Equipment Failure Prediction via Simulated Data Integration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing statistical models face challenges in accurately predicting equipment failure modes and degrees of failure when insufficient historical data is available, particularly in cases with sparse data or low sampling rates, leading to inadequate training and reduced predictive accuracy.
Innovation Solution
The system employs a physical model combined with domain knowledge and experimental data to generate simulated data, which is then used to train a statistical model, enabling the prediction of failure modes and degrees of failure even with limited actual data. This approach integrates historical and simulated data to normalize sensor data and improve predictive capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a statistical model is trained using historical data, then predictive accuracy improves, but insufficient historical data leads to inadequate training and reduced accuracy
Solution Approach 1:
The patent generates simulated failure data in advance through physics-based models before actual failure data is available. This preliminary action creates a foundation of training data that enables the statistical model to be trained even when historical failure data is insufficient, directly resolving the contradiction between needing sufficient training data and facing data scarcity
Solution Approach 2:
The patent creates copies of actual equipment behavior through physics-based simulations. These simulated data copies replicate the characteristics of real failure modes and sensor measurements, allowing the model to learn from synthetic representations when real data is unavailable, thereby improving predictive accuracy despite limited historical data
2Quantity of substance
If simulated data is generated using a physical model, then training data availability improves, but model complexity increases
Solution Approach 1:
The patent introduces a physics-based model as an intermediary between the equipment and the statistical model. This intermediary generates simulated data that bridges the gap between physical reality and computational analysis, enabling training without requiring complex direct observation of all failure modes while maintaining physical consistency
Solution Approach 2:
The patent utilizes physics-based models that simulate equipment behavior by changing parameters such as wear levels, material properties, and operating conditions. These parameter changes generate diverse training scenarios without requiring physical modification of the actual equipment, thus increasing data availability while managing complexity through virtual experimentation
3Measurement precision
If sensor data is normalized based on the difference between simulated and historical data, then predictive accuracy improves, but data processing complexity increases
Solution Approach 1:
The patent implements a feedback mechanism where the difference between simulated and historical sensor data is calculated and used to normalize incoming sensor data. This feedback loop continuously adjusts the normalization parameters based on the discrepancy analysis, improving predictive accuracy by aligning simulated expectations with actual observations while automating the complex processing through systematic comparison
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In some examples, a computing device may generate simulated data based on a physical model of equipment. For example, the simulated data may include a plurality of probability distributions of a plurality of degrees of failure, respectively, for at least one failure mode of the equipment. In addition, the computing device may receive sensor data indicating a measured metric of the equipment. The computing device may compare the received sensor data with the simulated data to determine a failure mode and a degree of failure of the equipment. At least partially based on the determined failure mode and degree of failure of the equipment, the computing device may send at least one of a notification or a control signal.