Sensors, Vision

3 steps to a successful machine vision project

Machine vision can add to a project’s quality and throughput. Heed these three ways to help a vision project succeed.
By Ian Visintine November 29, 2019
Courtesy: MartinCSI

Machine vision is a powerful tool that can be used to ensure the quality of a product, but it is often not trusted or even considered as a viable option due to bad experiences and unsuccessful projects. In most cases there are only a handful of reasons a vision project is unsuccessful; unrealistic expectations for a machine vision system, failure to properly analyze the application from the beginning, and letting a budget dictate the hardware instead of the requirements of the application. An unsuccessful project is not affected by one reason due to each one impacting another. However, if a project is unsuccessful, it is very likely it was due to one of these three reasons.

1. Unrealistic expectations for a machine vision system

Making sure there are realistic expectation for a machine vision application can be one of the easiest hurdles to overcome if the application is thought through and understood at the beginning. It is important to understand that not all vision systems are the same or capable of the same applications. Applications cannot be solved by using the favorite smartphone camera or a basic camera. Each application has specific needs that dictate the most important aspect of the vision system. If the application needs are not correctly identified or changed, it can be difficult and expensive to change the system’s capability once equipment has been purchased.

Expectation that need to be set and not changed after the design phase would be; inspection location, speed of inspection, camera working distance (WD), field of view size (FOV), area of interest size and inspection tolerances. One of these changes could make the wanted inspection not possible or require expensive hardware change to allow the inspection to still be possible.

An example is a change in the inspection location that causes a change in the mounting location, which leads to a change in the working distance. When the working distance changes, the user needs to recalculate the lens being used to maintain the same field of view and recalculate the resolution of the camera to ensure it is still capable of the inspection.

Another change that could have a drastic affect are changes in the inspection requirements. An inspection started out part presence of a large feature but changed to verifying that feature is within .001 of a millimeter to another feature. If the original camera selected had a low resolution due to the size of the feature, it will likely not have the resolution to measure .001 of a millimeter. This means a special lens is needed for an accurate measurement. Both changes would end up being costly due to the additions required.

Machine vision can add to a project’s quality and throughput. Courtesy: MartinCSI

Machine vision can add to a project’s quality and throughput. Courtesy: MartinCSI

This is why it is crucial to have a thorough understanding of the application needs by answering the following questions:

  • What is being inspected?
  • Where is it being inspected?
  • How it is being inspected?

Asking these questions allows accurate expectations to be set for what the vision system needs to be capable of.

2. Budget dictating hardware instead of application requirements

As with most projects, budget is a large limiting factor. However, when the budget for a vision project is not decided based on the application requirements, it is the greatest contributor to a vision project not going well or failing to meet expectations. The cost of vision equipment can vary a lot; so can the results of the same equipment selected. If equipment is only being selected based on cost, there is a possibility it will not be able to perform as needed.

For example, some might assume a low-end $2,000 camera can conduct a part presence inspection on a small bolt because it is being done on a similar bolt size in another part of the plant. As a result, the budget was based on the equipment used previously. However, the mounting location between the two applications, which are significantly different, was not considered.

The hope is it got caught during the proof of concept and would not cost any money. However, a similar application was already being done in the plant, so it was assumed a proof of concept was not needed for the new application.

Ultimately, it was not discovered until installation, and the equipment was not capable of getting the needed resolution or contrast due to the working distance of the camera and lighting solution. This situation resulted in the need for a new camera and a lighting solution change to complete.

If the budget does dictate what equipment will be used, then it is important to understand what is important for the application and have realistic expectations of what the equipment can accomplish.

3. Failure to properly analyze the application from the start of the project

Once an application has been conceived, had expectations set, and had some initial planning started, it is time to test equipment and verify what can accomplish the application. Verification should always be done so the calculation on paper and the idea of the application can be verified prior to buying equipment or installing the system. The easiest way is conducting a proof of concept by simulating actual inspection conditions or mocking up equipment at the inspection location during production. If done properly, this will verify all parts of the vision system.

If the users conducts the proof of concept offsite, it is important to test by reproducing actual inspection conditions as much as possible with actual parts. If the proof of concept is done improperly, it could lead to using the wrong equipment or not compensating for a potential pitfall.

The following items should be verified during the proof of concept: Working distance, field of view, equipment resolution, lensing, lighting, and inspection parameters.

Lensing, field of view, and resolution are a few things that are often overlooked when an application is conceived, and not verified before production testing. It will often be assumed how far away a camera will be mounted and with a general measurement of the part to get field of view, the lensing and resolution will be calculated on paper for a best guess. This will lead to problems when the calculated working distance is between the ideal spot for two lenses, and the mounting location has to be adjusted farther away than expected. With the move father from the part will the lens have any distortion as the user gets closer to the outside of the image, and will there still be a large enough number of pixels per millimeter to accurately conduct the inspection? The opposite can be true as well when the user needs to make an accurate measurement for a part, which will require an expensive lens to be size for the part. Now, the user has to move it closer, which means the field of view is no longer large enough to see the whole part.

The lighting solution is a part of the system that is easy to assume based on the type of inspection being done, but it can have unexpected results depending on the material or part color. Part defects also can have an unexpected result when viewed through the camera using the lighting solution. Filtering or color changes may be needed to reliably detect the change. It will be difficult to make the correct decisions about lighting until some mocked-up testing of the part is conducted.

A machine vision project can be successful by defining expectations, basing a selection on needs and conducting a proof-of-concept. Courtesy: MartinCSI

A machine vision project can be successful by defining expectations, basing a selection on needs and conducting a proof-of-concept. Courtesy: MartinCSI

A proof of concept goes a long way in helping a machine vision project be successful and meet expectations without going overbudget because assumptions were made, and different equipment was needed to complete the application as a result.

Three ways to help a machine vision project succeed

To give a vision project the best chance of succeeding:

  1. Define the needs of an application, set expectations for the application and maintain those expectations throughout the project unless they are proven unachievable.
  2. Set a budget based on the needs of the application and the return on investment (ROI) it will provide.
  3. Properly analyze the application to verify with actual parts in the simulated environment and work to identify any pitfalls that will make the project from meeting expectations. As previously stated, this step is crucial to the success of a project. It must be done without shortcuts or the information will lead to using good and bad parts, the simulated environment, and the equipment planned for the application.

If this is done correctly, the vision system ideas or concepts are no longer a guess or assumption that could lead to costly changes in equipment or a failure to meet expectations. Bad samples are often overlooked or not collected because they can be hard to get, but they are very important to make sure defects can be detected and verified.

Ian Visintine is senior project engineer, MartinCSI. Edited by Mark T. Hoske, content manager, Control Engineering, CFE Media, mhoske@cfemedia.com.

KEYWORDS: Machine vision

Set realistic expectations for a machine vision project.

Ensure proper budget for the project.

Consider the application from the start.

CONSIDER THIS

Have you considered machine vision? Can these tips help?


Ian Visintine
Author Bio: Ian Visintine is senior project engineer, MartinCSI.