Neural networks of engine fault diagnosis based on exhaust gas analysis

Abstract EN

This work uses the Artificial Neural Networks (ANNs) for fault diagnosis of a single cylinder four stroke gasoline generator type (Astra Korea AST11700).

One normal and fourteen faulty conditions are examined experimentally to produce a realistic data set, which is to be used for the training and validation of the ANNs.

The resulted data was in the form of exhaust gases and engine speed records for each case separately under different loading conditions.

After the learning process is completed, the ANN becomes able to make a diagnosis about the gasoline engine condition when new data is presented.

The data presented to the ANN system include a subset of engine faults which were selected and executed experimentally for this topic.

These include, faults in carburetor, air filter, spark plug, valves, piston rings, etc.

The results showed that the multi-layer training algorithm is sufficient enough in diagnose engine faults under different loading conditions.

It was found that the correlation coefficient values are 0.999 and 1 for the testing and training data, respectively.

The results obtained in this investigation showed that the ANN-based fault diagnosis system is capable of fault diagnosis with high reliability.