Unsupervised Method of Determining Cycle Times of Manual Assembly Processes

Abstract

In recent years, wrist-worn devices that contain inertial measurement units have become prevalent. The data from these sensors can be used to estimate characteristics of manual processes in a manufacturing environment. The goal of this research is to determine cycle times of manually performed assembly processes using data from wrist-worn inertial sensors. Specifically, this work explores an unsupervised method of analyzing time series data to extract patterns (motifs) which represent individual cycles of an operation. From here, cycle time is computed by applying knowledge of the frequency of data collection. Testing shows that the mean cycle times obtained from stopwatch are statistically indifferent from those obtained from the proposed approach. Furthermore, results suggest that the proposed approach is insensitive to high frequency noise in the data. These encouraging results warrant further investigation and more testing.