Adversarial Framework for Monitoring Unannotated Anomalies of Key Performance Indicators in Robot Applications

Addressing performance metrics of automation equipment and machines is a significant challenge imposed by the requirements of Industry 4.0 in flexible automation environments. Consequent issues are the unpredictable effects of anomalies in the system and the persistent problem of managing unannotated data. This paper addresses these issues by monitoring anomalies and values of five Key Performance Indicators (KPI) as Cycle Time, Cycles Completed, Wait Time, Utilization and Efficiency. These metrics are recognized as essential for intelligent operation and maintenance of robotic applications. Our framework implements single-class training of Generative Adversarial Networks (GANs) using normal operation patterns and calculates a weighted anomaly score based on the results of the GAN. Expected results encompass a reduced reliance on manual data labeling, enhancements in precision, and enable adaptability adjustments, thereby influencing operational monitoring and maintenance practices in flexible automation environments and contributing to the overall improvement of efficiency and intelligence in modern robotic systems.