Through our experience OneBoxVision have identified the key figures that will determine the cost of your web inspection system. These key figures are discussed in detail in this article.
When specifying a web inspection system, there are so many different technologies and issues to consider. Let’s understand some of the critical criteria.
The three key factors are:
- Line speed – does your line run at 100mpm or 10mpm?
- Product widths – what is the maximum width of product to run on that line?
- Minimum defect size – what is the smallest defect that must be detected?
Let’s create an outline and explain how each element affects the cost and the best practice for specifying a surface inspection system. The system below illustrates a typical line scan application. Assume the web width is 2m and the line speed is 400 mpm.
I first started designing such systems in Japan in 1988. I had only known standard cameras, such as we see in our smart phones. These we refer to as matrix cameras. The vast majority of machine vision applications use matrix cameras, but the vast majority of print inspection systems and surface inspection systems use line scan sensors. I was fascinated by this technology and now have spent close to three decades working with the technology and its variants. The rest of the article will take you through the necessary steps that will make you an expert in vision system design
It’s important to understand the technology if purchasing a system. Line scan cameras use a sensor that have a single dimension. The motion of the web or sheet then supplies the second dimension. The sensor is synchronized to that motion using an encoder, so for each tick that the encoder provides while its disk rotates, the sensor also delivers one scan.
The first step to solving any problem is to define the question and to make sure each element of that question is well understood. For a web inspection system, we need to define what we are inspecting.
Let’s ask the following questions:
- What are the defects that need to be detected? A clear list should be defined.
This list should be split according to a Pareto chart. There will be some key defects that represent the majority of the complaints and then a number of outliers. An outlier could be defined as a defect that would be nice to detect but not absolutely necessary.
- What are the characteristics and tolerances for each defect class? This includes width, height and contrast.
- What are the typical run time speeds for manufacturing? But also record the minimum and maximum speeds.
- What are the typical product widths? Also specify the minimum and maximum widths.
- What substrates are used? Match the defect list to the substrate. Note the percentage of each substrate used.
- Clearly define the installation envelope and the environment.
The goal should be to specify a system for the maximum width, speed and all substrates. Doing so can increase cost of the project considerably. It is important to investigate the cost of a system that will address the key products, defects, and requirements and to determine what the delta in cost to address the remainder is. Many projects never get off the ground as a result of over specifying.
How do vendors determine the number of cameras?
Let’s do the calculations. The web width is 2m and speed 400 mpm. The defect catalogue has been calculated and our minimum defect size is 250um.
The Nyquist principle applies to image processing, so a simple rule of thumb is that the resolution must be at least twice that of the smallest defect. This is not always true as certain lighting techniques can magnify defects. The contrast of a defect to the background also plays a role. However it is a starting point. Please note that detection is not the same as classification. To classify an item, four to five times the resolution is often required.
So for detection, we will need a resolution of 125um in both machine and cross directions as the smallest defect can be a spot so with an equal aspect ratio in both directions.
- So we need 2000mm/0.125mm = 16000 pixels across the web.
- The speed of the web is 400 mpm or 6660mm per second so we need cameras that can scan at 6660/0.125 = 54kHz.
- A good fit will then be four 4k cameras running at 70 kHz. Each camera will have a field of view of 500mm.
- Now we have a camera, let’s choose a lens. The length of this sensor is just over 40mm. Each pixel is 10um wide. If we then look at the desired resolution 125 um divided by the size of sensor we can calculated the desired magnification125/10 = 12.5.
- If we were to use a 50mm lens, the approximate working distance would be 50*12.5 = 625 to the camera sensor. So allow 625mm + 200mm as a vertical envelope.
- The length of light should be 2.2m allowing an overshoot of 100mm on either side of the web or sheet. The basic specification for this application will then be four 4K line scan cameras running at 70 KHz with a 50mm lens, and a 2.2 m back light.
- Let’s also consider the same defect specification, but we decided that 90% of the product was not 2m but 1.5m in width and the typical line speed never exceeded 250mpm. The same application could be completed using two 6k cameras running at 40 KHz. This could reduce the cost of such a system by as much as 30%.
- If width remained at 2m and speed at 400mpm but the minimum defect doubled to 500um, it would also have the same effect. It’s key to understand the effect of these three key criteria.
Do not over specify your system. Look carefully at typical figures. Understand the cost delta between typical and maximum. Look at quotes on the upgrade cost from typical to maximum. Often a project dies in costing due to setting the wrong goals. Learn more about collecting defects, classification, specifying key components and meeting business objectives by downloading our collection of whitepapers.