applications applications

Transcription

1 Vision Vision applications applications O2GAME with the capital of RCS : B NAF : 723 Z 20, rue du Fonds Pernant, ZAC de Mercières Tél : +33 (0) Compiègne Cedex France Fax : +33 (0) comm@o2game.fr

2 Table of contents Introduction Industrial problems Use of «Vision»? Constitution... 3 Advantages of applications of Vision Possibilities of the system «Vision» Some performances of measurements Potentialities Influenced fields... 6 Examples of applications Mesure of the positioning error of windshield Checking metal braid Detection of gravage of a metal rim car Checking meringues Quality/Enumeration/Conformity of seals (Control of plates blister) Classification / Traceability / Recognition color (Marking of tires) Measure of profiles / search for defects (Control of joints of pane) Detection of defects / recognition of forms (Bottles of perfume) Detection / aspect (defects of sealing of food boat-shaped) Detection / Geometry / Precision (Positioning of labels) Pattern recognition Characters recognition Reading of codes bars

3 Introduction VIS ION 1. Industrial problems The principal industrial problems that we meet are related to the quality controls, classification, verification and dimensional checks. Some of these problems require the intervention of an operator or require a mechanical installation, complex and not very evolutionary. 2. Use of «Vision»? An application of vision is necessary when no traditional mechanical solution is usable to solve a problem or when this last requires a repetitive and constraining human intervention (important cost and increased risk of error). 3. Constitution An installation of a process of vision comprises : o a camera embarking a processor, for the acquisition and the image processing. o a lighting adapted to the problem. o an industrial screen for the visualization of various information. Notices : No element comes directly into contact with the product manufactured.. Controls are done on-the-fly, in real time and seldom require a modification of the line or process. Moreover when that is possible, a camera known as intelligent is used. The system will not use a computer of the type PC

4 Advantages of applications of Vision 1. Possibilities of the system «Vision» The current power of the computers enables us to control several products/second. The range of video sensors (color cameras, black and white, linear, CCC sorting and frame scan) make it possible to analyze products manufactured uninterrupted (fabrics, pastes, products of extrusion machines...) or unit (bottles, parts of cars...) The observable fields extend from very small (cells, seals) to the objects of imposing size (seats of cars, tires...). The products can be stopped at the time of the analysis or in displacement. It is also possible to carry out a form or character recognition. Nota : A traceability can be included in the application because all the data forward by the computer (size, product code, productivity, number of defects and frequency, schedules of production...). These statistical data can be consigned in a data base. Each element is significant for a result of optimal control

5 For the realization of the application, O2game can use market products : LEUZE OMRON COGNEX DVT MATRIX VIS ION It is also possible to carry out applications to measure for the software as for the material. 2. Some performances of measurements By its performances, the system "Vision" enables to control : Quickly the presence of components on printed circuits (5 circuits/second). The presence and the state of seals contained in plates of 20 seals at a rate of 5 plates/second. 20 cm/second of profile of glasses (0,05 mm for the measurement of profile and 0,1 mm for the detection of defects), measure 10 times faster than with the conventional system by measurement under the microscope. Check of meringues at a rate of 10 a second.... and so on 3. Potentialities The system "Vision" rests on the combined use of the principal following elements : Lighting Camera Software So the possible evolutions of each element will allow new applications as well as improvements in the performances of the system in a total way for the applications already carried out.. Lighting For this element, the improvements could come from a better adaptation of the luminous spectrum with the problems to be treated. Camera The evolution of the camera goes in the direction where it will be possible to occur without computer to process the data, these last being treated by a program included in the camera

6 Computer For this element, progress should come from the speed of data processing as of the mass of information which could be taken into account. Software package The evolution of the software package should come from the integration of new theories of the image or data processing. What will enable to answer new problems. 4. Influenced fields An improvement of one of the elements of the system "Vision" can again allow a profit in the following fields : Productivity (by a profit of speed of data processing). Quality (by the improvement of the data processing). Reliability of measurements (by the improvement of the performances of the system) Use (by the improvement of the Man-Machine interface) Damping (by the fall of the primary cost of the system). Maintenance (by the integration of several elements in a more compact unit and more adapted to the industrial needs)

7 Examples of applications 1. Mesure of the positioning error of windshield Problem The goal is to provide a system which check the positioning error of windshields, in order to be able to stick to it a rear view mirror and a detector of rain. The precision of localization is 0,5 mm. Realization The camera is located at 90 cm from the windshield in order not to obstruct the robot which has to stick the rear view mirrors. This camera is slightly tilted. The installation does not require a computer because the camera has got a 400 Mhz processor. The resolution of the camera is 1600x1200 pixels making it possible to be able to cover a square surface larger than 30 cm on side. In order to limit the variations of ambient light, instead of locking up this device in one limps black, O2game decided to use a powerful light of neon type located below the studied windshield. Fig1 : Example of windshield - 7 -

8 Fig2 : screenshot of the program The figure 1 shows a model of windshield and the figure 2 another one. In this example the program seeks the circle, and calculates his center of gravity. The position of this point makes it possible to find the displacement which the robot have to do ( the calculation time is less than 1 second). Diagram of the line - 8 -

9 2. Checking metal braid Problem The program must check the state of a metal braid, that s to say : the width of the braid its opacity that it does not have a hole with a size larger than a fixed threshold s2 that there is not N holes with a size higher than a fixed threshold s1 < s2 there is no broken wire. Realization The program runs in 0,5 s. The research of the "holes" is carried out by a tracker of edge which makes it possible to obtain all the "holes" of the braid

10 3. Detection of gravage of a metal rim car Problem To locate the point where the valve must come on a rim of car. Realization Study, realization and setting in production of an automatic checkpoint and adjustment of the existing process allow : to detect the marking of a rim, to direct towards the concerned workstation the rim in a quite particular position, in order to be able to hore it in the desired location. In the event of non-attendance of marking in the targeted zone, the system immobilizes the process. Detected marking Rim to valve In order to direct the rim correctly, O2game chose to use a numerical camera charged to film the rim put on a punt forms revolving (1 turn in a second). The goal is to detect engraving and to then alert the mechanical system charged to direct the rim. The algorithm of engraving recognition runs in 15 milliseconds

11 4. Checking meringues VIS ION Problem To check if meringue is correct (height, width, defects...). Realization The edge of the meringue is extracted, and the algorithm checks if this edge is a circle. Correct meringue Bad meringue The program also checks the height of the meringue thanks to a laser. These calculations are done quickly at a rate of 10 meringues per second. O2game has done : a system which aligns meringues. Image processing algorithm allowing to check if meringue is good or bad a system ejecting bad meringues by pressure with air. System aligning meringues

12 5. Quality/Enumeration/Conformity of seals (Control of plates blister) Problem Plates blister of 20 seals of longitudinal form are produced. It is necessary to check before the installation of film covering the blister that it does not miss of seal or that none is damaged. Realization A dark bottom of colour was selected in order to obtain an optimal contrast with the seals, those are indeed of clear colour. A soft lighting was used in order not to cause reflections on the blister which would take to him also a clear colour. These reflections would artificially have like disadvantage of `inflate' the seals and thus to reveal them larger than they are actually. We could then miss the detection of a partially damaged seal. Notice : The processing time is 160 ms per plate. Fig.1 : Original image. The first seal is partially damaged. Fig.2 : After an segmentation algorithm, the various areas of interest are colored in order to highlight the recognized zones. Fig.3 Enumeration of the areas, classified here from 0 to 18, first is missing

13 6. Classification / Traceability / Recognition color (Marking of tires) Problem VIS ION It acts of a problem of classification of tires according to a coding by points of colours. This operation of recognition and detection required the presence of an operator 24h/24 and the errors rate was important (visual tiredness, repetitivity). The colours which one must detect are the white, blue, red, green, and yellow. The blue colour was difficult to detect because of little contrast with the black colour of the tire. Points of colours of approximately 5 mm of diameter are laid out on the side of the tire. They should thus be detected on a 10 cm broad zone and on all the circumference of the tire. The tire is presented flat and the points are laid out without position predetermined on surface to analyse. Several types of tires are produced, form and size so vary. Realization The scale of the tire (80 cm) obliges us to place the camera far from the object to be studied (1 m), this implies the use of a powerful lighting (4 benches of 2 tubes high frequency of 18 Watts each one). The first algorithm used in the software is a detection of the center of the tire in order to erase all numerically that does not form part of the tire (conveyor, carpet, interior of the tire). This operation saves processing time and avoids detecting parasitic points. The points are detected by a phase of segmentation and are classified by colour. The configuration recorded beforehand by the operator allows a shunting of the tire towards one of the three possible conveyors (good tires, to recycle, bad). Statistics are consigned to count the number of tires per conveyor. This traceability by type of tire makes it possible to follow the quality of the production. Fig.1 : Original image seen by the camera Figure 1 represents the tire on the conveyor on standby of the end of the vision treatment. It is marked by a yellow point, a white and a blue (southwest position)

14 Fig.2 : Tire after treatment Figure 2 represents the tire after it was treated. The center of the tire is materialized by a red cross and the various points are framed by their respective colour. The data of the traceability are updated and the tire is switched on one of the three conveyors. The processing time is to the 4 seconds the maximum per tire. This time is a function of the dimension of the tire to be analysed. Various screens of configuration make it possible to configure the camera, the analysis and the various conveyors to be chosen according to the combination of the points of colours met. Fig.3 : Screen of configuration of the camera

15 7. Measure of profiles / search for defects (Control of joints of pane) Problem To measure the variations of dimensions with 20th of millimeter on a profile of pane of vehicle and to detect defects (glares) with 10th of millimeter. That for several types of joints of panes. Realization The problem was solved with the use of a matric camera as well as white LED to light the joint. Measurement without contact can be done in several ways of which manual mode and the automatic mode (which corresponds to 100 manual measurements). This solution goes approximately 10 times more quickly than the conventional system by microscopic measurement. Surface quality of a joint of pane Prism realized by O2game In order to be able to observe various areas of the pane, a prism was carried out. Without this prism it would have been necessary to use several cameras

16 8. Detection of defects / recognition of forms (Bottles of perfume) Problem To control the presence and the position of a hooking element of a environment perfume bottle. Realization A frame grabbing of the bottle is carried out then the image is analysed on various zones in order to determine the presence and the good constitution of contours modelled by the system. Emission of an alarm in the event of detection of the one of the defects below : One chooses to be interested only in two zones of the image : - The first zone corresponds to the theoretical site of the stem. - The second zone corresponds to the theoretical site of the ball. Bad position of the stem Absence of the hook Absence of the ball An abnormal number of hooks

17 9. Detection / aspect (defects of sealing of food boat-shaped) Problem To ensure itself of the sealing of food boat-shaped in the flow of production. If the sealing is not carried out, the food boat-shaped must be put at the reject. Realization A camera carries out the frame grabbings of the boat-shaped with a lighting in black light to avoid the reflections on the line of production, then the image undergoes a dataprocessing treatment to locate contours then to highlight the defects (holes/claws...) of the surface quality of the cover. Image in black light Detected cut

18 10. Detection / Geometry / Precision (Positioning of labels) Problem To control the axial and longitudinal positioning of labels on bottles of products of bath. Realization The operation is done in two phases : The first consists of a training with a model. The second consists in comparing the parameters of positioning of the label of the bottle in progress with those of the bottle of reference (tolerances were parameterized out of machine). Notice : an algorithm of Template Maching is applied in order to locate the labels. These contours are then analysed to calculate the center and the axes of them. These parameters are compared with those which one wishes. The processing time is less than one second. Correctly labeled bottle Badly labeled bottle

19 11. Pattern recognition VIS ION O2game has several algorithms of pattern recognition which are adapted to various types of problems. These algorithms must be sufficiently powerful to adapt : to the changes of size to the rotations to the inclinations to the errors of localizations of the image to seek. The algorithms which answer several constraints require a more significant processing time. Example 1 : research of the number 3 in the picture below The picture of the number we are looking for is: Result : The algorithm runs in one second in a computer with 1 Ghz processor

20 Example 2 : Recognition of profile of car tire Realization : A grazing lighting is used to accentuate contours of the forms, and a PC with 1Ghz for calculations. The tire below will be valid if the program finds the pixels of the image "model to recognize". Tire Model to recognize Image result The red pixels represent the recognized pixels and blue those which are not recognized

21 12. Characters recognition VIS ION Nonconstant characters on ground shining and deformed, recognition by networks of neurons. Characters on ground shining of small size (hologram). Characters on brilliant metal plate. 13. Reading of barcodes Recognition of barcodes of bad quality and data matrix