Rapid Classification of Laser Induced Breakdown Spectroscopy of Titanium Alloys
Xu Cheng,Li Fang,Chen Feng,Zhang Deng,Deng Fan,Guo Lianbo
DOI: https://doi.org/10.3788/gzxb20225104.0430001
IF: 0.6
2022-01-01
ACTA PHOTONICA SINICA
Abstract:Titanium alloy is a kind of metal material which with a wide range of applications, including aerospace, rail transit, medical equipment and other fields. The appearance of titanium alloys with different brands is very similar, but they are suitable for different fields. Even the properties of titanium alloys with the same brand and different numbers are different , and confusion is easy to cause serious accidents. Therefore, it is urgent to study the rapid and accurate classification and identification of titanium alloys with the same brand. In recent years, laser-induced breakdown spectroscopy as a fast, real-time, in-situ, micro-loss, multi-element synchronous analysis of advanced detection technology, favored by researchers. Using the advantages of laser-induced breakdown spectroscopy technology, the titanium alloys with different national standard numbers under the same brand are analyzed, which can realize the rapid and accurate classification and identification of titanium alloys. The whole device used in laser induced breakdown spectroscopy is composed of laser, spectrometer, electronic control displacement platform, workstation, acquisition head, digital delay generator and several lenses and optical fibers. According to sequential analysis, the spectra of titanium alloy under various laser intensities and different delay times were collected by the device. Combined with previous literature research and experience, six characteristic spectra with high signal strength and high signal-to-noise ratio were selected. The optimal laser intensity and trigger delay were obtained by comparing the peak intensity and signal-to-noise ratio of the six characteristics. The TC4 titanium alloy spectrum collected under the optimal conditions is divided into training set and test set according to the ratio of 3: 1. The training set data is trained by a variety of algorithms, and then the test set is substituted into the trained model. Through the analysis of the two results, the advantages and disadvantages of the algorithm are determined, and the optimal algorithm is K-nearest Neighbor algorithm. The optimization of data is mainly carried out from three aspects: 1) By using 3 sigma method for data screening at 10 characteristic spectra, the inferior spectra with too large or too small peak values are eliminated, to avoid its impact on the results; 2) Through data normalization, reduce the impact of experimental environment and experimental parameters; 3) Through principal component analysis, the data dimension is reduced, a large number of redundant data is reduced, the classification accuracy is improved and the model training time is reduced. KNN model can optimize the parameters mainly include the number of adjacent points, distance measurement, and distance weight. The number of adjacent points is the core parameter of KNN, which determines the number of data used to determine the unknown points. Distance measure is the distance calculation method between two points. Under different distance measures, the data between two points are different. Distance weight is the relative importance of determining the distance between the known point and the unknown point. The three parameters are arranged and combined into the model to retrain , and the optimal parameter combination is determined by comparing the results of the training set and the test set. The optimal classification results are obtained through various work, and the classification and recognition of the same grade titanium alloy are realized. The results show that the classification accuracy of the same grade titanium alloy can he improved from 84.15% to 99.14% by combining data processing and model optimization, and the training time can be reduced from 1232.41 s to 83.91 s. The classification performance is significantly improved. The research results are expected to achieve rapid and accurate classification and identification of titanium alloys with the same brand , and have broad application prospects.