Research On Prediction Model Of Surface Roughness Based On Artificial Neural Network

In recent years,the ownership of car in China shows the tendency of growing rapidly,which indicates that the car has been integrated into the public life of people and become an indispensable travel tools instead of foot.The engine as a key power component of automobile has an important effect on meeting the different motion performances and the requirements of environmental protection.In this thesis,the cylinder block?a key component of engine?was chosen as the study object.However,when machining,because the cylinder block is belong to the porous part,it is necessary to do the fine machining first for the planes around the cylinder block to obtain the positioning datum of the hole machiningin the subsequent procedures.The work of this paper is aimed to construct a prediction model with higher precision and faster convergence speed,then we can get a set of optimized processing parameters corresponding to different surface roughness value with the help of this modelbefore machining.In this paper,an APSO-BP neural network prediction model based on natural selection mechanism was proposed to improve the BP prediction model with the shortcomings of slow convergence and low precision.The main work and conclusions are as follows:Firstly,the application points of BP prediction model was introduced and the characteristics of slow convergence and low precision caused by the existence of multi-peak error surface was analyzed.At the same time,in view of PSO algorithm having an excellent performance of searching for the optimized solution,a new BP prediction model with the PSO algorithm embedded in was constructed to get the initial weight and threshold value,which can ensure that the algorithm would complete the faster iteration in the better region of the error surface.However,the workflow of PSO-BP algorithm was also elaborated in detail.Secondly,the experimental platform and orthogonal experimental design method were introduced.By means of assigning the different levels for four research subjects,the data of L₇₅?3¹×5³?samples obtained by using the new repolarization method were analyzed.By comparing the Sig.values of four factors analyzed with SPSS software with the significance level??=0.05?of the system,the result shows that the rotating speed of grinding wheel and the speed of workpiece have the most significant effect on the dependent variables.In the meantime,the influence degree of different levels for each factor on dependent variable was analyzed in detail.Finally,this paper has selected the overall parameters of the BP prediction model and established the prediction model successfully.Proposed PSO-BP prediction model and weight improvement LPSO-BP,RPSO-BP,APSO-BP prediction model according to the result of analysis.The APSO-BP prediction model based on natural selection is proposed to improve the shortcoming of slow convergence of APSO-BP prediction model.The result show that the convergence step is 14,the maximum error rate of the 11 groups is 0.75%,the minimum error rate is 0.16%and the average error rate is 0.46%.The conclusion is that the APSO-BP prediction model based on natural selection has faster convergence and higher accuracy.It is more conducive to the process of promoting the intelligent development of modern enterprises by combining the bionic algorithms with engineering applications.