Neptuna: A Comprehensive Machine Learning Framework for Benchmarking Complex Multiphase Flows

2026-07-24 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Engineering & Applied Sciences · Depth: Expert, quick

Summary

Neptuna is a new, comprehensive machine learning framework designed for benchmarking complex multiphase flows, specifically addressing challenges in developing surrogates for compressible flows involving shocks and material interfaces. The framework introduces the first large-scale benchmark for shock-driven compressible multiphase flows, comprising 2.4 TB of high-fidelity 2D and 3D datasets featuring shock-induced bubble collapse and droplet breakup. Neptuna evaluates diverse surrogate model families, including convolutional, spectral, transformer-based, and pre-trained PDE foundation models. It also investigates composite losses, combining MSE with Sobolev, interface-aware, and structure-aware terms, alongside adaptive loss balancing using SoftAdapt and GradNorm. Evaluation employs pointwise, spectral, feature-focused, structural, and physics-informed metrics. Results indicate no single model excels across all datasets and metrics, but composite losses significantly enhance interface preservation and spectral fidelity, with SoftAdapt providing consistent improvements with minimal overhead.

Key takeaway

For machine learning engineers developing surrogates for compressible multiphase flows, you should integrate composite loss functions to significantly improve interface preservation and spectral fidelity. Consider implementing SoftAdapt for adaptive loss balancing, as it consistently enhances results with minimal computational overhead. Do not rely on a single model architecture; instead, evaluate diverse model families to find the optimal solution for your specific flow simulation challenges.

Key insights

Neptuna benchmarks shock-driven multiphase flows, showing composite losses and SoftAdapt improve ML surrogate fidelity, despite no single model excelling.

Principles

• No single ML model performs best across all multiphase flow datasets and metrics.
• Composite losses significantly improve interface preservation and spectral fidelity.
• SoftAdapt offers consistent improvements with minimal training overhead.

Method

The framework evaluates ML surrogates (convolutional, spectral, transformer, PDE foundation models) on 2.4 TB of 2D/3D shock-driven multiphase flow data, using composite losses and adaptive balancing.

In practice

• Implement composite losses for better interface and spectral fidelity.
• Use SoftAdapt for adaptive loss balancing with low overhead.
• Test diverse ML model families for complex flow simulations.

Topics

• Multiphase Flows
• Machine Learning Surrogates
• Fluid Dynamics Benchmarking
• Composite Loss Functions
• Adaptive Loss Balancing
• Neptuna Framework

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Surrogate models for Rock-Fluid Interaction: A Grid-Size-Invariant Approach — Grid-size-invariant neural networks offer efficient, scalable surrogate modeling for complex rock-fluid interactions.
• Hybrid Imbalanced Regression Through Unified Data-Level and Algorithm-Level Balancing — The framework unifies data- and algorithm-level balancing to improve imbalanced regression performance.
• REMAL: Residual Equilibrium Manifold Active Learning for Surrogate-Based Multidisciplinary Design Analysis — REMAL models the residual equilibrium manifold using multitask Gaussian processes and active learning for efficient fixed-point prediction in coupled engineering systems.
• Asymmetric-Loss-Guided Hybrid CNN-BiLSTM-Attention Model for Industrial RUL Prediction with Interpretable Failure Heatmaps — A hybrid CNN-BiLSTM-Attention model with asymmetric loss improves turbofan RUL prediction and interpretability.
• Optimal Transport Image Representation and Deep Covariance Alignment (CORAL) for Control Valve Stiction Detection — Combining optimal transport imaging with Deep CORAL effectively mitigates domain shift for robust control valve stiction detection in industrial settings.
• Deep Learning Surrogates for Emulating Stochastic Climate Tipping Dynamics — A dynamics-informed Temporal Fusion Transformer provides a 465x faster, differentiable surrogate for forecasting stochastic climate tipping events with high fidelity.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.