Facility State Learning with Measurement Drift Pre-Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control systems for plants face challenges in applying learned models for diagnosis across different situations due to model dependency on specific conditions and settings, making it difficult to adapt to changing feature data over time.
Innovation Solution
A learning apparatus and method that acquires measurement data, performs pre-processing to reduce data drift, and learns a model to determine facility states, allowing for reduced re-learning needs and processing loads when drift occurs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a model is educated using learning data obtained under specific conditions, then the model achieves high accuracy for those specific conditions, but the model cannot be applied to diagnoses in different situations or when data drift occurs
Solution Approach 1:
The patent applies preliminary action by performing drift removal processing on learning data before model education. The drift removal unit processes measurement data in advance to eliminate drift components, ensuring that the model is trained on drift-free data from the beginning. This preliminary processing prevents the model from learning drift patterns, thereby improving both accuracy and adaptability without requiring post-training adjustments.
2Loss of time
If drift removal processing is performed on learning data, then the model requires less re-learning when drift occurs, but the processing load during the learning phase increases
Solution Approach 1:
The patent shifts the processing load to the learning phase through preliminary action. By performing drift removal on learning data before model education, the system eliminates the need for frequent re-learning when drift occurs in operation. The increased processing load during the initial learning phase trade-off is acceptable because it prevents time loss during operational re-learning, which would occur more frequently without this preliminary processing.
3Use of energy by stationary object
If conventional techniques are used without drift removal, then the processing load during learning is low, but the model becomes dependent on specific conditions and requires frequent updates when drift occurs
Solution Approach 1:
The patent resolves this contradiction by performing drift removal as a preliminary action on learning data before model education. This ensures that the model learns from drift-free data, improving reliability and reducing condition dependency. The additional processing load during the learning phase is a necessary investment that significantly reduces the frequency and need for model updates during operational phases.
Data Source
AI summary
A learning apparatus is provided, which comprises: a learning data acquiring unit for acquiring learning data including measurement data obtained by measuring a facility and a state of the facility; a learning pre-processing unit for performing a pre-processing for reducing a drift of the measurement data in the learning data and outputting pre-processed learning data; and a learning processing unit for performing a processing for learning a model for determining the state of the facility from the pre-processed measurement data, by using the pre-processed learning data.