Hard Disk Failure Prediction Using Variation and Discrete Data Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hard disk failure prediction technologies fail to accurately predict the occurrence time of failures, leading to high false detection rates, inefficient disk replacement, and long processing times, resulting in service interruptions and data loss in data centers.
Innovation Solution
A method and apparatus using AI algorithms to screen hard disks on the verge of failure by calculating variation and discrete quantities of state data over a preset period, inputting this data into a training model to predict the probability of failure within a future time frame, thereby reducing false positives and improving prediction efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing hard disk failure prediction technologies pursue improvement in detection rate, then the ability to detect failed hard disks is enhanced, but the false detection rate increases causing healthy hard disks to be mistakenly determined as failed
Solution Approach 1:
The patent transforms raw state data into variation quantities and discrete quantities through mathematical transformations. This parameter change approach converts absolute values into relative changes, enabling the model to capture degradation trends rather than relying on fixed thresholds, thereby improving detection accuracy while reducing false positives
Solution Approach 2:
The patent introduces an intermediary processing layer between raw data collection and failure prediction. The variation quantity and discrete quantity calculations serve as intermediate representations that bridge the gap between raw state data and the final prediction model, allowing for more nuanced analysis of hard disk degradation patterns
2Ease of manufacture
If existing technologies can only predict whether a failure occurs without predicting occurrence time, then simple binary classification is achieved, but the time between predicted failure and actual failure is long (one or two weeks to one or two months) resulting in waste of lift cycle
Solution Approach 1:
The patent performs preliminary calculations of variation quantities and discrete quantities over a first preset time period before inputting data into the prediction model. This preliminary action captures early degradation signals and enables the model to predict not just whether failure will occur but also the timing, allowing for proactive replacement before the actual failure event
Solution Approach 2:
The patent transitions from static threshold-based prediction to dynamic trend analysis by calculating variation quantities over time. This dynamic approach allows the prediction model to adapt to changing degradation rates and provide time-to-failure estimates rather than static binary outcomes, enabling optimized replacement timing
3Quantity of substance
If prediction of hard disk failures is limited by large amount of data to be processed and limited processing capacity of processor, then comprehensive data analysis is performed, but relatively long processing time is needed (minutes to hours) resulting in low prediction efficiency
Solution Approach 1:
The patent extracts only the essential features from the large volume of state data by calculating variation quantities and discrete quantities. This extraction process filters out redundant information and retains only the most predictive features, significantly reducing the data volume that needs to be processed while maintaining prediction accuracy
Solution Approach 2:
The patent segments the prediction process into distinct stages: data collection over a first preset period, calculation of variation and discrete quantities, and final prediction over a second preset period. This segmentation allows for efficient processing at each stage and enables parallel computation, improving overall prediction efficiency
Data Source
AI summary
Disclosed are method and apparatus for predicting hard disk fault occurrence time of hard disk failure, and storage medium. The method includes steps of: screening a hard disk on the verge of failure from a plurality of hard disks according to state data acquired of hard disks; calculating variation quantity and discrete quantity of each piece of the state data of the hard disk on the verge of failure acquired over a first preset period of time, to obtain a first predicted data set; and inputting the first predicted data set into a first training model to obtain probability of occurrence of failure for the hard disk on the verge of failure over a future second preset period of time.