Research On Laser Ultrasonic Metal Surface Defect Recognition Based On Random Forest

Metal workpiece has been widely used in many fields,but in the process of its production and use,due to the uneven force and other factors,will cause the workpiece surface crack.The continuous use of workpiece will cause further deepening and expansion of cracks.Failure to timely check will have an impact on the reliability and safety of the workpiece,which will lay a safety hazard and cause a huge safety accident if it is serious.Therefore,the detection of metal surface defects is very important.Laser ultrasound has the advantages of high sensitivity and non-contact detection.In this paper,laser ultrasonic detection technology is used to study the location of micro-cracks on metal surface,the feature extraction technology of cracks and the identification technology of crack defects.Aiming at the defect location research,this paper uses the scanning laser line source detection method(SLLS)to scan the surface defects of aluminum plates,and combines the B-Scan data obtained by scanning to study the defect location problem,and compares the direct method and the indirect method.Two kinds of positioning methods were used to verify the accuracy of indirect method positioning through experimental data comparison.As for the feature extraction technology of surface crack,this paper studies the signal carrying the crack information.Firstly,the time-frequency characteristics of different types of crack and different depth crack are analyzed.Secondly,EMD method is used to decompose the defect signal to obtain the corresponding energy distribution.Finally,the input of the classifier is composed of the amplitude feature in time domain and the peak feature in frequency domain.For the identification of surface crack type and depth,this paper mainly studies the feasibility and effectiveness of crack identification based on random forest(RF)algorithm and three decision tree algorithms(ID3,C4.5 and CART).For the identification of crack type and depth,the transmission signals of different types of cracks and the reflection and echo signals of different depths and transmission signals were extracted respectively to build the identification model.Under each identification model,the identification performance of the four algorithms for crack type and depth was compared.In this paper,it is found that the random forest algorithm and the three decision tree algorithms can effectively identify the type and depth of cracks,and the recognition accuracy of the random forest algorithm is more than 90%,which is obviously better than the other three algorithms.Stability is also better than the other three algorithms.This paper provides a theoretical basis and a new research method for further research on the identification of surface crack defects.