Abstract:
The paper presents a new method for knock detection based on two neural networks. First a discrete Hopfield network extracts features from the structural vibration signal. Then the first and the forth coefficients of the autoregressive model and the maximal and minimal value of the signal are applied to a feedforward neural network in order to detect the knock. Once a cycle has been detected as a knock containing one, for the next cycle the engine can be protected in order to avoid further appearances of knock. For the feedforward neural network it was experimentally determined that four neurons in the hidden layer is the best solution for the knock detection.
