Misfire Detection Including Confidence Indicators Using a Hardware Neural Network

The complexity of automotive power train and control systems is necessitating the implementation of advanced techniques, in turn placing an increasing computational load on the ECU systems.

Misfire detection is a pattern classification problem involving the complex, non-linear interactions of good combustion/misfire event distributions for multiple input signals.

Building on previously reported developments in these areas, this paper describes practical advances in misfire detection techniques that, through the use of hardware neural network technology, provide measures of “confidence” in the decision, and a single diagnostic metric for arbitration and calibration of the solution for a current series production engine.