Spectral Data Preprocessing Of LIBS And Its Application In Alloy Steel Analysis

Laser induced breakdown spectroscopy(LIBS)has the advantages of simple operation,short analysis time and remote in-situ analysis,which can realize the quantitative detection of alloy steel.Affected by the detection environment,alloy steel matrix effect,detection system fluctuation and continuous background signals,LIBS spectrum baseline will drift,and the signals contain lots of noise,which affects the detection results.Baseline correction and spectrum denoising data preprocessing methods are proposed to eliminate these interference factors and improve the signals'quality.A multivariable quantitative detection method based on principal component analysis(PCA)and radial basis function(RBF)neural network is proposed.Combined with preprocessing method,it has higher detection accuracy and stability than the traditional single variable detection method,which provides a new way for LIBS quantitative detection.Setting up the LIBS experimental system,adjusting and optimizing the system's structure and experimental parameters.the preprocessing method of segmented Hermite cubic interpolation baseline correction data is proposed for baseline drift phenomenon,and different baseline correction methods are compared and analyzed.The results show that the baseline correction method proposed in this thesis can obtain smooth and convergent base signal,and can restrain baseline drift effectively.The source of spectral noise is studied and the corresponding noise reduction is carried out.Empirical Mode Decomposition(EMD)is used to optimize the adaptability and reliability of wavelet decomposition,and a wavelet-based EMD noise reduction algorithm is proposed to deal with the noise whose composition,influence path and action mode cannot be determined.Through simulation experiments and evaluation index parameters,and the stability of the spectral noise reduction method in this thsis is verified by repeated experiments.After preprocessing the original LIBS spectrum signals with denoising and baseline correction,the fitting correlation coefficient R~2of the internal standard quantitative detection is increased,and the root mean square error(RMSE)is decreased.By a number of cyclic inversion experiments,the data preprocessing method proposed can improve the stability of LIBS quantitative detection.On the basis of data preprocessing,a quantitative model based on PCA and RBF neural network is established.The accuracy and stability of 320 prediction sets are verified by RMSE and relative standard deviation(RSD).Compared with the internal standard method,the average relative error of the detection results is decreased.