Aircraft Engine Fault Accommodation via Neural Network Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex aircraft engine systems face challenges in fault accommodation due to their highly nonlinear nature, making it difficult to design optimal control strategies that maintain safety and performance metrics across various flight conditions and disturbance inputs, as conventional optimization methods are inadequate for real-time adjustments.
Innovation Solution
A method using predictive modeling and multi-objective optimization with a simulated machine and controller to generate a Pareto frontier-based solution space, allowing for real-time adjustments of actual engine control settings to accommodate faults, thereby maximizing desirable operational conditions and minimizing undesirable ones.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional optimization methods are used for fault accommodation in aircraft engines, then the control design can be simplified, but the methods fail to provide real-time adjustments for highly nonlinear systems
Solution Approach 1:
The patent segments the fault accommodation problem into two distinct phases: an offline training phase where the neural network is trained using historical data and simulation, and an online execution phase where the trained network provides real-time control adjustments. This segmentation allows complex computational work to be performed beforehand, enabling simple real-time operation without sacrificing accuracy or real-time capability.
Solution Approach 2:
The patent performs preliminary action by training the neural network offline before actual fault conditions occur. The system pre-processes training data, pre-trains the network architecture, and pre-computes optimal control strategies for various fault scenarios. This preliminary preparation enables the system to respond immediately to faults without requiring complex real-time computations, thus achieving both simplicity and real-time performance.
2Adaptability or versatility
If a large collection of computer modules with switching variables and logical elements is used in the engine controller, then the control system can handle complex nonlinear operations, but the control design space becomes high-dimensional and discontinuous
Solution Approach 1:
The patent replaces the traditional mechanical/control system approach (using switching variables, logical elements, and lookup tables) with a neural network-based computational system. Instead of using discrete switching logic and multiple computer modules, the system employs continuous neural network computations that naturally handle nonlinear relationships. This substitution reduces structural complexity while maintaining or enhancing adaptability to fault conditions.
Solution Approach 2:
The patent changes the fundamental parameters of the control system from discrete switching variables and logical states to continuous neural network weights and activations. This parameter transformation allows the system to represent complex nonlinear control strategies in a continuous, differentiable manner, eliminating the discontinuities inherent in switching-based control architectures while preserving the ability to handle high-dimensional control spaces.
3Ease of manufacture
If linear control system theories are applied to aircraft engine control, then the design process becomes more straightforward, but the theories are not relevant to the highly nonlinear nature of engine controllers
Solution Approach 1:
The patent substitutes linear control theories with neural network-based nonlinear control. The neural network is trained to learn the complex nonlinear relationships between engine states and optimal control actions, replacing linear approximation methods. This substitution maintains design processability through systematic training procedures while achieving high accuracy for nonlinear fault accommodation scenarios.
Solution Approach 2:
The patent changes the control approach from linear parameter adjustments to nonlinear parameter transformations learned by the neural network. Instead of relying on linear control laws that assume small deviations from operating points, the system uses neural network weights and activations that capture complex nonlinear relationships across the full operating range, including fault conditions. This parameter transformation enables both systematic design and high accuracy for nonlinear systems.
Data Source
AI summary
A method for multi-objective fault accommodation using predictive modeling is disclosed. The method includes using a simulated machine that simulates a faulted actual machine, and using a simulated controller that simulates an actual controller. A multi-objective optimization process is performed, based on specified control settings for the simulated controller and specified operational scenarios for the simulated machine controlled by the simulated controller, to generate a Pareto frontier-based solution space relating performance of the simulated machine to settings of the simulated controller, including adjustment to the operational scenarios to represent a fault condition of the simulated machine. Control settings of the actual controller are adjusted, represented by the simulated controller, for controlling the actual machine, represented by the simulated machine, in response to a fault condition of the actual machine, based on the Pareto frontier-based solution space, to maximize desirable operational conditions and minimize undesirable operational conditions while operating the actual machine in a region of the solution space defined by the Pareto frontier.