Small, fast runtime modules for probabilistic neural networks

An engine misfire detection algorithm based on the probabilistic neural network (PNN) has been developed using measured engine data. The PNN algorithm was used to develop a system that allows high classification accuracy while minimizing the dimensionality of the training database. An overall classification accuracy greater than 96% was achieved. The initial training data base consisted of 5060 feature vectors, each with 8 elements. The size of the training data was reduced, and therefore the runtime speed was increased, by approximately two orders of magnitude using maximum likelihood training. The classification accuracy was not significantly degraded by this reduction and overall accuracy remained approximately 95%