A novel model for monitoring wear state of a belt in grinding GH4169 alloy at multiple working conditions

Due to the non-renewable characteristics of abrasive belts, the belt wear will greatly reduce the grinding consistency and efficiency, especially for hard-to-machine materials such as GH4169 alloy. There are many factors that affect the belt wear, among which the nonlinearity is strong. Existing research is mostly based on fixed working conditions, whose applicability and accuracy decrease when the working conditions change. Regarding this issue, an online belt wear state monitoring method under multiple working conditions is established based on a transfer learning algorithm in this paper. Firstly, a multi-sensor integration robotic grinding system is developed. Based on this platform, GH4169 belt grinding signals are acquired, and the sensitive signals are screened based on correlation analyses, which are pre-processed by invalid signal truncation method and wavelet packet denoising to improve the signal-to-noise ratio. Furthermore, the sensitive signal features are extracted, and dimension reduction is completed by the Pearson correlation coefficient method. Then, according to the influence analysis of process parameters on signal feature distribution, the belt wear monitoring model based on the transfer learning strategy under the same linear velocity is established. Finally, in order to eliminate the influence of different working conditions, a feature standardization method considering process parameters (FSMCPP) is further proposed, and the online belt wear state monitoring method under multiple working conditions based on FSMCPP and transfer learning strategy is established. The experimental results show that the proposed method achieves the prediction accuracy of at least 90% under various working conditions. Further comparisons with three other typical intelligent algorithms show that the proposed method has certain advantages in prediction accuracy and convergence accuracy, indicating that the proposed method has good effectiveness and superiority in belt wear prediction under multiple working conditions.