Study On The Evaluation Model Of The Running State Of Heat-resistant Steel Based On Laser-induced Breakdown Spectroscopy And Chemometrics

In power plant,the heat-resistant steel of key equipment served at a high temperature and high pressure for a long time,and its running state seriously affects the safe production of equipment system.In conventional failure assessment of metal,the metal key position in equipment will be cut for detail off-line analysis or treated with on-site metallographic replica test during the maintenance period.Then combining the metallography inspection result with the regular mechanical property test,a comprehensive evaluation of the running state of steel can be obtain.The conventional method has lots of limitation in operation and time,hence it is a significant to develop the in-situ and rapid running state evaluation technique for equipment condition maintenance.During the service life of steel,there was no obvious change in the chemical composition in different state steel,but with evident degradation of mechanical property and the microstructure aging grade.These macro and micro damage of steel make obvious different in sample matrix,and the laser-induced breakdown spectroscopy(LIBS)technique was affected by matrix effect.Hence,LIBS technique was applied to evaluating the running state of steel in this work.The LIBS spectrum characteristic of different aging grade steel was analyzed first.Then two different feature dimension reduction methods,namely feature selection and feature extraction,were used to extract the important features of LIBS spectrum.After that,the important features were used to build the calibration hardness model with different regression algorithms.The results showed that the feature selection method can effectively extract the important features than feature extraction,and reduce the feature redundancy.Besides it,it is better to use the nonlinear algorithm than linear algorithm for the hardness calibration model in predicted result.Then,to magnify the influence of matrix effect in the LIBS spectrum of different samples,the spectrum was analyzed and treated with wavelet threshold denoising(WTD).14 different running state steel were used as model samples.Combining WTD and K-fold cross-validation support vector machine recursive feature elimination(K-SVM-RFE),the aging grade and hardness grade calibration models were built and discussed.The results showed that WTD pretreatment can effectively remove the noise signal in the spectrum,and benefit the followed feature selection process.The robust calibration model of aging grade and hardness grade can be obtained with the hybrid algorithm based on WTD and K-SVM-RFE.In addition,the influence of WTD pretreatment on selecting the feature subset in K-SVM-RFE was analyzed.Further,based on the indictors aging grade and hardness grade predicted by the hybrid algorithm,a risk assessment matrix was established and discussed.The effectiveness of the evaluation method was analyzed and verified.It revealed that the prediction error of single model can be avoided effectively by this evaluation method.Moreover,an accurate and reliable running state evaluation of steel can be evaluated by the method of assessment matrix established with the hybrid algorithm.Finally,the results of the whole research works were summarized and the directions for the further studies were suggested.