Physical System Event Prediction Using Personalized Sensor Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting events like failure in physical systems, such as the heart or aircraft engines, rely on collecting data from multiple similar systems to identify common patterns, which can be inaccurate due to diversity among systems.
Innovation Solution
A method that creates a personalized statistical model of a physical system's normal behavior based solely on data from that system, using sensors to collect streaming data and a clustering model like Kohonen Self-Organizing Map (SOM) to monitor for deviations from normal behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data from multiple similar physical systems is collected to identify common patterns, then the prediction system can be built with sufficient data, but the prediction accuracy deteriorates due to diversity among systems
Solution Approach 1:
The patent segments the prediction problem by creating separate prediction models for each individual physical system rather than using a single model for all systems. Each system's data is processed independently to capture its unique characteristics, thereby maintaining high prediction accuracy despite system diversity.
Solution Approach 2:
The patent applies local quality by tailoring the prediction model to each specific physical system's local characteristics. Each system receives a customized model trained on its own data, allowing the system to account for individual variations and maintain high prediction accuracy for diverse systems.
2Measurement precision
If a personalized model is created for each individual physical system, then the prediction accuracy is improved, but the data collection requirement increases significantly
Solution Approach 1:
Each physical system serves itself by generating its own training data and building its own personalized prediction model. The system uses its own historical data to train the model, eliminating the need to collect data from other similar systems. This self-service approach achieves high prediction accuracy while minimizing external data collection requirements.
3Adaptability or versatility
If data from diverse physical systems is used to build a generalizable model, then the model can apply to multiple systems, but the prediction accuracy deteriorates
Solution Approach 1:
The patent segments the generalization problem by creating separate models for each system rather than forcing a single general model. Each segmented model is optimized for its specific system, achieving high accuracy without sacrificing the ability to handle multiple diverse systems through the segmented architecture.
Solution Approach 2:
Instead of trying to make each system adapt to a general model (top-down approach), the patent inverts the approach by having each system develop its own specialized model (bottom-up approach). This inversion allows each system to maintain its unique characteristics while still being part of a broader framework that handles multiple systems.
Data Source
AI summary
A physical system receives and records measurements from a plurality of sensors for the physical system over a period of time and creates therefrom a statistical model of normal behavior of the physical system. The statistical model is applied to monitor the physical system for events or significant changes. The method detects events or significant changes in the operation or behavior of the physical system responsive to the monitoring of the physical system for events or significant changes. Events or significant changes in the operation or behavior of the physical system may cause a notification or alert, which can be sent a supervisory system.