Research On Clustering Diagnosis Theory And Method For Milling Condition Monitoring Based On Acoustic Emission Signals

The long life and high reliability requirements of aerospace equipment determine the machining process of such parts suffering high complexity,high risk and difficulty.Milling process plays an important role in ensuring the final working performance and the service life of these parts.On-line monitoring is an important way to ensure the high quality and efficient implementation of this process.This thesis research the monitoring theory and methods of the typical milling processes of aerospace structural components by means of the acoustic emission time-frequency analysis and clustering diagnosis based unsupervised pattern recognition.It is proved through experiments that acoustic emission signals are necessary in the monitoring of aerospace structural milling process.On this basis,a multi-resolution analysis method of acoustic emission based on discrete wavelet decomposition is established to extract the effective signal component of acoustic emission signals.A time-frequency spectrum acquisition method for acoustic emission signals based on the Short-time Fourier Transform was established.A method of normalization of timefrequency matrix is proposed to realize the characterization of the time-varying characteristics of the instantaneous frequency components of acoustic emission signals,which addresses the problem that the time-frequency spectrum is easily disturbed by signal energy fluctuations during milling.The feature set of acoustic emission signals based on time-frequency matrix is constructed systematically,including the energy based features,the time and frequency domain expansion features and the intrinsic features of time-frequency matrix.An unsupervised feature reduction method based on the cross correlation function and two filter-type feature selection indexes were constructed,and the optimal feature set was obtained.A clustering algorithm based on adjacent grid searching(CAGS)was created.The algorithm can automatically identify the number of clusters and has high execution efficiency since the computational efficiency is not affected by the number of samples.Compared with other grid-based clustering algorithms,the algorithm has higher clustering accuracy and robustness,and the input parameters have better interpretability.The effectiveness of the algorithm was evaluated using an international standard data set.The results show that the clustering algorithm proposed in this paper has the ability to handle noise,the ability to process large-scale data,the ability to process highdimensional data,the ability to handle complex shape clustering,the ability to handle clusters with large differences in density between classes,and the ability to handle clustering of linked cluster boundaries.It shows that this algorithm has universal advantages and is suitable for applications in engineering.Based on CAGS,an unsupervised condition monitoring method for monotonous multi-classification problems and two-class classification problems is established.It can be used in tool wear monitoring in corner milling and surface defects monitoring in nickel-base alloy finish milling.The evolution law of the distribution of clustering samples in the time-frequency feature space of acoustic emission with tool wear is discussed.Based on this,the cluster-density factor is proposed to characterize the tool wear conditions.The evolution law of time-frequency features values of acoustic emission with surface roughness and surface defects was analyzed.A two-step identification framework for surface defect monitoring in nickel-base alloy milling process was proposed and cluster monitoring and diagnosis of surface defects was realized.The research results of this paper help to realize the condition monitoring of aerospace structural parts with complicated profile and high precision requirements.It has important theoretical significance and practical value for improving the aerospace intelligent manufacturing level in China.