The Industrial Internet of Things (IIoT) is characterised by highly automated and universally networked industrial production processes.
A new generation of compact and mobile robots – called cobots – increasingly determines the scenario in production halls. Modern technologies contribute to making these digitised, value-added chains safer and more efficient. In this context, machine vision plays an important role. Machine vision solutions are capable of precisely recognising a wide range of objects in the production run based on optical features alone. As the “eye of production”, the technology is suitable for monitoring entire manufacturing processes, reliably identifying objects and precisely determining workpiece positions.
More and more, artificial intelligence (AI) technologies are becoming part of innovative machine vision solutions in the context of IIoT processes. Several machine learning and deep learning methods, such as convolutional neural networks (CNNs), are being used for this purpose. Through sophisticated analysis methods, the image data can be used to clearly categorise the objects to be identified and assign them to classes, based on a training process. As a result of the training, the images of the objects to be detected are labelled as to class. The models, or classifiers, trained in this way can then assign newly recorded images to the classes already trained. These technologies can result in very high and robust recognition rates for applications in numerous industries.
Optimising the results in quality management
This approach also allows the precise detection and exact localisation of defects in manufactured products. AI-based deep learning functions are capable of greatly reducing the effort involved in detecting errors. In this way, the algorithms can optimise the results in a company’s quality management system. Detecting all possible types of defects would require hundreds of thousands of images, all of which would have to be analysed manually. Such a time-consuming and costly method would be very difficult to put into practice. Thanks to deep learning, this process can be made significantly less labour-intensive. The algorithms are capable of independently learning certain errors and flaws and are thereby precisely defining the particular problem classes. As a result, a wide range of error types can be trained and reliably detected.
Companies that want to develop their own deep learning networks should be aware that this, too, is extremely labour-intensive. In order to gain valid recognition results, hundreds of thousands of sample images are necessary. Professional developers are needed to configure the networks. Furthermore, most of the sample images are protected by licenses, which means that they can be used only with the approval of the copyright holder. A better choice for companies is to purchase and use pretrained networks, such as those integrated in MVTec HALCON, for example. This kind of network already knows the basics of extracting general information from images. Thus, it must only learn how to use these generic image features on a specific application to distinguish between object-, feature- or error classes. The network has been trained with license-free images with industrial themes. This means that these pre trained networks may be used without restrictions.
Machine vision supports universally networked and highly automated processes in the IIoT.
Deep learning technologies can distinguish between numerous different defects.