Study On Defect Detection Method Of Metal Additive Manufacturing Based On LIBS

Additive Manufacturing(AM)technology is one of the most brilliant jewels embedded in the crown of the manufacturing field.In the past decade,with the development of metal AM technology,it has gradually been applied to the manufacture of high-precision metal components,becoming an important part of the global manufacturing industry.However,due to the unique manufacturing process of metal AM components,various defects are inevitably present,posing a threat to the safety of high-precision industries.Therefore,it is of great significance to detect and identify various types of defects in metal AM components,which can not only improve the utilization of materials but also ensure the safety of industry applications.However,traditional defect detection methods such as eddy current and ultrasonic testing have high detection accuracy,but they are limited in their ability to detect a single type of defect,have high time costs,and are complex in detection methods.Therefore,there is an urgent need to explore a fast and efficient detection method for identifying different types of defects in metal AM components.Laser-induced breakdown spectroscopy(LIBS)technology is a multi-element analysis technique based on atomic emission spectroscopy,which has the characteristics of rapid,real-time,low-damage,and adaptability to various environments.In this study,LIBS technology was used to obtain spectral data of different defects in metal AM components.Data processing and machine learning classification algorithms were applied to reduce the difficulty of analyzing spectral information and improve the accuracy and precision of component defect identification.The following research results were obtained:(1)A binary detection method for metal AM component defects based on LIBS technology combined with machine learning algorithms is proposed.Firstly,an experimental platform was set up to collect spectral data of metal AM component defects and defect-free samples.Secondly,denoising,baseline correction,normalization,and dimensionality reduction methods were used to process the collected spectral data of metal AM components,and the clustering effect was analyzed by PCA three-dimensional plot.The results show that the data points of defect-free samples are relatively clustered with obvious clustering effect,but some defects overlap with the data points of defect-free samples.To address this issue,four classification models(Naive Bayes,K-Nearest Neighbor,Decision Tree,and Random Forest)were constructed for metal AM component defects and defect-free samples,and the classification results of the four algorithms were compared and analyzed.The results show that the Random Forest model has the best classification performance,with training and testing set classification accuracies of 98.75% and 98.33%,respectively.This proves that the application of LIBS technology combined with machine learning algorithms in the field of metal AM component defect detection is feasible.(2)A multi-classification detection method for different types of metal AM component defects(cracks,bulges,and holes)based on LIBS technology combined with SVM algorithm is proposed and experimentally validated.Firstly,spectral data of metal AM component cracks,bulges,holes,and defect-free(control group)samples were obtained.Then,the smoothed data were processed by four methods: no processing,mean centering,first-order derivative,and second-order derivative,and the clustering effect was analyzed by PCA three-dimensional plot.The results show that compared with no processing and mean centering,the data processed by first-order derivative and second-order derivative have better clustering effect in the three-dimensional plot,but there are still some data overlaps,and the data processed by first-order derivative even have entanglement phenomenon.To address this issue,an SVM classification model was constructed,with PCA-reduced data as the input variables,and the model was trained and validated.The SVM classification results obtained by the four different processing methods were compared and analyzed.The results show that the classification accuracies of the training and testing sets are 97.08% and 96.67%,respectively,which proves that using first-order derivative processed spectral data of metal AM components to construct an SVM model for defect-free,cracks,bulges,and holes is appropriate.In summary,this study investigated the application of LIBS combined with machine learning algorithms for the detection and recognition of defect types in metal AM components,achieving rapid classification of different types of defects.This demonstrates the enormous potential of LIBS technology in defect detection of metal AM components,which is of significant importance in further promoting the development of metal AM technology.