Determination of Engine Misfire Location using Artificial Neural Networks

• Published in: International journal of vehicle structures and systems Vol. 11; no. 4; pp. 407 - 412
• Main Authors: Ghazaly, Nouby M., Abdel-Fattah, Muhammad, M. Makrahy, Mostafa
• Format: Journal Article
• Language: English
• Published: Chennai MechAero Foundation for Technical Research & Education Excellence 08.01.2020
• Subjects: Algorithms; Artificial intelligence; Artificial neural networks; Back propagation; Classification; Coils; Computer simulation; Cylinders; Defects; Fault diagnosis; Ignition systems; Induction coils; Neural networks; Propagation; Recognition; Spark ignition; Vibration analysis
• ISSN: 0975-3060; 0975-3540; 0975-3540
• DOI: 10.4273/ijvss.11.4.13

Misfire in spark-ignition engines is one of the major faults that affect the power produced by the engine and pollute the environment and may cause further engine damage. This paper presents an evaluation of an artificial neural network based performance system through three most popular training algorithms namely Gradient Descent, Lavenberg-Marquadt and Quasi-Newton to determine the misfire location. Misfire is simulated by removing ignition coil to that cylinder namely Cylinder 1,2,3,4 and Cylinders 1 and 2, 1 and 4 and 2 and 3 with three different conditions such as idle, 2000 rpm and 3000 rpm. The results showed that the Quasi-Newton is higher in recognition rate average of 98.19 % but it takes more time to train. The Lavenberg-Marquardt algorithm is also good with an average recognition rate of 96.09 % with the fastest performance than Quasi-Newton. The gradient descent algorithm requires the network size to be more complicated to perform well with least time and high recognition rate.