基于模式识别的正交试验数据分析及其应用

Orthogonal experiment data analysis based on pattern recognition and its application

针对现有正交试验数据分析方法有操作复杂和适应性弱的弊端,提出基于模式识别的最优判别平面(ODP)分析法。该法首先按Fisher准则而建立2个相互垂直的矢量,再将试验数据投影到这2个正交轴上,从而得到二维特征矢量,据此确定各因素对指标的影响程度。分析SIS热熔压敏胶性能试验数据结果,ODP分析法的结果与实际相符,且不受正交表指标数和有无空白列的限制,是一种简单而有效的正交试验数据分析方法,可在冶金化工领域推广。

There are some flaws for the existing orthogonal data analysis methods, such as complex operation process, weak adaptation. The optimal discrimination plane （ODP） analysis method of ortbogonal experiment based on pattern recognition was proposed. In this method, two orthogonal vectors were firstly built up based on Fisher＇s criterion, and then the experiment data were projected onto the two orthogonal axes, thus, the two-dimensional feature vectors were attained that can be utilized as features to determine the influence degrees of factors on technical indexes. For example, the experimental data of SIS hot-melt pressure sensitive adhesive properties were analyzed using this method, and the results are consistent with the practice whether there are blank lines or not in orthogonal table, and not restricted to the number of technical indexes. Therefore, the ODP analysis method is simple and eft＇ective to deal with the orthogonal experiment data, and it can be applied wide｜y to metallurgical and chemical fields.