Fault Diagnosis of Suspension Controllers Based on KAN-ResNet

The suspension controller is a critical component of high-speed maglev trains, and its real-time fault diagnosis and reliability are crucial. This method utilizes ResNet modules to build a deep learning model and uses Kolmogorov-Arnold Networks (KAN) as a ResNet classifier to improve fault diagnosis accuracy. A training sample set of multi-parameter data was established, including current, acceleration, voltage, resistance, running speed, and suspension gap of the suspension controller. ResNet was used to extract temporal and multi-dimensional features, and KAN was used to fuse these features for fault characteristics. The proposed method was validated using experimental data from a high-speed maglev suspension controller test platform. The results show that KAN-Resnet has a high accuracy of 98.38% in fault diagnosis and inference of suspension controllers, which is 5.54% higher than 7-layer-CNN. The inference speed reaches 804n/s, which is 42.80% higher than ResNet-50.