Fault Diagnosis of Planetary Gear Carrier Packs: A Class Imbalance and Multiclass Classification Problem

Fault diagnosis plays a key role in monitoring manufactured products for the purpose of quality control. Among the several fault diagnosis approaches, knowledge-based fault diagnosis, which uses signals from sensors and machine learning algorithms instead of a priori information, is widely employed to diagnose the status of products. In this paper, we propose a knowledge-based procedure to establish a fault diagnosis model. The model is aimed to diagnose planetary gear carrier packs, which have many fault types and an unbalanced number of samples in the sample classes, using transmission error. In the procedure, the best feature subset that contains the most important features is selected using two different feature selection processes. Several ensemble algorithms are used during the model training process. The imbalance ratio between classes of samples is addressed. The number of weak learners is automatically determined by a genetic algorithm. Finally, the performance of the proposed procedure is validated by comparison with other models trained without applying the proposed procedure. We observed that it is important to incorporate the class imbalance technique in the training process as it reduces the misclassification of faulty products as normal ones. This reduction is important in production quality control.