Sensor Fault Diagnosis Using an Indirect Fuzzy Feedback Linearization and Decision Trees

Sensor fault diagnosis is a cutting-edge research area in recent years. In the paper, an indirect fuzzy feedback linearization technique is suggested for sensor fault diagnosis. The proposed scheme has three main steps as following: The preprocessing unit is the first step to design an indirect observer in which the feature extraction technique is used to extract the root means square (RMS) data from resampled fuel to air ratio signals. The normal RMS resampled data is modeled using the nonlinear data identification technique. The combination of autoregressive with external input, Laguerre input-output filter, and fuzzy algorithm, FAL, is suggested for normal resampled RMS data modeling. After modeling the data, the combination of the FAL modeling method, feedback linearization technique, and fuzzy estimator, FAL-FF, is suggested for RMS resampled data estimation. The fuzzy technique is used to estimate the nonlinear function in feedback linearization observer and improve the accuracy of estimation for data classification. Next, the residual resampled RMS fuel to air ratio data is generated using the difference between original and estimated data. Finally, the decision tree is used to classify the resampled RMS residual sensors data for fault diagnosis. Regarding the results, the accuracy of sensor fault diagnosis using the proposed scheme is 96.7%.