基于BP神经网络的动态稳健参数设计

传统的动态稳健参数设计方法(田口方法)虽然在工业生产实践中展现了极大的方便,但是其本身也存在较大的改进空间.当调节变量不存在时,传统的田口方法难以实现;此外,田口方法只能根据所选取的参数水平得到最优参数组合,而这种所谓的最优结果有时并不符合实际的需要.首先构建BP神经网络模型,利用训练后的BP神经网络获得参数设计中质量特性、噪声因子以及各参数间的动态关系;然后,利用超拉丁方抽样,计算信号与特性参数间的斜率,并由此将动态稳健参数设计的寻优问题转化为相应的非线性规划问题;最后,利用次序二次规划(SQP)算法解决并优化动态稳健参数的设计。此外,我们选取了一个简单的数据案例对本文提出的方法的有效性进行了说明.

Dynamic Robust Parameter Design Based on a BP Network Approach

The Taguchi parameter design method,which is the traditional method people apply in robust parameter designs,has been widely accepted and recognized as an important tool for improving the quality of a product or a process.However,the statistical methods and optimization procedures proposed by Taguchi have much room for improvement.For instance,the Taguchi method may fail to identify an optimum design condition if an adjustment parameter does not exist.What's more,the optimal setting of a design parameter may sometimes be very impractical since it is determined only among the levels included in the parameter design experiment.In this paper,we develop a dynamic approach based on BP network to overcome the shortcomings of the Taguchi and existing alternative approaches.First,BP network is trained to capture the relationship between the characteristic,noise and signal parameters etc.Second,we calculate Latin hypercube samples of the signal and noise parameters to estimate the slope between the signal parameter and characteristic.Thus,we transform the parameter design problem into a nonlinear optimization problem and find the optimal settings using sequential quadratic programming(SQP).The effectiveness of the SQP approach is illustrated with an example.