A Recurrent Neural Networks based Method for Fault Diagnosis of Three-way Catalytic Converter

Three way catalytic converter (TWCC) is an important component in the process of vehicle exhaust after- treatment. The performance of TWCC is often affected by a number of serious faults that are not easy to be detected. Thus, intelligent fault diagnosis for TWCC is necessary. In our proposed approach, two recurrent neural networks(RNN) are implemented for extracting the fault features from the original experimental data and diagnosing the faults of TWCC respectively. The first RNN can decompose TWCC state data information structurally and complete the refined description of state information. The second RNN have a fully connected structure of feedback networks that can be used for fault classification. In this case, TWCC can be perfectly diagnosed by our proposed approach. Experiments also illustrate the merits and effectiveness of the proposed approach.