Non‐intrusive reduced‐order modeling using convolutional autoencoders

Abstract The use of reduced‐order models (ROMs) in physics‐based modeling and simulation almost always involves the linear reduced basis (RB) methods such as proper orthogonal decomposition (POD). For some nonlinear problems, RB perform poorly, failing to provide an efficient subspace for solution space. manifolds ROMs has gained traction recent years, showing increased performance certain problems over methods. Deep learning been popular this end through autoencoders providing a trial manifold In work, we present non‐intrusive ROM framework steady‐state parameterized partial differential equations that uses convolutional is augmented by Gaussian process regression (GPR) approximate expansion coefficients model. When applied numerical example involving steady incompressible Navier–Stokes solving lid‐driven cavity problem, it shown proposed offers greater prediction full‐order states when compared method employing POD GPR number dimensions.