Fault diagnosis of power-shift system in continuously variable transmission tractors based on improved echo state network

Abstract

For better reliability of tractors with continuously variable transmission, reported here is fault diagnosis of their power-shift systems. First, four hydraulic system faults are analyzed, i.e., pipe leakage, pipe blockage, a stuck solenoid valve spool, and a stuck clutch piston, and it is shown that these lead to clutch energy loss during power shifting and possibly even clutch burn-out. Second, fault simulations give more than 30000 groups of test data, and an improved dynamic time warping algorithm is used to segment the data samples automatically. Third, an improved echo state network is proposed to classify the above fault samples, and its performance is compared with those of traditional algorithms. Finally, the robustness of the proposed algorithm is tested using samples with noise superimposed. The results show that the accuracy, precision, recall, and F1-score of the improved echo state network are 96.87%, 96.31%, 97.11%, and 96.71%, respectively, making it equivalent in performance to a convolutional neural network with long short-term memory. However, the training speed of the former is significantly better than that of the latter, and even under the influence of strong noise, the proposed algorithm can still achieve an accuracy of 91.32%, which is better than the latter’s 86.62%. This is significantly better than traditional algorithms, thus demonstrating the better generalization characteristics and robustness of the proposed approach.