D. Michael McFarland / Zhejiang University of Technology
Huancai Lu / Zhejiang University of Technology
The machining and assembly errors of automobile seat horizontal adjusting mechanism (SAM), as well as tooth surface burrs, will lead to abnormal noise during the operation of the product, which will damage the driving and riding experience. Previously, this sound quality was detected by the subjective human ear, a traditional psychoacoustic index model, and an objective parameter evaluation index of sound quality.
In response toa demand for in situ sound quality detection of the SAM and the poor anti-noise performance of the evaluation model, a sound quality recognition model of the SAM with strong anti-noise ability was proposed based on small sample data.
By mixing various environmental noises with different signal-to-noise ratios as the input dataset, a convolutional neural network (CNN) was used to fit the generalized complex ratio masking matrix, and the signal-to-noise ratios of SAM acoustic signals under various environmental noises was increased to 28dB. The 2:1 mixed dataset of denoised samples and pure samples was used as the input of the recognition model. The recognition rate was 98% in the low-noise environment and 80% in the high-noise environment, which relaxed the requirement that the sound quality detection must be carried out in an anechoic environment.