Diesel engine condition monitoring using a multi-net neural network system with nonintrusive sensors

A multi-net fault diagnosis system, designed to provide power estimation and fault identification, is presented. Eight different faults: misfiring of each of the three cylinders, shaft imbalance, clogged intake and a leaking start plug for each cylinder were separately induced in the engine. Only the vibration and exhaust temperature sensors were used during the tests. The data obtained were applied to train a three-level multi-neural network system designed to estimate the load of the engine, its condition status (between failure and normal performance) and to identify the cause of the failure. Each level of the NN system uses different data to obtain the best individual and overall prediction accuracy. The results of this work prove that with the aid of neural networks, the load of a diesel engine and its condition can be stated based only on a few signals obtained from nonintrusive measurements, such as the exhaust temperature and surface vibration.