Abstract: Based on the issue that monitoring of wheel wear is difficult to be implemented directly during grinding process, a multi-feature optimization and fusion based random forest (MFOF-RF) algorithm was proposed to realize the accurate prediction of wheel wear. An experiment of cylindrical traverse grinding was performed and the power, acceleration and acoustic emission signals were collected and processed in order to extract a large amount of time-domain and frequency-domain signal features. Statistical criteria were used to adjust model parameters and choose best feature combination for the prediction of wheel wear. The results shown that the MFOF-RF model improved the prediction accuracy and diminished error more than 30% compared with the model with single feature.

Key words: grinding wheel wear; multi-feature optimization and fusion; feature selection; random forest