Study on the effect of wear and accuracy correction of the detecting arm of leaf spring caliper

Abstract

Leaf spring calipers are widely used as a new type of gas pipeline calipers with outstanding advantages. However, their detection accuracy is easily affected by the wear and tear of leaf springs. This paper proposes a solution to compensate for the wear error. A wear model of the detection arm is established, and a wear detection test bench is built according to the model. The wear-mileage correction function is obtained by fitting the experimental results of the wear test rig. This function can be directly applied to the pipeline inspection to compensate for the wear error. Based on the principle of compensating for wear errors, this paper proposes another scheme: A machine learning (ML) approach is used to predict the wear patterns of leaf springs, and the predicted wear patterns are used to compensate for wear errors. In the prediction experiments, the Least Squares Support Vector Machine (LS-SVM) method is used to predict the wear values of leaf springs. A 10-fold cross-validation technique minimizes the bias of the model, and a set of hyperparameter values is found by adjusting the hyperparameters to satisfy the error requirements. The prediction results of machine learning are validated by extending the experimental mileage method, and the validation results satisfy the accuracy requirements. The results obtained in this paper provide a brand new idea to improve the detection accuracy of other types of pipe detectors.