Quick Guide to Create Your Visual Inspection System

Vision inspection systems are more and more popular in the manufacturing industry. While manual inspection requires the presence of a person who passes judgment repetitively, a vision inspection system is fully automated and can save a lot of time.

The choice of the right software and artificial intelligence is, of course, crucial. But automation engineers also need to take into consideration what vision sensor is best suited for the task. Here is a quick guide with key considerations for choosing the right camera for your vision system.

Table of content: 

1. What is Visual Inspection?
2. Manual or vision system inspection?
3. 2D or 3D vision camera?
4. What am I inspecting?
5. What mobility do I need?
6. Do I need to see it in color?
7. Visual inspection with AI
8. Inspection case studies 

What is Visual Inspection?

In the industrial world, a visual inspection application involves the analysis of products on the production line for quality control purposes. Used in facility maintenance, it refers to the inspection of equipment and structures using any or all human senses such as vision, hearing, touch, and smell. It is a process that takes place at regular intervals, day by day. It has been demonstrated time and again that inspection leads to the discovery of most defects during production. Here are a few examples of applications: 

Source: Nanonets

Consideration nº1: Manual or vision system inspection?

Industrial inspection is about looking for the details that matter. One could argue that there are several limitations to using the old-fashioned way of inspection. Manual inspection requires training a person who will then have to perform assessments over and over again on the given object. The good part is you do not need any specific equipment except the naked eye of a trained inspector. 

Source: Istock

The eye will always be more technologically advanced than any mechanical camera. However, the "human machinery" is not perfect and can be easily fooled. According to research (Drury & Fox, 1975), manual inspection errors typically range between 20 to 30%, due to many factors: fatigue, pressure, isolation, bad lightning, noise, lack of experience, etc. These human errors can impact the performance of the whole production line and cause unnecessary costs. 

Automated Visual Inspection (AVI) can overcome these problems by making the whole procedure of visual inspection independent of any human involvement. Using automated systems typically surpasses the standard of manual inspection. Today, coupling the right software with the right sensor is the most effective solution for your visual inspection application.

Consideration nº2: 2D or 3D vision camera?

The fact that 2D machine vision systems have been deployed for a few decades now is a testament to their worth. 2D machine vision is a great tool for many automated applications, but it has some fundamental problems. 

Read more: Why 3D machine vision is better than 2D.

2D cameras only see the world in flat images. This means you can't inspect objects using all of their features, or measure them directly from the gathered data. It also means that your inspection software needs to apply complex algorithms and heuristics to try and guess what an object may look like from a single image. These assumptions often fail and lead to errors in your applications.

The lack of height and true shape information though is of course of no consequence for many reasonably simple applications. As a result, 2D (either area scanning or line scanning) machine vision systems are used extensively throughout the industry in a wide range of tasks.

A common misconception is that 3D sensors can not see barcodes and other printed markings. Wrong! It can if the camera is good enough. In this example, Zivid Two 3D color camera can see any label easily: 

Variety of objects with labels in a bin.

In the context of an inspection, the primary advantage of 3D over 2D is that it provides true volumetric height information. Coplanarity on varying surfaces and other height-sensitive devices can be detected without difficulty. With 3D, you can make highly accurate measurements of object dimensions. Using 3D inspection for coplanarity detection also provides a significant reduction in false calls versus the use of 2D inspection.

3D machine vision is being more and more applied to a broad spectrum of tasks where 2D capability falls short like quality control, detection of surface and assembly defects, object scanning, etc.

Comparison of intact vs defective metallic parts using 3D. See full point cloud HERE. 

For some professionals, combining 2D with 3D can be the ultimate solution in order to get the best of both worlds. This choice will depend on the nature of your inspection. 

Consideration nº3: What am I inspecting?

The magic of Automated Visual Inspection (AVI) is that you can inspect anything with the right equipment. For many industries, inspection is a priority as a defective product can cost you money (rework, loss of customers, etc). Here are some examples of industries making use of visual inspection: 

Source: NECAM

Ultimately, the choice of your sensor will depend on what you are actually inspecting, rather it is food, automobile parts, or other material handling items. If you need to see the very finest details, you will need a high-quality 3D camera. With Zivid Two+ and its outstanding accuracy coupled with point precision of 55 µm and best-in-class trueness error of < 0.2%, you will capture the details you are looking for.

Whether it's profiling wheels, checking the quality of a weld seam, or ensuring fastened bolts are in the correct position. See how incredibly detailed this point cloud of a motherboard is: 

Motherboard captured with Zivid Two

Consideration nº4: What mobility do I need?

A question you should ask yourself is "should the camera be fixed or robot-mounted?". If your robot needs to perform inspection and maintenance tasks in confined or dangerous working environments with remote access, you will need a sensor that is flexible, compact, and robust enough to sneak in anywhere.

Zivid 3D cameras are designed specifically to rise to this challenge and excel. Zivid Two works for both stationary and robot mounting, enhancing point cloud quality with a flexible capturing position:

• Remote and confined space access 
• Compact, lightweight, and fast 3D camera without reducing maneuverability or the robot payload
• Industrial grade: can work in tough environments
• Inspection from multiple angles and positions, avoiding blind spots, occlusion, and point cloud artifacts

A robot-mounted configuration can be incredibly valuable in the context of inspection. When you mount your 3D camera onto your robot, you have just given your robot the freedom to see anything within its reach. Now it can look at multiple scenes in its working area and you have 360 degrees of possibilities around the robot. Pretty useful if you have to inspect parts under a car or a train for example.

Consideration nº5: Do I need to see it in color?

We saw before that you can inspect more in detail with 3D. Among the many 3D sensors on the market, not all of them provide full HD color. Unlike monochrome 3D data, a full-color point cloud can provide significant value when handling same-shaped objects. Take the example of a tomato greenhouse: with full RGB colors, your robot will be able to recognize tomatoes that are ripe and differentiate them from unripe tomatoes as one will be red and the other will be green.

Another benefit of 3D coupled with RGB color is that your vision system can work better with difficult materials: plastic, ceramic, metal, cardboard, wood, colored, textured, light, dark or absorptive materials are easier to see. 

Zivid 3D cameras provide a full HD color point cloud with three-dimensional image processing. Check out the difference between this color vs greyscale point cloud:

Visual inspection with AI

Artificial Intelligence is turning out to be a game changer, with countless applications in nearly every domain. It is now making its way into the area of inspection in Manufacturing. Some companies like Nanonets specialize in artificial intelligence for inspection applications, helping manufacturers build and deploy deep learning models to develop the software behind AVI.

Deep learning technology uses neural networks containing thousands of layers that are adept at mimicking human-level intelligence to distinguish anomalies, parts, and characters while tolerating natural variations in complex patterns. In this way, deep learning merges the adaptability of human visual inspection with the speed and robustness of a computerized system. This can be quite useful if you wish to inspect parts but don't have the resources.

Inspection case studies

In a few words...

To conclude, there are many aspects to consider before choosing your Vision System for your inspection application: 3D or 2D, precision and accuracy, mobility (robot-mounted or stationary), grayscale, or color, and the choice of AI.

In any inspection case, your reputation depends on the system's ability to inspect and see the details, with a sub-millimeter level of high confidence, every time. For this reason, we recommend a high-quality 3D color camera that is lightweight, flexible, and robust enough to inspect anything, anywhere.