Background Image

Why computer vision has

applications beyond defect detection in manufacturing

BY KHIZER HAYAT

4 MINUTE READ

Defect detection is one of the most common use cases of computer vision in manufacturing. Modern Visual AI technologies rely on powerful cloud-based servers that allow them to rapidly ingest visual information for machine-learning training purposes. By training a computer vision system with hundreds of thousands or millions of images of specific types of product defects, these systems can learn to rapidly identify similar defects with a high degree of accuracy. Visual AI defect detection systems can identify flaws like bottles missing bottlecaps, cracks in pipelines, poorly painted surfaces, missing parts, broken items, misshaped items, cracked glass, cracked metal casings, and virtually any other type of errors that human visual inspectors identify.

However, there are a number of other simpler use cases for computer vision systems that can generate higher Return on Investment (ROI) for manufacturing organizations with faster payback periods. We will explore a few of those use cases in this post.

Defect detection use cases make it hard to generate high Return on Investment

One of the major challenges we’ve seen working on defect detection use cases in manufacturing is the ROI for manufacturers. Defect detection systems usually need to have accuracies above 99% (ideally greater than 99.5%) to truly generate a significant ROI for manufacturers. This is because these systems are usually meant to replace human inspectors and for that to happen the system’s accuracy needs to be near flawless to help manufacturing operations personnel feel comfortable with the solution, as Quality Control (QC) is a critical step in the production process. The closer the product is to completion the more critical the QC step is for production personnel, as any defective parts that go out to the customers can cause massive liability issues for the manufacturers.
The high accuracy for the computer vision system might be achievable in cases where there is little variation in the manufacturing process e.g. plastic bottle manufacturing but in cases where there is high variation, like automobile assembly, a lot of edge cases pop up during production making it extremely challenging to have a system produce accuracies above 99% continuously. In the high variation cases the computer vision systems usually fail to scale to production due to the issues mentioned.

Computer vision for measuring productivity is a much higher impact use case

Measuring productivity using computer vision can generate a much higher ROI for manufacturers as the threshold for accuracy is much lower (usually 90%) and the data generated can be used in real-time to make improvements. For example, computer vision can be used to run continuous time studies on a production line and whenever there is significant downtime on the line, the system alerts operations personnel who can take action immediately to fix the issue before it snowballs into loss of revenue.
Computer vision generates high ROI in this particular example because in a manufacturing environment, time studies cannot be run continuously due to personnel capacity i.e. it is impossible to have one person on every station running a stopwatch for 16 - 24 hours per day to measure the processing times. Usually the output at the end of each shift is measured at which point it is already too late to fix the issue. Thus using computer vision, operations personnel get continuous real-time visibility on all their lines with the ability to take action as a problem occurs.

i-5O’s computer vision system is helping manufacturers increase revenue

At i-5O our clients are using our computer vision system to track productivity and wait/idle times on their production lines. Our cameras are installed on all stations in a production line to track the processing times of every station and generate a real-time Value Stream Map (VSM) to help identify production bottlenecks. At the same time our vision system is being used to track products that are sitting idle for an extended period of time to alert operations personnel of potential issues with the product so they can take action. This has helped our clients increase revenue by over 26% on a month to month basis.

Try it for Yourself

Schedule your 1st Free Consultation and Start your AI Journey Today