Hybrid Data Augmentation for Industrial Plant Degradation Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for real-time performance degradation monitoring and prognostics in industrial plants are unreliable due to the lack of ground-truth measurements and complex relationships among process variables, especially in scenarios like sponge iron rotary kilns where accretion is difficult to observe and measure, leading to inaccurate diagnostics.
Innovation Solution
A hybrid data augmentation method that combines historical operation data from various sources, including sensors, environment, laboratory, and maintenance data, using both knowledge-based and data-driven models to estimate performance degradation, with soft sensors and machine learning models to generate high-confidence and low-confidence subsets, and validate performance scores for accurate prediction and diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If data-driven methods are used for performance prognostics, then the need for a priori knowledge is reduced, but the reliability of models deteriorates due to lack of significant measurements
Solution Approach 1:
The patent combines data-driven methods with knowledge-based approaches by integrating multiple data sources (sensor data, laboratory data, maintenance data) and combining them with domain knowledge about performance degradation patterns. This hybrid approach allows the system to leverage the ease of data-driven methods while improving reliability through knowledge-guided data selection and interpretation.
Solution Approach 2:
The patent introduces an intermediary layer of data augmentation that generates synthetic performance degradation data based on available historical data and knowledge-based models. This intermediary synthetic data bridges the gap between limited real measurements and the need for reliable training data, enabling data-driven methods to work effectively without requiring extensive ground-truth measurements.
2Measurement precision
If knowledge-based methods are used for performance diagnostics, then the accuracy of diagnostics can be improved through domain knowledge, but the initial effort and complexity increase significantly
Solution Approach 1:
The patent segments the complex knowledge-based diagnostic system into modular components: data collection modules, data preprocessing modules, model training modules, and prediction modules. Each module handles a specific aspect of the diagnostic process, making the overall system more manageable and easier to implement while maintaining diagnostic accuracy through the integration of domain knowledge in the model training phase.
Solution Approach 2:
The patent performs preliminary action by pre-processing historical data and pre-training models during system setup and maintenance periods. This allows the system to accumulate domain knowledge and refine diagnostic models in advance, reducing the complexity of real-time diagnostics while maintaining high accuracy through pre-computed performance indicators and trained prediction models.
3Measurement precision
If traditional sensors are used to measure performance degradation, then direct measurements can be obtained, but the ability to capture unseen behavior deteriorates due to limited sensor coverage
Solution Approach 1:
The patent creates copies of performance degradation patterns through synthetic data generation. By using knowledge-based models to generate synthetic degradation scenarios that mimic real-world patterns, the system captures unseen behavior and rare degradation modes that traditional sensors cannot observe, while maintaining measurement precision through validation against available ground-truth data.
Solution Approach 2:
The patent adds another dimension to performance measurement by incorporating multiple data sources (sensor data, laboratory data, maintenance data) and multiple performance indicators beyond what traditional single sensors provide. This multi-dimensional approach enables the system to capture complex degradation patterns and unseen behaviors that cannot be detected by conventional sensors alone.
Data Source
AI summary
This disclosure relates to a method and system for hybrid data augmentation for estimating performance degradation in industrial plant. Performance degradation in industrial plants cannot be measured by sensors or laboratory measurements and there are no methods to annotate performance degradation state. The embodiments of the present disclosure provide a knowledge-based data augmentation that use physics based information to model performance degradation. The disclosed method augments high fidelity data with knowledge-based methods into high and low confidence data which are used to calculate performance score of high confidence data. A physics-informed machine learning model is trained on high confidence data. The resulting model is then used to predict performance score for low confidence data. The model is further used for training prognostics and diagnostics models to predict and identify root causes responsible for performance degradation. The disclosed method is used for predicting ring formation inside the sponge iron kiln.