Background Image


The key to scaling computer vision in manufacturing



In 1908 the Ford Model T was introduced to the world. The vehicle was one of the first mass production vehicles, allowing Ford to achieve his aim of manufacturing the universal car however from 1914 - 1925 it was only manufactured in black.
However, in today’s world, a car can be customized in a plethora of ways from transmission type to color to the speaker system installed in the car. In fact, there are now up to 40% more components per vehicle compared to 20 years ago. This added complexity adds opportunities for defects, given the increased complexity of the process.
To combat this, manufacturers are increasingly investing in advanced technologies such as computer vision to continue improving their efficiency and productivity while maintaining a high variety of products. While manufacturers are looking to techologies, such as AI, to improve efficiency and productivity, it’s not always an easy fix; complex processes can make it difficult to scale across lines and plants easily.

Current technological solutions lack flexibility

Manufacturing plants are complex environments containing a large variety of different processes and shop floor layouts. What this means in the context of computer vision models is that there won't be a single computer vision model that can cater to all the different processes on the shop floor. Shop floor layout adds another layer of complexity because there aren’t always easy locations to install cameras for a model. This means that you'll need to use a different model to factor for an angle change.
What a lot of computer vision companies tend to do is create a single universal super algorithm to cater to all use cases and find it hard to scale beyond a single use case or line because the algorithm can’t cater to all the variations in a manufacturing plant. This leads to these companies being stuck in pilot purgatory with their customers and causes companies to pay more than they are saving etc.

Why flexibility is important for scale

Let’s take the example of an automotive production plant. In such a plant you can have a moving assembly line as shown in Figure 1 below and the vision system needs to be able to track operator actions on the moving assembly line.
However, there are sub-assembly stations close to the moving assembly line that are static and building sub-assemblies that are installed on the vehicles on the moving assembly line as shown in Figure 2 below.
For a computer vision solution to be scalable across an entire manufacturing plant like the one mentioned above, it has to be able to cater to both use cases without any drop in accuracy while maintaining high uptime and reliability. Furthermore, the vision system has to be able to cater to different camera angles and orientations across both sets of assembly lines since the layout and infrastructure of the plant might not make it feasible to have the same camera angle for every station on the plant.

How i-5O is tackling this challenge

At i-5O our team has spent the last decade deploying technology in a manufacturing environment and we built our system keeping the need for flexibility in mind.
What we have found is that for every process and camera angle in a production plant, there is a model that will deliver the highest level of accuracy. With this in mind instead of focusing on developing a single universal super algorithm, we’ve developed a library of computer vision algorithms tailored to specific manufacturing use cases that can be deployed at scale to cover multiple production plants.
Our platform makes data acquisition/processing quick and efficient thus allowing us to acquire more data at a faster rate and to continuously improve our ever growing library of algorithms. This approach has allowed i-5O to work with manufacturers in multiple verticals with different constraints and successfully scale to production.

Try it for Yourself

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