Myths vs. reality of AI-based visual inspection

In today’s fast-paced production environments, rejects aren’t just a bump in the road—they can bring entire processes to a standstill and increase costs. Even more critically, if a defective part is dispatched to a customer, the company’s brand reputation is at risk. To tackle these challenges by ensuring the most thorough and efficient inspection possible, part manufacturers are increasingly adopting AI-based visual inspection, like Inspekto from Siemens.

But while this technology promises potential, it’s often clouded by misconceptions. In this article, we'll address five common myths about AI-based visual inspection.

Myth 1: AI inspection is difficult to install

Inspekto is designed to make AI-based visual inspection simple and hassle-free. It is delivered with everything you need, containing a Siemens Industrial PC, a camera, a light, and all the required cables, which eliminates the need for additional orders or hunting for compatible components. Flexible installation options mean it works seamlessly in various setups, from production lines to robotic systems, even supporting multi-camera configurations for complex tasks. Inspekto can be fully integrated with PLC or MES/ERP systems, which makes it great for the inspection of incoming goods or the final quality inspection of finished products.

Once installed, AI guides you through the training process. The initial step is training the model using a data set of just 20 good samples, focusing on factors like quality, form, and accurate labeling. Bad samples are optional, not mandatory. Users get instructions via the Inspekto software for every step of the journey. There’s no guesswork, as it tells users which data to add and when. All in all, it usually takes less than a day to get everything going. These time savings are down to the fact that this is not a traditional machine vision solution requiring extensive customization by experts for each use case.

Myth 2: AI inspection is suitable only for non-reflective objects

Virtually every type of material can be inspected, such as plastic injection moldings, metal castings, coatings, mechanical assemblies, surface anomalies, incoming goods, and packaging and labeling. Take reflective parts, like a battery cell, as an example. Inspekto’s built-in AI simplifies the process by automatically adjusting lighting to capture the optimal image—whether the component is reflective or not. That’s in stark contrast to the past when parts were sent to external providers that tested many different combinations of lighting and cameras to achieve the best settings.

Myth 3: Machine vision experts are necessary for using the solution

Getting started with Inspekto is refreshingly straightforward. Initial programming requires just 20 good parts and basic domain knowledge. The system is intuitive, requiring neither a vision expert nor an AI expert. With only a few clicks, you mark the inspection regions on the good samples. The AI can then accurately identify small defects that would otherwise be missed by the human eye. And it does this at high speed – over 100 inspections per minute.

Myth 4: Each new use case requires setting up an entirely new project

After initial setup, you can easily add further products to inspect. You don’t need to start a new project or consult a machine vision provider. Our flexible approach allows you to easily accommodate changing demands, too. Take the example of a printed circuit board that has undergone some minor adjustments. Staff on the shop floor can quickly make the necessary changes, with finetuning as simple as the initial setup. There’s no need for new projects or to send out samples.

Myth 5: It takes significant effort to keep AI inspection systems going

Inspekto is a simple solution: implement it once and let it do its job. Factory operators can shift their resources elsewhere, such as focusing on other machines or production levels. Any defects that Inspekto identifies save users time and money, and operations become more efficient altogether.

This article was originally published as an Industry Stories article, which can be found here.

Discover Digitalization

Our LinkedIn newsletter for industry decision makers contains a broad variety of topics, including the Industrial Metaverse, Digital Twins, IT/OT Convergence, Artificial Intelligence, Cybersecurity, and more. In each of our issues, we share stories of digital transformations and give you premium insights on how to become a sustainable Digital Enterprise.

So, no time to wait – subscribe now to our newsletter and get excited about our next issue in February.

In the meantime, explore real-world Digital Enterprises on our Industry Stories platform or tune into Talking Digital Industries, the podcast for technologies and trends that drive industrial enterprises.