TL;DR
This paper introduces an "intelligent reduced-dimensional" Model Predictive Control (MPC) scheme for aero-engines. By intelligently selecting a single control variable at each step and coordinating its sequence, they significantly improve control quality over existing methods while drastically reducing computational load compared to standard MPC.
What Problem Does This Paper Address?
Aero-engines require sophisticated control systems to meet demanding performance and safety requirements. Model Predictive Control (MPC) is a powerful technique for this, capable of handling actuator constraints and optimizing performance. However, the computational burden of traditional MPC makes real-time implementation challenging, especially for complex systems like aero-engines. Existing methods to reduce this burden, like multiplexed MPC (mMPC), achieve speedups but often sacrifice control performance, leading to significant errors. This paper aims to bridge this gap by developing an MPC approach that is both computationally efficient and maintains high control quality.
Methods
The core of the proposed method is an "intelligent reduced-dimensional scheme" for MPC. This scheme has three key components:
- Reduced-Dimensional Control Sequence: Instead of considering all control inputs simultaneously, the optimization problem is reformulated to use only one control variable at each sampling instant. This drastically reduces the dimension of the optimization problem.
- Intelligent Control Variable Selection Algorithm: To overcome the inflexibility of simply cycling through control variables (as in mMPC), this algorithm intelligently selects the single control variable that is predicted to have the best control effect at the current sampling instant. This is done by predicting the range of outcomes for each control variable and evaluating a merit function.
- Control Sequence Coordination Strategy: This strategy considers the interactions between control variables across different prediction horizons. It ensures that the chosen control sequence is coordinated, aiming for a better sub-optimal solution than methods that treat each step independently.
These components are integrated into an MPC controller that solves a constrained optimization problem with significantly reduced complexity.
Datasets
Not discussed in the paper. The paper uses a component-level model of a twin-shaft turbofan engine constructed using the Simulink Toolbox for the Modeling and Analysis of Thermodynamics Systems (T-MATS), with component maps and performance data acquired from GasTurb.
Key Insights and Ideas
- Intelligent Selection is Key: The paper highlights that simply reducing the number of control variables considered isn't enough; which variable is chosen matters significantly for performance. Their intelligent selection algorithm is crucial for maintaining control quality.
- Coordination Improves Sub-Optimality: By coordinating the control sequence across prediction horizons, they achieve better results than simply optimizing each step in isolation, even with a reduced control dimension.
- Significant Trade-off Improvement: The proposed method demonstrates that it's possible to achieve substantial computational savings without the drastic performance degradation seen in previous reduced-complexity MPC methods.
Key Results
The proposed "intelligent reduced-dimensional MPC" (irdMPC) controller demonstrated significant improvements:
- Control Quality: Enhanced control quality by 34.06% compared to multiplexed MPC (mMPC).
- Time Consumption: Decreased average time consumption by 64.72% compared to standard MPC (sMPC).
- Performance: Achieved better control performance than mMPC and was closer to sMPC in terms of tracking and constraint management, with smaller Root Mean Squared Errors (RMSE) for the controlled parameter n₁.
Limitations and Future Work
- Linearity: The current methodology is suitable for linear MPC. The authors suggest that extending it to nonlinear MPC is a necessary direction for future research.
- Simulation-Based: The study relies on numerical simulations. The authors propose that hardware-in-the-loop experiments would be valuable for further verification.
Summary updated