FAULT RECOGNITION AND DIAGNOSIS IN INDUSTRIAL EQUIPMENT USING SENSORY DATA AND RECURRENT NEURAL NETWORKS

Abstract

Industries face a problem with unexpected failure of equipment and their parts, which often leads to operation stoppage and increased costs. In order to decrease abovementioned risks, it is needed to implement predictive maintenance techniques, by analyzing equipment condition in a real-time manner and evaluating its performance by applying data analysis techniques and machine learning algorithms. This Capstone project analyses bearings, commonly used in the rotary equipment and fault types associated with them. Retrieved vibration sensor data from University of Stavanger and Case Western Reserve University Bearing Data Center are preprocessed and analyzed in time-domain and frequency-domain for feature extraction. Extracted features are then used in the training and testing of Recurrent Neural Network (RNN) and Support Vector Machine (SVM) algorithms to diagnose the condition of the bearings and to identify the location of failure. As a result, the performance of RNN in comparison with SVM at different conditions was analyzed and further recommendations in this research field were provided.