A Novel Operation Sequence Similarity-Based Approach for Typical Process Route Knowledge Discovery

A typical process route essentially represents the commonly used process planning-related knowledge and can be modified to generate new process routes easily. Hence, its quality directly affects the performance of newly generated process routes and thereby the goodness of products. To effectively discover typical process route knowledge, a reasonable similarity measure and a clustering method specifically for process routes are required. However, existing operation sequence similarity coefficients often assign coarse-grained similarities, which leads to inaccurate clustering results. For the clustering problem, most researchers have not considered the practical constraints during typical process route discovery. In this paper, an operation sequence similarity-based discovery method is presented. First, the characteristics and information requirements of the operation sequence similarity problem are analysed, and a novel comprehensive similarity coefficient combined with a modified pseudo-longest-common-subsequence (pseudo-LCS) and Jaccard similarity coefficient is proposed based on this analysis and principal component analysis (PCA). This coefficient considers the precedence relationship, the number of common operations, and the operation similarity simultaneously to handle all the potential similarity situations. Second, two soft constraints, namely, quantity constraint and size constraint, are introduced in the traditional process route clustering problem to ensure the quality and validity of the discovered typical process routes. To solve this more practical problem and achieve a balance between these two conflicting constraints, the K-medoids method is improved with an adjustment mechanism to generate valid results under these two soft constraints. Finally, numerical illustrations are presented to verify the effectiveness of the proposed methods. The results show that compared with existing similarity coefficients, the proposed comprehensive similarity coefficient is more sensitive and much better at distinguishing the tiny difference between the process routes. In addition, the modified K-medoids method can perform much better than existing methods on process route discovery data sets under two conflicting soft constraints.