The health prediction of assembly robot based on feature fusion and weighted mahalanobis distance

Predicting the health status of assembly robots serves as an effective means to enhance the reliability of manufacturing systems. To precisely predict the health status of robots and thereby improve the safety and reliability of manufacturing systems, a health status prediction method for assembly robots based on feature fusion and weighted Mahalanobis distance(WMD) is proposed.Addressing the issues of slow processing speed and suboptimal feature fusion in robot high-dimensional monitoring data processing with SAE, a method of optimizing the hyperparameters and selecting the appropriate loss function of SAE is employed to achieve the deep fusion of features of robot monitoring data, and the manifold learning method is incorporated to select the health status-sensitive features.To address the limitations of existing robot health state prediction models, particularly the risk of false "healthy" predictions caused by the complexity of numerous monitored parameters, the Cox-Box transformation is applied to the WMD of sensitive features. A robot health state monitoring model is then designed based on the transformed WMD, enabling more accurate and reliable monitoring of robot health status. An assembly robot on a production line is used as an example to demonstrate the effectiveness of the proposed method in monitoring robot health status.