Investigation of Machine faults using Elman Neural Network and Decision tree

This paper demonstrates the application of an expert system aiming the diagnosis of rotating machineries to recognize and analyze the sources of abnormal vibration. A tailor made test rig is fabricated for the purpose of experimental investigation, and it is validated with the vibration severity chart. This paper discusses the approaches of the Elman neural network and Decision tree in the investigation of the predominant faults in rotating machineries. The various characteristics and operating conditions are determined. Using the neural network optimum results is achieved when the number of hidden neurons, learning rate and the momentum factor are 5, 0.1 and 0.9 respectively. In a decision tree, classification of faults is performed with the help of a cause-symptom matrix using the classification and regression trees algorithm. The functioning of the network and decision tree in diagnosing the rotating machinery was investigated based on the accuracy and convergence. The occurrence of faults such as unbalance, misalignment, bearing defects and looseness are predetermined. Expert system provides the information at right time when the operator is unavailable. The performance of the two approaches are compared and presented.