THE AUTO-SYNCHRONIZED WAVELET TRANSFORM ANALYSIS FOR AUTOMATIC ACOUSTIC QUALITY CONTROL

A new feature vector for automatic acoustic quality control is presented and applied to the classification of electric sliding sunroofs at the quality control point by analyzing the sliding noise. The specific features of the sound signals (signatures) which distinguish products of different quality are not explicitly sought, because these signatures are highly specific for each application. Instead, the property is used that the relevant information about the sound signals can be resolved by the ears of experts. The time-frequency resolution of the ear is approximated by a wavelet transform of the signals. As a new approach to the important problem of noise reduction the concept of auto-synchronized wavelet transforms is introduced which allows wavelet transforms (and more general time–frequency representations) to be averaged in the time domain without losing the time-resolved information in the signals. By this averaging process, statistical fluctuations (noise, parameter drifts) can be reduced significantly to reveal the characteristic features of the signals. The classification can then be performed by a next neighbour search on a training set. The concept of auto-synchronized wavelet transforms is developed in a mathematically formal way and the properties of the feature vectors obtained are studied by using artificial noisy signals before applying this method to the experimental data.