Study Of Steel Aging Estimation Via Laser-induced Breakdown Spectroscopy Coupled With Machine Learning

With the rapid development of the national economy,the safety and efficiency of the operations of industries such as power plant,metallurgy,chemical and machinery industry are very important to the economic modernization.And the number of high-temperature pressurebearing equipments used in these industries are increased,so the operation safety and reliability of the equipments are the critical factors of the safety and efficiency of the industrial production.During the long-term operation,the mechanical properties and metallographic structure of the heat-resistant steel of the equipment will gradually change due to the hostile environments(also known as material aging),it will seriously endanger the operation and reduce the lifespan of the equipment.The traditional steel aging estimation method is very inconvenient and timeconsuming.A specific part of the equipment needs to be cut off for metallographic analysis.And the existing nondestructive methods are mainly focused on the detection of the material macroscopic defect.Therefore,developing in-situ,faster and nondestructive steel aging estimation techniques is very necessary for equipment conditionmaintenance.Laser-induced breakdown spectroscopy(LIBS)is an appealing and prospective analysis technique based on atomic emission spectroscopy.It is capable of not only the simultaneous multi-elemental detection,but also reflecting the characteristic of physical and chemical properties of solid sample.Therefore,LIBS coupled with machine learning was applied to the steel aging estimation in this dissertation.Based on the spectral data acquired from the samples with different steel aging degrees,the temporal-spatial resolved plasma emission,the steel aging estimation models built with various meachine learning methods,and the spectral feature selection were studied and discussed.At first,based on the review of the research and development situation of the existing steel aging decection techniques,the orientation and contents of the research in this dissertation were determined.The studies about LIBS combining with meaching leanring methods were reviewed,and the studies about applying LIBS to the measurement of material properties were also reviewed.A high precision LIBS experimental system for analyzing the temporal-spatial resolved plasma emission was designed and constructed.A number of data processing methods based on machine learning were introduced in the LIBS application,and the mathematical principles and computational processes of these methods were descriped in detail.At last,the experimental results proved that the clustering analysis,dimensionality reduction,classification model and feature selection methods show great potential in LIBS application.In order to deeply understand the mechanism of LIBS steel aging estiamtion,the temporalspatial evolutions of spectral line images,line intensities,line intensity ratios,plasma temperatures of T91 steel specimens with different aging grades were observed by using grating spectrometer with intensified charge coupled device detector(ICCD).And the threedimensional surface topographies of ablated craters of specimens with different aging grades were analyzed.In addition,the spectral characteristics of T91 steel specimens with different aging grades were analyzed by using conventional grating spectrometer.Line intensities and the line intensity ratios that indicate the change of metallographic structure were used to establish SVM models,and the results using different variable sets were compared.The model was optimized by comparing different pulse number for practical effectiveness,and the robustness of the model was investigated in dealing with the inhomogeneity of steel composition.In order to deal with the complex high dimensional LIBS data from steel specimen,two representative feature selection methods including analysis of variance(ANOVA)and LGR filter were utilized to reduce the high dimensional LIBS data into fewer features for improving the performance of steel aging estimation models.Furthermore,a new layered interval wrapper(LIW)feature selection method was proposed for being more targeted toward LIBS data.The steel aging estimation models constructed by SVM and logistic regression(LGR)were used to evaluate the effectiveness of the feature selection methods,and the results indicated that LIW showed a greatest improvement for steel aging estimation.In addition,anomaly detection was used to remove the outlier in LIBS measuremen.And K-means clustering algorithm and principal component analysis were used to analyze the macro characteristics of spectral data.The studies mentioned above constructed a systematic and adaptive methodology for LIBS application,created a better understanding of physical mechanism in LIBS steel degradation analysis,and provided a good reference for practical measurement.