Machine vision metrology studies at Polytechnic
Efficient operation of an assembly line is one key factor in light assembly industry. In western countries, where the costs caused by the human workers are very high, increasing the level of automation has been seen as one of the means to improve the efficiency of the production and to cut production costs. However, when increasing the level of automation the normal operation of the assembly processes will become sensitive to technical factors, such as obtaining correct dimensional tolerances for the object parts to be jointed together and monitoring these dimensional tolerances during the whole assembly process.
Two and three-dimensional (2D and 3D) machine vision technology (metrology) is nowadays largely used to perform visual inspection in monitoring the dimensional tolerances of parts and subassemblies on the production lines. This machine vision metrology is selected to be one key technical area, where the focus of the studies at the Kemi-Tornio Polytechnic is laid. Today especially two application cases are considered in more details: the warpage measurement of the printed circuit boards (PCB) and the metrology measurements in an assembly application of installing a cover lens on the framework of a small size electronic unit.
Warpage measurement of PC board
The purpose of this study was to evaluate suitable measuring methods to measure the shape distortion of a PC board. The basic idea was to utilise optical measuring technology based on machine vision. Based on the survey which was made one method (so called MoirŽ method) was referenced in the literature. However the implementation of this method is complicated and expensive. So further studies were needed to find out more simple and inexpensive methods to measure the shape distortion of the PC boards. The project was a part of European Commission funded HPCN-TTN network programme.
Pointwise flatness measurements using a commercial laser distance sensor
The pointwise distance measurement based on a commercial laser distance sensor was found to be a suitable measuring method. The operational principle of the distance meter is triangulation. The implementation of this method requires connecting the sensor in a moving mechanical arm to position the laser beam at the desired places of the PC board. The positioning is guided using the CAD data of the board. By measuring the mechanical movement of the arm in two directions and the vertical distance from the moving sensor to the surface of the board the actual 3D shape of the board is obtained and the possible shape distortion can be seen from the results.
The results achieved with this measuring method meet the needs of the industry. That is the accuracy and the rate of measurements will be suitable. In Figure 1 the accuracy of the laser sensor is shown. The implementation of this method is straightforward.
Figure 1. Measured accuracy of two different laser distance sensors. The reference distance is from 1 mm to 10 mm.
Method based on the projection of a laser stripe on the board surface
The second method, which was evaluated in more details, was the projection of a laser stripe on the object surface. An experimental system was built to carry out the necessary measuring tests. In this method a typical video camera is looking the shape of the stripe on the board surface. The twisting of this stripe shows the shape distortion of the board.
The results of the experiments obtained with this measuring method meet the needs of the industry. That is the measuring accuracy and the rate of measurement will be suitable. Also the implementation of this method will be straightforward resulting in a simple measuring system.
The results of this feasibility study were such that two optical measuring methods were tested to measure the shape distortion of a PC board. Both of the methods were found to be most promising alternatives. The user requirements could be met with both of the methods (the accuracy and the rate of the measurement). The realisation of the methods will be straightforward. The industrial project partners, a PC board manufacturer and a PC board end-users are analysing the applications and setting up final requirements for the system.
Metrology measurements in assembly application
The purpose of this study
was to develop measuring methods for an assembly application shown in Fig.
2. The application contains two parts: cover lens and the framework of
an electronic unit. In order to fit cover lens to the framework of electronic
unit the boundaries of both objects must be measured with high accuracy
(better than 20 um).
During the study a laboratory environment suitable to study machine vision based metrology applications was established. This laboratory includes two different systems. In the first system analog monochrome CCD video camera is used. The optics is a manual zoom lens. The analog CCIR type video signal is digitised with a frame grabber. Four lamps emitting near monochromatic light are placed close to the camera, one on each side. Also a light box having the same monochromatic light is used.
In the second system shown in Fig. 3 a digital monochrome CCD video camera is used. The CCD sensor has 1300 x 1030 pixels. The optics is a 55 mm telecentric lens. The digital video signal is captured with the frame grabber. Lighting system based on red LEDs was developed. Using this lighting system an object can be illuminated using any combination of basic front, diffuse back and diffuse front lightings. The measurements are performed and results analysed using Matlab-software environment.
Figure 3. The digital camera environment with the lighting system specially developed as a result of the studies.
In order to use machine vision system for accurate dimensional measurements, camera calibration is needed to compensate the systematic errors and to relate the measurements obtained from the image to the real dimensions of the object. Using a number of calibration points whose world co-ordinates are exactly known and the corresponding image co-ordinates measured performs the calibration.
The calibration object used in our experiments has 9 x 7 white circular dots on black background with a centre-to-centre distance of 10 mm and 6 mm diameter of each dot. The centre-to-centre accuracy of the dots is better than 0,5 (m and the accuracy of diameter is (0,250 (m. Accurate measurement of the exact co-ordinates of all the control points is an important step in a calibration process. A sub-pixel feature extraction technique was used to compute the centres of the circular dots. The centre co-ordinates obtained are used as input data for camera calibration procedure.
Using analog video camera performed the experiments. The distance between the co-ordinates obtained by using calibration information and known reference co-ordinates is a basic measure of the accuracy of the calibration. For example, for the 63 control points used, the standard deviation of the distances between co-ordinates was 0,9 (m, the mean distance was 2 (m and the range was 4 (m. Fig. 4 shows the error between the positions of reference points and the positions projected by calibrated camera model.
In order to measure performance
of the calibration a calibrated camera was used to perform dimensional
measurement. In the first case, two circular dots were chosen from calibration
object. The distance between these dots was 10 mm and the centre-to-centre
accuracy of the dots was better than 0,5 (m. The centres of the dots were
obtained in sub-pixel accuracy and the distance between them was calculated.
In order to measure accuracy in different image areas the image area was
divided into 9 sub-areas. The measurement was repeated 10 times in each
sub-area and the object was rotated and placed again between measurements.
The measure for accuracy is the absolute deviation between measurement
value and the reference (true) value. The total accuracy in whole image
area is difference between maximum and minimum deviation values. The accuracy
in different image areas is shown in Fig. 5. The total accuracy of 4,4
(m in whole 100 x 80 mm2 image area was obtained.
Using analog video camera for accuracy measurement, the deviation was quite homogenous in whole image area. The accuracy was better than 10 (m in whole image area. It should be noticed, that in this study only circular features are used and the used sub-pixel edge detection algorithm applies only in this case. For this reason, obtained results are reliable only in these circumstances.
The digital camera should
be used in order to get better results. Digital cameras have the advantage
of a precise correspondence between pixels on the sensor surface and pixels
in the frame. Also digital output camera leads to a simpler imaging system,
because the conversion, transmission and resynchronisation of the analog
video signal is eliminated. The preliminary results already obtained reveal
that much more accurate dimensional measurements can be achieved using
the digital camera.
In this paper, machine vision based metrology studies at Kemi-Tornio Polytechnic is described. Two major applications were considered. One is focused to measure the shape distortion (warpage) of a printed circuit board (PCB). A feasibility study was carried out to find a suitable measuring method for this application. Two methods are reported in this paper both of the methods found to meet the demands of the customer. The other application area is the metrology measurements in an assembly application of installing a cover lens into the framework of an electrical unit. In the laboratory environment established for this application high measurement accuracy was achieved using the analog monochrome CCD video camera with the manual zoom lens system. However the preliminary tests reveal that the accuracy will still be better using the digital monochrome CCD video camera with the telecentric lens system.