Neural Inferential Sensing for Nonlinear Parameter Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional inferential sensing methods struggle to accurately estimate the time-varying parameters of complex, highly nonlinear systems, particularly in applications like autonomous navigation and spacecraft operations, due to underconstrained estimation problems and the lack of generalized parameter estimation methods.
Innovation Solution
A two-step process involving a filter operating in Failure Mode 2 and a multiple concurrent filter methodology, followed by an auxiliary neural network, to estimate time-varying parameters based on input-output behaviors, allowing for unlimited parameter estimation independent of the number of system outputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional inferential sensing methods (e.g., Kalman Filter variants) are used, then state and parameter estimation is supported for linearizable systems with few parameters, but the method cannot accurately estimate time-varying parameters of complex highly nonlinear systems with many parameters
Solution Approach 1:
The patent replaces conventional mechanical/mathematical filtering methods (Kalman Filters) with a neural network-based adaptive inferential sensor. The neural network learns complex nonlinear mappings from system inputs and outputs to estimate time-varying parameters, overcoming the limitations of linearizable state-space models and enabling accurate parameter estimation for highly nonlinear systems without requiring accurate reduced-order models
Solution Approach 2:
The patent transforms the estimation approach by changing from fixed-structure linear filters to adaptive neural networks with learnable parameters. The neural network dynamically adjusts its internal parameters during operation to adapt to time-varying system characteristics, enabling accurate estimation of multiple time-varying parameters in nonlinear systems where conventional methods fail
2Adaptability or versatility
If the number of time-varying parameters to be estimated increases, then the estimation problem becomes underconstrained, but conventional methods cannot handle underconstrained estimation applications
Solution Approach 1:
The patent implements a feedback mechanism where the neural network continuously receives system inputs and outputs, compares predicted outputs with actual outputs, and adjusts its parameter estimates accordingly. This closed-loop adaptive learning enables the system to reliably estimate multiple time-varying parameters even when the estimation problem is underconstrained, as the neural network leverages temporal correlations and patterns in the data that conventional methods cannot exploit
3Adaptability or versatility
If specialized ad hoc methods are used for highly nonlinear systems, then limited scope adaptive control solutions are provided, but broadly applicable inferential sensing is not achieved
Solution Approach 1:
The patent creates a universal adaptive inferential sensor based on neural networks that can estimate any number of time-varying parameters in any nonlinear system without requiring system-specific customization. The neural network architecture provides a general framework that adapts to different systems through learning, eliminating the need for specialized ad hoc methods for each application and achieving broad applicability across autonomous navigation, spacecraft operations, and other complex systems
Data Source
AI summary
The embodiments are directed to an inferential sensing system, methods and computer program product of an estimator for estimating parameters of complex nonlinear time-varying systems from scarce system output measurements. The estimator comprises a two-step process to accurately estimate the time-varying parameters of the time-varying system based on the input and output sample of the time-varying system. First, multiple filters in the high frequency processing loop, operating independently and concurrently process the input and output samples of the time-varying system to generate a hypersurface comprising time series objects. Each filter is restricted to adapt only a subset of the modeled time-varying parameters. The hypersurface comprising the time series objects is aggregated over several iterations of the high frequency processing loop. Second, the hypersurface is passed through a neural network in the low frequency processing loop to infer estimates of the time-varying system parameters.