A Lightweight Method Based on Adversarial Distillation for Online Wear State Recognition of Key Friction Pairs in Axial Piston Pump
Online wear state recognition of key friction pairs in the axial piston pump is crucial for ensuring the stability and reliability of the whole hydraulic transmission and control system.However,wear state recognition performance is always hindered by strong noise interference in real working conditions.Although wear state recognition models can enhance noise robustness through adversarial training,their large number of parameters limits the application on edge computing devices.To balance noise robustness and model scale,a lightweight method based on adversarial distillation for online wear state recognition is proposed.A student model and a teacher model based on one dimension convolutional neural network are designed.The student model generates adversarial examples as a knowledge transfer dataset and learns robust feature knowledge from the teacher model through knowledge distillation.Through fault injection,comparison experiments,and ablation experiments,the proposed method demonstrates strong noise robustness and lightweight model structure advantages.Edge deployment and online verification experiments show that the proposed method can accurately recognize the wear state of the axial piston pump in real time.
axial piston pumpwear state recognitionedge computingnoise interferenceknowledge distillationadversarial examples
万方数据
维普
