Deep learning based bearing fault detection and classification via diversified inputs

Electric motors are preferred in many industrial areas due to their stable performance and ease of use. However, bearing faults are among the most common problems within motor failures. The early and accurate diagnosis of bearing disorders is critically important for preventing costly failures and for operational efficiency. The development of fault diagnosis techniques enables malfunctioned parts to be replaced at low cost or to be repaired effectively on-site, thus extending the motor's lifetime and reducing maintenance expenses. Studies in the current literature have generally relied on basic categorizations and rudimentary methods for fault detection. These approaches often use a limited number of sensor data, which leads to limitations in the continuity and accuracy of diagnoses. Within the scope of this study, a special sensor board developed at the University of Dallas will be used to analyze eight types of data obtained from motors and bearings. Our goal is to accurately classify six different conditions, including five different types of bearing faults as well as a healthy state. Based on the experimental data obtained for this purpose, we will assess which data are more informative and predictable, and examine performance and accuracy rates with different input combinations and modeling techniques. Thus, the current study will present a new and more comprehensive methodology that addresses gaps in the existing literature and contributes to the advancement of the field in the diagnosis and monitoring of bearing faults. The results of this study will significantly contribute to increasing the reliability of electric motors used in industrial applications and to the optimization of failure management and maintenance strategies.