Quantitative Elemental Analysis Using Laser Induced Breakdown Spectroscopy And Neuro-Genetic Approach

Laser induced breakdown spectroscopy (LIBS) is a kind of emerging atomic emission spectrometry, which can detect all kinds of samples, such as solid, liquid, powder, gas and so on. It can also detect multiple elements simultaneously. Research works on elemental concentration on-site and real-time monitor by LIBS are highly valued by academic and industrial world.Laser plasma generation is complicated process. It is easily influenced by many factors, such as laser energy, triggering time of spectrometer, environment, sample preparation, matrix effect of sample and data acquisition method. These factors will affect the accuracy of quantitative analysis. Traditional calibration methods have already achieved some results to some extent, however, quantitative analysis accuracy always restricted by matrix effect and self-absorption.Artificial neural networks (ANN) has been employed in LIBS quantitative analysis for its nonlinear mapping ability. Researches indicated that ANN can reduce the influence of matrix effect and improve the detection accuracy effectively. However, back- propagation ANN (BP-ANN) is liable to trap in a local minimum value and used to converging slowly. In addition, the determination of original weight and threshold are lack of theoretic foundation, these parameters are always selected randomly. So, traditional BP-ANN can not ensure global optimization and then affect the quantitative analysis results.In this thesis, a quantitative elemental analysis technique based on LIBS and neuro-genetic algorithm was proposed. Firstly, genetic algorithm (GA) was used to make a thorough searching in the global space for original weights and thresholds of neural network, which can ensure them fall into the neighborhood of global optimal solution. Then, in order to improve the convergence precision, BP-ANN was used to finely train the network and find the global optimum or second-best solution with good performance. This algorithm can achieve high accuracy in elemental quantitative analysis. Elemental concentrations in some samples were successfully determined by the proposed method, which provide a feasible approach for LIBS quantitatively analysis.In this dissertation, firstly, the development and current status of LIBS technique, the establishment of laser plasma and the mechanism of its emitted spectrum were introduced in detail. In addition, several traditional LIBS quantitative analytical techniques and their shortages are discussed.Secondly, quantitative elemental analysis by combing LIBS with ANN was studied. The principle of ANN, BP-ANN and net structure design method were introduced. Quantitative analysis using BP-ANN algorithm was investigated in detail, including the determination of input spectroscopic data, the modeling methods of multiple elements simultaneous analysis and single element independently analysis.Thirdly, LIBS elemental quantitative analysis by neuro-genetic approach. The principle of GA was introduced. The modeling process of neuro-genetic and the detail method of LIBS quantitative analysis based on neuro-genetic approach were investigated in detail, including the combination mode of GA and ANN, fitness function selection and so on. Soil samples were analyzed by using the proposed combination method of LIBS and neuro-genetic approach. For comparison, the analytical results using traditional calibration method and BP-ANN method were provided.Finally, the repeatability of quantitative analysis method is briefly studied. A high repeatability spectral intensity signal acquisition method has achieved. The influence of the fluctuation of input data on the output of GA-ANN-LIBS analysis was discussed.