Using active thermography to inspect pin-hole defects in anti-reflective coating with k-mean clustering

Highlights

• Thermography is used to overcome illumination issues in visual inspection of film.

• De-trend filter is designed based on heat conduction and residual analysis.

• Pinholes as small as 0.03 mm can be detected effectively with uneven heat source.

• k-means clustering provides nearly noise-free edge detection.

• Pinholes from 4 mm to 0.03 mm can be detected cost-effectively.

Abstract

The study here demonstrates the capability of thermography to detect and characterize pinhole defects in a visually transparent anti-reflection (AR) film. The diameter of the pin-holes varies from 0.03 mm to 4 mm. Each inspected area was unevenly heated for 65 s. Thermal images were processed by the following steps: de-trend processing, primary edge detection based on Sobel approximation, and further edge separation based on k-means clustering. Then, the diameter of pinholes was directly measured from binary images. The proposed image processing enables thermography to detect 86 of 90 inspected areas correctly in position.

Introduction

The size of a pin-hole defect is one of the dominant factors in the quality control of an anti-reflection coating because the presence of pinholes lowers the usability, degrades the functionality and negatively impacts the user satisfaction. To the best of authors׳ knowledge, many manufacturers of AR coatings used for viewing applications have set limits on the maximum tolerable pinholes size [1], [2], [3], [4]. A commonly acceptable pinhole size for such applications is 0.08–0.1 mm [1], [2], [3], [4].

Nowadays, the visual inspection of AR films is still performed manually despite the fact that machine vision technology has existed for several years; furthermore, the manual inspection of defects is tedious and may cause a hazardous working environment [1], [2], [3], [4], [5], [6], [7]. Separately, other technologies [8], [9], [10], [11], [12], [13] such as atomic force microscopy (AFM) [8], functionalized near-field scanning optical microscopy [9], and photo-thermal related technologies [10], [11] have been developed for the detection of small defects in other fields. Their cost efficiency should be considered from the viewpoint of use for AR films for viewing application. Automatic visual inspection using a charge coupled device (CCD) camera provides a much quicker and convenient method to detect defects. In this approach [14], an illumination system and a CCD camera are used to obtain multiple images. Defects are then detected based on the intensity changes caused by them. However, the images captured using the CCD camera strongly depend on the viewing angles of the illumination source as the brightness differences caused by pinholes are large enough for the CCD method to capture only within certain view angles [5], [6], [7]. The viewing angle is the largest source of error when analyzing images. To improve the inspection capabilities, Tsai [15] and Kim [16] developed two methods to enhance the contrast between the defective and the non-defective areas.

To overcome the problems faced with CCDs, thermography, which detects defects based on changes in infrared radiation caused by them, has been proposed [17], [18], [19], [20]. The attenuation coefficient of many visually transparent materials, such as PETs [4], [21], [22], [23], is large in long-wave infrared radiation [24], [25]. In other words, the IR radiation intensity through the surface of such a film may change rapidly with the film thickness. The observed IR intensity through the film combines the emitted radiance, which is related to the heat conduction, and transmitted intensity, which is related to the intensity of the illumination source. The larger the attenuation coefficient, the more significant is the emitted radiance due to heat diffusion.

This study discusses the possibility of using a thermal image to detect the defects in AR coatings for a flat panel display and proposes image processing methods to measure the diameter of defects from the millimeter to sub-pixel scales. A de-trend process is established based on statics principles behind the infrared image measurement and on heat conduction theories behind thermography testing process. k-mean clustering [[26], [27], [28], [29]] is used to distinguish the edge pixel of a defective and a non-defective area. This study shows that proposed image processing method can recognize pinhole defects with sizes from 0.03 to 4 mm and predict their diameters within 0.23 mm.