Learning-Based Decision Making for Combustion Phasing Control in Multi-Fuel CI Engines with Latent Fuel Reactivity Estimation

2026-06-16 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Combustion Engine Control · Depth: Expert, quick

Summary

A new GRU-guided Reinforcement Learning framework addresses combustion phasing control in multi-fuel compression-ignition (CI) engines, which face challenges from uncertain, time-varying fuel reactivity (cetane number, CN). This framework formulates CA50 regulation under latent CN variation as a partially observable sequential decision problem. It learns a compact GRU-based representation of fuel reactivity from combustion history, conditioning both the actor and critic on this estimated signal. Evaluation using a Gaussian-process surrogate, trained on experimental multi-fuel engine data, demonstrates that the proposed method outperforms myopic/fixed-history bandit methods and observation-only DDPG, especially when CN evolves rapidly. The policy achieves stable CA50 regulation with a mean absolute tracking error below 0.25° CA at the training setpoint, producing smooth, physically consistent SOI and glow-plug-power actuation across unseen CN trajectories.

Key takeaway

For Machine Learning Engineers designing control systems for multi-fuel engines or similar dynamic systems with latent, time-varying parameters, you should integrate state estimation directly into your Reinforcement Learning policy learning. This approach, which conditions the policy on the same imperfect state information available at deployment, significantly improves control robustness and avoids train-deploy inconsistency, leading to more stable and physically consistent actuation.

Key insights

Aligning fuel-reactivity inference with control policy learning avoids train-deploy inconsistency in dynamic systems.

Principles

• Latent, continuously evolving fuel dynamics necessitate more than standalone estimation or generic recurrence.
• Training policies on imperfect, deployable state estimates improves robustness and consistency.

Method

The proposed GRU-guided RL framework learns a compact GRU-based representation of fuel reactivity from combustion history, then conditions both actor and critic on this estimated signal.

In practice

• Integrate GRU-based networks for latent state estimation directly into RL policy architectures.
• Evaluate control policies against unseen dynamic trajectories of latent variables.

Topics

• Multi-fuel Engines
• Combustion Control
• Reinforcement Learning
• GRU Networks
• Cetane Number
• Partially Observable MDPs

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

Related on AIssential

• Data-driven Control with Real-time Uncertainty Compensation for Multi-Fuel Engines — A data-driven framework uses GPR and uncertainty compensation for real-time combustion control in multi-fuel engines.
• Demystifying Hidden-State Recurrence: Switchable Latent Reasoning with On-Policy Reinforcement Learning — Explicit boundary tokens enable on-policy RL optimization and mechanistic analysis of hidden-state recurrence in latent reasoning models.
• Timescale Separation Enables Deep Reinforcement Learning Control of Rotating Detonation Engine Mode Transitions — Reformulating DRL in a moving reference frame enables effective control of multi-timescale systems like RDEs.
• Multi-Agent Reinforcement Learning from Delayed Marketplace Feedback for Objective-Weight Adaptation in Three-Sided Dispatch — Reinforcement learning can safely adapt dispatch objective weights in complex, real-world marketplaces using delayed feedback.
• Co-Learning Port-Hamiltonian Systems and Optimal Energy-Shaping Control — Co-learning pH system models and optimal EB-PBC policies via alternating optimization ensures stability and robustness in control.
• Two-Layer Linear Auto-Regressive Models Estimate Latent States — Two-layer linear auto-regressive models approximate Kalman filtering for latent state estimation in partially observed linear dynamical systems.

Open in AIssential →

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