Abnormality Diagnosis Using Simulation Model Mahalanobis Distance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality diagnosis systems cannot effectively diagnose issues in non-steady states due to the dynamic nature of monitoring targets, as they rely on pre-generated reference data that is not feasible in transient unstable operating conditions.
Innovation Solution
An abnormality diagnosing method and system that generates a simulation model of the monitoring target, measures internal state quantities, calculates a Mahalanobis distance using predicted and measured values, and diagnoses abnormalities based on this distance, allowing for diagnosis in both steady and non-steady states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reference data is prepared beforehand for abnormality diagnosis, then diagnosis can be performed in steady state, but diagnosis cannot be performed in non-steady state that changes dynamically
Solution Approach 1:
The patent applies dynamics by transitioning from static reference data to dynamic simulation models that adapt to changing operating conditions. The simulation model continuously generates predicted values based on current environmental and operating conditions, enabling the system to handle non-steady states where conditions change over time.
Solution Approach 2:
The patent changes the fundamental parameter from fixed reference data to dynamic predicted values generated by simulation models. By using simulation models that incorporate environmental conditions and operating parameters, the system can generate context-appropriate reference values for each specific operating state, enabling diagnosis across both steady and non-steady conditions.
2Measurement precision
If Mahalanobis distance is calculated using accumulated past data, then diagnosis works for steady state, but cannot diagnose non-steady state abnormalities
Solution Approach 1:
The patent applies preliminary action by using simulation models to predict what values should be observed under current conditions before actual measurements are taken. This allows the system to establish expected reference values in advance for each operating state, enabling immediate comparison with actual measurements without relying on historical accumulated data that may not represent current conditions.
Solution Approach 2:
The patent creates a virtual copy of the monitoring target through simulation models. This simulation model replicates the behavior of the actual system under various conditions, allowing the system to generate predicted values that serve as dynamic reference data without requiring accumulation of real-world historical data from the actual target.
3Ease of operation
If pre-generated reference data is used for diagnosis, then the process is simple for steady state, but cannot follow dynamic changes in non-steady state
Solution Approach 1:
The patent applies self-service by enabling the simulation model to automatically generate appropriate reference values based on current environmental and operating conditions. The system serves itself by using the simulation model to continuously update predicted values without requiring manual intervention to select or generate reference data for each changing condition.
Solution Approach 2:
The patent achieves universality by creating a diagnosis system that can handle both steady state and non-steady state conditions through a single simulation-based approach. The simulation model serves multiple functions: generating predicted values for comparison, adapting to different operating conditions, and providing a unified framework for abnormality detection across all operational phases.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(b)
AI summary
An abnormality diagnosing method includes a model generation step of generating a simulation model (3) of a monitoring target (2), an operation start step of starting an operation of the monitoring target (2), a measurement step of measuring an internal state quantity in the operating state of the monitoring target (2) and extracting a measured value (x^), a prediction step of inputting into the simulation model (3) same control input value (u) used in the operating state of the monitoring target (2) and calculating a predicted value (x) of the internal state quantity of the monitoring target (2), a Mahalanobis distance calculation step of calculating a Mahalanobis distance (MD) from a difference between the measured value (x^) and the predicted value (x), and an abnormality diagnosis step of diagnosing whether the operating state of the monitoring target (2) is abnormal based on the Mahalanobis distance (MD).