A Hybrid CBiGRUPE Model for Accurate Grinding Wheel Wear Prediction.

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence in Manufacturing
• Machine Condition Monitoring

Background:

• Effective monitoring of grinding wheel wear is crucial for maintaining process quality and economic efficiency in grinding operations.
• Existing methods for wear prediction often face limitations in accuracy and stability.

Purpose of the Study:

• To develop and validate a novel hybrid model, CBiGRUPE, for accurate prediction of grinding wheel wear.
• To compare the performance of the proposed model against established methods like CNN, BiGRU, and Transformer.

Main Methods:

• Extraction of time-domain features from spindle motor current signals of a surface grinding machine.
• Integration of Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Unit (BiGRU), and Performer encoder into the CBiGRUPE model.
• Optimization of model structure and hyperparameters using Bayesian optimization.

Main Results:

• The CBiGRUPE model achieved excellent performance with MAE of 3.041, RMSE of 3.927, and R² of 0.920.
• Demonstrated superior accuracy and stability in wear predictions compared to CNN, BiGRU, and Transformer models.
• Experimental validation confirmed the effectiveness of the proposed hybrid approach.

Conclusions:

• The CBiGRUPE model offers a highly accurate and stable solution for predicting grinding wheel wear.
• This research provides a foundation for implementing wheel wear compensation and optimizing dressing strategies.
• The study highlights the potential of hybrid deep learning models in advanced manufacturing monitoring.