External Thermal Monitoring for Transformer Local Heat Loss Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for monitoring complex assets in electric power systems, such as transformers or shunt reactors, struggle to detect local losses and malfunctions due to structural components that attenuate thermal measurements, leading to potential undetected issues.
Innovation Solution
A physics-informed data-driven processing technique that uses measurements from external locations on the asset tank wall, incorporating knowledge of thermal diffusion processes through partial differential equations, to determine parameters like temperature and heat in a spatiotemporally resolved manner, while filtering out unreliable measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If measurements are obtained from external locations on the asset tank wall, then the ease of operation is improved, but the measurement precision deteriorates due to structural components attenuating thermal measurements
Solution Approach 1:
The patent introduces a physics-informed machine learning model as an intermediary that processes external thermal measurements and infers internal asset conditions. The model acts as a mediator between the external measurements (which are easy to obtain but attenuated by tank wall structures) and the internal thermal states (which provide precise diagnostic information about local losses and malfunctions).
Solution Approach 2:
The patent replaces direct physical measurement (mechanical/thermal contact instrumentation inside the asset) with a computational system that uses physics-informed machine learning. Instead of placing sensors internally (which would provide direct measurements), the system uses external thermal cameras or sensors combined with a physics-informed neural network to infer internal conditions, substituting physical measurement with computational inference.
2Reliability
If physics-informed data-driven processing technique is used, then the reliability is improved, but the device complexity increases due to embedding differential equations
Solution Approach 1:
The patent merges physics-based modeling (differential equations describing thermal diffusion) with data-driven machine learning approaches into a unified physics-informed machine learning model. This combination allows the system to leverage the interpretability and physical consistency of differential equations while incorporating the adaptability and pattern recognition capabilities of machine learning, thereby improving reliability without requiring separate systems.
Solution Approach 2:
The patent creates a composite processing technique that combines two distinct approaches: physics-informed modeling (based on differential equations) and data-driven machine learning. This composite approach integrates the strengths of both methods, where the physics component ensures physical consistency and the data-driven component provides adaptability to specific asset conditions, resulting in a more robust system than either approach alone.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enhances the robustness and accuracy of detecting local losses and malfunctions by embedding physical relationships, allowing for timely alerts and control actions based on reliable thermal data without requiring internal instrumentation.
Implementation Method 1
the physics informed data-driven processing technique having embedded therein knowledge of a physical process (e.g., thermal diffusion) described by a partial differential equation (PDE)
Data Source
Figure 1
Figure 2
Figure 3
AI summary
To process measurements acquired using measurement instrumentation (41, 42), a processing system (50) is operative to perform a physics informed data-driven processing technique to determine at least one parameter related to an asset (20) of an electric power system (10). The processing system (50) is operative to perform a preprocessing (70) of measurements to generate input data, with the preprocessing (70) comprising identifying a subset of measurement locations for which the measurements are reliable for inputting to the data-driven processing technique and using measurements obtained at the identified subset for generating the input data. The processing system (50) is operative to estimate local heat loss, using the data-driven processing technique.