Grant: Discovery Grants Program - Individual
McGill University
$53,000.00
Bayesian optimization with automated machine-learned models for computational engineering design
EF 2024-25
Résumé
Numerical optimization and computational engineering design rely on quantitative information to make decisions: an engineering system is modeled computationally so that its design variables to be optimized are included in the model's input and the criteria of the optimization are derived from the model's output. These models have been developed traditionally using the laws of physics that govern a system's behavior in relation to it environment. The increasing complexity of engineering systems and the proliferation of available data through other means are shifting the modeling paradigm: we need, and often can, learn about systems from data using machine-learned models.
This paradigm is not entirely new: as the physics-based models mentioned above became more complex and computationally intensive, the engineering design optimization community first resorted to using so-called surrogate models that were built using data obtained from exercising the physics-based models. Since the latter are computationally expensive, the challenge is how to build sufficiently accurate surrogate models using limited amounts of data. Despite much progress, the curse of dimensionality is still prohibitive: the amount of data required for building adequately accurate surrogates grows exponentially with increasing number of design variables.
In addition, traditional computational design workflows involve many repetitive resource-intensive analysis tasks, whose execution, particularly in an industrial context, build a vast amount of domain knowledge available in form of data. This knowledge should be utilized to generate designs directly, instead of iteratively, in an automated manner, which could also lead to innovative design solutions.
In this context, the proposed research aims at 1) developing Bayesian optimization methods to address the curse of dimensionality and other model-based design optimization challenges such as non-computable functions and hidden constraints, 2) automating the optimal selection and configuration of machine-learned models, and 3) developing generative models that enable inverse design by means of artificial intelligence.
Machine learning (ML) and artificial intelligence (AI) are becoming an integral part of engineering. The proposed research will train future engineers to apply these methods in practice and lead future research. The ML-based and AI-driven methods and tools to be developed in this research will address persisting challenges of computational engineering design optimization and catapult its potency. This will enable innovative solutions to complex problems, providing Canadian industry, with which the applicant is already collaborating, with a competitive advantage. By introducing ML and AI into computational engineering design, this program will also contribute to Canada maintaining the lead it has built by investing in data science research and development.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2024-04794
Name nom
Kokkolaras, Michael
Department département
Mechanical Engineering
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
53000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1512
Committee name en
Mechanical Engineering
Committee name fr
Génie mécanique
Area of application code
1206
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Engineering
Area of application fr
Génie
Research subject code
2100
Research subject group en
Mechanical engineering
Research subject group fr
Génie mécanique
Research subject en
Mechanical engineering
Research subject fr
Génie mécanique
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Bayesian optimization with automated machine-learned models for computational engineering design
Keywords
computational engineering design; numerical optimization; distributed multidisciplinary optimization; bayesian optimization; derivative-free optimization; data-driven modeling; scientific and physics-informed machine learning; generative design; aerospace, energy, and healthcare systems applications; optimization applications of quantum computing
Application summary
Numerical optimization and computational engineering design rely on quantitative information to make decisions: an engineering system is modeled computationally so that its design variables to be optimized are included in the model's input and the criteria of the optimization are derived from the model's output. These models have been developed traditionally using the laws of physics that govern a system's behavior in relation to it environment. The increasing complexity of engineering systems and the proliferation of available data through other means are shifting the modeling paradigm: we need, and often can, learn about systems from data using machine-learned models.
This paradigm is not entirely new: as the physics-based models mentioned above became more complex and computationally intensive, the engineering design optimization community first resorted to using so-called surrogate models that were built using data obtained from exercising the physics-based models. Since the latter are computationally expensive, the challenge is how to build sufficiently accurate surrogate models using limited amounts of data. Despite much progress, the curse of dimensionality is still prohibitive: the amount of data required for building adequately accurate surrogates grows exponentially with increasing number of design variables.
In addition, traditional computational design workflows involve many repetitive resource-intensive analysis tasks, whose execution, particularly in an industrial context, build a vast amount of domain knowledge available in form of data. This knowledge should be utilized to generate designs directly, instead of iteratively, in an automated manner, which could also lead to innovative design solutions.
In this context, the proposed research aims at 1) developing Bayesian optimization methods to address the curse of dimensionality and other model-based design optimization challenges such as non-computable functions and hidden constraints, 2) automating the optimal selection and configuration of machine-learned models, and 3) developing generative models that enable inverse design by means of artificial intelligence.
Machine learning (ML) and artificial intelligence (AI) are becoming an integral part of engineering. The proposed research will train future engineers to apply these methods in practice and lead future research. The ML-based and AI-driven methods and tools to be developed in this research will address persisting challenges of computational engineering design optimization and catapult its potency. This will enable innovative solutions to complex problems, providing Canadian industry, with which the applicant is already collaborating, with a competitive advantage. By introducing ML and AI into computational engineering design, this program will also contribute to Canada maintaining the lead it has built by investing in data science research and development.