Explainable deep reinforcement learning reveals energy-efficient control strategies for turbulent drag reduction

2026-05-31 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Fluid Dynamics · Depth: Expert, quick

Summary

A novel method integrating Multi-Agent Deep Reinforcement Learning (MARL) and eXplainable Deep Learning (XDL) has been developed to reduce drag in wall-bounded turbulent flows. This approach compares three SHAP-guided strategies against baselines of direct wall-shear stress and opposition control. The most effective strategy combines SHAP attributions from two U-nets, one predicting skin-friction coefficient and another predicting wall pressure fluctuations. This combined SHAP method achieved a 34.44% drag reduction (DR) and 34.01% net energy saving (NES) with only 0.43% normalized input power. Relative to opposition control, DR and NES improved by 49.41% and 48.52% respectively. Furthermore, it reduced normalized actuation cost from 5.90% to 0.43% compared to the direct wall-shear-stress baseline, while simultaneously enhancing performance. Analysis indicates the energy-efficient policy uses pressure-gated actuation, primarily activating at near-zero wall pressure, on a timescale matching near-wall turbulent structures.

Key takeaway

For AI Scientists and Research Scientists developing fluid control systems, this research demonstrates a powerful approach. You should consider integrating eXplainable Deep Learning (XDL) attributions, specifically SHAP, into Multi-Agent Deep Reinforcement Learning (MARL) reward functions. This strategy can yield substantial improvements in drag reduction and net energy savings, reducing actuation costs significantly. Focus on combining multiple predictive signals, like skin-friction and wall pressure, to optimize control policies for energy efficiency.

Key insights

XDL-guided MARL significantly improves turbulent drag reduction and energy efficiency in fluid dynamics.

Principles

• XDL attributions can guide MARL rewards.
• Combining multiple XDL signals enhances control.
• Pressure-gated actuation is energy-efficient.

Method

The method combines MARL with SHAP-guided U-nets predicting fluid dynamics parameters (skin-friction, wall pressure) to compute rewards for drag reduction agents, leading to optimized control policies.

In practice

• Apply SHAP to U-net predictions for MARL rewards.
• Integrate multiple physical signals for control.
• Explore pressure-gated actuation for efficiency.

Topics

• Multi-Agent Reinforcement Learning
• Explainable AI
• Turbulent Drag Reduction
• Fluid Dynamics Control
• SHAP
• U-net Architectures

Best for: AI Scientist, Research Scientist

Related on AIssential

• TN-SHAP-G: Graph-Structured Tensor Network Surrogates for Shapley Values and Interactions — TN-SHAP-G efficiently computes Shapley values for graph-structured inputs using a graph-aligned tensor network surrogate.
• The Case for Model Science: Verify, Explore, Steer, Refine — The AI community needs a systematic "Model Science" to understand why complex models behave as they do, beyond benchmarking what they do.
• TriLens: Per-Layer Logit-Lens Entropy for White-Box Hallucination Detection — Internal model uncertainty, specifically per-layer logit-lens entropy, can effectively signal hallucination before final output.
• Beyond Task Success: Behavioral and Representational Diagnostics for WAM and VLA — WAMs improve object-level robot behavior, but architectural choices impact predictive representation and inference cost.
• Detection vs. Execution: Single-Bucket Probes Miss Half the Mamba-2 State Sink — Representational similarity in Mamba-2 does not imply functional equivalence for mechanistic interpretability probes.
• Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection — Fine-tuning VLMs with natural-language rationales significantly improves time-series anomaly detection performance and interpretability.

Open in AIssential →

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