Sound Quality Prediction Technology Research For Diesel Engine Radiated Noise Based On EEMD-HT And LSSVM

Diesel engine radiated noise quality directly affects people's physical and mental health as well as the product orientation,and it has become an important research subject in power machine vibration and noise control research fields.Consisting of the subjective and objective evaluation of diesel engine noise quality,some basic theoretical researches are carried out.The level score method is adopted to improve the subjective evaluation.Ensemble empirical mode decomposition(EEMD),Hilbert transform(HT)are used to extract the main objective features of diesel engine noise quality.Least squares support vector machine(LSSVM)is used to predict sound quality for diesel engine radiated noise;Prediction model for sound quality was also built.With sound quality prediction model,the subjective evaluation wasting time and energy is not needed.First,a subjective evaluation test on the diesel engine radiated sound quality is conducted using level score method.In the semi-dead room,this paper gathers four diesel engines noise signals of five different working conditions.The sample signals are preprocessed by Audience software and subjectively evaluated by level score method.The results are reliability analyzed by SPSS software.The loudness,roughness and sharpness are calculated by LMS software.The correlation analysis indicates that there is significantly positive correlation between subjective fidgety degree and loudness,roughness,sharpness.Then,a sound quality objective evaluation method for diesel engine radiated noise is proposed in this paper based on the EEMD and HT.The basic algorithms of empirical mode decomposition(EMD)and HT are studied.A group of simulation signals similar to diesel engine radiated noise signals are used to evaluate the recovery performances of EMD and EEMD methods.A frequency hopping simulation signal is used to evaluate the time-frequency analysis performances of CWT and HT methods.They verify that EEMD method can weaken the model aliasing problem,and HT can be more accurate positioning in time-frequency joint distribution.Therefore,the sound signals of diesel engine are sampled,filtered,decomposed by EEMD and transformed by HT;then the corresponding intrinsic mode function(IMF)components are weighted by critical frequency band with the instantaneous frequency,their energies are calculated as the objective evaluation features of sound signals.Finally,a sound quality prediction method for diesel engine radiated noise is proposed in this paper based on LSSVM.In order to compare prediction model,the multivariate linear regression model(MLR),BP neural network model,LSSVM model was built.What is more,by using LSSVM for training with the energy features of sound signals as input and the subjective evaluation values as output,a sound quality prediction model for diesel engine radiated noise was obtained.Meanwhile another prediction model for sound quality based psychology parameters was also built with psychology parameters as LSSVM input.The results of predictions on the diesel engine radiated noise show that the prediction model based on EEMD-HT and LSSVM in this paper get a higher accuracy,meaning it has much closer to subject feeling of people.