Prediction of contact resistance of electrical contact wear using different machine learning algorithms

Abstract

Abstract H62 brass material is one of the important materials in the process of electrical energy transmission and signal transmission, and has excellent performance in all aspects. Since the wear behavior of electrical contact pairs is particularly complex when they are in service, we evaluated the effects of load, sliding velocity, displacement amplitude, current intensity, and surface roughness on the changes in contact resistance. Machine learning (ML) algorithms were used to predict the electrical contact performance of different factors after wear to determine the correlation between different factors and contact resistance. Random forest (RF), support vector regression (SVR) and BP neural network (BPNN) algorithms were used to establish RF, SVR and BPNN models, respectively, and the experimental data were trained and tested. It was proved that BP neural network model could better predict the stable mean resistance of H62 brass alloy after wear. Characteristic analysis shows that the load and current have great influence on the predicted electrical contact properties. The wear behavior of electrical contacts is influenced by factors such as load, sliding speed, displacement amplitude, current intensity, and surface roughness during operation. Machine learning algorithms can predict the electrical contact performance after wear caused by these factors. Experimental results indicate that an increase in load, current, and surface roughness leads to a decrease in stable mean resistance, while an increase in displacement amplitude and frequency results in an increase in stable mean resistance, leading to a decline in electrical contact performance. To reduce testing time and costs and quickly obtain the electrical contact performance of H62 brass alloy after wear caused by different factors, three algorithms (random forest (RF), support vector regression (SVR), and BP neural network (BPNN)) were used to train and test experimental results, resulting in a machine learning model suitable for predicting the stable mean resistance of H62 brass alloy after wear. The prediction results showed that the BPNN model performed better in predicting the electrical contact performance compared to the RF and SVR models.

Keywords

• electrical contact wear
• contact resistance
• machine learning (ML)
• neural network