Feature Selection for Fault Diagnosis Using Fuzzy-ARTMAP Classification and Conflict Intersection

In automotive industry the safety of cars behavior is monitoring using computers. The information acquired on the bus communication is often redundant and not relevant. Therefore in the case of faults detection and isolation based on machine learning model, we need to reduce the number of variables according with their relevance and allowing taking decision in real time. In this paper, we propose a new approach for feature selection using fuzzy-ARTMAP classification and conflict characterization in fault diagnosis process. This approach is realized in two stages. In the first one, we classify the unfaulty functioning data of system using the fuzzy-ARTMAP classification. In the second stage, a conflict is accounted between features of test data based on the hyper-cubes resulted in the first stage. Two features are in conflict if her intersection does not belong to the model elaborated by fuzzy-ARTMAP classification. This approach is applied with success in automotive application in which the relevant features are detected and isolated.