Storage Anomaly Detection via Hierarchical Feature Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection techniques in storage devices are ineffective in processing large volumes of data from multiple sources, failing to accurately identify anomalous behavior and associate it with specific system components, leading to difficulties in predicting and addressing potential failures.
Innovation Solution
A system that collects signal data from storage devices, determines hyper feature representations, computes scores for each statistic, generates reduced hyper feature representations, and identifies anomalous behavior using storage device scores, enabling the detection of outlier behavior and aggregation of unusual scores across multiple time series.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing time series anomaly detection techniques are used, then independent time series can be flagged as anomalous, but the features extracted are difficult to interpret and associate with specific system components
Solution Approach 1:
The patent segments the analysis by introducing hierarchical levels: first analyzing individual time series features, then grouping them into component-level aggregations, and finally system-level summaries. This segmentation preserves interpretability at each level while maintaining detection accuracy across the entire system.
Solution Approach 2:
The patent introduces intermediate representations that bridge raw time series data and final anomaly decisions. These intermediate features serve as interpretable mediators that connect detected anomalies to specific system components, making the detection process transparent and actionable.
2Reliability
If multiple time series are monitored for each storage device, then more comprehensive anomaly detection is possible, but the volume of data makes it challenging to process and isolate unusual behavior
Solution Approach 1:
The patent merges multiple time series from the same storage device into aggregated representations that capture collective behavior patterns. By combining related metrics and analyzing them together, the system reduces processing complexity while maintaining reliable anomaly detection across all monitored parameters.
Solution Approach 2:
The patent transforms the data structure by adding hierarchical dimensions: organizing individual time series into component groups, then into system-level aggregates. This dimensional reorganization reduces the effective data volume at each processing stage while preserving anomaly detection capability across all original metrics.
3Measurement precision
If current anomaly detection solutions are implemented, then unusual values can be detected, but the solutions cannot automatically identify specific components requiring attention or provide reliable insights about anomalous behavior
Solution Approach 1:
The patent implements feedback loops where anomaly detections at the time series level propagate upward to component and system levels, and conversely, system-level anomalies drill down to identify specific affected components. This bidirectional feedback enables automatic component identification and provides actionable insights about the root causes and implications of detected anomalies.
Data Source
AI summary
The technology described in this document is, among other things, capable of efficiently monitoring storage device signal data for anomalies. In an example method, signal data for a plurality of non-transitory storage devices is collected. The method determines a hyper feature representation from the collected signal data and computes, using the hyper feature representation, scores for statistics associated with the non-transitory storage devices. The method further determines a reduced hyper feature representation aggregating the scores for each of the statistics associated with each of the non-transitory storage devices; generates, using the reduced hyper feature representation, storage device scores for the non-transitory storage devices of the plurality, respectively; and identifies one or more non-transitory storage devices from among the plurality of non-transitory storage devices exhibiting anomalous storage device behavior using the storage device scores.