Abstract
Modern internal combustion engines (ICEs) require complex calibration of numerous actuators to balance fuel efficiency with stringent emission standards. This paper introduces a semi-automatic framework that combines adaptive measurement sampling with Integer Linear Programming (ILP) to optimize engine lookup tables. By replacing traditional grid-based testing with an intelligent sampling and discrete optimization approach, the researchers achieved a 15–20x speedup in calibration time while ensuring compliance with complex standards like EURO 5.
Introduction
The calibration of modern ICEs is increasingly difficult due to the "curse of dimensionality." Engineers must manage dozens of actuators—such as fuel injection timing and air valve angles—to create Engine Control Unit (ECU) lookup tables that map engine speed (RPM) and torque to optimal settings. This process faces two primary hurdles: the exponential growth of required tests in high-dimensional spaces and the need to satisfy conflicting constraints between fuel consumption, pollutant limits (NOx, CO, etc.), and drivability.
Traditional model-based methods often fail to capture the non-linear and non-smooth behavior of engines. To address this, the authors propose a three-stage semi-automatic pipeline designed to streamline the calibration process without sacrificing accuracy or engine safety.
Methodology
The proposed framework consists of three core components:
- Adaptive Meshing & Routing: Instead of a fixed grid, the system utilizes "smart sampling" to concentrate measurements in regions where engine behavior is highly sensitive. A graph-based approach using Dijkstra’s algorithm determines the most efficient path for the test bench, minimizing idle time.
- Adaptive Space Compressor: To manage noise and redundancy, a "space compressor" algorithm filters raw measurement data into a representative subset. This ensures that data points are well-distributed for local polynomial fitting.
- Discrete Optimization (ILP): The engine's operating field is divided into "stacks" (small RPM/torque rectangles). An ILP solver selects exactly one measured data point from each stack to construct the final lookup table. This allows for a globally optimal solution that treats emission limits and actuator variation speeds (drivability) as hard mathematical constraints.
Experimental Evaluation
The researchers evaluated the framework using the AVL Engine Model, a sophisticated simulation of a diesel engine featuring 8 dynamic actuators and 18 sensor outputs. The results were validated against standard driving cycles, including the New European Driving Cycle (NEDC) and a RANDOM cycle, to assess real-world applicability.
Results and Discussion
The semi-automatic method demonstrated significant efficiency gains, requiring 15 to 20 times fewer measurements than traditional uniform grid approaches. For instance, a full EURO 5-compliant engine map was obtained in roughly 107 hours of simulated measurement time—a task that typically takes weeks on physical test benches.
Beyond speed, the optimization successfully adhered to legal pollutant limits. The study highlighted the necessity of wide-field calibration; maps optimized solely for specific test cycles performed poorly in uncovered regions, emphasizing the robustness of the ILP approach.
Conclusion and Future Work
By formalizing drivability as a Lipschitz continuity condition and treating calibration as a discrete optimization problem, the authors have provided a tool that significantly reduces the manual workload of engine engineers. While the current study was conducted on high-fidelity simulations, future work will focus on deploying this framework on physical combustion engine test benches to validate its performance against real-world variables like ambient temperature and fuel quality fluctuations, particularly as standards evolve toward the even stricter EURO 6 requirements.
Summary updated