Research On Wood Acoustic Vibration Signal Based On Wavelet Packet Analysis And Neural Network

Vibration is the basis of transmission and radiation in vibration energy and sound energy of wood or wood composite materials.Only through the transmission and radiation of vibration signals can people hear the sound of musical instruments.Vibration signals collected from wood or soundboards contain a large load of information.How to analyze them in depth is of positive significance in the study of acoustic vibration characteristics of wood.At present,acoustic parameters such as dynamic elastic modulus E,acoustic attenuation coefficient σ,E/G,specific dynamic elastic modulus E/ρ,acoustic radiation quality constant R,acoustic conversion efficiency v/(σ·p),and acoustic impedance ω are mainly used to evaluate the acoustic vibration characteristics of the soundboard wood of musical instruments,and the analysis and research on the vibration signal of the soundboard wood has not been in-depth enough.In order to explore the relationship between the characteristics of wood vibration signals and its acoustic vibration performance,this study combines wavelet packet analysis methods to quantitatively study the paulownia and spruce wood vibration domain signals of musical instruments,and extracts the characteristic laws of the signals and correlates them with wood acoustic parameters,so as to provide a new idea for the evaluation of musical instrument wood from the perspective of signal analysis.The main contents and conclusions of this research are as follows:(1)Based on the vibration theory of the beam,two-ended free boundary conditions were used,and the dynamic elastic modulus E,sound attenuation coefficient σ,E/G,and specific dynamic elastic modulus E/ρ of paulownia and spruce were measured by an FFT analyzer,as well as Acoustic parameters(such as the acoustic radiation quality constant R,the acoustic conversion efficiency v/(σ·ρ)and the acoustic impedance ω).The acoustic signal was collected to obtain the basic data.(2)In the actual measurement of vibration signals,noise may be caused due to the shielding effect of the measuring instrument itself and the unstable power supply voltage of the surrounding environment.Therefore,it is a very important link to eliminate the noise component of the signal.This study based on the wavelet threshold method to de-noise the vibration signal of the test piece.The energy ratio and the root mean square error of the de-noised signal were used as evaluation indicators to quantitatively analyze the de-noising effect of the signal.The analysis results showed that:db4 wavelet base using the soft threshold algorithm of the wavelet transform heuristic threshold principle to perform 3-layer wavelet decomposition on the noisy wood vibration signal is most suitable for the preprocessing of the signal in this study,under which the energy ratio of the signal after noise reduction was maximized(0.9979),and the root mean square error smallest(0.4748).(3)Based on the wavelet packet analysis method,the pre-processed vibration domain signal is analyzed.By comparing the characteristic information of the wood vibration signal and the relationship between its acoustic vibration performance,it can be concluded that after the signal preprocessed,the correlation between the signal characteristics and the various acoustic parameters of Spruce and paulownia showed an increasing trend,,with a maximum increase of 8.64%;time-domain characteristic indicators(root mean square value,crest factor,kurtosis factor)and the energy rate of wavelet packet characteristics(0～500 Hz,500～1000 Hz frequency band)were both related to the acoustic vibration parameters of wood(dynamic elastic modulus E,sound radiation quality constant R,sound conversion efficiency v/(σ·ρ),acoustic impedanceω,the attenuation coefficient σ and the energy loss per cycle tanσ/E)significantly.The correlation coefficient is 0.702～0.942.The characteristic index of the signal was well related to the acoustic conversion efficiency v/(σ·ρ)and the energy loss per cycle tanσ/E.(4)Based on the Matlab using the BP neural network algorithm,a classification model for predicting the acoustic quality of wood vibration signal characteristics information was constructed,and the built BP neural network was used to verify the acoustic quality level prediction of musical instrument wood,The correct recognition rate of acoustic quality classification prediction of paulownia and spruce wood was 93.3%.The verification result showed that the verification error obtained by predicting the acoustic quality level of wood through the characteristic information of the vibration signal was relatively smaller,and the result can fully illustrate that the network’s has good predictability in the classification of the acoustic quality of wood.