Abstract:

Abrasion grinding wheels lower the material removal rate and increase the grinding temperature. To avoid the negative influence of the mentioned issues on the grinding process, replacing grinding wheels in time was necessary. This paper proposed a method predicting the wear and service life of grinding wheels by applying artificial neural network, through which grinding wheels could be replaced timely and reasonably. The basic theory of the method as follows: A series of wear parameters was abstracted from the axial acceleration signal of grinder motor shaft. By applying Z-score transformation, the wear parameters’ dimension was removed and comparability was improved. After that, according to mean gain, the XGBoost algorithm was employed to filter the most relevant features that are strongly correlated with the wheel's service life. A fusion strategy integrating wear time and wheel wear volume was adopted as the criterion for assessing the wear and service life. Constructing a neural network model that mapped the filtered feature parameters to both the wear volume and the grinding wheel thickness, followed by network training and validation. Experimental results showed that, in train and validation set, the judgment accuracy and prediction precision of this method was basically the same. The judgment accuracy of grinding wheels’ wear was 87.8%, with errors primarily concentrated in the transition interval when wear changed. The prediction precision of the grinding wheels’ service life was -5.3% in validation set. Also, this method showed a good versatility under various grinding process parameters.

Key words: rail grinding, grinding wheel wear, neural network, wear evaluation, residual life prediction