Research On Acoustic Monitoring Of Laser Paint Removal Effect Based On Wavelet Packet Transform And K-means Clustering

Laser paint removal is one of the important applications of laser cleaning.However,in the actual production application,the surface cleaning state of the object in the process of laser cleaning largely depends on the observation and judgment of human eyes,and the lack of effective online monitoring means greatly restricts its intelligence.In the process of laser paint removal,the interaction between laser beam and paint layer produces plasma sound wave,which changes with the change of paint layer in the cleaning process,which makes it possible to use acoustics to monitor the cleaning state of substrate surface.In order to realize the acoustic online monitoring of paint removal effect,an experimental platform is set up in this paper.Through the collection and analysis of paint removal sound signals,the acoustic monitoring model of laser paint removal effect is studied by unsupervised machine learning method,and the overall monitoring strategy of laser paint removal acoustics is studied,which can realize the effective monitoring of laser paint removal effect.The main research contents are as follows:(1)The principle of acoustic monitoring of laser paint removal effect and the superiority of unsupervised machine learning method in acoustic monitoring are discussed.For laser paint removal with diverse surface cleaning conditions,it is difficult to characterize some cleaning conditions artificially,so unsupervised machine learning method has quantitative advantages in monitoring paint removal effect.According to the requirements of acoustic monitoring of paint removal effect,a laser paint removal acoustic monitoring experimental platform is set up,which can provide real-time,stable and accurate paint removal acoustic signal data for paint removal acoustic monitoring.(2)In order to follow the law of acoustic signal itself to the greatest extent,the best acoustic signal data was obtained from the aspects of sample preparation,the criteria for judging the surface cleanliness of the sample and the experimental flow,and the whole surface acoustic signal was segmented by micro-element plane,which was analyzed by a more refined wavelet packet transform,and the characteristic parameters were extracted from the reconstructed signal to represent each micro-element surface acoustic signal.(3)In order to realize efficient and unsupervised cleaning condition monitoring,principal component analysis(PCA)and K-means clustering algorithm are adopted in the paint removal effect detection model,and a reference library suitable for clustering detection and analysis is trained.Four kinds of clean states were determined in the experiment,and the micro-element surface proportion method was used to effectively detect the clean state of the sample surface.The detection effect of this model reached the expected requirement,and the detection accuracy of the most important "clean" state reached 95.98%.(4)In this paper,the overall process of acoustic monitoring for paint removal is designed,and considering the efficiency of cleaning and monitoring,the problems of hysteresis and damage caused by high laser energy density cleaning are solved by strategy research.Finally,according to the research results of this paper,the overall monitoring strategy is formulated to realize all-round monitoring of laser paint removal effect.