Hybrid Diagnostic Modeling for Anomaly-Triggered Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing diagnostic methods for facilities, such as machine tools, face challenges in accurately determining the timing for updating diagnostic models due to varying influencing factors, leading to potential deterioration in accuracy and inefficient resource consumption, and struggle with selecting the right data for model updates under storage capacity restrictions.
Innovation Solution
A diagnostic device that includes a physical model base diagnosis unit, a mathematical model base diagnosis unit, a model update necessity determination unit, and an update notification unit, which automatically determines the need for model updates based on differing diagnostic results from physical and mathematical models, and efficiently stores necessary information for improved performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the diagnostic model is updated frequently to maintain high accuracy, then the diagnosis accuracy is improved, but the consumption of calculation resources increases and the balance of learning data deteriorates
Solution Approach 1:
The system implements a feedback mechanism where diagnostic results from both physical model-based and mathematical model-based diagnosis units are compared. When discrepancies are detected, this triggers a feedback loop that initiates model update processes. This feedback-based approach ensures updates occur only when necessary (when accuracy degradation is detected), avoiding unnecessary frequent updates that would waste calculation resources while still maintaining diagnosis accuracy.
Solution Approach 2:
The diagnostic system dynamically adjusts its operation mode between using pre-stored models and performing model updates based on real-time diagnostic conditions. The system transitions from a static state (using fixed models) to a dynamic state (updating models) only when diagnostic result comparisons indicate the need for updates. This dynamic approach optimizes resource consumption by avoiding continuous updates while maintaining the ability to adapt when necessary.
2Measurement precision
If the diagnostic model is updated at unnecessary timing to maintain accuracy, then the diagnosis accuracy is improved, but the balance of learning data deteriorates and calculation resources are excessively consumed
Solution Approach 1:
The comparison mechanism between physical model-based and mathematical model-based diagnostic results serves as a feedback indicator for model accuracy. Updates are triggered only when this feedback indicates a discrepancy, ensuring updates occur based on actual performance needs rather than arbitrary schedules. This prevents unnecessary updates that would disrupt learning data balance.
Solution Approach 2:
The system performs preliminary comparison of diagnostic results before initiating model updates. This preliminary action (comparison step) determines whether updates are actually needed, preventing premature or unnecessary updates that would disrupt the balance of learning data while maintaining readiness to update when genuinely required.
3Adaptability or versatility
If all possibly patterns are set before shipment to ensure comprehensive diagnosis, then the adaptability is improved, but the device complexity and calculation resource requirements become unmanageable
Solution Approach 1:
The diagnostic system segments the diagnosis function into two parts: a physical model-based diagnosis unit that handles known, predictable abnormality patterns using physics principles, and a mathematical model-based diagnosis unit that handles learned patterns. This segmentation allows comprehensive coverage without requiring one massive complex model, as each segment handles specific types of patterns efficiently.
Solution Approach 2:
The system uses universal physical laws and principles in the physical model-based diagnosis unit that can apply to multiple different abnormality patterns without requiring specific preset models for each pattern. This multi-functionality approach allows comprehensive adaptability while keeping the physical model relatively simple and manageable.
4Measurement precision
If the storage capacity is increased to hold more learning data for model updates, then the model update quality is improved, but the device cost and physical size increase
Solution Approach 1:
The system extracts only the essential and necessary diagnostic data for model updates, rather than storing all possible data. The comparison mechanism identifies specific cases where model updates are needed, and only relevant data for those specific update scenarios is stored and processed. This extraction approach maintains high model update quality while minimizing storage requirements.
Solution Approach 2:
The system performs partial model updates based on specific triggered conditions rather than comprehensive continuous updates. Only the necessary portions of the model are updated when discrepancies are detected, rather than retraining the entire model with all available data. This partial action approach maintains update quality while reducing the amount of data that needs to be stored and processed.
Data Source
AI summary
A diagnostic device diagnoses a state of a test object. The diagnostic device includes a physical model base diagnosis unit, a mathematical model base diagnosis unit, a model update necessity determination unit, and an update notification unit. The physical model base diagnosis unit is configured to perform the diagnosis using a feature quantity. The mathematical model base diagnosis unit is configured to perform the diagnosis using a machine learning model. The model update necessity determination unit is configured such that if diagnostic results by the physical model base diagnosis unit and the mathematical model base diagnosis unit differ, the model update necessity determination unit determines that at least one of a physical model or a mathematical model is necessary to be updated. The update notification unit is configured such that if the model update necessity determination unit determines that the update is necessary, the update notification unit notifies it.