Browse Prior Art Database

Method to Automate the Classification of Defect Maps

IP.com Disclosure Number: IPCOM000016333D
Original Publication Date: 2002-Nov-18
Included in the Prior Art Database: 2003-Jun-21
Document File: 1 page(s) / 41K

Publishing Venue

IBM

Abstract

A software application has been written that reads HDD polar plot defect maps from hard drive surface testing and applies the Hough Transform (HT) to detect shapes such as lines, spirals, and circles. All polar plots are currently analyzed manually, and as such, are not done on a large scale. Typically, the rate of HDD data log processing is about 180 drive defect map files/day/person. This translates to about 2160 (180 12 surfaces max) surfaces analyzed/day/person or 17280 (180 12 heads 8 error planes) error planes/day/person. The automated prototype maps at a rate of 28800 maps/day/ (850MHz P3 512MB Win2K JVM 1.4) with a 100% discrimination for annular shapes. This easily outpaces by orders of magnitude the rate at which defective disks are generated. The Hough Transform (IBM 1959) is a standard image processing tool for parameterized shapes like lines, circles, and ellipses. It is robust to noise and missing data and is scale and rotation invariant. A count of all image space pixels belonging to an instance of a shape's parameters are mapped into a parameter space accumulator array. Recognition then becomes a peak finding process in parameter space. To enhance peak detection, convolutions can be applied before and after the HT to mitigate quantization errors in Hough space and to enhance image space. As virtually all HDD defects are of a circular, spiral, or linear nature, the Hough Transform works well for the HDD process. Potentially, any testing tool that generates defect maps and has defects that can be described with a parameterized equation can use this transform along with its associated smoothing and inference engine algorithms. The classification data is another feedback statistic that can be used to monitor the nature of manufacturing defects rather than just quantity number of errors pixels detected. 1

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 78% of the total text.

Page 1 of 1

Method to Automate the Classification of Defect Maps

A software application has been written that reads HDD polar plot defect maps from hard drive surface testing and applies the Hough Transform (HT) to detect shapes such as lines, spirals, and circles. All polar plots are currently analyzed manually, and as such, are not done on a large scale. Typically, the rate of HDD data log processing is about 180 drive defect map files/day/person. This translates to about 2160 (180 * 12 surfaces max) surfaces analyzed/day/person or 17280 (180 * 12 heads * 8 error planes) error planes/day/person. The automated prototype maps at a rate of 28800 maps/day/ (850MHz P3 512MB Win2K JVM 1.4) with a 100% discrimination for annular shapes. This easily outpaces by orders of magnitude the rate at which defective disks are generated.

The Hough Transform (IBM 1959) is a standard image processing tool for parameterized shapes like lines, circles, and ellipses. It is robust to noise and missing data and is scale and rotation invariant. A count of all image space pixels belonging to an instance of a shape's parameters are mapped into a parameter space accumulator array. Recognition then becomes a peak finding process in parameter space. To enhance peak detection, convolutions can be applied before and after the HT to mitigate quantization errors in Hough space and to enhance image space.

As virtually all HDD defects are of a circular, spiral, or linear nature, the Hough Transform works wel...