Physics-Informed Neural Networks with Hard Constraints for Inverse Design

Abstract

\bfA \bfb \bfs \bft \bfr \bfa \bfc \bft . Inverse design arises in a variety of areas in engineering such as acoustic, mechanics, thermal/electronic transport, electromagnetism, and optics. Topology optimization is an important form of inverse design, where one optimizes a designed geometry to achieve targeted properties parameterized by the materials at every point in a design region. This optimization is challenging, because it has a very high dimensionality and is usually constrained by partial differential equations (PDEs) and additional inequalities. Here, we propose a new deep learning method---physics-informed neural networks with hard constraints (hPINNs)---for solving topology optimization. hPINN leverages the recent development of PINNs for solving PDEs, and thus does not require a large dataset (generated by numerical PDE solvers) for training. However, all the constraints in PINNs are soft constraints, and hence we impose hard constraints by using the penalty method and the augmented Lagrangian method. We demonstrate the effectiveness of hPINN for a holography problem in optics and a fluid problem of Stokes flow. We achieve the same objective as conventional PDE-constrained optimization methods based on adjoint methods and numerical PDE solvers, but find that the design obtained from hPINN is often smoother for problems whose solution is not unique. Moreover, the implementation of inverse design with hPINN can be easier than that of conventional methods because it exploits the extensive deep-learning software infrastructure.\bfK \bfe \bfy \bfw \bfo \bfr \bfd \bfs . inverse design, topology optimization, partial differential equations, physicsinformed neural networks, penalty method, augmented Lagrangian method \bfA \bfM \bfS \bfs \bfu \bfb \bfj \bfe \bfc \bft \bfc \bfl \bfa \bfs \bfs \bfi fi\bfc \bfa \bft \bfi \bfo \bfn \bfs . 35R30, 65K10, 68T20 \bfD \bfO \bfI .

A Meta-Model to Predict the Drag Coefficient of a Particle Translating in Viscoelastic Fluids: A Machine Learning Approach

“…In addition, for a ML model to be fully predictive under any new or unseen conditions (e.g., flow features), physics must complement the model. This can only be achieved using deep learning models, known as physics-based neural networks [ 57 ] or physics-guided neural networks [ 58 ] that mimic an infinitely deep model. Incorporating physics in DNN is essential in the field of particle-laden fluid flow, because it is not a data-oriented domain (i.e., large datasets can be hardly found).…”

MIONet: Learning multiple-input operators via tensor product

“…Scientific machine learning (SciML) field grows rapidly in recent years, where deep learning techniques are developed and applied to solve problems in computational science and engineering [11]. As an active area of research in SciML, different methods have been developed to solve ordinary and partial differential equations (ODEs and PDEs) by parameterizing the solutions via neural networks (NNs), such as physics-informed NNs (PINNs) [48,30,35,37,47], deep Ritz method [43], and deep Galerkin method [40]. These methods have shown promising results in diverse applications, such as fluid mechanics [38], optics [4], systems biology [44,5], and biomedicine [14].…”

An Optimal Control Method to Compute the Most Likely Transition Path for Stochastic Dynamical Systems with Jumps

“…As the existing methods in the optimal control problem can not be directly applied to our case, we then develop a neural network method to approximate the state and control. Similarly as the idea of the soft constrain method [34], we construct a loss function which makes the state and control satisfy the constrains, and further we obtain the desired most likely transition path when the cost functional attains its minimum value. As a validation for our method, we first apply it to the Maier-Stein system under Gaussian noise with two different parameters, and compare our numerical results with the most likely transition path that is computed though the adaptive minimum action method.…”