitor the effective working time of welder in real-time is required to overcome the low use of effective welder time in the Shipbuilding Project in the Indonesian Shipyard. It is made possible by using a wearable sensor tri-axial accelerometer, gyroscope, and magnetometer. In this research, sensors are used to recognize typically hand motion of welder during welding activities: preparation, welding and cleaning slags, respectively in three welding positions 1G, 2G, and 3G. Initially, observations were made to recognize the relationship between welder activities and hand motion. Second, raw data containing hand movements from the welder is captured in the form of time-series signals using inertia sensors for various different activities. Third, the raw data of measurements for those activities is extracted and analyzed to identify significant features such as mean, root-mean-square, power spectral density using the welch method (autocorrelation, spectral peak, and spectral power). Finally, typical activities of welder are classified using the resulting feature data with Multi Layer Perceptron. The validation of results shows that the algorithm is capable to recognize the hand motion activities of the welder.