Using a genetic algorithm to select a subset of quality metrics as input to a disk drive failure prediction algorithm
View Patent ↗A subset of quality metrics as input to a disk drive failure prediction algorithm (DFPA) may be selected using a genetic algorithm. The DFPA is executed for an initially selected generation of subset quality metrics using quality metric values stored in a reference data base. At least one DFPA setting is adjusted and the DFPA executed again for the selected subset. After training the DFPA, the best DFPA setting is saved for the selected subset. A fitness score is generated for the selected subset, representing an accuracy of the DFPA relative to failure indicators stored in the reference data base. At least one genetic operator is applied in response to the fitness scores to generate a new generation of subsets. The process is repeated until a best subset of quality metrics and corresponding DFPA setting are found.
1. A method of using a genetic algorithm to select a subset of quality metrics as input to a drive failure prediction algorithm (DFPA), the method comprising the steps of:
(a) generating a reference data base of quality metric values and a corresponding failure indicator for a subset of disk drives out of a family of disk drives;
(b) selecting from a group of a first number of quality metrics an initial generation of subset quality metrics, wherein each subset comprises a number less than the first number of quality metrics and each quality metric can take on one of a number of quality metric values representing a quality of a disk drive;
(c) executing the DFPA for a selected one of the subsets using the quality metric values stored in the reference data base;
(d) adjusting at least one setting of the DFPA;
(e) repeating steps (c) through (d) at least once and saving a DFPA setting for the selected subset;
(f) generating a fitness score for the selected subset, wherein the fitness score represents an accuracy of the DFPA relative to the failure indicators stored in the reference data base;
(g) repeating steps (c) through (f) for each subset to generate a fitness score and corresponding DFPA setting for each subset;
(h) applying, at least one genetic operator on the subsets in response to the fitness scores to generate a new generation of subsets, wherein the genetic operator is selected from the group consisting of a crossover operator, a mutation operator, and a replication operator;
(i) repeating steps (c) through (h) at least once; and
(j) saving a subset of quality metrics in response to the fitness scores and saving the corresponding DFPA setting for later use by a user.
2. The method as recited in claim 1 , further comprising the step of generating a mean absolute error in response to a difference between an output of the DFPA and the failure indicators stored in the reference data base.
3. The method as recited in claim 2 , wherein steps (c) through (e) are repeated until the mean absolute error falls below a predetermined threshold.
4. The method as recited in claim 1 , wherein the group of the first number of quality metrics comprises a primary set of quality metrics and at least one predetermined function operating on the primary set of quality metrics to generate a secondary set of quality metrics.
5. The method as recited in claim 4 , wherein the predetermined function comprises a logarithm function.
6. The method as recited in claim 4 , wherein the predetermined function comprises a hyperbolic tangent function.
7. The method as recited in claim 1 , wherein the drive failure prediction algorithm comprises a neural network.
8. The method as recited in claim 7 , wherein:
(a) the neural network comprises a plurality of processing elements;
(b) each processing element comprises a plurality of weights, and
(c) the at least one DFPA setting comprises at least one weight.