Real-time Linear Model Scheduling for Engine Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating linear models for dynamic systems, such as aircraft engines, face challenges in real-time computation due to high computational load and limited processing power, making offline linear model generation with online scheduling more feasible for real-time control applications.
Innovation Solution
A method involving offline determination and filtering of linear models at selected operating conditions, followed by generating lookup tables based on these models, and creating algorithmic software for online scheduling using scheduling variables, ensuring accurate and efficient linear model generation for control, estimation, and detection purposes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If linear models are generated online for real-time control applications, then model accuracy and adaptability to current operating conditions are improved, but computational load increases beyond available processing power
Solution Approach 1:
The patent pre-computes and stores linear model parameters in lookup tables during an offline phase, before real-time control is needed. This preliminary action transfers the heavy computational burden from online execution to offline preparation, enabling fast real-time model retrieval without exceeding processing power constraints.
Solution Approach 2:
The patent segments the continuous operating space into discrete regions defined by scheduling variables, with each region associated with a pre-computed linear model from the lookup table. This segmentation allows the system to switch between simplified models based on operating conditions, avoiding the need for continuous complex computations while maintaining adequate model accuracy.
2Productivity
If controller gains are computed offline and implemented as scheduled gains, then real-time computational requirements are reduced, but adaptability to system changes and health deterioration is lost
Solution Approach 1:
The patent introduces scheduling variables that dynamically select or interpolate between different pre-computed linear models based on current operating conditions. This dynamic adaptation allows the controller to respond to system changes and health deterioration while still using the efficient lookup table structure, combining offline computation benefits with online adaptability.
Solution Approach 2:
The patent changes the parameters of the linear models by organizing them in lookup tables indexed by scheduling variables representing different operating conditions. This parameter organization enables the system to adapt to changing conditions by selecting appropriate pre-computed models, maintaining adaptability without requiring full online model regeneration.
3Adaptability or versatility
If linear models are generated at substantially any given operating point, then model applicability across the operating envelope is improved, but computational complexity increases
Solution Approach 1:
The patent introduces scheduling variables as additional dimensions to organize linear models in a structured lookup table format. This dimensional organization allows the system to cover the entire operating envelope by indexing models along multiple scheduling variable dimensions, achieving broad applicability while keeping the complexity manageable through systematic organization rather than exhaustive computation.
Data Source
AI summary
A method of generating linear models for a physical system of interest is implemented in steps including, first, determining offline, a set of linear models for the physical system of interest by linearization of a nonlinear computational model of the physical system of interest at selected operating points or from desired data; second, analyzing offline, accuracy of each linear model and eliminating inaccurate linear models therefrom to provide a residual set of linear models; third, generating offline, linear models corresponding to grid points of one or more lookup tables based on the residual set of linear models; fourth, associating offline, lookup table grid points with selected scheduling variables, and fifth, generating algorithmic software for the physical system therefrom such that linear models for the physical system generated offline form the basis for online scheduling of linear models.