Optimization Of Laser Process Data And Software Development Based On GA-BP Hybrid Algorithm

With the further development of laser processing technology in the physical manufacturing industry,both laser processing methods and process parameters are becoming more and more complicated so that the optimization methods of the optimal process parameters is also increasingly difficult.This article aims at how to find the optimal laser process parameters quickly,accurately and economically.Based on the reserching of Back-Proragation(BP)Algorithm and Genetic Algorithm(GA),the two methods are combined and improved,which is applied to the exploration of laser processing technology parameters.Meanwhile,the improved GA-BP hybrid algorithm is realized by programmed and applied to software development.In this subject,the main research work has the following aspects:Firstly,the basic theory of BP neural network algorithm and traditional GA-BP algorithm is introduced,and the defects in the process of calculating optimization are deeply analyzed.Then,based on the traditional GA-BP algorithm,the BP neural network algorithm module and the genetic algorithm module are improved respectively.The concrete improvement is as follows:1)In order to overcome the failure of the implicit layer neurons in the training process due to the premature saturation output,we adopts a new type of error function;In this paper,we also add a magnification factor based on the weights and thresholds of the output neurons,aims at avoiding that the threshold is no longer adjusted;In addition,compared with the traditional "single sample training" strategy,this article uses the "batch training" method,because it is more objectively reflect the relationship between the sample data.2)The traditional roulette selection strategy is based on the chromosomal string fitness ratio and can not reflect the principle of equal selection of chromosomes,so this paper adopts the improved roulette selection strategy based on non-linear sorting;in order to avoid destroying the chromosomal string the original good convergence trend in the cross or mutation operation process due to genetic operation is too large or too small,this paper uses the adaptive crossover rate and adaptive mutation rate;In addition,in order to further improve the convergence speed of genetic algorithm,the "optimal preservation strategy" and "climbing algorithm" are added to the genetic algorithm.Secondly,the traditional GA-BP algorithm and GA-BP hybrid algorithm are realized by C language programming,which can dynamically adjust the neural network model structure.And the training performance of GA-BP algorithm and GA-BP hybrid algorithm is compared by training the neural network model through laser hardening experimental data under temperature control.It is concluded that the improved GA-BP hybrid algorithm are significantly better than that of the traditional GA-BP algorithm in terms of training speed,increased by about 80%;and the predicted parameters of laser process is also very accurate,which average relative error is 2.035%.Both of them confirm that the improved GA-BP hybrid algorithm is reasonable and effective.Thirdly,in order to enhance the engineering value of the research results,this paper uses Labwindows/Cvi software development platform to develop two with the use of the software based on the two steps of artificial neural network processing data-training and forecasting.Two software are named respectively the laser process neural training system and the laser process prediction client system.Through testing the performance of them at run time,the results show that they can meet most of the demand in exploring or predicting the laser processing data.