Fault diagnosis of diesel engine information fusion based on adaptive dynamic weighted hybrid distance-taguchi method (ADWHD-T)

In the case of the fuzzy correlation between variables of faults due to the complex monitoring information of diesel engines, neither Mahalanobis distance (MD) nor Euclidean distance (ED) can effectively diagnose by features, in this paper, a novel Adaptive Dynamic Weighted Hybrid Distance-Taguchi method (ADWHD-T) is presented to diagnose diesel engine faults. The method used Adaptive Dynamic Weighted Hybrid Distance (ADWHD) to fuse data of many sensors into the single system-level performance index. The ADWHD is an adaptive and dynamic weighting of MD and Standardized Euclidean distance (SED). The adaptive weights are adjusted according to the distance scale of MD and SED. The dynamic weight coefficients are calculated by the correlation coefficient of characteristic variables to consider the correlation and independence of characteristic variables. The diagnosis results are derived according to the optimized and adjusted fault threshold of ADWHD defined by 3 sigma method. In view of the dimension reduction optimization of characteristic variables, combining Taguchi method (T), ADWHD-T provides one systematic method for determining the key parameters of characteristic variables to solve the cost problem of multi-sensor analysis. Aiming at the real-time diagnosis, offline-online modeling and real-time fault diagnosis program based on ADWHD-T are designed. Quoting real-time data from diesel engine benches verifies the effectiveness of the scheme. Compared with MD and MD-T methods, ADWHD-T could promote diagnosis efficiency, enhance classification accuracy and expand its application range in fault diagnosis.