Virtual Sensors for Missing Industrial Machine Measurements
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial machines often lack appropriate sensors for certain process parameters, and even when sensors are present, measurement data may not be available due to sensor failures, data connection issues, or heterogeneity in machine configurations, posing challenges for real-time monitoring and control.
Innovation Solution
A neural network is trained using historical data from reference machines with sensors for the desired process parameters, enabling the network to provide parameter indicators for parameters not directly measurable, through unsupervised domain adaptation and transfer learning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensors are installed to measure process parameters, then measurement data availability improves, but device complexity and cost increase
Solution Approach 1:
The patent creates virtual sensors that generate measurement data through computational models and neural networks, copying the function of physical sensors without the hardware complexity. The virtual sensor system replicates measurement capabilities by processing data from existing sensors and applying machine learning models to infer parameters that would otherwise require additional physical sensors.
Solution Approach 2:
The patent replaces physical sensor systems with an information-processing system based on neural networks and computational models. Instead of installing additional mechanical/electrical sensors, the system uses software-based virtual sensors that substitute hardware measurement functions with algorithmic data generation.
2Measurement precision
If physical sensors are used for measurement, then direct measurement capability is achieved, but reliability decreases due to sensor failures and data connection issues
Solution Approach 1:
The patent prepares backup measurement capabilities by training virtual sensor models in advance using historical data from reference machines. When physical sensors fail or data connections are lost, the pre-trained virtual sensors can immediately provide estimated measurements without interruption, cushioning against the reliability issues of physical sensor systems.
Solution Approach 2:
The virtual sensor system acts as an intermediary between physical sensors and the control system. Instead of directly relying on physical sensor outputs, the system uses virtual sensors that process and validate data through computational models, providing an intermediate layer that improves reliability by compensating for physical sensor failures.
3Reliability
If multiple sensors are installed across heterogeneous machine configurations, then measurement coverage improves, but adaptability decreases due to machine-to-machine differences
Solution Approach 1:
The patent creates universal virtual sensor models that can adapt to multiple machine configurations through transfer learning. The neural network models are trained on data from reference machines with various configurations and can then be applied to target machines, providing a single adaptable measurement solution that works across heterogeneous machine types without requiring configuration-specific sensor installations.
Solution Approach 2:
The virtual sensor system adapts to different machine configurations by changing its internal parameters through transfer learning and fine-tuning. The neural network models adjust their weights and biases based on target machine characteristics, allowing the same virtual sensor framework to accommodate variations in machine geometry, operating conditions, and sensor arrangements.
4Adaptability or versatility
If human operators perform manual measurements, then measurement flexibility is maintained, but productivity decreases and safety risks increase
Solution Approach 1:
The virtual sensor system enables self-service measurement capabilities where the system automatically generates measurement data without human intervention. The neural networks continuously process sensor data and produce parameter estimates autonomously, eliminating the need for operators to perform manual measurements while maintaining measurement flexibility through configurable virtual sensor models.
Solution Approach 2:
The patent replaces manual human measurement operations with automated virtual sensor systems. Instead of operators physically accessing machines to take measurements, the virtual sensor system uses computational models and data processing to automatically generate measurements, improving productivity while maintaining adaptability through software configuration.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An industrial machine (123) may not have a sensor for a particular parameter, so that a computer uses a neural network (473) to virtualize the missing sensor. The computer trains the neural network (373) to provide a parameter indicator (Z') of a further process parameter (173, z) for the industrial machine (123) with steps that comprise receiving measurement time-series with historical measurement data from reference machines, obtaining transformation rules by processing the time-series to feature series that are invariant to domain differences of the reference machines, transforming time-series by using the transformation rules, receiving a uni-variate time-series of the further process parameter (z), and training the neural network with features series at the input, and with the uni-variate time-series at the output.