Anomaly Detection Windowing for Device Failure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems for devices face high processor utilization and increased latency due to the large amount of operational data stored, making it difficult to detect device failures or impending failures efficiently.
Innovation Solution
A method that periodically receives operational data, determines unhealthy operation, calculates the distance from a predetermined threshold, and stores only the most recent data points within a window, deleting older data to minimize storage use, optimize failure detection accuracy, and reduce latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all historical operational data is stored and analyzed, then failure detection accuracy is improved, but data storage capacity is excessively consumed and processor utilization increases leading to latency
Solution Approach 1:
The patent extracts only the essential feature from historical data - the distance of measured values from abnormal thresholds - rather than storing complete operational datasets. This extraction principle reduces storage requirements while preserving the information needed for anomaly detection by storing only the relevant deviation metrics in a data store
Solution Approach 2:
The patent transforms operational data into a different parameter representation by calculating distances from abnormal thresholds. Instead of storing raw operational parameters, the system stores transformed parameters (distances) that directly indicate anomaly severity, enabling efficient storage and faster processing while maintaining detection accuracy
2Measurement precision
If all historical operational data is stored and analyzed, then failure detection accuracy is improved, but processor utilization increases leading to increased latency
Solution Approach 1:
The patent extracts only the essential feature from historical data - the distance of measured values from abnormal thresholds - rather than storing complete operational datasets. This extraction principle reduces storage requirements while preserving the information needed for anomaly detection by storing only the relevant deviation metrics in a data store
Solution Approach 2:
The patent transforms operational data into a different parameter representation by calculating distances from abnormal thresholds. Instead of storing raw operational parameters, the system stores transformed parameters (distances) that directly indicate anomaly severity, enabling efficient storage and faster processing while maintaining detection accuracy
3Productivity
If a window of most recent reporting periods is used, then data storage is minimized and processor utilization is improved, but too much historical context may be lost
Solution Approach 1:
The patent implements a dynamic window mechanism where the number of reporting periods included in the window is not fixed but determined based on the distance of the most recent measured value from the abnormal threshold. When distances are large (indicating significant anomalies), the window expands to include more historical periods, preserving necessary context. When distances are small, the window contracts to minimize storage and processing requirements
Data Source
AI summary
A computer-implemented method, a system, and at least one computer-readable storage device are provided. A measured value related to operation of a device is periodically received for each respective reporting period. If the measured value indicates unhealthy operation of the device, a distance of the measured value from a predetermined abnormal threshold is determined and stored in a data store. Based on the determined distance, a number of most recent reporting periods included in a window is determined. All stored distances determined from measured values received before the window of the most recent reporting periods are deleted from the data store. An occurrence of a device failure is determined based on the stored distances in the data store after the deleting. In response to determining the occurrence of the device failure, an action is performed.