| Summary: | Everything that has a beginning also has an ending and so does a hard disk drive, which is a crucial subunit of a computer. The failure of a hard disk drive may cause serious data loss and inconvenience. In spite of hard disk drive being such a crucial subunit of a computer, limited research has been done on hard disk drive failure mechanisms, diagnostics, and prognostics. By knowing in advance an impending hard disk drive failure, can we not only avoid data loss but also
minimize computer down time and wastage of time and money. Through the proposed prognostics technology we can also extend the length of hard disk drive usage and delay their replacement. We extensively investigated different degradation signatures for a hard disk drive that characterize its aging and failure. We identified reported uncorrect, hardware ECC recovered and read write rate as effective degradation signatures. Next we develop a neural network model to assess the current
health and estimate the remaining useful life of a hard disk drive. We collected 320,800 data points by conducting experiments on 13 hard disk drives in an accelerated degradation mode. We used more than half of these data points for computing the neural network parameters and the rest for evaluating the accuracy of model predictions. The overall prediction accuracy of the model was found to be around 88.51%. This means, we can assess the health of a hard disk drive correctly 88 times
out of 100 instances.
|
|---|
