Abnormality Degree Calculation Using Device-Specific Normal Distributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormality detection methods struggle with low detection accuracy when normal data is insufficient or when the normal distributions of similar devices differ, as they assume identical distributions for devices of the same kind.
Innovation Solution
An abnormality degree calculation system that uses a concept classification assignment unit, feature value vector extraction, likelihood calculation, loss calculation, model update, re-learning necessity determination, and abnormality degree calculation to improve accuracy by distinguishing between devices with different normal distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only normal data from the target device is used, then detection accuracy is maintained when distributions are consistent, but sufficient detection accuracy cannot be achieved when normal data is insufficient
Solution Approach 1:
The patent merges normal data from the target device with normal data from similar devices to create a combined learning dataset. This merging approach allows the system to achieve sufficient detection accuracy even when the amount of normal data from the target device is limited, by supplementing it with data from similar devices.
Solution Approach 2:
The patent creates a universal learning approach that can function effectively with varying amounts of normal data from the target device. By designing the learning unit to accept and process normal data from both the target device and similar devices, the system achieves multi-functionality in handling data scarcity scenarios.
2Device complexity
If a unified normal distribution is assumed for devices of the same kind, then the system is simple to implement, but false positives increase when devices have different normal distributions
Solution Approach 1:
The patent implements a dynamic learning approach where the system automatically adjusts the segmentation coefficients and learning weights based on the characteristics of the input data. This dynamic adaptation allows the system to maintain simplicity in implementation while improving detection accuracy by automatically recognizing when devices have different normal distributions and adjusting accordingly.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An abnormality degree calculation system 1 includes a concept classification assignment unit 11 that assigns a predetermined concept classification based on an identification number of a target device, a feature value vector extraction unit 12 that extracts a feature value vector based on sensor data of a sensor corresponding to the target device, a likelihood calculation unit 13 that calculates a likelihood of the feature value vector by using a machine learning model obtained from a learning database DB2, a loss calculation unit 14 that calculates a loss by using a loss function defined as a function of the likelihood, a model update unit 15 that updates the model by using the loss and a model, a re-learning necessity determination unit 16 that determines whether re-learning is necessary from the calculated likelihood when an abnormality of the target device, is detected, and an abnormality degree calculation unit 17 that calculates an abnormality degree when it is determined that the re-learning is unnecessary.