中文摘要：

对于作用关系复杂,而且质量特性拥有多个极值的制造过程,现有质量改进方法只能实现参数的局部优化,产品质量仍有较大改进空间.本文采用支持向量机（SVM）作为复杂作用关系过程的近似模型,提出基于支持向量聚类（SV）与序列二次规划（SQP）的参数全局性优化方法.首先建立了复杂过程的SVM近似模型;而后根据ε管道理论,通过对聚类过程谱系图的分析,确定了聚类的最小相似度水平及合适的聚类数目,将过程各极值点邻域内的支持向量分别聚为一类;最后由各聚类中心出发,并行进行SQP寻优以发现过程的多个极值.仿真研究表明,所提方法能够全面反映过程的极值分布,实现参数的全局性优化;寻优结果与实际极值的绝对偏差及相对偏差的平均值分别为0.15和1.28%,并且偏差的大小与过程极值的数目无关,说明方法具有较高的精确度和稳定性;此外,通过支持向量聚类,不仅保证了SQP寻优结果对于过程全部极值的遍历性,而且将寻优的次数降低了50%以上,提高了寻优效率.

英文摘要：

For manufacturing processes whose output quality characteristics have multi-extremes and whoseinput-output relationships are complex, existing approaches of quality improvement can only obtain a local optimimization of the process parameters. Therefore, there is still great potential for product quality improvement. By adopting support vector machines （SVM） to approximately model the complex relationship processes, this article proposes a support vector clustering and sequen- tial quadratic programming （SQP） based approach for global optimization of process parameters. Firstly, a SVM based approxi- mation model for the complex processes is set up. Then, based on s-tube theory, the minimal similarity level and hence the ap- propriate number of clusters are determined through analyzing the clustering dendrogram, and the support vectors in the neighbor- hood of each quality characteristics＇ extremes are clustered into one group. Lastly, by using the geometrical centers of each clus- ter as initial points, the multi-extreme values of the quality characteristics are found through concurrent SQP optimization. The simulation study shows that the proposed approach can effectively reflect the distributions of quality characteristics＇ extremes and achieve global optimimization of the process parameters ; the average absolute deviation and the average relative deviation of the optimization results from the actual extreme values are 0.15 and 1.28 %, respectively, and these deviations are independent of the number of the quality characteristics＇ extremes, as demonstrates the high accuracy and stability of the approach. Moreover, after clustering the support vectors, not only the ergodicity of SQP optimization results for all of the process extremes has been en- sured, but also the number of SQP optimization has been decreased by at least 50%, as increases the optimization efficiency of the approach.