Some time ago, I shared theoretical results. Today I saw an article based on the actual situation of the car factory. You can look at the application of machine vision in the automotive industry from different angles.
1 Preface
At the heart of Industry 4.0 is data. With the popularization of smart equipment and smart terminals and the application of various sensors, production data and product data in the automotive industry will be collected. These data will permeate the entire life cycle of enterprise operations, value chains and even products. Therefore, as the cornerstone of the traceability system in the automotive industry, the accuracy of these data must be guaranteed.
At present, there are many brands of automobile tires, and different brands of tires and wheel hubs have their own differences. In the process of manual assembly, wrong installation often occurs. This paper explains that after the tire assembly of the whole vehicle is completed, the matching detection of the four tires, wheel hub models and vehicle model information is carried out through the visual matching error prevention system.
2 Principle of visual error proofing detection
Visual inspection is to use industrial cameras to replace human eyes to complete functions such as recognition, measurement, and positioning, and convert images of captured objects into data for the system to compare, process, and analyze with the system's preset standard graphic vision system, and qualified products are released. , Objects that do not meet quality standards are tracked and rejected to ensure compliance with their manufacturer's quality standards and industry regulations. The visual inspection system can replace manual detection of barcode characters, cracks, packaging, surface layer integrity, and dents, etc., which can effectively improve the detection speed and accuracy of the production line, greatly improve output and quality, reduce labor costs, and prevent human errors. Misjudgment caused by eye fatigue.
3 Application of tire matching visual error prevention detection
3.1 Background requirements
At the tire assembly station of the vehicle assembly line, manually scan the code to determine the model to be installed. After the tire assembly is assembled, move to the next station, and use the code scanning information of the previous station to determine the corresponding wheel hub and the model that needs to be detected visually. Tires: Visually detect the 4 wheels and tires of the vehicle to judge whether they are consistent, and if they are inconsistent, they will be prompted through the warning lights.
3.2 Detection overview and principle analysis
The one-dimensional code on the side label of the front of the car is recognized and read to obtain the model information, and the corresponding model tires and wheel hubs established in the database are called according to the model information for information comparison.
3.2.1 Appearance contour feature detection
The first error-proofing detection is carried out on the tire through the appearance profile features of the wheel shape and spokes, as shown in Figure 1.
3.2.2 Dimensional inspection of the hub
If the rear wheel type is close to the first detection, the second error-proof detection is carried out by measuring the size of the hub, as shown in Figure 2.
3.2.3 Tire OCR information detection
Optical Character Recognition (OCR) means that an electronic device (such as a scanner or digital camera) checks characters printed on paper, determines their shape by detecting dark and light patterns, and then uses character recognition methods to translate the shape into computer characters. The process of text, that is, for printed characters, the text in the paper document is converted into a black and white dot matrix image file optically, and the text in the image is converted into a text format by recognition software for further editing by word processing software Processing technology, as shown in Figure 3.
Take a picture of the tire tread, segment the OCR image area, identify the tire OCR information, identify numbers and letters, and judge whether the tire type matches the vehicle model information.
3.3 Detection process
3.3.1 Processing method
Machine vision non-contact front and rear wheel time-sharing dual camera shooting online real-time detection, identification and reading of model label information on the front side of the front of the car, as shown in Figure 4.
3.3.2 Trigger mode
The photoelectric sensor is triggered, the sensor is set up at the front of the car, the occlusion signal is detected, the trigger signal is given, the label recognition and reading camera at the side of the car is triggered, the label is photographed, and the one-dimensional code on the label is started to be recognized. After the recognition is successful, the position of the bumper on the side of the front of the car is captured in video, the vehicle’s driving position is judged, and the detection of the front and rear tires and wheel hubs is triggered.
3.3.3 Processing flow
After the vehicle is transported to the tire hub detection station, after the photoelectric trigger signal is given, it will recognize and read the vehicle model information on the one-dimensional code of the label according to the video or picture taken on the side of the front of the vehicle. If the label fails to be recognized, the red light of the alarm light will always be on , cancel the alarm with one key after manual intervention, and continue to wait for the next vehicle to be detected; if the recognition is successful, turn on the video shooting mode, and judge the position of the vehicle on the detection line in each frame of the video according to the vehicle type information, and trigger the trigger after reaching the predetermined position. The tire rim detection station takes pictures of the tire rims, matches the wheel rim template of the car model with the front and rear wheel pictures, if the matching fails, the red light will flash and alarm, if the matching is successful, it will recognize the OCR information on the tire, and judge whether the tire matches the car model, and if it does not match, the red light will flash One alarm, if successful, pass, the green light is always on, waiting to detect the next car.
3.4 Overall Design Scheme
3.4.1 Installation diagram
Three cameras are arranged asymmetrically on both sides of the tire assembly station. Camera 1 and camera 2 respectively detect the tire hubs on the left and right sides of the car, and camera 3 takes pictures to identify and read the model label (one-dimensional code identification) system. The 3 groups of cameras are equipped with 3 groups of light sources for supplementary light processing. The three views (top view, side view, and front view) shown in Figure 5 show the installation layout of the assembly station vision system from three directions.
3.4.2 Left and right asymmetric structure + front photoelectric sensor structure
The left wheel tire detection structure: the left 20 million rolling shutter camera (camera 2) is equipped with a flash light supplementary light detection system for wheel tire type and size.
Right wheel tire detection structure and vehicle model label and vehicle positioning shooting structure: 20 million rolling shutter camera (camera 1) on the right side with flashlight supplementary light shooting, real-time detection system of wheel tire type and size; 20 million rolling shutter camera (camera 1) 3) It is equipped with a flashlight supplementary light camera to identify and read the vehicle model label (one-dimensional code identification) system.
The trigger structure of the front photoelectric sensor: double infrared sensors are installed on the upper and lower sides of the front, and the dual infrared photoelectric sensors can accurately detect whether the vehicle enters the tire hub detection position, and trigger the tire hub detection system to start after confirming that the vehicle enters.
3.4.3 Operation process
Front cover trigger: enter the tire hub detection position, the front photoelectric sensor trigger structure monitors whether the vehicle enters the tire hub detection position, when the front of the vehicle blocks the opposite beam photoelectric sensor, a level (TTL) signal is generated, the rising edge is triggered, and the running time is counted. Then start the model label detection position to take pictures. Label shooting: the trigger signal triggers the model label recognition on the right side to read the one-dimensional code, 20 million rolling shutter camera (camera 3) video shooting for 10 s, frame by frame analysis, to reach the field of view ROI range (1 400 mm×400 mm) , covering the label position on the side of the front of all models. The identification and reading process of the label one-dimensional code is as follows.
a. Continuously identify the images and read the model number until two consecutive pictures identify the same model number, stop the recognition, compare the model number according to the database, and call out the corresponding model information;
b. If there are no 2 consecutive pictures to identify the same model number, but the same model number is identified twice, the model number shall prevail, stop the recognition, compare the model number according to the database, and call out the corresponding model information;
c. If the identification number of each picture is different, the number recognized for the second time shall prevail, and the yellow light will be on to give an alarm, and the warning information will be recorded. Model information;
d. If there is no one picture that can be identified and read out the number, the red light will always be on and the alarm will be recorded, and the alarm code will be 0004. Record the detection log, and cancel the alarm with one key after manual intervention. If there is no manual intervention, until the front of the next vehicle blocks the photoelectric sensor, the red light of the warning light goes out.
3.4.4 Position Judgment
If the tag parsing is successful, according to the vehicle model information, the vehicle position is judged frame by frame in the video, and when it reaches the predetermined position, it triggers the shooting of the front wheel detection photo, and then when it reaches the predetermined position, it triggers the shooting of the rear wheel detection photo.
3.4.5 Shooting at the detection position of the left and right front wheels
After the left and right wheel detection position is triggered and the front wheel detection is triggered, 10 consecutive shots are taken within 5 s, covering a field of view of 1 400 mm × 1 000 mm.
a. If the matching is successful, judge whether the center point of the hub circle coincides with the center point of the tire. If they coincide, directly identify the tire OCR information and compare it with the tire information in the model database corresponding to the model number. If the comparison is successful, the green light will flash once;
b. If the hub model fails to match, continue to match the next picture. If all 10 pictures fail to match, the red light will always light up and alarm, and the alarm information will be recorded. All failed, alarm code 0003, waiting to detect the rear wheel;
c. If the centers of the wheels and tires are not coincident, the red light will always be on for an alarm, and the alarm information will be recorded. The alarm code for the left wheel matching failure is 0500, and the right wheel matching failure alarm code is 0600. 2 rounds failed with alarm code 0300. At this time, the whole vehicle detection is completed, and the red light of the alarm light is always on until the front of the next vehicle blocks the photoelectric sensor, and the red light of the alarm light is off; d. If the OCR information comparison fails, the red light is always on and the alarm is recorded, and the left wheel matching fails Alarm code 0010. If the right wheel fails to match, the alarm code is 0020, if both wheels fail, the alarm code is 0030, and the rear wheel is waiting to be detected.
3.4.6 Statistics of system testing results
The main interface of the system displays the real-time photos of the four wheels, the label recognition results and the wheel hub judgment results, and gives an alarm prompt for the abnormal matching data, and displays and counts the alarm codes, as shown in Table 1.
3.4.7 Detection pause
Save the stop time, trigger time of the front of the car, and vehicle travel position when stopping, and start video shooting directly after restarting to judge the vehicle travel position, and judge whether the vehicle has passed the next detection station after the shutdown, if not, according to the vehicle travel position If the item information is not obtained during the completion of the stop, if it is passed, the vehicle detection has not been completed.
3.4.8 Communication method
The alarm adopts the USB interface, and the USB port of the industrial computer outputs alarm red light, pass green light and warning yellow light signals. A flashing red light means that the front and rear tires and hubs are unqualified in the single inspection, a solid green light means that all four wheels of the vehicle are qualified, and a solid red light means that the tires and hubs of the whole vehicle must have unqualified items. Yellow light means that the label is not read accurately, and you need to pay attention to the position where the label is pasted.
3.4.9 Detection conditions
a. Establish a model database in advance, which includes vehicle model, label information, vehicle front length (front face to front axle), model width, axle length, wheel base, wheel brand, wheel model, wheel hub matching template model, tire OCR and other information;
b. The distance between the sticking position of the label of all models and the front of the vehicle is within the specified range, the tolerance range is ±300 mm, and the tolerance range of the height from the ground is ±100 mm, so that after the front of the vehicle reaches the trigger, it can be photographed by the label detection camera within 5 s Image field of view coverage;
c. The left and right shaking deviation of the vehicle on the mechanized catenary spreader is ±2 mm;
d. Ensure that the height deviation of the chassis of all models from the ground is ±10 mm.
3.4.10 Vision detection algorithm description
a. By collecting the hub sample model, a model library is established through the outline of the hub and the outline of the spokes;
b. When the hub arrives at the camera position, the vision system receives the VIN (Vehicle Identification Number) code sent by the MES (Manufacturing Execution System), and then calls the model corresponding to the VIN code in the model library;
c. The camera collects pictures, and performs image matching between the collected pictures and the transferred model. If the matching is successful, it will flow to the next station. If the matching fails, an audible and visual alarm will be issued;
d. For those with similar outlines, measure the size of the hub again to achieve the effect of error prevention;
e. Tire error prevention is mainly through marking different feature points on the tire surface for visual recognition.
4 Conclusion
With the continuous improvement of manufacturing process capabilities, the demand for machine vision continues to increase, the types and technologies of vision products continue to improve, and the application status will also change from the initial low-end to high-end. With the intervention of machine vision, the replacement of standardized products by user personalized solutions and services is also the development direction of machine vision in the future.
Interested students can go to Zhiwang to download this paper.
Article source: Xu Yuehong. Talking about the application of automobile tire matching visual error-proofing detection [J]. Automobile Technology and Materials, 2022,09
At present, the combination of machine vision and the automotive industry is very popular, and it is also a very popular section. Here I would like to recommend a domestic machine vision platform—Coovally, which is a machine that includes a complete AI modeling process, AI project management, and AI system deployment management. visual platform. The development cycle can be shortened from months to days, and the development, integration, testing and verification of AI vision solutions can be accelerated. Help improve the enterprise's AI stack software development capabilities, so that advanced AI systems can be popularized at a lower cost and faster. "Package its own AI capabilities" for business personnel to use, so as to realize "teaching people how to fish". At present, Coovally, which runs codes, has covered multiple application fields, including manufacturing quality inspection, geological disaster monitoring, power industry equipment monitoring, medical special disease diagnosis, smart transportation, smart parks, etc., helping manual replacement of barcode characters, cracks, packaging, and so on. Whether the surface layer is complete or not, and the detection of depressions can also achieve high precision.