Storage Device Failure Prediction Through Attribute Mutation Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting storage device failures fail to consider mutations in attributes and do not accurately predict failure modes or severity, especially when factors differ from those in training data, and are not robust to noisy data.
Innovation Solution
A method that determines a matrix of differences between actual and predicted attribute values, using similarity measures and attribute weights to predict failures based on healthy and failed storage devices, allowing for finer granularity and severity prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning-based binary classification method is used, then failure prediction can be performed, but the model cannot predict failures based on factors different from training data
Solution Approach 1:
The patent transforms the prediction approach by changing from direct failure prediction to prediction of attribute mutations. It introduces a matrix of differences between actual and predicted attribute values, and uses mutation rarity weights to adapt to various failure modes without retraining the model for each specific failure type.
2Measurement precision
If machine learning-based anomaly detection method is used, then unusual patterns can be identified, but accurate prediction of failure modes cannot be achieved
Solution Approach 1:
The patent segments the failure prediction problem into two parts: first detecting anomalies through attribute mutation detection, then classifying the type of failure by comparing the mutation pattern against a database of known failure mode characteristics. This segmentation preserves both anomaly detection capability and failure mode information.
3Reliability
If existing failure prediction methods are used, then whether failure will occur can be predicted, but severity of failure cannot be predicted
Solution Approach 1:
The patent adds a new dimension to failure prediction by introducing the concept of attribute mutations and their rarities. Instead of only predicting binary failure occurrence, the system evaluates the severity based on how rare and significant the detected mutations are, providing a fine-grained severity assessment.
4Device complexity
If traditional prediction methods are used, then computation can be simplified, but robustness to noisy data is insufficient
Solution Approach 1:
The patent incorporates a feedback mechanism where the system continuously monitors attribute values, compares them with predicted values, and adjusts the mutation rarity weights based on the detected deviations. This feedback loop enhances robustness to noisy data by learning from patterns over time while maintaining a relatively simple system architecture.
Data Source
AI summary
A method for predicting a failure of a storage device includes: determining a matrix of differences between actual values of a plurality of attributes of the storage device obtained during a time period and predicted values of the plurality of attributes of the storage device for the time period; and predicting whether the storage device will fail based on the matrix of differences.