Abstract
The Assembly Sequence Planning (ASP) problem is a complicated task that is still performed manually in most advanced industries. It consists of finding the best or optimal sequence to assemble a certain product, given its CAD design. Although it seems simple at first, its complexity drastically increases with increasing numbers of product parts, so that complexity is very high for most actual industrial products. Many ASP planners have been developed by researchers to automate this problem, but most of these are not practical and general enough to deal with actual industrial products. One of the main disadvantages is that these planners perform an extensive search of all possible sequences in order to choose the optimal solution. Another rather important disadvantage is that most generated sequences are linear, i.e. one part is assembled at a time, or have very simplistic plan generation. Sebaaly and Fujimoto (1996) introduced a novel approach to overcome the first disadvantage by applying genetic algorithms. A best-so-far solution is reached without searching the complete set of possible candidates, and the search is performed on a sequence population basis rather than on parts basis. However, this method is restricted to generating linear sequences only. This paper addresses improving that approach to generate more general solutions, by introducing an assembly fuzzy graph representation that can represent both linear and non-linear sequences. The sequences search space is thus extended to include all feasible combinations of linear and non-linear assembly operations. From the set of assembly rules and constraints of a certain product, a set of assembly stages is defined, such that every assembly operation is assigned to a certain stage according to its position in the set of constraints. A fuzzy relation is then defined as a grade of connectivity between product parts. Based on this relation, a fuzzy graph connecting the product parts is generated. This graph can represent both linear and non-linear sequences. After that, the algorithm of Sebaaly and Fujimoto (1996) is improved to deal with the new search space. The new modified algorithm is applied to a practical example from industry where the applicability and capability of the new algorithm are confirmed.
