Misfire detection of a turbocharged diesel engine by using artificial neural networks

• Published in: Applied thermal engineering 2013-06, Vol.55 (1-2), p.26-32
• Main Authors: Liu, Bolan, Zhao, Changlu, Zhang, Fujun, Cui, Tao, Su, Jianyun
• Format: Article
• Language: English
• Subjects: Applied sciences; Artificial neural networks; Diesel engine; Diesel engines; Energy; Energy. Thermal use of fuels; Engines; Engines and turbines; Equipments for energy generation and conversion: thermal, electrical, mechanical energy, etc; Exact sciences and technology; Experimental study; Heat transfer; Mathematical models; Matlab; Misfire detection; Modeling; Neural network; Neural networks; Theoretical studies. Data and constants. Metering; Training; Turbochargers
• ISSN: 1359-4311
• DOI: 10.1016/j.applthermaleng.2013.02.032

This study presents a novel misfire detection model of a turbocharged diesel engine by using artificial neural network model. An explicit back propagation neural network has been developed to identify diesel combustion misfire according to the general engine operating parameters. The parameters are selected by using engine fault mode tree analysis. The proposed neural network model has been implemented in MATLAB/Neural Network Toolbox environment. Experimental study then has been performed on a V6 turbocharged diesel engine to get the parameters for both network training and validation purpose. Initial results show that misfire can be captured in most cases, however some mis-detection could happen though the mean square error of the model is satisfied. Furthermore, the in-cycle engine speed variation, a deductive parameter of transient engine speed, is added into the training data, which promotes the final results to full correct detection with no exception. The current study provides a new way to detect the happenings of misfire of turbocharged diesel engine. •We build a neural network model for diesel engine misfire detection.•We make fault topology of diesel engine misfire to compose the training vectors.•Detection veracity depends on adding more related information of training vectors.•Experimental research shows that the detection algorithm can identify engine misfire successfully.