Subjective response simulation of brake squeal noise applying neural network approach

A non-linear neural network (NN) model is developed for simulating human subjective annoyance response to brake squeal noise as a function of both the occurring tonal frequency up to 18 kHz and the corresponding amplitude. The proposed approach produces a robust, multi-layered NN structure of the subjective response surface for the full squeal frequency range, which is more efficient and accurate than classical sound pressure weighting networks. It is also shown to be superior to the Zwicker loudness (calculated in sones using the standardized ISO 532-B algorithm) which completely ignores higher frequency contents. Measured interior noise and driver's subjective evaluations from numerous vehicle-braking tests are applied to train and evaluate the proposed NN model. The fully trained network is capable of predicting the expected subjective rating and identifies low response sensitivity regions. The predicted index that is essentially an estimation of squeal intensity may be used to evaluate brake design alternatives with respect to subjective annoyance.