Diagnosis of sensor faults in a combustion engine control system with the artificial neural network

Abstract

The work presents the investigations carried out on a spark-ignition internal combustion engine with gasoline direct injection. The tests were carried out under conditions of simulated damage to the air temperature sensor, engine coolant temperature sensor, fuel pressure sensor, air pressure sensor, intake manifold leakage, and air flow disturbances. The on-board diagnostic system did not detect any damage because the sensor indications were within acceptable limits. The engine control system in each case changed its settings according to the adaptive algorithm. Signal values in cycles from all available sensors in the engine control system and data available in the on-board diagnostic system of the car were recorded. A large amount of measurement data was obtained. They were used to create a statistical function that classifies sensor faults using an artificial neural network. A set of training data has been prepared accordingly. During learning the neural network, a hit rate of over 99% was achieved.