How does Machine Vision Transform Industrial Automation?
Machine vision has been around for a long time since the 1960s, the technology has really started off only around late 70s as an integral part of the automation inspection process.
In 1977, Professor David Marr of MIT introduced a groundbreaking theoretical framework for vision, known as the computer vision theory, which divides vision into three levels: objectives and strategies, representations and algorithms, and hardware implementation. This marked a significant milestone in the development of machine vision, setting the stage for subsequent theoretical innovations and practical applications that have driven the advancement of industrial intelligence and digitalization.
In recent years, with the rapid development of industrial automation, machine vision technology which has its convenience, accuracy, rapidity, and intelligence. It utilizes the advantage of high-resolution imaging, fast data processing, wide coverage, and continuous operation. It has been widely used in various fields of industrial production, and it has received more and more attention as a modern means of detection.
Credit: Azo Materials
What Can Machine Vision Do in Modern Manufacturing?
Machine vision technology is evident to have a lot of practical and essential uses in a number of fields. Starting from the identification of objects to the identification of defects or analyzing the packaging material, its abilities are phenomenal. These systems can also be programmed for object classification, color detection and recognition, color verification, pattern detection and recognition, and pattern matching. However, machine vision is extremely useful where barcodes need to be read in structured environments. Due to this versatility, machine vision is used in process automation and robotic guidance in almost all types of industries and businesses around the world.
The Role of Industrial Cameras in Machine Vision
Industrial cameras are essential ”eyes” components of machine vision systems, with the function of capturing light signals and converting them into structured electric signals. They directly define the quality of the shown images (or data) by the whole system and can significantly influence the evaluation of a vision system’s performance.
Optical lenses are as important to machine vision systems as the lens is to the human eye; in other words, they are a critical component. Their main role is to manipulate light beams and form the image of the object on the photosensitive plane of the image sensor in the machine vision system. The lenses used are found to have a direct impact on the performance of the machine vision system. Thus, proper lens selection and its integration into the machine vision system design are crucial steps.
With the steady development of machine vision in practice, the standard for higher resolutions, sensitivity, and precision in industrial cameras is heightened. These cameras must meet the standards of up-to-date automation, where accuracy and reliability are essential. The need and application of machine vision have thus directly encouraged advanced developments in the fields of industrial cameras and vice versa.
Components of Industrial Cameras
1. Image Sensor: The key component of an industrial camera, usually containing CCD or CMOS, and used to record visual information successfully.
2. Lens: responsible for directing the reflected light from objects onto the sensor to obtain good-quality images.
3. Optical filter: critical for the limitation of low light that would not be of great benefit to the imaging process and allowing only moderately relevant light in.
4. Lighting: ensures sufficient lighting to produce the right environment for image production, which is important in producing an accurate image.
Read More: Classification Of Cameras In Machine Vision Systems
Things you should know about industrial camera
Given the fact that industrial cameras are almost never a standalone install, but rather a part of a larger inspection system, the biggest difference between selecting a camera for a machine vision process and a day-to-day scenario is that for factory automation processes, the camera, the inspection object and the light source are all static and therefore you’d have to calculate the setup. In order to get the right setup, you need to know the technical specifications of your industrial camera based on the distance between the camera and the object, the shape of the object, the angle and positioning of the light source and how much space the whole setup takes to make sure the whole inspection system works smoothly.
FOV
Field of view (FOV) is based on the angle of view extending from the lens to the surrounding scene and may be measured either vertically or horizontally. Typically measured in degrees (angle) to accurately convey the extent of the visual coverage.
FOV determines how large of an area your camera covers. The field of view (FOV) is the angular extent of the observable world that is seen at any given moment. In short, it determines how much observable area that a lens system is capable of imaging. It represents the section of the object that occupies the sensor of a camera.
Telecentric lens or fixed focal lens
Telecentric lens is a special type of industrial lens designed to fix the parallax issue that plagued the traditional lens. The unique thing about telecentric lens is that their imaging magnification is unchanged under a certain range of working distance. This feature is highly valued in factory automation since the distance between camera and inspection objects can vary (although usually slightly) given the slightly different surfaces, shapes or other factors. Telecentric can make sure the resulting images are always the same without distortion so that the inspection software would not make any errors.
Fixed focal lens, or sometimes called prime lens, is a type of lens with a fixed focal length which makes them light weight, low volumn and low cost, they are also much less of an issue when it comes to chromatic aberration/color distortion; they are usually massively depolyed in factory automation, so much so that in some area they are also called FA lens.
There are four main types of fixed focus lenses. The first includes ultra wide angle lenses (with a focal length equivalent to 24mm or less in the 35mm format) and wide angle lenses, both of which can capture a wide range of scenes. The second type includes standard lenses (equivalent to 50 mm) and mid telephoto lenses (equivalent to 85 mm; often used for portraits, so they are also called “portrait lenses”), which can capture a perspective close to what the naked eye can see. The third category includes telephoto lenses and super telephoto lenses, which can take close-ups of distant objects. Finally, there is the macro lens, which can take close-ups of small objects from a closer shooting distance.
Read more: A Beginner’s Guide to Telecentric Lenses and Their Applications
For fixed focal lens, the closer the inspection object is to the lens, the larger the resulting image (usually speaking). Furthermore, the FOV would increasingly gets bigger as the inspection distance gets longer which creates problems such as distortion, lower resolution, etc. Therefore, fixed focal lens are more suited for scenarios where the inspection object distance is static and the inspection object plus the camera are on the same level, thus making sure that the imaging size and resolution are always the same without distortion.
For telecentric lens, the imaging size of the object stays the same regardless of inspection distance so long as it’s within the working distance of the lens. This feature gives telecentric lens an advantage where the inspection distance is not static but constant resulting image size is required. However, since the resulting image size is constant for telecentric lens, you would need a bigger lens in diameter than, or at least as big as, the inspection object.
Sensor size
You can calculate the sensor width and height based on camera resolution and pixel size.
Take a camera of 1280*960 resolution with a pixel size of 3.75μm for instance:
1280* 3.75μm = 4800μm=4.8mm
960*3.75μm=3.6mm
The sensor size would be 4.8×3.6mm.
Usually speaking, the higher the sensor size, the larger the pixels, and therefore the better the resolution. For industrial cameras, it’s better to calculate your exact imaging size for maximum cost efficiency.
Read more: DTCM Series
How to pick industrial cameras for your automated inspection?
First thing you need to ask yourself: What image resolution do you need?
If we are talking about inspecting a static/constant working area meaning that the distance between camera and inspection object stay the same and on the same level, the higher the resolution of the camera you have, the higher the detection accuracy you will get. Of course, your inspection software algorithm will also affect accuracy here.
For instance, if you aim to inspect an area, or the size/shape of an object or its positioning, an area scan camera would be the top choice. However, if you are inspecting a line of objects on a conveyor belt, or something that is very lengthy, or a cylinder shaped object, usually you may want to pick a line scan camera.
Read more: How Machine Vision Can Help In EV Battery Inspection
Read more: Understanding the Basics of Telecentric Scan Lenses
Area scan camera case
Let’s assume that your algorithm’s detection accuracy is one pixel, and your inspection scenario is:
The length of your workpiece is 10mm.
And the detection accuracy is 0.01mm.
Then the length of the resulting imaging is: 10/0.01=1000 pixels
Say that we need to inspect possible deficit/flaw/dent on that object with a width of, say, 8mm; and the field of view (FOV) we want to shoot is 12*10mm, then the minimum resolution of the camera should be: (12/0.01)*(10/0.01)=1200*1000, which is about a 1.2 million pixel camera, that is to say if at least one pixel is responsible for flaw detection, then we need a minimum resolution of 1.2 million pixels. However, your whole inspection process will be extremely unstable for a “one pixel for one flaw” system, because any interfering pixel may be mistaken for a flaw. Therefore, it is recommended to allocate 3 to 4 pixels for one flaw on average in order to improve the accuracy and stability. Usually, a 3 million pixel camera will do the job.
Line scan camera case
For line scan cameras, you have some basic things to consider: Web width, resolution, accuracy, motion speed and line rates.
For example, if the web width is 1600mm, and the accuracy is 1mm, while the motion speed is 22000mm/s. Your required resolution would be: 1600/1=1600 pixels. Which means you would need a 2k camera. And the accuracy would be 1600/2048=0.8mm. Since 22000mm/0.8mm＝27.5KHz, which means you need a 2048 pixel 28kHz camera.
Determine the FOV of your line scan camera:
FOV of a typical line scan is (almost) one dimensional – only the width. The web width and focal length of a line scan camera are closed related, and you can calculate them like this:
Field of view (FOV) = [pixel cell size] x [number of pixels] x [working distance] / [focal length]
For instance, if a line scan camera sensor has a pixel length of 10μm, and its resolution is 2048 pixels, the inspection distance is 160mm, and the focal length of the lens is 55mm, then that line scan has a FOV of 10um x (1/1000mm) x 2048 pixels x 160mm / 55mm = 59.58mm
Note: 10um x (1/1000mm) is the unit conversion.
Determine the length resolution of your line scan camera
Even though line scan cameras only work on the width dimension, the resulting image is in 2D with length and width. So how do we determine the length resolution of a line scan camera?
The length resolution of a line scan camera is tied to its line frequency, the motion speed of the inspection object. The equation goes like this:
Image length resolution (mm/pixel)=motion speed(mm/s)/line frequency of the camera(Hz)
To ensure the best imaging results, you may want to match both the length and width resolution of your line scan camera.
Conclusion
To sum up, it is deemed that machine vision technology has become significant in industrial manufacturing currently. They assist us in the enhancement of quality and effectiveness, thereby cutting costs and human effort, increasing production rates, and decreasing defects. Terms like telecentric lenses and, in general, the application of machine vision are increasingly found. It’s the capability to solve challenging questions and the possibility to advance solutions in automated quality control and robotics even more. When integrating the approaches of machine vision, artificial intelligence, and machine learning, new opportunities for automating processes and achieving accurate automation arise in modern industry. This is getting us into a new age of industrial revolution, specifically smart manufacturing.
