It is reported that researchers from Foshan University, the Institute of Chemical Defense of the Academy of Military Sciences, the National Defense Technology Key Laboratory of Equipment Remanufacturing Technology of the Armored Forces Academy, and Chengdu State owned Jinjiang Machinery Factory have summarized and reported the latest progress of machine learning in defect detection and prediction of laser cladding technology. The related paper titled "Recent advances in machine learning for defect detection and prediction in laser cladding process" was published in Next Materials.
a key:
1. Summarized typical defects and their formation mechanisms.
2. Summarized typical laser cladding monitoring techniques.
3. Summarized typical machine learning algorithms used for defect classification and recognition.
As a fundamental component of artificial intelligence, machine learning has attracted much attention in the field of laser cladding in recent years. Machine learning has had a significant impact on various aspects of laser cladding technology by using algorithms to analyze data, identify patterns and patterns, and make predictions and decisions. The appearance of defects during the cladding process poses a huge challenge to the quality and performance of the cladding layer. Solving the reliability and repeatability issues of cladding quality is the primary focus of laser cladding technology. With the help of data-driven machine learning algorithms, defects can be monitored and detected throughout the entire laser cladding process. In addition, these algorithms provide a way for feedback adjustment, parameter optimization, and reduction of cladding defects in the cladding process, making it a cutting-edge research field.
This article provides an overview of the types and formation mechanisms of defects that occur during laser cladding, explains the signal characteristics, and elaborates on the monitoring principles and methods used in the laser cladding process. In addition, the progress of machine learning methods in signal feature extraction, defect classification, and predictive modeling during laser cladding process was comprehensively discussed. Furthermore, common machine learning models and algorithms for defect detection have been summarized. The research results highlight the effectiveness of machine learning algorithms in detecting defects in laser cladding coatings, while establishing the correlation between feature signals, coating defects, and cladding processes. At present, supervised learning algorithms dominate in this research field. However, unsupervised and semi supervised learning algorithms are receiving increasing attention in the field of laser cladding process monitoring due to their low requirements for data annotation. Overall, the research results clarify the key focus and direction for future exploration of machine learning methods in the field of laser cladding technology.
Keywords: laser cladding; Machine learning; Defect detection; artificial intelligence
Figure 1. Application of machine learning in laser cladding defect detection.
Figure 2. Formation mechanism of pores in laser cladding layer.
Figure 3. Formation mechanism and main distribution areas of cracks in the cladding layer.
Figure 4. Defects caused by particle splashing, such as surface roughness and spheroidization.
Figure 5. Geometric deformation of laser cladding samples, (a) Geometric deformation: wavy defects, (b) Transformation point cloud of wavy defects.
Figure 6. Signal acquisition method for molten pool in laser cladding process.
Figure 7. Laser cladding process image acquisition system, (a) device structure, (b) melt pool photo, (c) feature extraction.
Figure 8. Machine learning framework for laser cladding defect detection and process performance optimization.
Figure 9. Abnormal powder feeding detection process based on image recognition.
Figure 10. Feature extraction of melt pool thermal imaging data and defect detection of laser cladding thin-walled parts, (a) Feature extraction of melt pool thermal imaging data, (b) Defect detection of laser cladding thin-walled parts.
Figure 11. Fusion pool feature extraction and pore defect prediction model based on convolutional neural network (CNN) method.
Figure 12. Depth learning prediction method for pores and cracks in laser cladding layer based on acoustic emission signals.
Figure 13. Prediction framework for mechanical properties of thin-walled samples based on convolutional neural network (CNN).
Figure 14. Fusion pool classification framework based on semi supervised learning method.
This article provides a comprehensive overview of the application of machine learning algorithms in the field of laser cladding defect assessment. In depth exploration was conducted on common defects and their formation mechanisms in laser cladding processes, summarizing the acoustic, optical, and thermal signals generated during the cladding process, and elucidating the corresponding relationship between these signals and cladding defects. In addition, the classification and characteristics of machine learning algorithms were reviewed, and their applications in processing signals during laser cladding processes were summarized. Based on literature analysis, the following conclusions are drawn:
1. The laser cladding process is complex, and the defects generated directly affect the quality of the cladding layer. The formation mechanism and distribution pattern of defects are influenced by various factors, and there are interactions and evolution between different defects. At present, researchers worldwide have conducted research on defects such as pores and cracks through multi-scale experiments and simulations. However, the understanding of the mechanism of related defects and their impact on the quality of laser cladding is still insufficient, and a more comprehensive approach is needed to promote research on laser cladding.
2. Establishing a quantitative evaluation system for the relationship between process signal defect quality in laser cladding is a key challenge to ensure the stability of laser cladding quality. At present, various sensors such as acoustic, optical, and thermal sensors have been applied to monitor laser cladding processes and study the relationship between signals, processes, defects, and quality. However, due to limitations in sensor accuracy and defect feature extraction efficiency, establishing quantitative relationships between processes, signals, and defects remains challenging. Therefore, developing an online laser cladding process monitoring and defect feature extraction technology that integrates multiple sensors and signals is crucial for obtaining comprehensive and accurate cladding information and defect status. Realizing real-time quality monitoring throughout the entire process is an important development direction for laser cladding process monitoring.
3. Machine learning algorithms have been applied to defect detection in laser cladding. Usually, features are extracted from signals, cladding processes, and defect features to construct a dataset, and machine learning algorithms are used to establish the relationships between these features. However, current monitoring research mostly focuses on single or small area cladding layers, and small datasets may lead to overfitting, reducing the accuracy of actual defect detection. Therefore, it is necessary to design a defect detection database specifically for laser cladding processes. In addition, selecting appropriate machine learning algorithms is crucial for detecting different defects in laser cladding processes. Different algorithms have their own advantages in processing image data or sensor signals. Convolutional neural networks (CNN) are more suitable for processing defective image data, while support vector machines (SVM) are suitable for multi classification problems involving sensor signals or images. K-means clustering is widely used in unsupervised and semi supervised learning.
Based on the above summary, in order to promote laser cladding technology as a key innovative technology in the fields of manufacturing and remanufacturing, the following prospects are made for the application of machine learning methods. Laser cladding technology has broad application prospects in the manufacturing industry, and the introduction of machine learning technology can effectively improve the efficiency of laser cladding and reduce coating defects. At present, most of the reported literature focuses on supervised learning algorithms, which have high requirements for data annotation and require a lot of time and cost. Therefore, unsupervised and semi supervised learning algorithms have attracted attention in the field of laser cladding process monitoring, and new models continue to emerge and show great potential. In addition, through machine learning technology, automatic control and online monitoring of laser cladding equipment can be achieved. By analyzing a large amount of laser cladding datasets, optimizing and adjusting parameters such as laser power, scanning speed, and powder feeding rate, the automation and intelligence of laser cladding processes can be achieved, improving production efficiency and reducing costs. Machine learning technology will bring more innovation and development opportunities to the field of laser cladding, helping it to be widely applied in the manufacturing industry.
Source: Yangtze River Delta Laser Alliance
