應用機械視覺於硬碟磁頭表面瑕疵檢測

Application of Machine Vision for Hard-disk Head Inspection

邱勝郁(Sheng-Yu Chiou)

指導教授 ： 吳明川

國立臺北科技大學/機電學院/機電整合研究所/碩士(2012年)

https://doi.org/10.6841/NTUT.2012.00002

摘要

為了達到更高的儲存等級，硬碟(Hard Disk Drive)磁頭和硬碟片的間距需減少至幾奈米的間距，以提高磁通密度，而磁頭在硬碟片間的碰撞機會也因此增加，使得硬碟穩定性及使用壽命面臨考驗。現今硬碟穩定性的評估方式主要以硬碟運轉後其磁頭滑塊空氣軸承 (Air Bearing Surface, A.B.S )表面髒污的狀況區分等級。現階段仍以人工的方式透過光學顯微鏡（Optical Microscope）取像後人工判別等級。其缺點在於沒有量化的分級程序，容易受主觀因素左右，誤判的結果可能造成重工的浪費或客戶的抱怨，且區分等級的過程中，需花大量時間檢查、評估及比較參考樣本後才能做決定，十分浪費人力且枯燥乏味。
本研究針對磁頭表面提出一套基於機械視覺的自動化的磁頭分級系統。定位部分，我們提出輪廓定位法，分別定位ABS及其前端導流板，並將輪廓加粗以降低背景邊緣的影響。瑕疵檢測部分，採背景偵測的策略找出理想的閥值範圍，分割ABS部分之瑕疵；前端導流板部分，則以方向性的影像梯度偵測瑕疵，以解決背景變異太大及磁頭製造上本身的差異等問題。最後利用倒傳遞類神經(Back-Propagation Neural Network)訓練範例樣本以找出分級系統的分級最佳權重分配並進行測試。性能測試結果，符合實際業界等級判別的合理範圍，總檢測時間約為3秒。

關鍵字

空氣軸承表面 ； 倒傳遞類神經 ； 光學顯微鏡

並列摘要

In order to achieve a higher storage rank for hard disk drive(HDD), the head-disk spacing must be reduced to several nanometers to enhance the magnetic-flux density, but head-disk contacts during flying can not be avoided ,which could cause unstable flyability and reduce the service life. Nowadays the method of evaluating flyability of hard disk driver is by ranking contamination level of air bearing surface (ABS) after head-disk operating for constant time. In the industry, The level ranking is still by artificially inspecting the pictures of head surface image captured from Optical Microscope (OM). It has some shortcomings, such as it’s not a fixed standard method for objective judgement, and the misjudgement possibly cause the waste of reproduction or the customer complaint, and the process must take the massive time to inspect defect and compare reference samples before judgement, that waste the manpower, and is arid.
In this research ,we provide an automatic head ranking system based on machine vision.In location ,we present contour location algorithm to locate region of ABS and trailing edge of head and reduce interferences of background edge by thicking the contours;In defect inspection ,we present background detection to get ideal threshold to segment defect region on ABS, and use directional gradient to detect defects on trailing edge region of head, by this way to redue the influence of various background and manufacture error. Finially, we use back-propagation neural network to train the reference samples to get the best weight adjusting of the ranking system. the performance test result shows the system is acceptable for industry application and take about 3 seconds for ranking a head.

並列關鍵字

Air Bearing Surface ； BP Neural Network ； Optical Microscope

參考文獻

[33] 席友亮、李芳繁，以機器視覺分級文心蘭切花之研究，農業機械學刊，2001，17~20頁。

[41] 林建成，人臉表情自動辨識系統之研製，碩士論文，國立台北科技大學自動化工程所，台北，2005。

[3] 簡培倫，硬碟機碟片表面型態的磨潤研究，碩士論文，元智大學機械工程學系，桃園縣，2004。

[4] J.white, "Air Bearing Slider-Disk Interface for Single-Sided High Speed Recording on a Metal Foil Disk," Transactions-American Society of Mechanical Engineers Journal of of Tribology, vol.129, 2007, pp.562-569.

[7] 曾紀綱，應用機器視覺方法於晶圓表面瑕疵檢測之研究，元智大學，碩士論文， 工業工程與管理學系，桃園縣，2006。