Smart & autonomous quality assurance
12 November 2020
Quality assurance (QA) technology is essential when adjusting manufacturing operations, to guarantee that a new process doesn’t mean a drop in quality. Here Yonatan Hyatt explains how this technology is pushing the boundaries of artificial intelligence (AI) to respond to market needs
Machine vision plays a critical role in quality assurance (QA), allowing manufacturers to identify which products have been made to specification and which contain defects — an inevitable consequence of any manufacturing process. By using machine vision solutions for quality inspection, manufacturers can quickly and efficiently identify defective products, while avoiding expensive, slow and unreliable manual checks.
Challenges of machine vision projects
The success of machine vision in industrial manufacturing is due to the fact that it provides faster, more accurate and more cost-effective QA than manual visual inspection. In Germany, for example, the average value added by each employee in automotive engineering is €596,000 per year. By assigning employees to tedious inspection tasks that add no value, the manufacturer misses out.
Moreover, manual inspection carries an error rate of approximately 25%. In challenging applications, such as parts with many small components and tight tolerances, the error rate might be even higher, or their sheer complexity may mean manual inspection is not an option at all.
Machine vision solutions allow manufacturers to overcome these problems, but they also come with numerous disadvantages. The first is that plant managers have no direct control over them, since traditional machine vision solutions are developed ad-hoc by a systems integrator. The integrator is responsible for selecting the right components — lenses, frame grabbers, software and more — as well as for designing, integrating and testing the solution on the production line.
Cost is another obstacle, since traditional solutions start at a minimum price of €20,000 and can cost up to €150,000 per point of vision, or more. Finally, these solutions require long periods of downtime to be installed and trained, and are inflexible in that they are designed to inspect only one specific product at a set location.
Autonomous machine vision (AMV)
To overcome the cost, flexibility and complexity challenges holding manufacturers back from using machine vision technologies on a large scale, Inspekto developed Autonomous Machine Vision (AMV), a new category of machine vision for quality inspection.
Autonomous Machine Vision is a hybrid technology that merges computer vision, deep learning and real-time software optimisation technologies. Unlike traditional machine vision projects, systems in the AMV category are plug & play, self-contained products that any user can simply install on the production line without relying on a machine vision expert.
The user does not need to have any expertise in data annotation or labelling, since the system will learn the characteristics of the object to inspect in a fully autonomous way, with minimal human intervention.
Learning like a human
During set up, the user simply switches on the controller and ensures that the field of view (FOV) covers the location to be inspected. The user then places a good sample item in the FOV and uses a mouse to mark a region of interest for the system to detect defects in.
Unlike traditional QA solutions, which require hundreds or even thousands of good and defective sample products to be trained, Autonomous Machine Vision systems only require an average of 20 to 30 good sample items, and no defective ones.
Autonomous Machine Vision flips the parameters of traditional QA. Instead of memorising what a defective item looks like, systems learn — much like a human being would — what a good one is supposed to look like.
Once in operation, an AMV system compares each image with the ones memorised during set up, verifying both the shape tolerances and surface variations to identify any defects. The system will then communicate the location of a defect to a human operator or to a programmable logic controller (PLC) so that the defective product can be quickly removed from the production line, so time and resources are not wasted completing a product that is already faulty.
The image acquisition process
The autonomy in Autonomous Machine Vision is the result of several AI engines working in tandem. The algorithms developed by Inspekto make AMV systems self-setting, self-learning and self-adjusting.
A Video-Sensor-Optimisation AI engine automatically adjusts the illumination and camera parameters to the product being inspected and the environmental conditions, so the system can obtain the best possible image, with perfect focus, depth-of-field, exposure level and dynamic-range settings. This capability means that unlike traditional solutions, AMV systems easily adapt to new lighting conditions and can operate effectively at any time of day.
A Detection-and-Alignment AI engine automatically locates products in the 3D space. This means that the system will recognise products even if they appear in a different location or orientation then the ones it memorised. Consequently, the system will only notify the operator of true defects, as opposed to false alarms from phenomena such as movement, orientation or lighting changes.
While in operation, AMV systems keep learning — if a system flags an abnormality that the manufacturer deems as non-defective, the system will learn and not flag the same defect again in the future.
Systems in the AMV category are not product-specific, they are universal. Since they are not tailored-made to inspect a specific item, they can be used for items produced in every industry, with any handling method.
AMV systems can communicate with the manufacturer’s PLCs via all common industrial protocols including PROFINET, Ethercat and Modbus and can be installed in every position along a Bosch profile, present in most manufacturing lines.
This flexibility, along with the minimal initial investment that AMV systems require, naturally encourages what Inspekto calls Total QA — visual quality assurance at every stage of production.
Total QA enables manufacturers to reduce scrap and substantially boost yield and productivity, by removing defective products from the line as soon as they become faulty. Over time, Total QA allows manufacturers to identify areas where defects are introduced more often, so that they can be optimised.
Leading manufacturers like Bosch, Pepsico, BSH, Mahle, Geberit, and Dailer are already using Autonomous Machine Vision, and Inspekto’s products now have a global footprint, reaching plants across Europe, the US, Mexico, Thailand and Turkey. In the post pandemic industry scene, more manufacturers across the world may choose to join these global players in embracing Autonomous Machine Vision. Opting for flexible and affordable machine vision technology may help them cope with the need to inspect a greater variety of locally produced items, supporting manufacturers efforts to onshore production.
Yonatan Hyatt is CTO and co-founder of Inspekto