AI/ML enhancement of a quasi-Newton acceleration scheme for computational multi-physics

Multi-physics considerations are important for many engineering applications. For example, we may want to know at what rate heat is transferred between two different systems or how a fluid and solid interact with each other for applications such as turbo-machinery, parachutes, or blood flow through an artery. Engineers often rely on numerical methods to solve these problems, which combine multiple single-physics solvers, and resolve the physical interactions between the various solvers using an iterative approach. Unfortunately, this approach can often be numerically unstable. A generic technique which can be applied to stabilize the iterations is a data-drive quasi-Newton approach. However, the method can be prohibitively expensive in some situations. There is also a reliance on linear algebra techniques using past information to attempt to capture data which is often nonlinear.

The proposed project will investigate the use of AI/ML technology to compress the operating space of the quasi-Newton algorithm and provide a more natural filter for data which is no longer relevant to the algorithm.

Faculty Supervisor:

Rajeev Jaiman

Student:

Partner:

ANSYS Canada Ltd.

Discipline:

Engineering

Sector:

Information and cultural industries; Professional, scientific and technical services

University:

The University of British Columbia

Program: