1
Feature Fusion Method for Rapid Corrosion Detection on Pole Tips

Due to its high magnetic moment, FeCo film is a type of ferromagnetic material commonly used for recording pole tips in modern hard disk drives (HDDs). These FeCo pole tips are prone to corrosion because of the corrosive environment and wide pH variations during HDD production. The machine vision is then utilized for automatically auditing of these corroded pole tips. The developed image-processing approach comprises three steps: extraction of top-shield region, fusion of extracted features and decision-making. The key step, fusion of extracted features, employs two types of features which are area-based and contour-based features. The second feature is extracted by image filtering with a specially designed filter kernel. It is capable of extracting the position of corrosion from the top shield of the pole tip. The experiments show that the algorithm reveals corrosion with high accuracy, precision, specificity, as well as sensitivity, and the overall processing time satisfies the industrial environment.

1.1. Introduction

In magnetic recording drive, a recording head consists of both write and read elements embedded at the end of an air-bearing slider. The strength of the magnetic field generated by write heads is determined by the magnetic flux density inside the head poles. The Fe65Co35 alloy used in today's write heads has the highest saturation magnetic flux density, of 2.45 Tesla (T), compared to any existing magnetic materials. Thus, the head pole material has changed over the years from permalloy (Ni8, Fe19), with a magnetic flux density of 1.0 T, to Ni45Fe55, with a magnetic flux density of 1.50 T, to the Fe65Co35 alloy. In addition to the write element, both Fe and Co are also applied in the construction of giant magnetoresistive (GMR) read sensors in HDDs.

The alloy Fe and Co magnets have outstanding magnetic properties, but these properties are greatly deteriorated by corrosion [CHE 12]. This is caused by a corrosive environment, where pH varies over a wide range [MAB 12]. In order to provide protection against such deterioration, a protective film (PF), typically sputter-deposited carbon on adhesion film, is formed on the air-bearing surface to protect the pole tips [WAN 11, TAN 09]. The PF is a single layer, or a bilayer, of an outer film of carbon formed on an adhesion film of silicon. In order to provide enhanced corrosion resistance, Flint et al. [FLI 06] study the modification of the magnet alloy itself by modifying the magnet composition. It simultaneously provides both good corrosion resistance and high coercivity.

The aforementioned methods display corrosion resistance on the thin film material. Nevertheless, FeCo pole tips are susceptible to corroding on the surface. Figure 1.1 shows the different corrosion configurations on the top shield of the pole tips. The components of FeCo pole tip, as shown in Figure 1.2, have the thin films at the top shield and the bottom shield. The top shield consists of S3 and S4 layers and the bottom shield consists of S1, S2 and P1 layers. The detection of corrosion performed by the expert is based on the following criteria. The corrosion, or pitting, typically occurs outsides the free-zone, within the top shield, at a size of less than 0.2 mm, and a quantity of less than 3 points (at 200× microscope). However, if the corrosion or pitting is inside the free-zone or bottom shield, the pole tip is rejected.

During the HDD production process, these corroded specimens are rejected prior to assembly by relying on human visual inspection. Specifically, inspectors perform 100% inspection using microscopes with a magnification ratio of 1200× for many hours without adequate resting time. Such inspections force the operators to be highly concentrated, inspecting the small product for nonconformities. Due to the nature of the industrial production, a large quantity of FeCo pole tips must be inspected in a short time. Taking into account the fatigue of the operators, it is nearly impossible to maintain the desired inspection accuracy over long working hours. Hence, a more reliable solution is needed to detect the imperfections in pole tips.

In this chapter, a novel approach to detect corrosion on FeCo pole tip is presented. The developed approach aims to improve the accuracy, specificity, robustness, as well as rapidness of detection in order to handle the problem under the time constraint imposed by the industrial environment. The approach starts by extracting the top-shield region of the FeCo pole tip from the input images. A specially designed filter is then applied to the top-shield region to highlight possible corrosion on the bottom contour of the top shield. Only the feature values that follow the criteria are used for decision fusion with the second feature (the areabased feature) to detect a defect. The experiment with several images showing different corrosion configurations demonstrates the high accuracy and specificity with low computational cost. The algorithm is readily implementable in industrial productions lines.

Figure 1.1. FeCo pole tip images, some of which bearing the corrosions indicated by the arrowheads. a) The corrosion occurs inside the top shied area. b) The corrosion occurs on the bottom contour of the top shield. c) The corrosion occurs at the positions as in (a) and (b)

Figure 1.2. FeCo pole tip and its components revealing the thin films within the top shield and the bottom shield

Processes in diagnostics and fault detections are commonly composed of several steps including signature detection, decision-making, and final feedback to user interface system. Signature detection characterizes features from various sensor signals including mechanic-based sensors, image [ACC 11, TAN 12, TSA 11, LI 12b], speech [KIM 11], optical-based sensor [SIL 11] and current [CHO 11, PON 11]. A review of the literature revealed only limited previous research focused on visual inspection of pole tips in the HDD manufacturing. Earlier, image processing algorithms have been implemented for automated defect detection in different materials. These include fabric [CHO 05, CUI 08, KUM 08, MAK 09], liquid crystal display (LCD) [LI 13, LI 11], metal stencil [CHO 07, BEN 13], steel [JAG 08, ZHU 07, LI 06, LIN 09], cork parquet [FER 09], and food [GIN 04]. These approaches have met with success to some extent; however, the previous success depends on the peculiar objectives of a production system, and on products to be inspected. There is no generally accepted industrial standards for automated inspections. In this chapter, the extracted features are specific for corrosion detection on FeCo pole tip image, which is very tiny and considered as a different type of surface to those previously studied. The chapter is organized as follows: section 1.2 discusses an effective approach for corrosion detection and section 1.3 shows the results of the algorithm tested with several images.

1.2. Algorithm for corrosion detection

The image processing approach introduced in this chapter comprises three major steps: extraction of top-shield region, extraction of features characterizing corrosion and decisionmaking for classification. Since the top shield of FeCo pole tips is the corrosion area advised by the expert, it is necessary to determine the top shield area in the image prior to corrosion detection. The extracted top shield is then examined to characterize area-based and contour-based features. While the two features are sensitive enough to pick up each type of corrosions, a single feature could not capture the information which is available on other modality. Hence, in order to increase the specificity of the algorithm, at the final stage, decision fusion rules are applied to further distinguish the possible corrosion regions.

1.2.1. Extraction of top-shield region

In Figure 1.1, it is clear that the corrosion is at the top shield of the pole tip image. This top shield area can then be used to define the region of interest. The extraction of the top shield area comprises three steps: template matching, locating top-shield region on the test image and binary conversion of the selected top-shield image. Prior to the automated top-shield region detection, a single template image, of approximately 150 × 1250 pixels containing only the top-shield region, is manually extracted from a perfect pole tip image chosen by a professional inspector. As this template contains only the region of interest, the selected image covers the top shield of the pole tip, as shown in Figure 1.3(a). This image is kept as a reference template for test images.

In the following step, the matching of a reference template and a test image is obtained. The grayscale value at each pixel of this image is correlated to that in the reference template, from which the corresponding pixel locations are mapped to the test image. The template matching method is based on the calculation of cross-correlation function (CCF). It can be defined as follows. Let {f[m, n]} be the grayscale test image of size 2048 × 2048 pixels, and {w[m, n]} be the grayscale template image of size 150 × 1250 pixels. Then, we can have:

where m and n are the coordinates m, n ? {0,1, .,2047}, and s, t are integers [GON 87, LIA 10]. The mapped image can be considered as the probability of finding a pixel with a particular gray value in the template. Hence, from this...