Predicting disk drive failure at a central processing facility using an evolving disk drive failure prediction algorithm

• US 8,316,263 B1
• Filed: 11/08/2004
• Issued: 11/20/2012
• Est. Priority Date: 11/08/2004
• Status: Active Grant

• Current Assignee: JP Morgan Chase Bank, N.A. (JP Morgan Chase & Co)
• Original Assignee: Western Digital Technologies Incorporated (Western Digital Corporation)
• Inventors: Gough, Ross E., Rivkin, Steven Neal
• Primary Examiner(s): MCCARTHY, CHRISTOPHER S
• Application Number:
  US10/983,982
• Time in Patent Office: 2,934 Days
• Field of Search: 714/47, 714/47.3, 706/21
• US Class Current: 714/47.3
• CPC Class Codes: G06F 11/008 Reliability or availability analysis
• CPC Class Codes: G06N 3/086 using evolutionary algorithms, e.g. genetic algorithms or genetic programming

Abstract

A method of predicting disk drive failure at a central processing facility using an evolving drive failure prediction algorithm (DFPA) is disclosed. A set of quality metric values are transmitted from each of a plurality of remote disk drives to the central processing facility. The DFPA is executed at the central processing facility in response to the quality metric values to detect an impending failure of at least one of the remote disk drives. The DFPA is evolved at the central processing facility in response to a reference data base of quality metric values and a corresponding failure indicator. The processes is repeated so as to improve the accuracy of the DFPA over time.

Claims

1. A method for improving disk drive failure prediction at a central processing facility, the method comprising:

receiving a set of quality metric values and a failure indicator from one of a plurality of disk drives that are remote to the central processing facility; and

evolving a disk failure prediction algorithm (DFPA) comprising a neural network to detect an impending failure of at least one of the remote disk drives, the evolving comprising;

(a) applying a function to a set of primary quality metrics to generate a set of secondary quality metrics;

(b) using a genetic algorithm to select a subset of the secondary quality metrics;

(c) applying the quality metric values corresponding to the selected subset of secondary quality metrics to the neural network to generate an output indicative of the fitness of the secondary selected subset of secondary quality metrics to predict drive failure, the applying comprising;

applying, to the inputs quality metric values, a process element function to generate the output; and

comparing the output to a reference value based at least in part on the received set of quality metric values and failure indicator; and

(d) repeating steps (b) and (c) at least once to determine a subset of secondary quality metrics to be used in the DFPA.

2. The method of claim 1, wherein the neural network comprises a process element function that comprises a summation of inputs scaled by a set of weight factors.

3. The method of claim 2, wherein the evolving further comprises:

determining if a difference between the generated output and a reference value meets a threshold;

if the difference meets the threshold, saving the selected subset of secondary quality metrics and weight factors; and

if the difference does not meet the threshold, repeating the step of applying the quality metric values corresponding to the selected subset of secondary quality metrics to the neural network with a different set of weight factors.

4. The method of claim 2, wherein the weight factors are derived from a referenced database.

5. The method of claim 1, wherein the function in step (a) comprises a logarithm base 10 function.

6. The method of claim 1, wherein the function in step (a) comprises a hyperbolic tangent function.

7. The method of claim 1, wherein the quality metrics comprise two or more of:

head/disk interface characteristics, read channel settings, error correction parameters, and servo control parameters.

8. An apparatus for improving disk drive failure prediction at a central processing facility, the apparatus comprising:

a processor configured to execute instructions causing it to;

receive a set of quality metric values and a failure indicator from one of a plurality of disk drives that are remote to the central processing facility; and

evolve a disk failure prediction algorithm (DFPA) comprising a neural network to detect an impending failure of at least one of the remote disk drives, the evolving comprising;

(a) applying a function to a set of primary quality metrics to generate a set of secondary quality metrics;

(b) using a genetic algorithm to select a subset of the secondary quality metrics as inputs to processing elements of a neural network;

(c) applying the quality metric values corresponding to the selected subset of secondary quality metrics to the neural network to generate an output indicative of the fitness of the selected subset of secondary quality metrics to predict drive failure, the applying comprising;

applying, to the quality metric values, a process element function to generate the output; and

comparing the output to a reference value based at least in part on the received set of quality metric values and failure indicator; and

(d) repeating steps (b) and (c) at least once to determine a subset of secondary quality metrics to be used in the DFPA.

9. The apparatus of claim 8, wherein the neural network comprises a process element function that comprises a summation of inputs scaled by a set of weight factors.

10. The apparatus of claim 9, wherein the evolving further comprises:

determining if a difference between the generated output and a reference value meets a threshold;

if the difference meets the threshold, saving the selected subset of secondary quality metrics and weight factors; and

if the difference does not meet the threshold, repeating the step of applying the quality metric values corresponding to the selected subset of secondary quality metrics to the neural network with a different set of weight factors.

11. The apparatus of claim 9, wherein the weight factors are derived from a referenced database.

12. The apparatus of claim 8, wherein the function in step (a) comprises a logarithm base 10 function.

13. The apparatus of claim 8, wherein the function in step (a) comprises a hyperbolic tangent function.

14. The apparatus of claim 8, wherein the quality metrics comprise two or more of:

head/disk interface characteristics, read channel settings, error correction parameters, and servo control parameters.