Adaptive Learning Model Correction for Noisy Ground Truth Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data-driven and physics-based models in manufacturing systems face accuracy issues due to corrupted ground truth data, particularly in industrial sintering plants, where wear and tear lead to noisy wind box temperature profiles, affecting the models' ability to adapt to new operating regimes and learn underlying physical processes effectively.
Innovation Solution
A processor-implemented method and system for adaptive learning that corrects distorted ground truth profiles using a Profile Deviation Index (PDI) and iteratively refines them to meet quality thresholds, integrating simulated data from soft sensors to ensure accurate model retuning, and selects instances for effective data-driven and physics-based model updates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ground truth data is used for model retuning, then model adaptability to new operating regimes is improved, but measurement precision deteriorates due to corrupted data from wear and tear and air ingress
Solution Approach 1:
The ground truth data is segmented into multiple profiles, each representing different operating conditions or time periods. This segmentation allows the system to identify and isolate corrupted profiles while retaining valid ones, thereby maintaining measurement precision while enabling model adaptability through selective use of data segments.
Solution Approach 2:
An intermediary validation mechanism is introduced between the corrupted ground truth data and the model retuning process. This intermediary layer assesses data quality, identifies corruption patterns, and filters or corrects problematic data before it reaches the model training stage, thus preserving both data accuracy and model adaptability.
2Productivity
If adaptive learning is performed using corrupted ground truth data, then model updates are achieved, but manufacturing precision deteriorates due to learning from noisy data
Solution Approach 1:
Data validation and quality assessment are performed as preliminary actions before the adaptive learning process begins. By pre-screening ground truth data for corruption and noise, the system ensures that only high-quality data enters the model training pipeline, thereby maintaining manufacturing precision while enabling frequent model updates.
Solution Approach 2:
A feedback mechanism is implemented where the system continuously monitors prediction accuracy and data quality metrics. When degradation is detected, the system adjusts the adaptive learning process by filtering problematic data or adjusting training parameters, thus maintaining manufacturing precision while preserving productivity through continuous model updates.
3Adaptability or versatility
If all measured ground truth profiles are used for retuning, then comprehensive model coverage is improved, but device complexity increases due to processing distorted data
Solution Approach 1:
Corrupted or distorted profiles are extracted and separated from the valid ground truth data set through automated validation. This extraction process removes problematic data elements that would increase processing complexity, while retaining comprehensive coverage through selective inclusion of valid profiles for model retuning.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
In applications such as adaptive learning of physics-based and data-driven models associated with industrial plants, the models are corrected periodically by taking into consideration the dynamic changes occurring in plant conditions and related data. However, accuracy of adaptive learning depends on accuracy of ground truth data being used as reference data. The disclosure herein generally relates to data preprocessing, and, more particularly, to a method and system for ground truth profile correction and instance selection. The system performs a ground truth profile correction for ground truth profiles having a Profile Deviation Index (PDI) value exceeding a threshold of distortion, to reduce the PDI value, and in turn reduce the distortion in the ground truth profiles. Further, the system performs a data instance selection to identify and remove outliers, and the data that remains after the data instance selection may be then used for applications such as model generation or retuning.