Sensor fault detection, isolation, accommodation and unknown fault detection in automotive engine using AI

Sensor fault detection, isolation (FDI) and accommodation has been investigated along with detection of unknown faults for an automotive engine. Radial basis function (RBF) neural networks are used for fault diagnosis. The RBF network is trained off line with K-means and batch least squares (BLS) algorithms. No fault and fault data are simulated in Matlab for four different sensors e.g. throttle angle position, crankshaft speed, and inlet manifold pressure and temperature sensors. All the sensors are investigated for ten percent positive and negative bias faults and also for unknown faults. Simulations show satisfactory results for FDI. Further, the fault accommodation for three sensors is also investigated using predictive neural networks and the results with acceptable levels of errors are achieved.