Study On Evaluation And Improvement Of Sound Quality Of Motorcycles Based On Neural Networks

Motorcycle's sound quality, an important criterion for consumers to choose vehicles, is one of the most important indicators of NVH performance, has been attracting more and more attention of consumers and producers. The new concept of noise control is that noise controlling should not only reduce noise sound pressure level, but also can adjust the sound characteristics of products, to achieve selective noise control, highlight the unique characteristics of this product, making the product's sound capture subjective experience requirements of consumers.Firstly, the preference subjective evaluation of 44 effective noise samples, which were selected from 10 different 150cc motorcycles at drive's binaural position, was carried out with the paired comparison method in this paper. And then, the preference subjective scores of each noise samples were acquired. The three main characteristic factors(energy factors, modulation factors, high frequency factors) were extracted by analyzing psycho- acoustic parameters of the 44 noise samples through the principal component analysis (PCA) and the three principal component values of the sound samples was calculated out. The neural networks method was used in the motorcycles sound quality evaluation, and the neural network model was trained thought setting the principal component analysis results of 44 noise samples as the input of neural network while the people's subjective feeling as the output, and the generalization ability of the model was verified. It shows that: the model is able to establish mapping relations well between sound quality parameters and the people's subjective feeling.Secondly, the subjective evaluation of motorcycle's sound samples which were filtered by different orders and frequency bands respectively was carry out, the influence of orders filtering and frequency bands filtering on motorcycle's sound quality was studied based on the statistical analysis of the evaluation results. It shows that: the motorcycle's sound quality can be significantly improved as long as the half order of energy of sound is remained only at 1500 rpm; the motorcycle's sound quality can be improved as long as enriching the order of energy of sound samples at 2000 rpm, 3000rpm, 4000rpm, 5000rpm; the motorcycle's sound quality can be significantly improved as long as the sound energy of low-middle frequency is filtered at 1500rpm, 2000rpm; the motorcycle's sound quality can be improved as long as the sound energy of high frequency is filtered at 3000rpm, 4000rpm, 5000rpm. Finally, the main distinctions of drive's binaural sound energy on different frequency bands between the test motorcycle and the benchmark was found out thought 1/3 octave comparative analysis. The main assemblies and its reasons which lead to motorcycle's poor sound quality were analyzed on the basis of the results of sound intensity experiment and pertinence improvements were given out. Comparison was made between prediction results of the evaluation model and subjective evaluation test on the sound quality level of the motorcycle with optimized muffler. It shows that: sound quality evaluation model has practical value, while the motorcycle's sound quality can be improved by optimized muffler.