Method for physical system anomaly detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in physical systems face challenges such as unreliable sensor data, outdated system diagrams, high costs of simulation, lack of training examples for machine learning, and fragility of rule-based approaches due to system changes.
Innovation Solution
A method involving a process graph representing physical systems, physics rules, and machine learning models to generate candidate physics models, validate them with sensor data, and detect anomalies using supervised and unsupervised learning techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simulation approaches are used for anomaly detection, then measurement precision is improved, but device complexity and cost increase significantly
Solution Approach 1:
The system segments the anomaly detection process into multiple components: (1) generating candidate physics models from process graphs and physics rules, (2) validating candidates against sensor data to produce valid models, (3) training machine learning error models on validation errors, and (4) using unsupervised learning to detect anomalies in production. This segmentation reduces the complexity of any single component while maintaining overall detection accuracy.
Solution Approach 2:
The system performs preliminary actions by pre-generating and validating physics models before actual anomaly detection. The candidate physics models are created and validated against historical sensor data in advance, producing a set of valid physics models and training error models beforehand. This preliminary preparation reduces the complexity and computational burden during real-time anomaly detection.
2Productivity
If machine learning approaches are used for anomaly detection, then productivity is improved, but reliability deteriorates due to lack of training examples
Solution Approach 1:
The system introduces physics models as an intermediary between sensor data and machine learning analysis. The physics models serve as a mediator that generates meaningful validation errors from the comparison between predicted and actual sensor readings. These physics-based errors provide high-quality training examples for the machine learning error models, ensuring reliability even when training data is limited. The physics rules act as a bridge that transforms raw sensor data into informative training signals.
3Measurement precision
If expert or rule based approaches are used, then measurement precision is improved, but device complexity and maintenance cost increase
Solution Approach 1:
The system changes the parameters of the rule-based approach by automatically generating physics models from process graphs and physics rules, rather than manually configuring complex rule sets. The validation process automatically adjusts model parameters by comparing predictions against sensor data within error tolerances. This parameter-driven approach maintains the precision of physics-based detection while reducing the complexity of rule configuration and maintenance.
4Measurement precision
If traditional simulation is used to handle unreliable sensor data, then measurement precision is improved, but loss of energy increases due to computational expense
Solution Approach 1:
The system extracts and applies only the essential physics rules relevant to each process graph, rather than running comprehensive simulations. By selecting and applying specific physics models that are valid for the given system configuration, the method reduces computational energy consumption while maintaining the ability to detect anomalies in sensor data. The error models further refine this by focusing computational effort only on predicting and analyzing deviations from physics-based expectations.
Data Source
AI summary
A method for detecting anomalies in a physical system generates from a set of physics rules and a process graph representing the system a set of candidate physics models that assign physics rules to portions of the process graph representing sensors. Candidate physics models are rejected if an error between the models and sensor data exceed a predetermined error tolerance. Supervised learning is used to train a machine learning model to predict an error between the physics models and the sensor data. The predicted error and predicted sensor measurements from the physics models are then used to detect anomalies using unsupervised learning on a distribution of error between the predicted sensor measurements and the sensor data.