A Closed-Loop Evaluation Method for Industrial Robot Performance Driven by Health Data

Abstract

Evaluating robot performance through health data is important for industrial robots in production. In this article, a closed-loop evaluation method based on low-cost health data is proposed to realize the evaluation of robot performance and contribute to robot optimization. Specifically, we first define a closed-loop evaluation structure between a robot system and its performance. Based on this, a multicluster cross-fusion algorithm based on K-means clustering is proposed to extract patterns of the performance based on the direction and magnitude of control errors instead of metrics to describe the robot performance. Then, objective and interpretable criterion sets are designed based on the strategies for robot system maintenance. The evaluation model based on hybrid game theory is proposed to realize the evaluation of the robot overall performance with the pattern of performance and criterion sets. Meanwhile, a closed-loop evaluation method based on an autoregressive nonlinear neural network and Bayesian model is used to evaluate the robot joint performance by mapping the performance to the robot joint. The effectiveness of the method is verified using an actual DoF robot. The results show that the method is sufficient for the evaluation of robot performance and optimization guidance.