A Collaborative Robotic Approach for Inspection and Anomaly Detection in Industrial Applications

Abstract

Inspection and quality assurance is an important step in manufacturing systems, including newly manufactured or re-manufactured parts. Currently, there is a heavy reliance on the knowledge of experienced workers in interpreting the data from inspection sensors and detecting anomalies. Using robots to perform automated inspection becomes challenging in high-mix settings, where the work-pieces to be inspected change frequently and require the robot to be re-programmed. In this paper, we propose a human-robot collaboration approach, where part of the work involving fixturing, sensor attachment and work-piece handling is done by the human, whereas the data collection, processing and anomaly detection is done autonomously using AI techniques. Our inspection algorithm is a generic approach using dilated convolutional neural network (DCNN) based multivariate time series predictive analytics. We demonstrate our approach on a gearbox inspection application, where we use time-series data streams captured from vibration sensors mounted on the gearbox. We have conducted experiments to demonstrate the effectiveness of the proposed DCNN solution for anomaly detection in a human robot collaborative assembly system.