Dependency modelling in complex system design

1. Topic and background of this Special Issue

Dependency modelling techniques support complexity management by focusing attention on the elements of a system, or the parts of a design problem, and the dependencies through which they are related. The dependency modelling perspective highlights structures in systems and their environments. It can lead to a better understanding of the implications of connectivity on different aspects of system performance, and may help to structure design problem-solving more effectively. This can assist in understanding, designing, optimising and maintaining complex systems – such as products, processes and organisations.

Since the ‘Design Structure System’ was first introduced in 1981 (Steward 1981), a great deal of progress has been made exploring how different forms of dependency modelling can support complex system design. Much has changed in the three decades since Prof. Steward's seminal paper was published. Many approaches have been developed based, for example, on Design Structure Matrix (DSM), Multiple-Domain Matrix (MDM) and network analysis techniques. Tools and methods have also been transferred into practice.

This special issue on Dependency Modelling in Complex System Design aims to take stock of some new developments and trends. It was conceived by members of the Organising Committee of the 12th International Dependency and Structure Modelling (DSM) Conference, which was held in Cambridge during July 2010. The Editors felt that there was scope for a journal special issue in the same research area as the conference, but with a focus on collecting mature archival publications rather than reports of work-in-progress. They also discovered that a special issue on this rapidly developing research field had never previously appeared. An open call for papers was approved by the Journal of Engineering Design, and released in September 2011.

Fifty abstracts were received leading to 35 full-paper submissions. Due to space constraints, only seven articles could be published – an acceptance ratio of 1:5. All submitted papers were peer-reviewed by at least three reviewers. In selecting the final articles, all manuscripts and reviewers’ comments were carefully considered by the Editors. Inevitably, many difficult decisions were needed, and some high-quality submissions could not be included.

2. The articles and their main themes

The Editors believe that the articles published in this special issue provide an excellent cross-section of current research relating to dependency modelling in complex system design. Some of the articles are grounded in case studies, whereas others develop theoretical insights. Some articles propose and evaluate new methods to support different aspects of design, while others formulate and test hypotheses about complex system design. The domains of product, process and organisation are all considered.

As well as the overall topic of dependency modelling, five key themes span two or more of the articles. These themes are: (1) modularity; (2) architecture synthesis, design and assessment; (3) co-evolution dependencies across domains in product development; (4) design process information dependencies; and (5) coordination. Most of the articles draw upon DSM methodology to cast light on these issues: functional analysis, clustering analysis, discrete-event simulation and optimisation techniques are also used.

The first article, by ElMaraghy and AlGeddawy, introduces a new method for designing product family structures. Entitled New dependency model and biological analogy for integrating product design for variety with market requirements, the article explains how each market segment targeted by a product family might be possible to satisfy through alternative component combinations. ElMaraghy and AlGeddawy consider the problem of selecting a family structure from these possible combinations, such that modules shared across as many variants as possible can be formed. A new procedure is introduced that adapts the cladogram classification technique used in biology and combines it with dependency modelling to generate an optimal product family and tree-structured modularisation scheme. This method is explained and illustrated through a detailed case study of washing machine variants. Implications for production planning are also highlighted.

The second article continues the theme of combining modules to form architectures, focusing on individual designs rather than product families. In the article, entitled Multiple-domain design scorecards: a method for architecture generation and evaluation through interface characterisation, Jankovic et al. introduce a new method to assist a design team in generating and assessing architecture concepts. Their method captures alternative technical solutions for each functional module in a design, along with the interfaces and how they may change according to which combination is chosen. All possible combinations of sub-solutions are automatically generated according to the interface dependencies. A structured process comprising a number of Design Assessment Cards is proposed to help the design team rank the large number of concepts identified, according to challenges that are often encountered during design. A case study in the oil industry found that the approach yielded ‘better’ concepts than those which the team had previously identified during creativity workshops.

Tilstra et al. also consider product architectures, focusing on modelling and assessing the architectural characteristics of existing designs. Their article, A high definition design structure matrix (HDDSM) for the quantitative assessment of product architecture, introduces an enhanced hierarchical DSM representation and modelling process that can assist in the rigorous decomposition of a product into interacting subsystems and components. This ‘high-definition’ DSM includes a typology for specifying component dependencies, extending the functional basis with the aim to better support analysis of designs that already have a physical embodiment. Tilstra et al. argue that this enhanced interaction basis can assist in the study and comparison of architectures across products and by different modellers. Two metrics are introduced that use the new dependency model to reveal different aspects of product architecture.

The fourth article considering architecture and modularity is Comparative analysis of coupling modularity metrics by Hölttä-Otto et al. In this paper, the authors focus on determining the degree of modularity of a design. A number of modularity metrics have been proposed to address this issue; Hölttä-Otto et al. review the existing approaches and design an experiment to compare a selected sub-set. The metrics that are compared each generate a single number to represent the modularity of an architecture comprising components that are grouped into modules, along with the network of dependencies between components. In the experiment, each such metric is tested on 24 hypothetical models that capture different ‘degrees’ of bus architecture, integral architecture and modular architecture. The paper concludes that existing metrics are not consistent in their indication of modularity, and that many do not yield sensible results when confronted with basic modularity features (such as buses) found in real designs. Furthermore, the authors show that many metrics are strongly affected by ‘irrelevant’ factors such as the number of components in a design. These findings may have important implications for future research on modularity metrics.

The fifth article, by Le and Panchal, is entitled Analysis of the interdependent co-evolution of product structures and community structures using dependency modelling techniques. Picking up the theme of co-evolution that was discussed in the first article, these authors consider the interaction between the structure of dependencies in an open-source product and the interrelationships between members of the loosely coupled community who contribute to its development. Several versions of an open-source software product, Drupal, are analysed to test hypotheses relating to mechanisms causing co-evolution of product and community structures. Dependencies between source files are extracted from the software versioning system for each of four consecutive versions of Drupal, and dependencies in the community structure are extracted from posts on the online forum for Drupal developers. Le and Panchal find that, in this case, changes in the product structure tend to be followed by similar changes in the community structure – while the converse is not necessarily true. They suggest that the introduced method could provide insight for coordinating the open-source development of software and hardware products, by helping to identify component clusters for which the corresponding communication links are not yet in place.

This topic of coordination is revisited in the sixth article, Optimal design processes under uncertainty and reciprocal dependency, contributed by Suss and Thomson. These authors focus on the information flow characteristics of a project, proposing a simulation model for studying the iterative exchange of information between interdependent, concurrent tasks arranged into phases. The model represents progressive development of a design in terms of the reduction in uncertainty levels associated with information transferred between tasks. Interdependencies cause ‘churn’ iteration as information is passed back and forth between tasks and an integrating function, while the probability of phase-gate evaluations revealing rework is related to the uncertainty levels when each gate is reached. Suss and Thomson use their model to analyse coordination mechanisms in hypothetical processes comprising either sequentially dependent or fully interdependent tasks. A number of improvement approaches are considered and their effects analysed, including: increasing concurrency between tasks; reducing communication delays and implementing ‘scrum’ methods. Recommendations are made to enhance coordination according to characteristics of the project at hand.

The final article, Enabling exploration in the conceptual design and optimisation of complex systems, combines the information dependency perspective with a computational view on design space exploration. In the article, Nunez et al. consider the issue of solving conceptual design problems represented as computational workflows. Nunez et al. focus on the situation where a design problem turns out to be infeasible, such that the designer needs to reformulate it in order to reach an acceptable solution. This occurs if the workflow cannot be used to find a feasible design, given the configuration of inputs, outputs and constraints, and their required or desired values in the case at hand. Nunez et al. develop a method to resolve this problem, integrating a dependency network of models and parameters in the computational workflow with an isocontour-based approach that assists in visualising the space of design parameters and constraints that result from a particular configuration. The network model helps to reformulate a design problem by automating the process of ‘reversing’ selected inputs and outputs of the workflow, while the isocontour approach allows designers to assess whether the resulting design space topology would allow feasible solutions. The method is applied to a case study of aircraft conceptual design, comprising 97 parametric models interconnected by 125 parameters.

To summarise, the Editors expect that the articles collected in this special issue should be of interest to researchers working on aspects of dependency modelling applied to products, product families, processes, and organisational structures in complex system design. We would like to thank all authors who submitted manuscripts, and the researchers who gave their time to write over 130 anonymous reviews. Special acknowledgement is due to the Editor-in-Chief, Professor Alex Duffy, for his patience and support during the preparation of this special issue.