13 May 2019
Traditional machine control technologies are based on the work of application-specific engineers. To tackle a certain task a control engineer would need to understand it in terms of physical requirements, work on a traditional physics-based solution for it and transfer the knowledge obtained into source code. In an AI-based approach, the engineer will be more focused on data. The challenge to be solved is first described through data collection and then the engineer works with an abstract data-based representation of the physical problem. Therefore, the data engineer may not need to understand the physical details of a problem to generate a data-based solution for it. “Some machine builders are already successfully integrating AI into their machines, said Fabian Bause, product manager TwinCAT at Beckhoff Automation. “Most are focused on AI-enhanced vision applications. Machine vision is a key application for AI because of the very powerful neural networks available for image recognition. In addition, successful networks are available in the AI community such as AlexNet, which can be shaped to suit similar applications by learning transfer. AI-based anomaly detection is also in use with some machine builders, because for training, this approach only requires data to describe the normal behaviour of a machine.” Bause expects that the highest expected impact will be in applications such as AI-enhanced predictive maintenance, collaborative robots, enhanced productivity, quality control and process optimisation. “In terms of collaborative robots and process optimisation, machine builders will be able to solve certain tasks in a more precise, efficient and flexible way compared to classical approaches, he said. “Machine users will benefit from less downtime due to improved predictive maintenance, an increased production output because of prediction algorithms that can detect faulty processes and correct them early in time, and also benefit from robust and reliable end-of-line testing of the produced goods,” concluded Bause. Different approaches According to Tim Foreman, european R&D manager at Omron Industrial Automation, there are different approaches to the application of AI technologies in the production process: “Whether you take the wide analysis – like comparing factories or machines – or you take the deep analysis – like drilling down to the microsecond level to see what is really happening in your machine – a human can be very creative and is skilled enough to recognise images, but not at continuous microsecond speed or in data sets with ‘more than a few numbers’. It is in these situations that the machine can really add value and take our creativity to the next level,” said Foreman. “Taking into account these challenges, Omron has focussed on the second area: how technology can extend human capabilities by drilling deep down into the machine. Using our machine AI technology, a machine builder can now continuously monitor and analyse the detailed behaviour of the machine. From this, it is possible to detect anomalies happening or even trends indicating maintenance needs. Such technology can also be used during the design and commission phases to help reduce debugging times. For a machine user, these same benefits will lead to less unforeseen down-time of the machine and at the same time prevent outflow of ‘out-of-spec’ products from unforeseen events.” Reduced effort Daniel Smalley, product manager SCADA at Siemens believes that the use of AI, with all its different capabilities, will reduce programming and engineering efforts, making control logic more agile and flexible towards changes in the machine environment. He goes on to say that it will make production processes more configurable enabling end users to respond to the exact requirements of their customers. “This is why it is necessary to focus on introducing scalable and highly customisable AI applications that can be distributed over all levels of our automation portfolio from controllers, to our Industrial Edge and Mindsphere IOT platforms,” he said. “This can offer benefits for both the machine builder and the machine users. For the machine builder, it is possible to predict machine failures before they happen, focus development on machines rather than infrastructure and lower the expertise required to operate the machines. For the machine user, benefits include increased availability of machines and lower complexity in the operation of machine and condition monitoring”. New possibilities Klaus Petersen, marketing director Factory Automation EMEA, Mitsubishi Electric Europe B.V. argues that the implementation of AI technologies in machine controllers is extending traditional machine control architectures with more advanced data processing, learning and decision-making capacity. “The objective is to deliver increased productivity, efficiency, reliability and accuracy, as well as opening up new possibilities for machine control” he said. “AI can, for example, be a driver for increased productivity. Today, machines are built to work within defined margins of capability – perhaps to allow for different loads or speeds or safety ranges. AI technology, using deep learning algorithms within the control system, could enable machines to be driven right up to and even beyond today’s margins, significantly boosting productivity without compromising reliability. “Applying AI principles to individual machine processes can already help to reduce auto-adjustment times, synchronise increasingly complex systems and offer helpful suggestions to operators,” continued Petersen. “It can even enable autonomous decisions to be made based on measured data in real-time, further optimising the process.” Making reliable predictions based on experience, evidence and guidelines is a fundamental function of human intelligence and AI is no different. It can contribute toward more effective predictive maintenance, monitoring the condition of components to enable replacement before damage occurs, so preventing unplanned downtime. According to Petersen, deep learning algorithms are now pushing the boundaries further, calculating with more accuracy how long a component can run before replacement. Maybe even compensating for delivery times on replacement parts by slowing the machine down slightly to increase longevity rather than stopping the production line completely. AI could become a driver for increased efficiencies right across the production environment, moving into the realm of big data analysis. Petersen said: “AI technologies enable different machine states to be recorded and analysed in real time to recognise the current machine status, detect potential faults on the horizon, and immediately offer recommendations for actions to the machine operator or autonomously initiating remedial actions. “Reaping the maximum benefit from this development will depend on the use of control systems that not only embed these technologies, but which also provide higher levels of connectivity. Only once the full spectrum of data sources on the plant floor can be connected to edge computing platforms for efficient processing – for example and on to MIS/MES and ERP systems – can the full benefits of AI be realised. This level of integration enables a far greater range of KPIs to be analysed and so can be used to drive improvements in overall equipment effectiveness (OEE). “What we see, then, with control systems built around AI technologies are machines that are self-learning and self-optimising. The importance of AI to the machine control market cannot be overstated,” concluded Petersen.
Print this page | E-mail this page
This isn't a paywall. It's a Freewall. We don't want to get in the way of what you came here for, so this will only take a few seconds.
Register Now