COMPARISON BETWEEN ARTIFICIAL NEURAL NETWORK AND RANDOM FOREST-BASED MULTI-SENSOR FUSION FOR PREDICTING CBN WHEEL CONDITION

Abstract

Grinding is a machining process commonly used in industry to achieve a desired level of product quality (e.g., surface finish). The condition of the grinding wheel is one of the features of the grinding process that has the largest effect on the quality of the finished product. This research proposes a methodology to predict grinding wheel condition based on surface roughness (Ra). In this methodology, two different regression models based on the multilayer perceptron artificial neural network (MLPANN) and random forest (RndF) machine learning algorithms were first compared based on their R ² test values. Then, a study on the best performing models was conducted to identify a sensor (or combination of sensors) that best predicts tool condition. The results showed that the RndF-based regression model performs marginally better than the MLP-ANN based regression model. It was also found that the tangential component of force plays a significant role in determining Ra.