激光修整青铜金刚石砂轮预测模型与优化研究

Prediction model and optimization study of laser trimming bronze-bonded diamond grinding wheel

为了根据实验规律获得更多激光修整青铜金刚石砂轮的工艺参数,采用反向传播(BP)神经网络和粒子群遗传混合优化算法(PSO&GA),建立了激光修整青铜金刚石砂轮的预测模型。首先通过分析激光修整的原理得出砂轮型面角度、激光偏转角度、入射角度和光斑重叠率为主要影响参数,并以砂轮型面角度误差和峰谷(PV)值为评价指标修整了192组工艺实验数据;建立了4×9×2的3层BP神经网络预测模型,通过PSO&GA混合优化算法对预测模型进行训练优化;最后选取16组实验数据测试BP神经网络预测模型,预测结果比较准确;并对比了梯度下降法(GD)、粒子群优化算法(PSO)和遗传算法(GA)的BP神经网络的训练效果。结果表明,经PSO&GA-BP预测模型角度误差预测偏差在0.2°以内,PV值预测偏差在1.6μm以内,相较于其它优化算法,收敛速度更快、精度更高。该研究为激光修整青铜金刚石砂轮提供了良好的预测模型。

In order to obtain more process parameters for laser dressing bronze diamond grinding wheel according to experimental rules,this paper uses back propagation(BP)neural network,particle swarm optimization and genetic algorithm(PSO&GA)to establish a prediction model for laser dressing bronze diamond grinding wheel.Firstly,by analyzing the principle of laser dressing,the grinding wheel profile surface angle,laser deflection angle,incidence angle and spot overlap rate were obtained as the main influencing parameters,and 192 sets of process test data were trimmed with the grinding wheel surface angle error and peak-to-valley(PV)value as the evaluation index.Then,a 4×9×2 three-layer BP neural network prediction model was established,and the predictive model was trained and optimized by the PSO&GA hybrid optimization algorithm.Finally,16 sets of experimental data were selected to test the BP neural network prediction model,and the prediction results were more accurate,and the training effects of the BP neural network by gradient descent(GD),particle swarm optimization(PSO)and genetic algorithm(GA)were compared.The results show that the angle error prediction bias of the BP neural network trained by the PSO&GA hybrid optimization algorithm is within 0.2°,and the prediction deviation of PV value is within 1.6μm,and compared with other optimization algorithms,the BP neural network has a faster convergence speed and better convergence accuracy.It provides a good predictive model for laser dressing of bronze diamond grinding wheels.