OpenPyStruct: Open-Source Toolkit for Machine Learning-Driven Structural Optimization

Abstract

OpenPyStruct is a first-version open-source toolkit that provides finite element model based optimization frameworks for generating training data and machine learning models for global structural optimization of indeterminate continuous structures. The key machine learning feature of OpenPyStruct is its ability to optimize subject to single or multiple arbitrary simultaneous loading and/or support conditions. The framework utilizes multi-core CPU and GPU-enhanced implementations integrating OpenSeesPy forward solvers in structural optimization, leveraging PyTorch for accelerated computations. Accompanying machine learning scripts enable users to train high-fidelity predictive models using transformer architectures with diffusion modules, physics-informed neural networks (PINNs), convolutional operations, and contemporary machine learning techniques to analyze and optimize structural designs. By incorporating state-of-the-art optimization tools, robust datasets, and flexible machine learning resources, OpenPyStruct aims to establish a scalable – fully transparent – engine for structural optimization by engaging the structural engineering community in this open-source project.