STR/03/044/PM Laser Scanning for Surface Analysis of Transparent Samples - An Experimental Feasibility Study E. Lea Abstract An experimental investigation of a surface analysis method has been carried out. The method uses a scanned laser line to detect scratches on glass surfaces. The defects are detected as differences in the scattered light, which are recorded with a digital camera. A problem with this method is to distinguish between light from the back and from the front of the sample, since transparent samples are being investigated. In this investigation, a flat sample from a display device is used to demonstrate the method. Subsequently, measurements are made on curved samples with significant roughness on the back surface. It is found from the measurements that this method can usefully extract surface details and distinguish between the front and the back of transparent samples. Keywords: Surface analysis, Laser scanning, Light scattering useful when only one surface is prepared to optical quality, and roughness at any other surface should not be included. This could be the case for instance when inspecting glass moulds. It should be mentioned here that the defects of interest here are only those, which can affect the visual optical performance of the components. This is because the components are used for visual optics. Defects which cannot be detected from light scattering, i.e. refraction and reflection, are therefore not of interest for this type of measurement. Such defects could be crystal dislocations or impurities, which could have significant influence in for instance microelectronics or photonic devices. The method investigated here is physically similar to the manual methods currently used for defect detection and can be a useful method to implement in an automated system. 1 BACKGROUND It is often of interest to inspect the surface of transparent materials for scratches, digs and other imperfections. The surface quality can be important for instance in display devices, lenses and moulds. In practice, the inspection process is often performed manually, which is slow, expensive, and it is likely to be less consistent than an automated system. It is also not practical to record the specific findings of such manual inspections. Laser scanning for geometrical analysis has been used in surface studies [1], 3-dimensional biological studies [2] as well as for determination of 3-dimensional structures inspected during manufacturing processes [3]. This study uses laser scanning to extract surface detail from transparent optical devices. An important issue becomes to extract information from only one surface. However, since the specimens are transparent, most of the light will actually travel though the specimen and reflect and scatter more strongly at the back of the specimen. If this method allows details to be extracted from only one surface at the time, this will be useful when one wants separate records of the front and back surfaces of specimens. It will also be 2 OBJECTIVE The following experiments are performed as a feasibility study of laser line scanning for surface analysis of transparent samples. Two issues are of particular interest in this study. Firstly, it must be possible to extract useful surface information. Secondly, it is required to extract information from only the front surface, and not from the back surface of the specimens. The study first uses a substantially flat specimen to investigate issues about the basic laser scanning and image recording are investigated. Subsequently, examples are given from curved samples where the roughness on the back surface is significant, while the defects on the front surface are minor. This is to evaluate whether small details can be extracted in the presence of other rough surfaces. 3 METHODOLOGY Notice that the study comprises a number of small improvements to the measurement technique. Initially the method is simply a sequence of manually recorded and assembled images, while in the end the laser scanning is computer controlled and the entire recording is made with 1
one image. The improvements are presented in sequence in this section. For the sake of clarity, a number of recorded images are presented throughout this section, rather than in a separate section at the end. 3.1 Experiment 1: manual recording For the first experiment, the set-up was as shown in Fig. 1. A laser beam was sent through a cylindrical lens, which projected a line of light onto the sample. The light source used here was a red (632 nm) Helium Neon laser from Melles Griot. An Optronics MicroFire camera with a 1600 by 1200 pixel resolution was set up to image the sample from above. The angle of the incoming light was set so that the reflections from the front and the back of the sample could be easily distinguished. In this experiment, a screen was placed on the actual sample to remove light scattered form the back surface of the sample. Fig. 2. Photography of the first experimental set-up. Fig. 1. A laser source was set up as shown to shine light upon a transparent sample. Scattered light from the front and back of the sample were collected with a camera. A cylindrical lens was used to generate a laser line in a plane normal to this picture. A photography of the first experimental set-up is shown in Fig. 2. For the first experiment, there was no motion stage, and separate images were recorded. The sample was moved manually between each recorded image. Fig. 3 shows an image of a single line. Fig. 3. A single image recording of light reflected off a transparent sample. In this particular picture the scatter from the back of the sample was not covered up. The back surface scatter is the higher intensity line seen to the right. By recording a series of images with the line in different positions, the image shown in Fig. 4 could be generated. The image manipulation program GIMP was used to extract the images. In this experiment the laser line was shifted by changing the angle of the laser. This is not ideal, as the angle at which the light meets the sample is not constant. 2
Fig. 5. Experimental set-up used in experiment 2 and 3. The difference between this set-up and Fig. 1 is the screen mounted above (and not in contact with) the sample. The sample and camera were mounted on a motion stage (not shown) allowing them to move together horizontally towards the laser. Fig. 4. A map of the surface was generated from a series of images. Although surface detail can be seen in this map, it is difficult to extract clear surface information. The scanned area was about 1.5 cm wide, and the extracted section shown here represents approximately 400 by 600 pixels. It is seen in Fig. 4 that there are distinct line patterns in the assembled image, although a real map of the surface is difficult to produce. Notice that the reflections from the back surface were covered up, so that they do not appear in this picture. 3.2 Experiment 2: using a motion stage The first experiment proved that surface roughness could be detected, although it was difficult to extract the pure surface information. In the second experiment, both the camera and the sample were mounted on a linear motion stage. The experimental set-up is shown in Fig. 5. The set-up is very similar to Fig. 1, except for the screen mounted above the sample in a fixed position, for the purpose of removing the scattered light from the back surface. This set-up allowed the camera and sample to be mounted on the same motion stage and move relative to the fixed laser line and back reflection screen. It was first attempted to move only the laser line. However, it was found that it was easier to move the sample and camera together, and have the laser fixed, since a screen could then more easily be set up to remove the light from the back surface of the sample. The screen had to be at a set distance from the laser line. The difference from the first experiment was that the angle of the incoming light remained the same for all samples, and that the distance between the lines was more precise. Again, an aggregate image was generated from a series of recorded lines. It was found that with this experimental set-up, a fairly smooth map could be generated, as shown in Fig. 6. Although distinct surface details can be seen in Fig. 6, this is a slow and tedious process, and the information is not easily extracted. 3.3 Experiment 3: single image recordings with laser line scanning Both the previous experiments show that there is information about surface roughness in the recorded images. However, they are slow processes in terms of recording and image assembly, and the information is quite noisy and without fine detail. In an attempt to improve on the image quality, the experiment was repeated using only one image, where the laser line scanning was done in one smooth sweep while this one image was recorded. The speed of the motion stage was computer controlled and was set to have a constant speed throughout the scan. Fig. 7 shows an example of a scan using a 19.9 seconds exposure time. It is a section of an unmodified single image recording, and it is apparent that scratches of different dimensions as well as spots and digs are easily recorded in the image. Fig. 8 shows the same image with slightly modified colour and intensity settings, but with no functional image analysis. This could be done in a similar way by optimising the camera settings. 3
Fig. 6. Surface map assembled from a series of laser line images from the second experiment. Several lines are detected across the sample, although the image is fairly noisy. The scanned area was about 1.5 cm wide, and the extracted section shown here represents approximately 400 by 600 pixels. Fig. 8. The same image as in Fig. 7 with slightly modified colour and intensity settings. 3.4 Experiment 4: evaluating curved samples with significant back surface roughness The first parts of this investigation have showed that it is definitely possible to record surface defects by laser line scanning. It has also been demonstrated that surface information from the front of the sample can be extracted separately from the back surface information. This is especially important for lens moulds, which are often polished on only one side, while the back of the mould is rough. In addition, the surface of a lens mould is typically not flat. The same method as used in experiment 3 was applied to glass lens moulds, since some had significant roughness on the back surface. The images were recorded with and without the screen (see Fig. 5), to show the effect of the screen in eliminating the information from the back surface. Fig. 7. A single image as recorded, using a camera shutter speed of 19.9 seconds, during which the laser line was scanned across the illuminated area. The scanned area was about 2.0 cm wide, and the extracted section shown here represents approximately 400 by 600 pixels. Figs. 9 & 10 show the images from a region of about 20 mm by 20 mm, recorded with and without a screen. This region was chosen since the roughness on the back surface was significant, while there were a couple of scratches on the front, which were barely visible by manual inspection. 4
The experiments performed in this study show that surface detail can be gathered by laser line scanning, and it is possible to extract only the front surface information from transparent samples. These two points were the main objectives of the study. No mathematical analysis was included to estimate the smallest detail, which could be detected. The reason for this is that the method simulates fairly closely the way manual inspection is performed. Basically, the sample is inspected visually while shining light onto it. Defects, which do not show up during such an inspection, will clearly not affect the optical performance for visible applications. It should be possible to integrate this measurement into a dedicated device, in which case both the positioning system, control applications and the camera could be implemented in a much simpler way. Fig. 9. Scan of a 20 mm region of a concave glass mould. These defects, which could barely be detected by manual inspection, can be detected by this method. 5 CONCLUSIONS An experimental evaluation of a scanning laser line method for detecting surface defects was performed. It was shown that scratches of different dimensions as well as spots and digs could be recorded. The experimental set-up was chosen such that the reflections from the back of the transparent sample would be virtually eliminated from the recorded image. No significant attempt was made to optimise the scanning speed, camera gain or shutter speed. It is likely that the parameters could be optimised to give much better surface information. Fig. 10. Same scan as in Fig. 9, but without background screening. The details from Fig. 9 are still visible, but it is impossible to tell whether they are on the front or on the back of the sample. The experimental set-up is quite straightforward, and it should be possible to implement a dedicated system for this kind of imaging. It might even be practical with a portable system. This would allow transparent components such as lenses or cathode ray tubes to be scanned during production or storage, so that the specific data recorded for every produced item could be stored and used for quality control and reference for shipped components. Figs. 9 & 10 give an experimental demonstration of the effect of the screening, showing that details from the front surface can be extracted even when there is quite significant roughness on the back of the sample. 4 RESULTS & DISCUSSION 6 INDUSTRIAL SIGNIFICANCE Optical components for visible optics, such as TV monitors, objective lenses and lens moulds for spectacle lenses are often manually inspected. This study shows that it should be possible to construct measurement systems for this kind of inspection. In addition to television screens, computer monitors and spectacles, which have been common for a long time, the last few years have even shown cameras being integrated into most new 5
mobile phones. It is quite likely that both the number of devices with optical components as well as the quality requirements for these devices will grow. Automated inspection systems for optic will then both be a factor to reduce costs and a means for quality control and reference for shipped components. REFERENCES [1] J. Engelhardt and H. Ulrich, Optical device for scanning a beam in two axes that are substantially perpendicular to each other, European patent no.: EP 0 950 208 B1, 3 July 2002. [2] C. Hitzenberger and A. Baumgartner, Method and appratus for recording threedimensional distribution of light backscattering potential in transparent and semitransparent structures, US patent no.: 6,288,784 B1, 11 September 2001. [3] T.A. Good et al., Automated sytem and method for identifying and measuring packages transported through a laser scanning tunnel, US patent no.: 6,457,642 B1, 1 October 2002. 6