Study On Laser Cutting Quality Detection And Evaluation Based On Machine Vision

Laser cutting is a kind of processing technology which used most widely in the laser processing industry.The quality of laser cutting,especially the roughness of the cutting surface,will directly affect the further processing of the workpiece,and ultimately determine the performance and quality of the product.With the development of industry,the progress of science and technology and the use of new processing technology,the precision of parts and components requirements are higher and higher,so the need for laser cutting quality monitoring is more and more extensive.At present,the detection methods of laser cutting quality include traditional manual detection and sensor detection.These detection methods have many problems.Firstly,manual detection requires personnel with relevant professional knowledge to observe or measure,which is inefficient.Secondly,the traditional sensor monitoring systems usually use sound and light sensors,which have some limitations such as high cost,low accuracy and no quantitative measurement.Machine vision is a non-contact detection method based on image processing technology,it is easy to operate,own high efficiency and flexibility,and can improve the automation of manufacturing facilities.Moreover,machine vision can acquire a large amount of information quickly and integrate information easily.It is one of the basic technologies to realize equipment intelligent.On the basic of studying the above problems,firstly,the paper purpose the evaluation index of laser cutting quality – roughness and the monitoring object of machine vision--laser cutting sparkle jet based on study of experiment.Then,the camera calibration,image segmentation and image feature extraction were completed on the machine vision detection platform.Finally,the mathematical model of laser cutting quality visual inspection was found by using statistical methods.It is of great significance to ensure the cutting quality,improve production efficiency,optimize the laser cutting process parameters and intelligent equipment.The main work and results of this paper are as follows:(1)On the basis of experimentally and theoretically studying the distribution of laser cutting quality characteristics,the paper proposes the laser cutting surface bottom edge's roughness as the quality evaluation index,the laser cutting sparkle jets' signal characteristics as monitoring objects,and then develop a quality detection and evaluation method with using visual single as independent variable and using roughness as dependent variable.(2)A side-axis visual detection platform was built in order to obtain effective visual signals.Aiming at the errors caused by camera installation and lens distortion,monocular camera calibration was completed through modeling and analysis.In order to restrain the noise interference in the process of digital image acquisition and transmission,the image smoothing algorithm was studied and verified by experiments to select an appropriate algorithm for the paper.(3)In view of the characteristics of laser-cut spark cluster images and the complex environment in the industrial scene,the highlight luminescent image is easily disturbed by halo noise,and a multi-scale image segmentation algorithm of spark cluster based on color and wavelet texture is proposed.Firstly,In this algorithm three-level wavelet decomposition by Daubechies and transformation from RGB to HSV color space are carried out for the RGB image collected by the camera.Wavelet high-frequency components are used to describe texture features,and H channel and V channel of HSV color model are used to describe color features.Secondly,the color-texture feature matrix T was constructed.In order to improve the operation speed of the algorithm,the matrix elements were compressed by block,and k-means initial clustering segmentation was carried out in the compression scale space.Then by calculating the gradient Angle and modulus,the edge detection of the image segmentation results in compressed scale space is realized and mapped to the original scale space to construct the feature matrix.Finally take the initial clustering center as the local clustering center and use k-means clustering method again to obtain the image accurate segmentation results at the original scale.The experimental results show that the proposed algorithm can achieve accurate and effective segmentation of the laser cutting sparkle jets' camera image.(4)In order to quantitatively describe the characteristics of lase cutting sparkle jet signals,a robust feature descriptor of visual signals was calculated by modeling and analyzing sparkle clusters,and a feature extraction algorithm of laser cutting sparkle jet signals was proposed.The algorithm based on the improved connected components labelling method labelled the brightest part of laser cutting sparkle jet and core jet part and maximum jet part.According to the established image geometric model the quantitative description of the machine vision monitoring objects like angle of sparkle jet and image gray value moments are calculated by use of PCA and fast convex hull.The experimental results show that the characteristics of the sparkle jet signal extracted by the paper algorithm are invariant in rotation,translation and scale,and the relative error is very small.The quantitative detection and evaluation of laser cutting quality can be realized by the descriptor of the signal characteristics.(5)The corresponding relationship between the laser cutting quality and the characteristics of the spark jets visual signals were revealed through the experimental study on the visual detection of the sparkle jet behavior of laser cutting and the experimental study on the cutting surface quality(roughness)of different process parameters.Within a certain range of process parameters,the gray value moments result in the region with the highest luminance of the spark jet changes is an inverted u-shaped curve with the change of the cutting speed.When the gray moment is the largest,the roughness of the cutting surface bottom edge is the smallest.Now,the Angle(?)of the sparkle jet is close to the vertical,and the cutting speed is the optimal speed.(6)By statistically analyzing the data from the experiment of machine vision and cutting quality,a multiple regression mathematical model of laser cutting quality visual detection was established based on the improved stepwise regression method,and a laser cutting quality evaluation method was proposed.The experimental results show that the prediction error of the visual quality detection model is very small,which verifies the validity of the mathematical model and the feasibility of the evaluation method.The paper utilizes the advanced machine vision technology to implement laser cutting quality detection and evaluation within a certain range of laser process parameters.This has important theoretical significance and application value to shorten the visual detection algorithm development time,real-time monitor the laser cutting quality,save labor costs,improve production efficiency and laser cut equipment intelligent degree.The present dissertation has 103 figures,17 tables and 169 references.