Answers are grounded in this paper and cite their sources.
This study presents a proof-of-concept deep learning-based methodology for predicting dynamic flow fields in chemical reactors, addressing the high computational cost of traditional Computational Fluid Dynamics (CFD). The key novelty of this work is the coupling of Proper Orthogonal Decomposition (POD) and Deep Neural Networks (DNNs) to predict both spatial modes and temporal coefficients for new operating conditions in Confined Impinging Jets (CIJs), operating in a self-sustained chaotic flow regime. The methodology follows a four-step workflow: (i) dimensionality reduction of CFD data through POD, (ii) supervised learning of POD flow features using DNNs, (iii) prediction of POD modes and time coefficients for new conditions, and (iv) reconstruction of the full flow field. Two DNNs were developed: one to predict spatial POD modes (DNN_ ϕ ), and the other to predict the corresponding time coefficients (DNN_ a ), both based on standard Multilayer Perceptron (MLP) architectures. The optimized networks enabled reconstruction of 88% of the flow energy while preserving similar mixing scales. While CFD simulations require several days on High-Performance Computers (HPC), the trained DNN generates equivalent predictions within a few seconds, highlighting its potential as a computationally efficient alternative. As a proof of concept, this work provides a foundation for deep learning-based prediction of complex chaotic flows in opposed jets mixers. This proof of concept opens the door to the development of AI-based models for chemical reactor analysis and design. Future work will focus on extending the methodology to a wider range of operating conditions, flow regimes, and reactor configurations, as well as integrating physics-informed learning approaches to improve accuracy and generalization.
No takes yet. Share an insight, caveat, or question.
Brandão et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: