The recent developments in the research area highlight a significant shift towards integrating deep learning (DL) and machine learning (ML) techniques with traditional engineering and materials science methodologies to address complex problems. A common theme across the studies is the emphasis on uncertainty quantification (UQ) and the development of models that not only predict outcomes with high accuracy but also provide robust confidence intervals to inform decision-making processes. This trend is evident in the application of recursive Bayesian neural networks (rBNN) for constitutive modeling of sands, the use of surrogate-based multiscale approaches for analyzing thermoplastic composites, and the implementation of uncertainty-aware digital twins for robust model predictive control. Additionally, the incorporation of domain-specific knowledge into deep learning models has been shown to significantly enhance predictive performance, especially in data-scarce scenarios. The development of physics-informed neural networks (PINN) and generalizable operator learning methods like MeshONet further underscores the field's move towards more efficient, accurate, and scalable computational models. These advancements are not only improving the predictive capabilities of models but are also enabling the exploration of new materials and the optimization of engineering designs with unprecedented speed and accuracy.
Noteworthy Papers
- A recursive Bayesian neural network for constitutive modeling of sands under monotonic loading: Introduces a novel rBNN framework for uncertainty-aware constitutive modeling, demonstrating superior predictive accuracy and robust confidence intervals.
- Surrogate-based multiscale analysis of experiments on thermoplastic composites under off-axis loading: Presents a surrogate-based multiscale approach using PRNN, offering insights into macroscopic inhomogeneity and improving agreement with experimental data.
- Uncertainty-Aware Digital Twins: Robust Model Predictive Control using Time-Series Deep Quantile Learning: Develops a robust MPC framework with TiDE for real-time, uncertainty-aware decision-making, showcasing less-conservative UQ and enhanced constraint satisfaction.
- Using Domain Knowledge with Deep Learning to Solve Applied Inverse Problems: Highlights the importance of domain-specific knowledge in enhancing deep learning models' predictive performance in data-scarce scenarios.
- Statistical Design of Thermal Protection System Using Physics-Informed Machine learning: Proposes a Bayesian approach with PINN for rapid and accurate estimation of material properties, significantly reducing computational time.
- MeshONet: A Generalizable and Efficient Operator Learning Method for Structured Mesh Generation: Introduces MeshONet, a generalizable method for structured mesh generation, achieving significant speedup and enabling generalization without retraining.
- A topology optimisation framework to design test specimens for one-shot identification or discovery of material models: Offers a topology optimisation framework for designing test specimens, enhancing the robustness of material model calibration.
- Stochastic Deep Learning Surrogate Models for Uncertainty Propagation in Microstructure-Properties of Ceramic Aerogels: Develops an integrated deep learning framework for predicting ceramic aerogels' properties, enabling robust UQ in predictions.
- DoMINO: A Decomposable Multi-scale Iterative Neural Operator for Modeling Large Scale Engineering Simulations: Proposes DoMINO, a novel ML model for engineering simulations, demonstrating scalability, performance, and accuracy.
