Training Data Selection for Anomaly Detection False Negatives
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In anomaly detection devices, sensor data from the initial operation stage or aged deterioration stage may be incorrectly classified as normal, leading to erroneous detection where the fault detection target is incorrectly identified as normal despite being abnormal, even after retraining the learning model.
Innovation Solution
A training data selection device that identifies and selects sensor data related to false negative detections, excluding it from retraining to ensure accurate classification by using a learning model that has learned normal data distribution, thereby reducing erroneous detections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sensor data from initial operation stage or aged deterioration stage is used as training data, then the learning model can be retrained to improve detection coverage, but erroneous detection occurs where abnormal conditions are incorrectly classified as normal
Solution Approach 1:
The patent extracts and removes false negative detection data from the training dataset. The training data selection device identifies sensor data that was incorrectly classified as normal despite being abnormal, and excludes this data from retraining the learning model, thereby preventing propagation of classification errors while maintaining beneficial training data.
Solution Approach 2:
The patent implements a feedback mechanism where detection results are used to evaluate and select training data. The system detects false negatives by comparing detection results with actual conditions, feeds this evaluation information back to the training data selection process, and uses the feedback to refine which data should be used for retraining, creating a continuous improvement loop.
2Quantity of substance
If all sensor data within normal period is used for retraining, then training data quantity is maximized, but data quality deteriorates due to inclusion of false negative detections
Solution Approach 1:
The training data selection device extracts false negative detection data from the pool of sensor data within normal period by evaluating detection results. This extracted data is then excluded from retraining, ensuring that training data volume is maximized while quality is maintained by removing only the problematic portions.
Solution Approach 2:
The patent applies local quality control by evaluating and selecting training data based on detection result quality rather than uniformly treating all normal period data. Each dataset is assessed locally for false negative conditions, and only high-quality portions are selected for retraining, creating heterogeneous training data with varying quality characteristics.
Data Source
AI summary
A training data selection method includes: acquiring pieces of sensor data indicating an observation result of a fault detection target from a sensor to observe the target; giving each piece of sensor data to a learning model that has learned a distribution of pieces of sensor data when the target is normal, and acquiring each piece of detection data indicating whether the target is normal or abnormal from the learning model; and acquiring identification information for identifying which piece of sensor data is related to false negative detection data indicating that the target is normal although the target is abnormal, and selecting, based on the identification information, a piece of sensor data related to detection data indicating that the target is normal as training data used for retraining of the learning model from pieces of sensor data other than the sensor data related to the false negative detection data.