Thanks to technologies like artificial intelligence, computer vision, deep learning, and IoT tracking and analyzing production quality becomes much easier, accurate, and effective. These technologies allow production and manufacturing companies to save thousands of dollars on defective products. Today, we are going to show you how production quality works thanks to AI and how manufacturing companies can make the most of this technology.
Everything started just a few years ago with the Industry 4.0 revolution. This fourth industrial revolution brought intelligent technologies to the manufacturing sector. As a result, modern factories are smart and efficient. Machines and devices are assembled not by human workers but by trained bots. The vast majority of repetitive tasks are now automated, and production plants are filled with AI-powered solutions, devices, and applications. They have one goal–to make production quicker and more effective.
And partly, that’s an answer to the question about production quality with AI consulting. You see, AI-fueled bots and machines simply don’t make as many errors as humans. They are never tired, they never feel bad, and can operate at maximum capacity 24/7. This means that production lines supported with smart solutions are much less prone to defects and errors. But let us start from the beginning.
Today, we live in times of automation and digital manufacturing. Almost every production line is automated and supported by intelligent algorithms. But the intensive development of production began in the 1950s when computer numerical control (CNC) was introduced to the factories. Then, in the 1990s, computers were implemented that were able to render complex 3D designs using computer-aided design (CAD) software and translate them into CNC instructions using computer-aided manufacturing (CAM). At the same time, the maturation process for product lifecycle management (PLM) software has begun.
Today we are at the very center of an intensive transformation of production. The introduction of technologies such as digital twins, IoT (Internet of Things), augmented reality (AR), and robotics enables the digitization of the entire production process. As a result, we are seeing unprecedented increases in efficiency, shortened lead times, and the development of completely new business models.
Nowadays, digital manufacturing (involving the implementation of digital technologies into production) is already changing production plants. Technological advances allow enterprises to use real-time data analysis and thus optimize their processes and maximize efficiency. Digital production also enables:
- Elimination of bottlenecks
- Inventory reduction
- Quality improvement
- Shortening the time of introducing the product to the market
- Meeting customer needs more effectively
- Increasing the number of manufactured products
And today, we want to focus on quality improvement. As it happens, thanks to digital manufacturing and AI, production quality becomes much more effective and accurate.
Read our case studies in the manufacturing industry.
Intelligent solutions on the production lines
DIGITAL TWINS AND THE INTERNET OF THINGS
We talked about digital twins at least twice in the past months. And the fact is, this technology has tremendous potential, especially in the manufacturing sector. In short, digital twins are utilized to create a digital version of a product or device. What’s their role in production quality AI? Well, this solution allows manufacturing companies to get valuable insights concerning:
- Improving their operations
- Increasing production efficiency
- Reducing production costs
- And, this is what we are particularly interested in, discovering issues or glitches before they happen.
All of that is possible because digital twins are far more than just a digital model of the physical object. They are accurate digital replicas that are based on IoT sensors. Thanks to this solution, digital twins receive constant, real-time data coming from these sensors and, in effect, from the products themselves. So yes, with digital twins, manufacturing companies can observe how their products behave in the real world, and they have all that information readily available on their computer screens! Currently, digital twins are used chiefly to increase the performance, effectiveness, and quality of the production process but to enhance end products as well. Digital twins can also be used to improve operations of manufacturing machines, production lines, and even entire plants.
From the technical point of view, digital twins combine IoT and data analytics to provide a more detailed view of the product or production line. As a result, manufacturers can optimize costs and improve their solutions.
Let’s start with the most basic industry 4.0 technology that comes in handy in production quality AI. Machine learning is based on algorithms that learn themselves based on extensive initial training. This way, you can train ML algorithms to distinguish defective products from good ones. First off, you need to train your algorithm. In order to do so, you need images of both good and bad products so that your algorithm knows what to look for and how to spot the key areas to examine. Next, machine learning algorithms focus on each element and identify anything different from the good base example. And because these algorithms learn themselves, over time, they get more and more effective. And it goes further, as one algorithm can find diverse defects.
A bit further in this article, we are going to show you how machine learning algorithms can be improved using the more advanced forms of ML–deep learning and lifelong learning.
COMPUTER VISION PRODUCTION QUALITY
Let’s start with computer vision. It’s an AI-related technology that’s based on cameras constantly monitoring everything that happens within the production line. And thanks to these smart cameras, manufacturers have the opportunity to introduce significantly improved quality inspection at high speeds and low costs. Simply because cameras are cheaper and more effective than humans. Moreover, they can do many inspections simultaneously.
At this point, it is vital to mention machine vision as well. Machine vision is a subset of computer vision used primarily in the manufacturing industry. Just like computer vision, machine vision solutions are based on cameras and sensors, but this technology is adjusted strictly to the manufacturing sector’s needs. Machine vision algorithms are capable of:
- Detecting deviations (many of these systems can automatically initiate corrective actions!)
- Detecting quality defects
- Processing tons of visual information simultaneously
MV apps are designed primarily to work with human operators and make them more accurate and effective. This means that machine vision plays a critical role in quality assurance in the production process.
BMW: Autonomous Machine Vision System
In late 2020, BMW announced that they started working with German-Israeli company Inspekto in order to introduce the Autonomous Machine Vision system in their plant in Steyr, Austria. BMW Group Plant Steyr has purchased four systems, followed by a pilot phase, which has resulted in a noticeable improvement in quality and reduction of false detection instances upon the employment of two systems. They use Inspekto G70, a ready-to-go and out-of-the-box visual quality assurance system. G70 replaces tedious integrator-centric solutions with a self-learning, simple, and standalone customer-centric system. According to BMW, by eliminating pseudo-defects, the system has allowed them to avoid the extra loop of manual double-checks.
IMAGE ANALYSIS PRODUCTION QUALITY
Shortly put, image analysis comes hand in hand with computer vision. As the name suggests, these apps and algorithms analyze every picture they receive (e.g., coming from the CV cameras) and alert workers when an error or defect occurs. It all happens within one and the same process.
Machine learning algorithms can evaluate all image data that flows into an AI-based system to pinpoint defects or deviations and provide corrective measures. Results are analyzed to provide risk assessments of the line’s process and the overall program. When the product is already in production, the AI-fueled defect detection systems continuously monitor the production line.
IBM WATSON IOT FOR MANUFACTURING
At the heart of IBM’s solution is an advanced image recognition and cognitive analysis system teamed together with continuous machine learning. Now, thanks to IBM Watson, model managers and data scientists can use their combined expertise to put together a dataset of known defect images depicting good and defective elements for comparison with images captured from the manufacturing floor.
IBM perfectly knows that any particular product might have dozens of images representing a variety of typical faults and glitches, which are classified by the model manager to allow for easy recognition of them. Once the images and their classifications are classified, they are next sent to a data scientist, who trains the cognitive model to recognize them using the NVIDIA Deep Learning GPU Training System (DIGITS). Take a look at this video provided by IBM to understand how their IoT for manufacturing solution works:
And here’s another problem that needs addressing. Today, more than half of product quality checks include some form of visual confirmation. This confirmation is essential as it ensures that parts are in the correct position and everything is as requested. However, such confirmations are time-consuming because glitches and errors can happen to any product, and they can be of any sort and size. The solution provided by IBM compares super high definition images from the manufacturing floor against a library of images displaying known defects to detect faults in parts, components, assemblies, and products. As a result, visual inspections are accelerated and facilitated, making the entire endeavor more straightforward.
Predictive maintenance is another interesting field where production quality AI thrives. Suppose you’re working on a new machine or device. Such products usually require proper maintenance to keep them safe and fully operational, especially after they leave the production plant. Thanks to predictive maintenance techniques, you can lower maintenance costs and improve the product’s quality. As a result, your manufacturing company can offer an extended warranty and quickly repair your products when any problem occurs.
Predictive maintenance is extensively used in sectors where product defects and glitches can cause serious problems, leading to significant expenses. To mention a few, these sectors commonly use predictive maintenance:
One of the companies using predictive maintenance is Siemens. This company sees predictive maintenance as one of the key areas where AI can help production companies reduce costs. When predictive maintenance applications are implemented, operating data is constantly gathered, cleaned, analyzed, and stored. Therefore, as a result, predictive algorithms can monitor the condition of critical components of a specific device or machine. Siemens even introduced something called Predictive Services. Their solution is divided into three modules. Different components are analyzed and data acquired, depending on the industry and application:
- Assessment: They assess the current situation based on machine data, automation hardware, network situation, and similar factors. Then, they generate a detailed connectivity concept on the basis of this assessment.
- Connectivity: The connectivity concept serves as the framework for installing various components in order to acquire the necessary operational data.
- Analytics: Siemens’ experts evaluate the data collected and provide clients with informative reports on the status of their plants and potential causes of error. Siemens’ experts optimize adaptive algorithms in order to detect anomalies that indicate potential errors reliably.
Deep learning in production quality AI
As you know from our other blog posts, deep learning is an advanced form of machine learning. Thanks to a neural network, these algorithms have enhanced learning capabilities and don’t require as extensive initial training as ML apps. And the fact is, deep learning takes production quality AI to the next level. You see, with the traditional approach (comprising machine vision), it’s human experts’ role to indicate which aspects of the product ought to be taken into consideration. These aspects can, for instance, comprise color, size, curvature, and many more elements. This method is simple and effective. However, there are many situations when we simply need more advanced solutions. Consider bottle caps. These parts are small and pretty much the same. However, even the slightest glitch can cause the entire beverage to spill.
In such complicated situations, deep learning ups the game. That’s because the deep learning algorithms learn themselves. They don’t need to indicate which parameters are important. They learn from data as they analyze the production process and its outcome. In this analyzing and learning process, DL algorithms create their own implicit rules that determine the combination of features that define the requested quality.
Of course, the initial training is still required. After all, these algorithms have to know what they are looking for and what are the potential errors. That’s why you need both perfect and defective products to teach the algorithm what to look for. Once it’s trained, it can look for glitches and defects all on its own. However, even here, there is room for improvement. What do we have in mind?
At this point, we’d like to turn your attention to something called lifelong learning systems. These systems are still based on neural networks but have the ability to add new information to their networks without the need to stop working. Lifelong learning systems are based on so-called L-DNNs–lifelong deep neural networks. And interestingly, these networks don’t need images of defective products or predefined rules. The manufacturer only has to provide some images of the 100% correct products, and every deviation is noticed automatically.
L-DNNs require just a few images in order to develop a thorough understanding of the product. And once it’s done, they can work independently, marking every deviation. If some changes need to be implemented, they can be done without stopping work. This way, any changes to the rules that define an operative object can also be made in real-time. And this means there are no delays in the production process!
It’s a US-based company working on L-DNNs. When it comes to the manufacturing world, Neurala offers so-called Vision Inspection Automation (VIA) software that consists of two modules:
- Brain Builder: Enables creation of AI models. Brain Builder is a cloud-based solution that enables rapid AI prototyping, allowing users to annotate and train L-DNNs at the same time.
- Inspector: Allows for seamless integration on the factory floor, including collecting images from machine vision cameras and providing outputs over an industrial network directly to the PLC, controlling the applicable machine to run the vision AI models.
Neurala systems can be trained with only good product images (unlike traditional deep learning models that need both good and bad examples). Thanks to working with good samples only, Neurala algorithms can produce an anomaly signal when a glitch or defect happens, regardless of its nature.
As you can see, there’s quite a lot going on concerning quality control AI. Intelligent solutions allow manufacturing companies all over the world to improve their production processes, spot potential anomalies, and glitches and even prevent them from happening in the future. And thanks to predictive maintenance, high-quality products, and devices can serve much longer in perfect condition.
If you’d like to find out how AI-related technologies can improve your production processes – drop us a line. Addepto is an experienced AI consulting company, and manufacturing is one of our specializations. We are at your service!
 Sam Scane, Manufacturing Global, BMW: Quality Control Through Artificial Intelligence, Nov 13, 2020, https://manufacturingglobal.com/ai-and-automation/bmw-quality-control-through-artificial-intelligence, accessed May 19, 2021.
 Jen Clark, IBM.com, Cognitive inspection: IBM Visual Insights, July 4, 2017, https://www.ibm.com/blogs/internet-of-things/cognitive-visual-insights/, accessed May 19, 2021
 Predrag Jakovljevic, technologyevaluation.com, Siemens Sees AI as Integral to Industrial Digitalization, Oct 25, 2017, https://www3.technologyevaluation.com/research/article/siemens-sees-ai-as-integral-to-industrial-digitalization.html, accessed May 19, 2021