AI leveraged to improve automotive design process

Researchers have found that machine learning and artificial intelligence (AI) can significantly reduce costs and time in the product design phase within the automotive sector, not only in the actual generative design of the product, but also in the predictive analysis of whether consumers will be attracted to certain designs.

With the research focused on the automotive industry, the findings were revealed in a new study, published by INFORMS journal Marketing Science. The peer-reviewed article is called ‘Product Aesthetic Design: A Machine Learning Augmentation.’ The authors of the study are from Yale University, MIT and Northwestern University.

Source: Getty Images

“It’s well understood in the automotive industry that aesthetics are critically important to market acceptance. An improved aesthetic design has demonstrated that it can boost sales 30 percent or more,” said the study authors. “That’s why automakers are known to invest over USD$1 billion in the design of a single model.”

The current automotive design process relies mostly on both the conventional human development of designs and prototypes, along with in-person testing of possible designs with actual consumers. These consumer evaluations feature the A/B testing of alternative designs in laboratory test markets. The industry calls them “theme clinics,” where hundreds of targeted consumers are recruited and brought to a central location to evaluate aesthetic designs.

Multiple aesthetic designs per vehicle

Consumers are asked to rate the designs based on established benchmarks, such as scales for ‘sporty, appealing, innovative and luxurious,’ among other characteristics. Automotive manufacturers typically invest more than $100,000 per theme clinic for one new vehicle design. Since there are multiple aesthetic designs per vehicle, and over one hundred vehicles in its product line, General Motors alone spends tens of millions of dollars just on theme clinics.

“Through our research, we have found ways to augment the traditional product development process with machine learning tools that address both the generation of the design itself, and the testing of possible consumer acceptance or rejection of the design,” said the study authors.

Allows designers a tool to morph potential designs

The authors added, “We have developed a generative model that creates new product designs and allows designers a tool to morph potential designs more efficiently and effectively. And we have created a predictive model that helps identify those designs with high aesthetic scores. The predictive model is designed to screen newly proposed aesthetic designs so that only the highest-potential designs need to be tested in theme clinics.”

The study authors created their models using data from an automotive firm, using images of 203 SUVs that were evaluated by targeted consumers, and 180,000 high