Research On Abnormal Sound Detection Algorithm Of Machine Tool Based On Feature Fusion And OCSVM

With the vigorous development of China’s real economy,the number of production orders for many products has increased dramatically,and the demand for production equipment CNC machine tools has also increased dramatically.Due to the complex structure of CNC machine tools and the precision structure of parts,it is easy to be affected by various factors when working.In addition,the use of CNC machine tools for a long time without stopping leads to the occurrence of machine tool failures,resulting in abnormal work of the machine tools,and affecting product quality.Aiming at the difficulty of machine tool fault diagnosis and detection,a method of fault diagnosis and SVM fault classification using audio fusion features combined with One Class SVM(OCSVM)is proposed.The normal working and nine kinds of abnormal working audio of the machine tool are collected as data sets for analysis.Experiments show that the machine tool fault can be diagnosed accurately based on audio feature fusion and OCSVM,the diagnostic accuracy can reach 96.1%,and the classification accuracy can reach 93.3% through SVM.The main contents of this paper are as follows:(1)Collection and pretreatment of machine tool working audio.In view of different kinds of audio characteristics during machine tool operation,the microphone made by Brüel & Kj?r(BK)Company is used to replace the traditional vibration sensor to obtain audio signals.This non-contact acquisition method is more suitable for the working environment of the research target.The collected audio is preprocessed to reduce noise interference,such as pre-emphasis,framing and windowing.(2)Feature extraction,feature dimension reduction and feature fusion are carried out for the working audio signal of machine tool.Mel Frequency Cepstral Coefficient(MFCC)and Linear Predictive Cepstral Coefficient(LPCC)were used.LPCC extracted features from the audio signals of machine tools,and then compared three dimensionality reduction effects,including T-distributed random neighborhood embedding(t-SNE),uniform manifold approximation and projection(UMAP)and principal component analysis(PCA),to select the method with better effect for dimensionality reduction,so as to achieve the feature purification of audio signals and shorten the running time.Finally,conduct feature fusion after using standard normalization of features to expand the audio feature information and get more key features of an audio signal.(3)Intelligent detection and classification.The modular method is used to detect and classify the feature data.First,based on the abnormal sound detection of OCSVM,judge whether the audio is normal or abnormal.If abnormal audio is found,the data is classified for the second time.Through the comparison of support vector machine(SVM)and K nearest neighbor(KNN)methods,a method with a good classification effect is selected to determine the fault category of the machine tool.The experimental results show that the overall detection and classification effect is more than 93%,which can effectively detect abnormal points in batch samples,timely feedback to reduce losses and achieve the expected effect of engineering practice.(4)The construction of the detection system platform.Use Python language to build a graphical user interface(GUI)operation platform on the Py Charm platform using the PyQt5 module,including the expected template,implementation process and practical operation.After the actual detection of the data,the operation is simple,it can test the effectiveness of different algorithms and detect and classify different data sets.