Prediction On Micro-hardness Of TiAl Alloy Based On Soft Computing

TiAl alloy has become a kind of hot material due to its high hardness and other excellent properties,the properties of micro-hardness is one of the criteria to measure the hardness of TiAl alloy,modeling can guide the experiment,reduce the test frequency and also save cost.However,due to many influencing factors and complex relations,it is difficult to establish the prediction model through traditional methods,and most forecasting work fails to take into account the characteristics of fuzziness and data non-uniformity.Therefore,it is still some defect in micro-hardness prediction.In this paper,a BP neural network micro-hardness prediction model was established based on the experimental data of capital alloy and the neural network in soft computing.And from two aspects of fuzziness and technical route to optimized the prediction model,finally consider sample non-uniformity characteristics prediction model is established,to improve the performance of the forecasting model to reduce the prediction error.The main work are as follows:(1)Select appropriate indicators to participate in the establishment of the prediction model and data preprocessing.Parameters were selected from external and internal factors,firstly TiC(wt%),preparation temperature T and B(wt%)were selected for external factors.The internal factors include non-linearity,fuzziness and non-uniformity.After filtering and removing invalid data,165 pieces of data are retained.In order to reduce the prediction error caused by large sample size difference,mapminmax function was used to normalize the model.(2)The prediction model of TiAl alloy micro-hardness based on BP neural network was established.First,determine the input node of the prediction model is 3 and the output node is 1.According to the empirical formula,the value range of the number of hidden layer nodes in the BP neural network is calculated.The predictionprogram is written on the software platform of MATLAB,and it is determined that the network structure of the BP neural network is 3-7-1.The average absolute value and mean variance of the prediction results are calculated to evaluate the prediction model.(3)Optimize the prediction model of BP neural network.On the one hand,considering the fuzzy characteristics of the data sample,on the other hand,by optimizing the parameters of BP neural network to improve network prediction ability,select T-S fuzzy model and genetic algorithm to optimize the BP neural network to reduce the prediction error.The prediction results of T-S fuzzy neural network and genetic algorithm optimizes BP neural network prediction model were compared between single BP neural network prediction model.(4)Establish micro-hardness prediction model based on fuzzy clustering analysis and neural networkAccording to the non-uniformity characteristics of microhardness data samples,combined fuzzy clustering analysis with BP neural network,a combined prediction model was established to predict micro-hardness.Through fuzzy cluster analysis the sample data is divided into several degrees higher similarity between sets of data,as the training sample,according to the similarity between samples with the clustering forecast model is set up as a test standards.The effect of fuzzy cluster analysis on micro-hardness prediction was analyzed.The results showed that the BP neural network forecasting model is better described relationship between micro-hardness and its affecting factors the mean square error is 11.92.After optimizing BP neural network through T-S fuzzy model and genetic algorithm,the mean square error of the prediction model increased by 1.68% and 34.90%.The mean square error of the prediction model established after fuzzy clustering was 4.94,which was 58.56% higher than that of the single BP neural network prediction model.Although the optimized model cannot achieve complete prediction,its results are of great significance.