A flexible information model for systematic development and deployment of product families during all phases of the product realization process is crucial for product-oriented organizations. In current practice, information captured while designing products in a family is often incomplete, unstructured, and is mostly proprietary in nature, making it difficult to index, search, refine, reuse, distribute, browse, aggregate, and analyze knowledge across heterogeneous organizational information systems. To this end, we propose a flexible knowledge management framework to capture, reorganize, and convert both linguistic and parametric product family design information into a unified network, which is called a networked bill of material (NBOM) using formal concept analysis (FCA); encode the NBOM as a cyclic, labeled graph using the Web Ontology Language (OWL) that designers can use to explore, search, and aggregate design information across different phases of product design as well as across multiple products in a product family; and analyze the set of products in a product family based on both linguistic and parametric information. As part of the knowledge management framework, a PostgreSQL database schema has been formulated to serve as a central design repository of product design knowledge, capable of housing the instances of the NBOM. Ontologies encoding the NBOM are utilized as a metalayer in the database schema to connect the design artifacts as part of a graph structure. Representing product families by preconceived common ontologies shows promise in promoting component sharing, and assisting designers search, explore, and analyze linguistic and parametric product family design information. An example involving a family of seven one-time-use cameras with different functions that satisfy a variety of customer needs is presented to demonstrate the implementation of the proposed framework.

Over the past 20 years I have focused on the synthesis process, the process of creation, of bringing together ideas and objects to fulfill and enable needs. Synthesis is a core engineering activity, the partner in the design and analysis cycle of engineering (and other) design. Synthesis is the basis of innovation, the enabler of creating that which is new. Over the past 20 years, research and practice in the area of synthesis and innovation has become more intricate, more complex, and more complete in its breadth of exploration and depth of understanding and delivery. Research in the area that was started over the past 10 to 20 years is now being commercialized, beginning to impact the way design is practiced. The state of the art of algorithms, theories, and processes for the computing basis of synthesis research today can be found in Formal Engineering Design Synthesis (Antonsson & Cagan, 2001). This discussion is not a review of the literature or state of the art, but rather my views of the field of innovation, its emergence into a scientific study, areas of focus for future research, and some of my experiences in each of these areas.

Agent models provide a generalized approach to the design of systems that autonomously monitor and affect an environment. Societies of agents that reason and communicate about an environment can achieve complex emergent behavior to facilitate and augment human activities. This paper introduces artificial agent technologies as a means by which the behavioral complexity of virtual worlds can be extended to provide the functionality needed to support collaboration in multidisciplinary design teams. Three key roles are identified that artificial agents can play to provide this functionality: support for multiple views of designed objects, support for the expression of relationships between designed objects, and compatibility with existing design tools. The implementation of a society of agents fulfilling these roles within a virtual world based, multidisciplinary design tool called DesignWorld is described. The increased behavioral complexity and functionality of DesignWorld's underlying virtual world is demonstrated using the results of multidisciplinary design experiments with DesignWorld.

Artificial Intelligence for Engineering Design, Analysis and Manufacturing has accomplished a lot, but we still have far to go in building a community. I was reminded of this earlier this year when I attended a workshop on a topic within the scope of AI EDAM's coverage. In addition to engineers and educators, workshop speakers included a linguist studying languages used by professionals, an ethnographer studying CAD users and a psychologist studying innovation. All three social scientists started their presentations essentially the same way:

Architecture, engineering, and construction team members, while collaborating on building projects, rely on past experiences and content through the use of project design archives (whether in paper or digital format). Underutilization of potential knowledge in the decision-making process of data, information, and knowledge reuse is limited by access to these archives, because of sheer size, decontextualized content, and inconvenient access and presentation. This paper presents an integrated solution called CoMem–iRoom that leverages two technologies Corporate Memory (CoMem) and interactive Room (iRoom) developed at Stanford. CoMem–iRoom addresses critical limitations (content, context, visualization, and interactivity) constraining the process of collaborative exploration toward knowledge reuse and decision making.

This paper presents an ontological framework for situated design teams in which the team is both the subject and the object of designing. Team designing is modeled using the set of processes provided by the situated function–behavior–structure framework. This is a formal basis for understanding the drivers for change in the product to be designed and in the design team. We specifically focus on changes in a team's structure that emerge from interactions among individual team members and subteams.

From its inception the journal Artificial Intelligence for Engineering Design, Analysis and Manufacturing recognized that designing is the precursor to analysis and manufacturing by placing it at the front of the list of areas it covers. Designing distinguishes itself from other aspects of engineering by its goal of changing the world within which it operates: designers are change agents. This characteristic makes designing a difficult task even for humans let alone for machines, because most of our knowledge and the means to acquire it assume that the world is given to us and what we need to do is characterize it. The Journal provided one of two continuing publication outlets for artificial intelligence (AI) in engineering at the time. This created the opportunity to have a focal point for the publication of archival research in the area, research that had previously appeared in disparate locations. It also offered the potential to develop a coherent research area where future research could build on previously published research.

Artificial intelligence (AI) emerged from the 1956 Dartmouth Conference. Twenty-one years later, my colleagues and I started daily operational use of what we think became the first application of AI to be used in practice: the PUFF pulmonary function system. We later described the design and initial performance of that system (Aikins et al., 1983; Snow et al., 1998). Today, easily recognizable descendants of that first “expert system” run on commercial products found in medical offices around the world (http://www.medgraphics.com/datasheet_pconsult.html), as do many other AI applications. My research now focuses on integrated concurrent engineering (ICE), a computer and AI-enabled multiparticipant engineering design method that is extremely rapid and effective (Garcia et al., 2004). This brief note compares the early PUFF, the current ICE work, and the modern AI view of neurobiological systems. This comparison shows the dramatic and surprising changes in AI methods in the past few decades and suggests research opportunities for the future. The comparison identifies the continuing crucial role of symbolic representation and reasoning and the dramatic generalization of the context in which those classical AI methods work. It suggests surprising parallels between animal neuroprocesses and the multihuman and multicomputer agent collaborative ICE environment. Finally, it identifies some of the findings and lessons of the intervening years, fundamentally the move to model-based multidiscipline, multimethod, multiagent systems in which AI methods are tightly integrated with theoretically founded engineering models and analytical methods implemented as multiagent human and computer systems that include databases, numeric algorithms, graphics, human–computer interaction, and networking.

Why has integration and interoperability of design and manufacturing knowledge proven so difficult? There is a common puzzle that appears in your Sunday newspaper or The New Yorker magazine.

Triggered by expert systems technology, artificial intelligence (AI) was a silver bullet in the early 1980s. AI seemed to be able to perfectly solve various problems that involved any intellectual activities. For instance, MYCIN (Shortliffe, 1976), developed at Stanford, gave a strong impression that medical doctors could have been soon supported by a clever consultation system, resulting in more accurate diagnoses. Inspired by these, also in engineering fields, a variety of experimental systems for fault diagnosis, planning, selection, and design were developed, which demonstrated promising possibilities of their applications.

Achieving the dimensional integrity for a complex structural assembly is a demanding task due to the manufacturing variations of parts and the tolerance relationship between them. Although assigning tight tolerances to all parts would solve the problem, an economical solution is taking advantage of small motions that joints allow, such that critical dimensions are adjusted during assembly processes. This paper presents a systematic method that decomposes product geometry at an early stage of design, selects joint types, and generates subassembly partitioning to achieve the adjustment of the critical dimensions during assembly processes. A genetic algorithm generates candidate assemblies based on a joint library specific for an application domain. Each candidate assembly is evaluated by an internal optimization routine that computes the subassembly partitioning for optimal in-process adjustability, by finding a series of minimum cuts on weighted graphs. A case study on a three-dimensional automotive space frame with the accompanying joint library is presented.

Biomimetic, or biologically inspired, design uses analogous biological phenomena to develop solutions for engineering problems. Several instances of biomimetic design result from personal observations of biological phenomena. However, many engineers' knowledge of biology may be limited, thus reducing the potential of biologically inspired solutions. Our approach to biomimetic design takes advantage of the large amount of biological knowledge already available in books, journals, and so forth, by performing keyword searches on these existing natural-language sources. Because of the ambiguity and imprecision of natural language, challenges inherent to natural language processing were encountered. One challenge of retrieving relevant cross-domain information involves differences in domain vocabularies, or lexicons. A keyword meaningful to biologists may not occur to engineers. For an example problem that involved cleaning, that is, removing dirt, a biochemist suggested the keyword “defend.” Defend is not an obvious keyword to most engineers for this problem, nor are the words defend and “clean/remove” directly related within lexical references. However, previous work showed that biological phenomena retrieved by the keyword defend provided useful stimuli and produced successful concepts for the clean/remove problem. In this paper, we describe a method to systematically bridge the disparate biology and engineering domains using natural language analysis. For the clean/remove example, we were able to algorithmically generate several biologically meaningful keywords, including defend, that are not obviously related to the engineering problem. We developed a method to organize and rank the set of biologically meaningful keywords identified, and confirmed that we could achieve similar results for two other examples in encapsulation and microassembly. Although we specifically address cross-domain information retrieval from biology, the bridging process presented in this paper is not limited to biology, and can be used for any other domain given the availability of appropriate domain-specific knowledge sources and references.

Neural network (NN)-based constitutive models have been used increasingly to capture soil constitutive response. When combined with the self-learning simulation (SelfSim) inverse analysis framework, NN models can be used to extract soil behavior when given field measurements of boundary deformations and loads. However, the data sets used to train and repeatedly retrain the NN models are large, and training times, especially when used in SelfSim, are long. A diverse set of stress–strain data is extracted from a simulated braced excavation problem to train a NN-based constitutive model. Several methods for reducing the data set size are proposed and evaluated. Each of these methods selectively removes training data so that the smallest amount of data is used to train the NN. The Gaussian point method removes data based on its position in each finite element in the model. The lattice method removes data so that all remaining points are evenly spaced in stress space. Finally, the loading path method compares the stress–strain history of each Gaussian point and removes points with similar loading histories. Each of these methods shows that a large amount of the training data (up to 94%) can be removed without adversely affecting the performance of the NN model, with the loading path method showing the best and most consistent performance. Model training times are reduced by a factor of 20. The performance of the loading path method is also demonstrated using stress–strain data extracted from a simulated laboratory triaxial compression test with frictional ends.

The design of a new car is guided by a set of directives indicating the target market, specific engineering, and aesthetic constraints, which may also include the preservation of the company brand identity or the restyling of products already on the market. When creating a new product, designers usually evaluate other existing products to find sources of inspiration or to possibly reuse successful solutions. In the perspective of an optimized styling workflow, great benefit could be derived from the possibility of easily retrieving the related documentation and existing digital models both from internal and external repositories. In fact, the rapid growth of resources on the Web and the widespread adoption of computer-assisted design tools have made available huge amounts of data, the utilization of which could be improved by using more selective retrieval methods. In particular, the retrieval of aesthetic elements may help designers to create digital models conforming to specific styling properties more efficiently. The aim of our research is the definition of a framework that supports (semi)automatic extraction of semantic data from three-dimensional models and other multimedia data to allow car designers to reuse knowledge and design solutions within the styling department. The first objective is then to capture and structure the explicit and implicit elements contributing to the definition of car aesthetics, which can be realistically tackled through computational models and methods. The second step is the definition of a system architecture that is able to transfer such semantic evaluation through the automatic annotation of car models.