Dempster-Shafer-Based Sensor Fusion Approach for Machinery Fault Diagnosis

Artificial intelligence (AI) has played an increasingly important role in condition monitoring and machinery fault diagnosis in power generation plants. However, the accuracy and reliability of any AI-based machinery fault diagnosis is highly dependent on the quality and quantity of the input data fed to the AI model. The hypothesis of this paper is that AI-based fault diagnosis can be further improved by taking into account all the available sensor inputs of the machine. In short, the more sensor inputs fed into the AI model, the more accurate and reliable the outcome of the fault diagnosis. This paper proposes an application of Dempster-Shafer (DS) evidence theory for sensor fusion to improve the accuracy of decision-making in machinery fault diagnosis, by fusing all the available vibration signals measured on different axes and locations of the test machine. Vibration signals from different axes and locations of a machinery faults simulator were collected by multiple accelerometers simulating various machinery health conditions, namely healthy, unbalance, misalignment and foundation looseness. The accuracy of fault diagnosis using a different number of sensor inputs was then investigated. Analysis results showed that by combining more sensor inputs using a DS-based algorithm can improve fault detection accuracy from an average of 63% to 83%. In conclusion, the multi-sensor fusion algorithm can be applied to increase the accuracy and reliability of AI-based fault diagnosis.