Machine Learning-Driven Detection of Repetitive Manufacturing Processes Using Radar Sensor

Abstract

This paper presents a non-invasive system for detecting repetitive manufacturing cycles using pulse-coherent radar and machine learning. The Acconeer A111 radar sensor, combined with an Arducam USB camera, is integrated within a ROS2-based data acquisition framework. The system operates in Envelope and Sparse radar modes, optimized for tracking static and dynamic motion. A YOLO-based model analyzes radar heatmaps to detect repetitive cycles automatically. The approach was validated through controlled experiments and in an industrial setting. Results demonstrate the system’s potential to accurately detect production cycles without modifying existing machinery, highlighting its potential for real-time process monitoring and optimization.