PI: Warren Seering, Department of Mechanical Engineering, MIT
The process of design is central to the delivery of new products and product variants. Design is frequently taught and even practiced as a rather ad hoc set of coupled procedures. A major weakness of design as practiced is the consequent failure to identify novel solution options at the beginning of the process that, if discovered, would have proven superior to those that were considered. Innovation is stifled because of incomplete exploration of options. Patents, current and expired, are rich with design information. Often patents are reviewed at the end of the design process if at all. We propose to develop methods for extracting relevant design information from existing patents to guide the search for design concepts at the beginning of the design process. These methods will lead to a more complete exploration of design alternatives and support a more effective strategy for converging to the design option selected for development.
- Project Title: Extracting and Employing Intellectual Property Constraints in the Space of Design Alternatives
- Principal Investigator: Warren Seering, Department of Mechanical Engineering, MIT
- Grant Period: September 2017 – August 2018
Accomplishments and results of the project
The objective of this research is to clarify mechanisms for searching through the space of possible product concepts to identify the most promising product candidates for addressing a need. Our intention is to discover what we can about how engineers search through the space of design options and, consequently, to identify areas of the search process that merit improvement.
There are few rigorously structured design methods for conducting this early search phase of product development. We have elected to explore the potential of one of these methods, set-based concurrent engineering, as a process for identifying promising product options. Set-based concurrent engineering, also called set-based design, relies on the propagation of constraints to rule out possible product concepts and so to define and limit the space of acceptable options. The defining of constraints also serves as a way of capturing the rationale for making design decisions. To understand the potential of this method, we conducted four related research activities. The first consisted of a structured literature review to codify the strategies in use for applying set-based design methods. The second was comprised of 22 small case studies to categorize types of constraints currently used to eliminate design options from consideration. The third involved an examination of the approach taken by a representative model-based design system for pruning the space of candidate solutions. The fourth was an analysis of the potential for using patent claims as constraints to guide design choices.
For the first phase of the work, we conducted a structured literature search that yielded 72 technical documents addressing the use of set-based design. A review of these papers showed that the processes they describe can be seen as having three phases; mapping of the space of all possible solutions for the design task, integrating the perspectives of various engineers or teams by intersection of the constraints specified by each, and gradual convergence as more design information is acquired and consequently more options can be ruled out as being infeasible or dominated by others.
Proposed processes for mapping the space of possible solution options varied from informal to formal. Informal approaches included such methods as brainstorming; think of as many options as you can. More formal approaches included conducting of searches to find all examples of existing designs for comparable systems. These formal approaches were more often proposed for cases in which the family of options shared common design elements, ships for example. The wider the variance in option configurations, the less formal the proposed methods. From our analysis, it seems clear that identifying the bounds of the set of all feasible solutions to a design problem remains a difficult challenge, and correspondingly an important research opportunity.
Quite a bit of good work has been done on defining procedures for articulating constraints on the set of feasible solutions and then propagating these constraints through the set. A key characteristic of set-based design is that constraints representing the priorities of various design team subgroups can be imposed on the set of solutions independently and without the need for consensus. In a novel contribution, the Navy team that designed the Ship-to-Shore Connector, now in production, required that each constraint applied to the solution space be ‘owned’ by a member of the design team. This provided accountability for eliminating design options and set up a communication path for negotiation of constraints.
Little consensus was found on specific methods for driving the set-based design process to converge. Design options can be removed from the feasible set because they violate a design constraint or because they are dominated in performance by other options. Early in the process, representations of design systems and their components tend to be more conceptual and so uncertainty is introduced into the process of determining which concepts cannot be brought into compliance with constraints. There are corresponding risks associated with eliminating options on the basis of dominance. As the design process proceeds, and resolution of design representations increases, propagation of constraints will continue to eliminate options and dominance can be determined with less uncertainty. An open issue, and one for future research, is that of employing risk management procedures to inform the process of removing design options from the feasible set as more design information becomes available.
To develop insights into the types of constraints that can be applied in defining and pruning the solution space, we evaluated the processes of 22 design teams as they pursued solutions to system level design tasks. The teams were made up of engineers and engineering managers most of whom are pursuing a masters degree in engineering management at MIT; about half are employed concurrently and participate by distance. The design tasks were provided by industrial collaborators. The students, in teams of 4 or 5, spent about three months on these capstone projects. Each team was asked to define, as best they could, the full set of solution options that merited consideration. Then they were asked to document their reasons for eliminating options from consideration as the design process proceeded. In reviewing the reports that the teams generated, we identified more than 100 constraints in 30 categories that led to removing design options from consideration. These constraint categories can be bundled into three groups; market constraints, technology constraints, and company constraints. Included in the category of market constraints were factors such as cost, operational safety, and alignment with the expectations of the customer. Technology constraints included technology readiness, environmental compatibility, and scalability. Company constraints included alignment with corporate strategy, projected profitability, and risks associated with the supply chain. In each case, these constraints served as justification for eliminating solution concept options from consideration while allowing other concept options to remain in consideration. In numerous cases, the teams dismissed options without fully understanding if the concept when developed could satisfy the constraints. In so doing, they accepted the risk that the solution they eventually arrived at could be less than optimal and possibly not competitive. The reason typically given for accepting this risk was lack of sufficient time or human resources to evaluate the culled options thoroughly.
To gain some insight into how the cost of considering options might be reduced, we established access to a model-based engineering program developed at Audi for evaluating product architecture options for new families of automobiles. The software accepts as input the descriptions of the primary components; engines, transmissions, suspensions, and so on; for more than 200 existing automobile models from various manufacturers and then generates more than 50,000 possible product family architectures based on these components. Then, given a set of specifications for the family of automobiles being designed, the software determines the architectures capable of supporting the various product options. The program can interpolate among dimensions of existing components to define new components with correspondingly new performance capabilities, so the number of actual vehicle options to be considered is essentially infinite, and as a consequence efficiency of search is a key performance characteristic of the program. The authors of the software were not aware of set-based design principles, but the structure of the software was generally fairly well aligned with the process guidelines that had emerged during our literature review. As an exception, evaluation of the program’s structure showed that performance constraints, like knowledge of the types of transmission that can be used with each type of engine, have been applied in an ordered way to most effectively constrain the number of options being considered at any given point during the constraint propagation process. This element of the structure is not aligned with existing guidelines for use of set-based design. Current work on set-based design suggests that all constraints can be applied in parallel. An insight that we drew from this study is that though constraints may be propagated in parallel, it could be more efficient, and so require fewer human resources, to define an order, grounded in the structure of the design being generated, in which to apply the constraints to the set of candidate designs. This is another area in need of further study.
One design method, developed primarily in Russia, is TRIZ, a Russian acronym for the “Theory of Inventive Problem Solving”. When the TRIZ method is employed, it generally serves to generate a set of candidate product concepts. It has been known to reveal large sets of options including ones considered to be novel. A particularly rich pool of knowledge regarding the performance of product concepts of a given type can be found in the claims made in patents for comparable concepts. So far as we know, those who employ TRIZ have not yet used the claims of patents as constraints to eliminate candidate options. Our work suggests that patent claims from active patents are in fact well-structured to serve as constraints on a set of candidate product concepts and could be used productively in this way early in design processes like TRIZ as part of an ordered constraint propagation protocol to pare the set of options that are worthy of further consideration. At the same time, claims from patents that have expired can serve to enrich the set of feasible product concepts to be considered further. The use of patent claims to constrain a solution set, and particularly the use of claims to enrich the set of feasible options both proved promising and worthy of additional study.
Impact on Skoltech
The field of new product development is underdeveloped at this point in Russia. Russia’s top 20 exports, accounting for a substantial majority of export revenue, include no finished goods. An important element of Skoltech’s mission is to support the initiation of systems and processes for developing new and competitive products for export and internal use.
An objective of this work has been to examine and rate potentially valuable methods for developing new products. An understanding of these methods will be of particular value to Skoltech students as they learn to design successful products. The results of this research contribute to the knowledge base of structured design processes to be taught at Skoltech and consequently to the quality of the products designed by Skoltech students. The results are also of use to corporate sponsors seeking more effective ways to define successful product concepts.
It is our hope that this work’s Skoltech co-PI will visit MIT in the coming year to continue this research and to develop related educational modules.
Two masters students and one Ph.D. student participated in this research project. One of the masters students completed a thesis as part of the project and graduated in June. The other two will graduate in the coming year. One journal paper and one conference paper based on this research are in preparation.