Research On Abnormal Sound Source Identification And Location Based On Machine Learnin

Under the background of combustion industry,from the perspectives of safety production and energy efficiency improvement,the theory of abnormal sound source identification and location is applied to the research of combustion diagnosis technology.The abnormal audio signal in combustion state is identified through the classification model of machine learning,and its location is determined by combining the sound source location method.According to the different forms of fuel and air mixing,the flame types can be divided into premixed flame with complete mixed combustion and diffused flame with incomplete mixed combustion.In this paper,microphones are used to collect premixed flame audio and diffused flame audio when butane and air are mixed in Bunsen burner.In order to increase the diversity of data samples,air flow sound of Bunsen burner flameout is collected as data supplement.Firstly,by preprocessing the audio data in three working conditions,the two-dimension spectrogram is obtained.In order to further display the differences of the three kinds of audio,the two-dimension spectrogram is converted into a three-dimension spectrogram,which is used as the input in the classification model of support vector machine to construct the audio recognition model of burning state.According to the experimental results,the premixed flame audio can be significantly different from the other two kinds of audio,because the premixed flame audio has a stable combustion process and little fluctuation of sound amplitude.The three dimensional spectrograms of diffused flame audio and flamed air audio are somewhat similar due to inadequate mixing of combustion and chaotic sound information,but they are different in time dimension and frequency dimension.Therefore,the premixed flame audio can be better distinguished from other abnormal audio,which realizes the identification of abnormal audio in this paper.After the identification of combustion audio signal,the diffused flame audio is used as the abnormal sound source to estimate its location.In this paper,time delay estimation method and hyperbola knowledge are combined to realize spatial localization.What the microphone receives is the original information of the sound signal.It is necessary to conduct generalized cross-correlation processing on the original data,so as to obtain the relative delay between the microphones,and then estimate the location of the sound source according to the geometric relationship between the sound source and the microphone array.From the experimental data obtained,it can be seen that the basic localization effect can be achieved when the sound source localization is estimated within the optimal collection distance of the microphone.