A Study of Failure Prediction of Gearbox Rolling Bearings of a Bar Finishing Rolling Mill Using Artificial Neural Network

This study addresses the issue of premature bearings failures in the gearbox of a finishing bar rolling mill at the Gerdau steel plant in Barão de Cocais by applying machine learning techniques. The methodology combined K-means clustering with an artificial neural network (ANN) classification model for predicting motor current patterns and failure. The clustering method adopted was the Hartigan- Wong algorithm configured with four groups. The ANN was structured with eight neurons in the hidden layer and five neurons in the output layer, employing ReLU and sigmoid activation functions, respectively. The error metric used was categorical-crossentropy, and the optimizer adopted was Adam. Four test groups were conducted on the ANN model, varying the activation function in the output layer, the number of neurons in the hidden layer, the error function, and the network optimizer. The best-performing model achieved 99% accuracy, with sensitivity and specificity of 0.9871 and 0.9942, respectively, confirming its effectiveness in anomaly detection. A second dataset reinforced the hypothesis that bearing failures increases torque and, consequently, motor current. The proposed approach proved effective in detecting this relationship, establishing a reliable tool for predictive maintenance and preventive shutdown planning.