Virtual Sensor Using Generative Adversarial Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex technical systems often have physical variables that cannot be measured directly or accurately due to inaccessible positions or ambient conditions, and simple simulation models for these processes are not readily available, making it difficult to estimate or predict these variables efficiently.
Innovation Solution
A computer-implemented method using a generative machine learning model, specifically generative adversarial networks, to estimate and predict physical variables of a technical system by generating values based on measured data from similar systems, acting as a virtual sensor to provide continuous and accurate data without the need for direct measurement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physical sensors are used to measure physical variables, then measurement accuracy is improved, but device complexity and cost increase due to the need for multiple sensors in inaccessible positions
Solution Approach 1:
The patent creates a virtual copy of the physical sensor measurement capability through a generative machine learning model. Instead of deploying additional physical sensors in inaccessible positions, the system trains a generative model on data from accessible sensor positions, enabling the virtual sensor to estimate values at inaccessible locations by learning the spatial and temporal relationships from training data collected when the system is accessible.
Solution Approach 2:
The patent replaces the mechanical/physical sensor system with a computational system. The generative machine learning model substitutes for physical sensors by computing estimated values based on inputs from accessible sensors, thereby eliminating the need for complex physical sensor deployments in difficult-to-reach locations while maintaining measurement capability.
2Measurement precision
If complex simulation models are used to estimate physical variables, then measurement precision is improved, but computational complexity and time requirements increase
Solution Approach 1:
The patent performs preliminary action by training the generative machine learning model in advance during a training phase when the technical system is accessible. The model learns the complex relationships between physical variables from training data collected during this phase. Once trained, the model can rapidly generate estimates during operation without requiring complex real-time computations, thus achieving both accuracy and speed.
Solution Approach 2:
The patent creates a simplified computational copy of the complex physical processes through the trained generative model. The model captures the essential relationships and patterns from complex simulation or measurement data during training, then uses this learned knowledge to generate accurate estimates efficiently during operation, avoiding the need to run complex simulations in real-time.
3Productivity
If simple simulation models are used for quick calculations, then computational speed is improved, but measurement precision deteriorates due to inability to capture complex physical processes
Solution Approach 1:
The patent changes the parameters and structure of the simulation model by using a generative machine learning model with multiple layers and nonlinear transformations. The model architecture includes input layers, hidden layers with activation functions, and output layers, allowing it to capture complex nonlinear relationships in the physical processes while maintaining computational efficiency for real-time predictions.
Data Source
AI summary
An apparatus, such as a virtual sensor, for measuring a value of a physical variable of a technical system includes a generative machine learning model that is trained, e.g., on the basis of generative adversarial networks (GANs for short), in such a way as to take at least one value of a first physical variable of a technical system as a basis for generating and outputting at least one value of a second physical variable of the technical system. The apparatus is configured in such a way as to use the generative machine learning model to generate and output a value of a second physical variable of the technical system on the basis of a measured value of a first physical variable of the technical system.