Sound quality prediction for power coupling mechanism of HEV based on CEEMD-HT and RVM

Abstract

The sound quality evaluation model based on complementary ensemble empirical mode decomposition (CEEMD), Hilbert transform (HT) and relevance vector machine (RVM) is proposed to predict the sound quality of acoustical signals at power coupling mechanism of a hybrid electric vehicle (HEV). The technique is applied with wave filtering and CEEMD of the acoustical signals acquired at power coupling mechanism of HEV to achieve the intrinsic mode function (IMF) components. Built upon this is the calculation of the instantaneous frequencies of the IMF components by HT. Then, the critical frequency bands are used as the weight to calculate the weighted energy values of the IMF components. The weighted energy values are used as the new inputs of the sound quality evaluation model. Afterward, the subjective evaluation experiment of the acoustical signals at the power coupling mechanism is carried out based on pairwise comparison method. The subjective evaluation values are used as the outputs of the evaluation model. Finally, the new evaluation model is established based on RVM. In addition, the second RVM model is built for sound quality evaluation with the psychoacoustic objective parameters as the inputs. In this paper, the sound samples acquired under steady- and unsteady-state operating conditions are tested, respectively, in two models to obtain the prediction results. The prediction result suggests that the prediction accuracy of the evaluation model based on CEEMD-HT is higher than the evaluation model based on psychoacoustic objective parameters, and the prediction accuracy of the steady sound samples is higher than the unsteady sound samples.