Prediction of bolted joint looseness based on feature weighted-multiple kernel support vector machine

Given that traditional theoretical diagnosis is difficult to use for the accurate prediction of the looseness of bolted joints, this paper will use the support vector machine (SVM) method to solve this problem. The paper establishes a database by studying the loosening mechanism of bolted joints and by collecting the connection information of the fastening points of the vehicle chassis. Key influence features are determined by the information gain value. The average prediction accuracy of the standard SVM model is 76.94 %, which is better than the theoretical diagnosis method. To achieve optimization, the feature weighted support vector machine and the multiple kernel function support vector machine (MKL SVM) are established, and the average prediction accuracy is 82.18 % and 84.19 %, respectively. The fusion weighted support vector machine (FW-MKL SVM) is built by combining the advantages of two weighted optimization SVMs. The average prediction accuracy is 94.5 %, which meets practical engineering requirements.