INVERSE DESIGN OF MANUFACTURING PROCESSES USING PHYSICS-INFORMED NEURAL OPERATOR LEARNING

Machine learning-based approaches, especially neural network models, have achieved remarkable success in diverse scientific applications, where the deep neural operator (DNO) enables the learning nonlinear operators between infinite-dimensional function spaces and is able to approximate governing equations of time-dependent engineering processes. This paper presents two novel applications of DNO learning to accelerate the design of composite materials and their manufacturing processes. First, we develop a DNO-based surrogate model for finite element simulations of transient mechanical response of 3D interpenetrating phase composites (IPCs) fabricated via additive manufacturing. These IPCs consist of randomly structured metallic phases with different elastic moduli. DNO is trained using a sequence-to-sequence incremental learning approach based on 5,000 finite element simulations of an IPC beam subject to random strain loads, resulting in a 98% prediction accuracy on transient stress at various cross-sections of the IPC beam under dynamic loading. The trained DNO enables real-time prediction of mechanical properties, significantly reducing the design iterations for composites. Second, we introduce an enhanced physics-informed DNO (PIDNO+) model for predicting the particle size distribution in biomass comminution processes. PIDNO+ integrates a population balance model into a hybrid DNO+ architecture that incorporates both material and processing parameters. This integration enables high accuracy with a limited small dataset while preserving physical consistency. The shared parameter network with the embedded physical model allows for differential operation in parameter space and enables inverse design of manufacturing parameters to meet target outcomes. Both DNO applications demonstrate the potential of operator learning for fast and accurate surrogate modeling in composite material design and accelerated optimization of manufacturing processes.