混合神经网络用于滚珠丝杠热误差预测

Application of Hybrid Neural Network in Predicting Thermal Error of Ball Screw

为了减小数控机床热致定位误差影响,提高机床加工精度,这里建立了径向基神经网络(RBFNN)和时间序列(ARIMA)混合模型的变权值热误差预测方法。综合两个单一模型对数控机床热误差进行预测,利用逆向辨识优化算法分别获取两个单一模型的优化权值,得到变权值混合模型,使得热误差预测精度得到提高。将这里混合模型与RBFNN模型和ARIMA模型分别进行对比分析,结果表明混合模型(RBFNN-ARIMA)的预测精度明显优于单一RBFNN和单一ARIMA模型,证明了此算法的有效性。

Improving the machining accuracy of CNC machine tools is conducive to reducing the influence of thermal positioning error.This paper establishes a variable weight thermal error prediction method for a mixed model of Radial Basis Function Neu-ral Network(RBFNN)and Time-Series(ARIMA).The thermal error of the CNC machine tool is predicted by integrating two single models,the optimal weight of the two models is obtained by using the reverse identification optimization algorithm,and obtain the variable weight hybrid model,so that the prediction accuracy of thermal error is improved.To verify the feasibility of the model in this paper,the proposed model in this paper and the RBFNN model,and the ARIMA model are experimentally veri-fied and compared.The results show that the prediction accuracy of the mixed model(RBFNN-ARIMA)is significantly better than the single RBFNN and single ARIMA models,which proves the effectiveness of the proposed algorithm.