Fast, accurate, automated vision inspection

Machine Builder Provides Vision System That Can Handle Three Parts Per Second With 99.97% Accuracy. Its Name Is 'Vision System'

By Brian Altman, Altman Manufacturing

2 of 2 1 | 2 > View on one page

"We use this pre-production process to create a customer presentation that illustrates how a product could be inspected given the technology available," Altman explains. "In production it points to the defect, describes how it was found and the algorithm we used, and explains how a defective part would be sorted. We can show customers how effective a solution will be before they order the machine, similar to a proof of concept. This ability is especially important when developing a vision system. Until we have a lighting scheme and know which lens will be used, we can't be sure if a solution will function as we expect. Working with Teledyne Dalsa to show customers difficult applications ahead of time has been very important to our business. Customers understand in advance what a vision system will do and how it will work, so there is no ‘vision creep,' that is, future requests for inspections beyond the capabilities of a machine."

Right Light, Right Camera, Right View
The machine Altman built for the medical-component manufacturer (Figure 1) uses Teledyne Dalsa's GEVA Vision Appliance, eight Genie high-resolution 1600 cameras and iNspect vision application software. The system is set in advance based on which parts are to be inspected, specific lighting schemes and camera zones that are developed for each unique part. For example, one part might require inspection using an overhead light, while another might need an under light, and a third could require a dark-field inspection. Each part and lighting scheme is also associated with a particular field of view, so one part will be inspected with cameras one, two and three, but a second part will be inspected with cameras one, four and six. To simplify an otherwise complex process, cameras are set so that no elaborate focus techniques are required.

Since the machine can accommodate multiple parts, the parts are hand-fed by an operator, who places them on the belt. The machine then orients each part so that it moves under Camera One, which checks that the correct part is presented properly. As a part moves down the conveyor, it crosses on to a glass disk, and is viewed under the appropriate lighting scheme and with the designated camera regime (Figure 2). Images are then inspected using Teledyne Dalsa's iNspect software. If a part doesn't meet the pre-set requirements, it's discarded. The image of the defective part is stored so that the defect can be tracked, and a counter identifies into which bin the part was sorted and why.

The vision system is integrated with other elements of the production process by an Ethernet connection. As data is compiled, operators can be notified quickly if the number of discarded parts meets a preset threshold, which might indicate an issue in the manufacturing process.

Parts that pass inspection are sent on to packaging or shipping. The customer produces only Elastomer rubber components, so finished parts are sent to a different facility for assembly.

"With this automated inspection system, our customer inspects more than 100,000 of the same type of part in each run and transitions seamlessly from part to part, inspecting as many as 20 different parts in a day," Altman notes. "Thanks to the efficacy of this vision system, the machine can inspect three parts per second with 99.97% accuracy, a rate of speed and accuracy that was unimaginable before." Altman is quick to add, however, that the real value of the vision system goes beyond the percentages: The most important consideration is whether or not the customer is able to identify as many defective parts as possible based on the algorithms and criteria they've defined.

Altman credits Teledyne Dalsa's GEVA Vision Appliance with ensuring the scalability of the solution. "The GEVA Vision Appliance enables the solution to be completely adaptable," he says. "We can incorporate eight different cameras for eight unique fields of view, but then choose which ones will turn on and when for a completely flexible implementation."

Even given the complexity of the solution overall, Altman notes that the iNspect vision application software simplified the design and deployment of the automated inspection process. "Customers do not have to be technically skilled to use it," he says. "The software works with real-world dimensions and drag-and-drop algorithms for an easy-to-use operator interface. We provide the original ‘recipe' and back-end script, so there is no need for an operator to have the expertise to write code. A user can easily switch from part to part, or create a scheme to inspect a new part using a previously scripted ‘blank' menu item and the drag-and-drop algorithms."

In fact, Altman says that while training for other types of programs could take weeks, the training for iNspect was completed in just over a day, and he was able to deliver a level of training that ensured that the operator he trained could train someone else.

In addition, the vision system requires minimal maintenance, just periodic cleaning of the lenses and optics. But if additional support is needed, the customer can reach out to either Altman or Teledyne Dalsa.

Flexibility for the Future
Altman never had any doubt that the Teledyne Dalsa solution was ideal for this application, but his confidence was proven again after the equipment had been in operation for more than two years. "This vision system was ideal in that it allowed us to incorporate multiple cameras and gave us the ability to control them with easy-to-use software," he notes.

"Recently, the customer needed to inspect a completely different and unique part, a slit in a Duckbill (check valve), and they were able to set up the appropriate lighting scheme and fields of view on their own — without any help from me. They're highly satisfied with the solution, and the results are proven. Eliminating manual inspections alone has led to a significant return on investment, and the speed and accuracy of the new equipment has increased productivity rates overall. Plus, we've demonstrated that the solution can scale to meet all their future demands."


More from Control Design:

Download our related ebook: Sensors and Vision State of Technology



2 of 2 1 | 2 > View on one page
Show Comments
Hide Comments

Join the discussion

We welcome your thoughtful comments.
All comments will display your user name.

Want to participate in the discussion?

Register for free

Log in for complete access.


  • Above, the article states that "Since the machine can accommodate multiple parts, the parts are hand-fed by an operator, who places them on the belt" Also above, a throughput of 3 parts per second is mentioned. This seems inconsistent. Does the throughput depend on how quickly the operators can place the parts on the conveyor? Three parts per second seems unreachable if that's the case. Also, one of the goals of automated inspection is to save labor. Manual loading seems to defeat that goal. What am I missing?


RSS feed for comments on this page | RSS feed for all comments