Modular analysis using the Design Structure Matrix (DSM) identifies the interactions between groups of components, and clusters them into modules in order to achieve competitive advantages in the product design processes. In this paper, a hybrid approach, based on multidimensional scaling (MDS) and clustering methods, is applied to component DSM for product architecting. The motivation is to facilitate better modularizations that enhance different product attributes in various product lifecycle stages. An experimental framework is developed to evaluate the performance of MDS clustering. Three MDS methods and four ubiquitous clustering methods are compared to reveal the most suitable for DSMs. The experimental results with several examples demonstrate that the effectiveness of MDS clustering, and show the superiority of non-metric MDS, SMACOF (Scaling by MAjorizing a Complicated Function) and hierarchical/cosine methods. These methods are capable of partitioning the product architecture into a set of modules and outperform the Newman-Girven algorithm, which has been extensively applied to DSM clustering. The proposed method is capable of partitioning the product architecture into reasonable modules. In addition, it can produce the optimal modules for any number of clusters, which is favourable especially when the cluster number is a higher managerial decision.

The terms module and modularity are not part of the technical taxonomy in any of the systems engineering standards, which do not regard a module as a part of the formal system breakdown structure. In this paper, we redefine the term module as a unit composed of a set of components with a set of specific interfaces. This unit serves one or more purely nonfunctional goals, such as flexibility, evolvability, manufacturability, testability, and maintainability. According to this definition, a configuration item, a subsystem, an assembly, a subassembly, or a component can all be regarded as modules as they can serve nonfunctional goals. The important assertion here is that a module's boundaries do not necessarily coincide with those dictated by the functional or spatial system decomposition and hierarchy. The aim of this paper is to lay the foundations for the future standardization of various engineering design processes based on modularity for nonfunctional benefits. A clear definition of the terms module and modularity can assist systems designers and developers to optimize the value of a modular system. This research highlights the present inconsistencies in the field of modular system design and puts forward some critical questions, which will shape the future research into this field.

Stochasticity in assembly processes is often associated with the processing time and availability of machinery, tools and manpower, however in this paper it is determined by probability of an assembly task successful completion which here is referred to as task reliability. We present a mathematical model for optimising the expected assembly cost, and consider two scenarios: the first a situation where a failure of one assembly task requires rework of that task alone; second a situation in which a failure in the midst of the process requires resumption of previously completed tasks. In the worst case scenario the assembly process must restart from the beginning. We show that the first scenario is insensitive to sequencing unless there are set-up costs. In the second scenario the process is sensitive to tasks’ sequence. We present a heuristic that argues for accomplishing more uncertain tasks (with less reliability) earlier in the process to decrease the expected cost of assembly, and show that in a mutually dependent assembly process, when tasks’ reliabilities are similar, the cheaper tasks should be executed earlier in the process.