A New Intelligent Fusion Method of Multi-Dimensional Sensors and Its Application to Tribo-System Fault Diagnosis of Marine Diesel Engines

Marine diesel engines, a critical component to provide power for entire ships, have been received and still need considerable attentions to ensure their safety operation. Vibration and wear debris analysis are currently the most popular techniques for diesel engine condition monitoring and fault diagnosis. However, they are usually used independently in practice, and limited work has been done to address the integration of data collected using the two techniques. To enhance early fault detections, a new fault diagnosis technique for the marine diesel engine has been proposed by the information fusion of the vibration and wear particle analyses in this paper. A new independent component analysis with reference algorithm (ICA-R) using the empirical mode decomposition based reference extraction scheme was adopted to identify the characteristic source signals of the engine vibration collected from multi-channel sensors. The advantage of this approach performed at a data fusion level is that the ICA-R can extract only the relevant source directly related to the engine fault features in one separation cycle via incorporating prior knowledge. The statistical values of the recovered source signals were then calculated. The above vibration features, along with the wear particle characteristics, were used as the feature vectors for the engine fault detection. Lastly, the improved simplified fuzzy ARTMAP (SFAM) was applied to integrate the distinctive features extracted from the two techniques at a decision level to detect faults in a supervised learning manner. Particularly, the immune particle swarm optimization was used to tune the vigilance parameter of the SFAM to improve the identification performance. The experimental tests were implemented on a diesel engine set-up to evaluate the effectiveness of the proposed diagnosis approach. The diagnosis results have shown that distinguished fault features can be extracted and the fault identification accuracy is satisfactory. Moreover, the fault detection rate of the integration approach has been enhanced by 16.0 % or better when compared with using the two techniques separately.