Using OCR to read ID plates in automotive production

In the complete production process, every single car body must be absolutely unmistakably assigned to the respective vehicle order and to its corresponding individual accessories. Extremely reliable image processing systems read the vehicle identification plates at the production lines of the Ford Werke AG in Cologne, Germany.

Even small mistakes along the production lines of the Fiesta and Fusion models could cause serious problems. If an error in the exact assignment of a vehicle were possible, then a five-door model could be identified as a three-door model or also the other way round. If a robot would then, for example, be working on the roof seam of the mixed-up vehicle, then this would really lead to a “crash”, and the line would come to a standstill for quite some time. More than 45 vehicle types must be distinguished and individually identified. Absolutely safe identification is, therefore, essential for a smooth production process and for the faultless quality of the vehicles.

For more than a year, two reading systems installed at the Ford facilities have been recognizing the ID plates in the assembly shops X and Y with the highest degree of reliability. In three-shift operation, and presently producing about 1,900 vehicles per day, every ID plate is unmistakably identified with the highest degree of safety.

Extremely safe and reliable functions along the complete range of the production facilities are of utmost importance in Cologne. And, the regulations and functional specifications for the suppliers of automation systems are correspondingly rigorous in the automotive industry.

Highest demands
In the past, the car bodies in the production process were equipped with radio-controlled data carriers. But after a few hundred runs through the car body pre-treatment and the paint shop, the data carriers resembled shapeless lumps. Cleaning of these data carriers was not possible. This and other reasons spoke in favor of searching for new system solutions. Ford engaged RESA System Engineering GmbH to provide a modern system solution with a more sophisticated technology. This company provides solutions in the field of control and automation engineering.

Harald Maaß from RESA in Cologne, who is responsible for the new system solution, comments on this issue: “The only option under discussion from the very beginning was to employ OCR (optical character recognition) to read the characters. And for this purpose, we needed a competent image processing partner who was able to meet the restrictive demands."

The intention was to have a reading system read the type plates with the inscribed characters. These plates must remain durably connected to the car body. Ford demanded a reading reliability of 99.9 percent. Every car was meant to be guided over this OCR identification. An error of 1 per 1,000 during steady production would, however, already mean that at least one problem would occur per day. This had to be excluded by all possible means. And that is why the second decimal place of the reading reliability rating is still a quite important factor regarding the effectiveness of the production process and its costs. The demand was, therefore, to improve the 99.9 percent in direction toward 100 percent as far as possible. The demand for 100 percent availability of the OCR reading system was, of course, also added.

A variety of different requirements had to be fulfilled in this context. The type plate and the fastening surroundings of the raw car body in the raw production department consisted of shiny, strongly reflective metal. The pre-treatment of the car body through electro-galvanic surface treatment using the dipping method created a phosphate coating with strong light-absorbing properties.

It was, therefore, necessary to be able to cope with both situations with extreme reliability.

Various providers of image processing systems were given small sample plates, and they were asked to offer solutions for the problem. The solution AIT Göhner GmbH from Stuttgart had to offer stood out considerably from the other competitors with regard to guaranteed reading reliability. The safe and sound mode of operation of the vision software was the key factor for the decision in favor of the system solution from Stuttgart.

Efficient vision algorithms
Reading characters by means of OCR belongs to the most advanced forms of identifying imprinted plain characters. The highest degree of certainty in reading requires considerably more computer power due to the highly sophisticated algorithms used in the evaluation process. Such particularly advanced tasks are still primarily restricted to PC-based systems. Opposed to validating the readability of the trained characters by means of OCV (optical character verification) inspection, in OCR inspection, every single character is compared with the sample characters of the alphabet and the numerals. This leads to an up to 40-fold increase of the complexity in computation of the algorithms. The characters engraved into the type plate must fulfil the requirements for clear readability within a narrowly configurable quality threshold. It is necessary to be able to cope with variations in processing quality and with the reflection behavior of the individual metal-bright engraved characters.

The gray value correlation procedure generally used in image processing reaches the limits of its applicability very quickly if the highest degree of precision is in demand, and influences such as varying illumination and contrast conditions are to be handled. The OCR recognition software used at the Ford plant is based on the outstanding abilities of the geometry-based vision software PatMax by Cognex. Unlike the gray-value correlation, these patented techniques use the basic geometrical structures of objects in a three-step procedure. First, the most important individual features in the image of the object – such as edges, measurements, shapes, angles, bends and shades – are separately identified. The spatial relations between these central features of the trained pattern are compared with the real-time image. The position of the object and its qualities are then unmistakably determined with utmost precision from the analysis of the geometrical information of the features, and also of their spatial relation to each other. Features, such as the contours of the characters with low contrast, can thereby be recognized with a considerably higher degree of reliably, precision and speed. This procedural method up to CAD-reference models also provides a very important contribution in the extremely reliable recognition of characters.

PatMax can, for example, immediately recognize where further features are to be found on the basis of a partial contour, despite distortion, displacement/shifting or coverage. It is not necessary anymore to first conduct a linear analysis of the complete image. This simplifies feature-finding, and it makes the vision system very fast, flexible and extremely precise.

PatMax can also handle very high resolutions down to the sub-pixel range and the sure and safe determination of angles on objects down to 0.02 degrees. The vision tool is invariant with regard to positioning, orientation and varying scales of an object. The simultaneous examination of contour and structure of the object image also compensates changing lighting and contrast conditions. This extremely reliable vision tool is included in the efficient VisionPro software package of Cognex. As a particularly comfortable development and programming environment, it particularly also permits simple and fast integration of individual processes into the complete flow of production and in the communication system of the enterprise.

Reliability from scratch on
A speedy and reliable complete solution was important for this image-processing task because after the plant shut down for vacation, the production line at the Ford plant in Cologne starts up in three-shift operation, five days a week. That is why the reading station had to be up and running without problems from the very beginning. The PC-based vision system with the MVS 8500 high-performance frame grabber works with a vibration-proof industrial PC, and a second PC is held in stand-by as a redundant system.

Regarding the installation phase, Maaß states: “After the training phase, the system ran with a reading reliability as good as 100 percent dependability.”

Several problems were, however, to be solved beforehand. The riveting point of the plate serves as a reference point for image processing. Since the numerals may, in some instances, show displacement to the reference point, the reading process must nevertheless be performed with absolute reliability if this should occur. Incorrect placement of the plates will immediately trigger an error message, and the car body will be removed from the line to a repair area.

In raw condition, the ID plate, and the car body as well, displays a metal-bright surface. These conditions provide very strong reflections. Even insignificant alterations of the position of the plate will let the human eye perceive totally different reading conditions. The reading reliability was an issue for the vehicle type Fiesta, and later on for the Fusion model. This required the optimal customization of the illumination by means of fluorescent lamps and the training of the characters.

Strong light-absorbing conditions were created by the pre-treatment using the electro-galvanic dipping technique, and that is why a reflector light was used here. The reading station installed after the pre-treatment has a very important function for the further processes. The read ID data (Carin Number) is sent to the Siemens control S7 and the central PAS (Production Administration System) and compared there. If it should nevertheless occur by some influence, that an F is sent to the PAS from the reading station instead of an H, an error message will then be immediately returned due to the comparison and the assessment that this does not make any sense. The PAS will then send the data for a label according to the type of vehicle and its accessories, which is glued onto the bottom side of the floor of the vehicle. A control scan and a comparing check with the PAS is carried out after the application of this very important label. This course of procedures ensures that no errors such as confusions or mix-ups regarding the vehicle equipment occur and that only faultless quality remains in the production process. This label is needed at another 20 production stations for the definite identification of vehicles and their equipment. This complete procedure at the reading station must be carried out within a cycle time of 37 seconds.

Regarding the production “without hiccups”, Maaß says: “After about one year on the production line, we can say that we have not only met our specifications, we have clearly exceeded them in the positive sense of the word. Our reading reliability and the availability have proven themselves to be important contributions to a smooth and efficient process."

To learn more about this application, visit www.cognex.net.

 


Around 1,900 vehicles are manufactured per day in the three-shift operation at Ford’s plant in Cologne, Germany. Every identification plate must be definitely identified by the image processing system (camera at left, below) with the highest degree of reliability.

 

 


The ID data is processed in the PC-based image-processing station with production control, and then it is passed on to the central PAS.

 


The absolutely sure identification of every single car body is essential for a smooth production process and the faultless quality of the vehicles. Approximately 45 types of vehicle are distinguished and individually identified.

Subscribe to Machinery Lubrication