Automobile Concept Design and Artificial Intelligence

Type de publication:

Conference Paper

Source:

Gerpisa colloquium, Paris (2018)

Mots-clés:

Artificial intelligence, Automobile Development, Conceptual Design, Machine Learning.

Résumé:

Research question (purpose)

About twelve years ago, major automobile manufacturers were claiming electric vehicles were not viable due to battery technology, lack of customer interest, and economics. Then came Tesla and nowadays most OEMs are actively developing their own electric models and bespoke platforms.

Still, most new automobiles reaching the market today are barely incremental products. Moreover, a new automobile takes about three years of development and must remain in the market for around six years. It should be competitive for nine-years (from concept) but it is not uncommon for just launched products to be matched or surpassed by new competitors coming soon afterwards. There maybe a paradox in traditional automobile concept design. Most supporting tools - such as market research, benchmarking, QFD, and brainstorming - are rooted in the present and the past. They are analytical and data interpolative. But new product development needs to create value for the future  - it should be predictive and extrapolative.

Although the product development phase (from project approval) is an orderly and  systematic set of activities, its front-end (preceding project approval, mainly conceptual design) remains largely unstructured and vague - reason why it is called the “fuzzy” front-end. It is performed in a loose and “ad hoc” basis. Distressingly, 85% of  project cost is generated at the front-end (ROSENFELD et al, 2006). Combination of systematic domain knowledge management and artificial intelligence may be part of the answer to those challenges. It may also be an opportunity for developing countries to leapfrog their knowledge and technology gaps vis-a-vis advanced nations (MANYLKA et al, 2015).

 

Methodology (design)

The theoretical framework for this paper is knowledge management and knowledgement management systems theory. Aladi and Leidner (2001) envision a firm as a system that creates, organizes, transfers and applies knowledge. Information becomes knowledge once it is cognitively processed and becomes actionable in individuals’ minds. Information technology - in the form of knowledge management systems - plays a key role to create knowledge-based assets, which become the foundation for competitive advantage. Another theoretical foundation is Design Science Research, which aims the creation of artifacts to solve problems, with  both rigor (scientific soundness) and relevance (applicability) (DRESCH et al, 2015).

Literature review encompasses Knowledge Management, Design Science, Product Development, Product Innovation, Lean Development, Artificial Intelligence, and Machine Learning. Cases from the automobile and aerospace industries - Toyota, Nissan, Tesla, Tata Motors/Jaguar-LandRover, Geely/Volvo/Lync@Co/Polestar, Lockheed Martin – as well as analogous cases from the technology (Amazon, Google) and finance industries are studied to investigate facts and trends in product concept design and application of artificial intelligence.

 

Main results (findings)

Ward and Sobek II (2014) describe product development as a process of learning – i.e., investigating and evaluating all possible alternatives and solutions for solving the given problem. However, automobiles come in a wide range of segments, prices, sizes, needs, usages, tastes, geographical conditions, legal requirements and cultural flavors. It is difficult to map all the variables in the human mind and to make sense of them into coherent and integrated concepts. Concept design is usually performed heuristically and businesses struggle to keep track of knowledge and to use it effectively. There is variability in individual knowledge, skills, and experience. Often, a firm’s knowledge may not be available, be inaccessible or even be forgotten and lost (ALAVI; LEIDNER. 2001). Mistakes happen very often and there is loss of income and customers. A structured concept design approach may reduce those problems, by organizing  information, guiding the exploration of alternatives and integrating solutions ((ULRICH; EPPINGER. 2015).

Alan Turing wrote the original academic paper (1950) on Artificial Intelligence and established the standard criteria for AI: capacity to emulate the activities of the human brain. Machine Learning is a category of AI that “learns” or evolves with “experience” (data). It represents a significant improvement over previous knowledge representation algorithms, since it circumvents the “Polanyi’s Paradox” – knowledge that is tacit but cannot be fully explained, making it impossible to be coded (BRYNJOLFSSON; MC AFEE. 2017). Artificial Neural Networks (ANN) are capable of capturing complex nonlinear representations, handling vast reams of data and are among the most powerful machine learning tools.

According to Simon (1996), a concept is an exclusive set of features (instances) – collectively, they do not belong to other sets (concepts). An ANN  can be  used to summarize those features into a mathematical model (CHEN; WANG; JANG. 1998). Simon identifies two requirements for good predictions: a sound theoretical understanding of the problem and good data about the problem’s initial conditions. The task is not to forecast but to construct alternative scenarios and to analyze their sensitivity to errors on both theory and data. An ANN performs by recalculating and adjusting weights of the feature variables in its equations, as new data is received. But a person must understand the theory and supply the feature sets (scenarios). That implies the ideal combination for concept design is the interaction between a person’s domain knowledge and machine processing power. Through the effort to make sense of data, by analyzing and reevaluating the meaning of features and weights, the man-machine iterative process may boost learning and lead to new (product) insights. Humans remain in charge but become more efficient and effective ((BRYNJOLFSSON; MC AFEE. 2017).

 

Implications

Advanced or free-style chess, a man-machine team way of playing, was created by Garry Kasparov after being defeated by IBM Deep Blue in 1997. It is based on the premise the man-machine combination will always best either of them playing alone (EVERETT, 2015).

The man-machine integration may not only enhance the concept design task but also lead to the rethinking and reinvention of the concept design process itself. Since market-specific data, knowledge and particular expertise tend to stay closer to the native markets, that opens up the possibility for emerging countries to recapture part of the design responsabilities from the advanced ones.

<!--EndFragment-->

Copyright© Gerpisa
Concéption Tommaso Pardi
Administration Géry Deffontaines

Créé avec l'aide de Drupal, un système de gestion de contenu "opensource"