Five steps to implement AI on the factory floor
Increasing computing power, growing data volumes and the increased use of sensors means the discussion about artificial intelligence (AI) on the factory floor is gaining momentum. Adaptive algorithms offer enormous potential for the further developments required for Industrie 4.0, such as predictive maintenance and networked production. In this context, AI can help increase overall equipment effectiveness (OEE) to reduce costs and increase productivity.
However, the challenge faced by businesses is many of the cloud-based AI solutions place enormous demands on infrastructure and IT. These solutions work with vast amounts of data that is laborious to prepare and take advantage of. In addition, system concepts for mechanical engineering are often complex and specially tailored to the respective requirements. A reliable use of typical AI algorithms is only possible through extensive testing, constant optimization and over-dimensioning – which many companies shy away from.
Five ways for getting started with AI
There are many AI solutions suited for use in industrial automation. These include open source software and applications based on machine learning (ML). Robotics and automation providers are currently developing AI solutions that help small and medium-sized businesses in particular to use AI effectively and efficiently. The following tips will help you get started:
- Upgrade data management capabilities. Manufacturing companies are often more conservative when it comes to new technologies, as they work with machines that must run for 20 years or more. This should not mean these companies they have to lose out when it comes to AI. It is important investigating the benefits AI and ML will bring to the industrial environment and to over any reluctance to invest in these technologies. Companies should ensure they can work with large amounts of data and advanced algorithms, the two cornerstones of artificial intelligence.
- Outline central project questions and approaches. Important questions at the beginning of an AI project include: Which problems and challenges should be tackled? Which strategy and technology are best suited, and are they adaptable and expandable for a variety of projects and use cases? Which managers and employees should be brought on board? Is there the necessary expertise within the company or is there a need to involve external experts? How can a new machine with an integrated data science approach be planned and implemented?
- Define clear and measurable goals. The primary goal of AI deployment is to increase quality and process efficiency, for example through improved predictive maintenance to avoid equipment downtime. The AI-based solution should therefore aim at measurable and noticeable improvements in OEE. It is important to note even an optimization of only a few percentage points can lead to considerable increases in efficiency and cost reductions. AI in machine maintenance, for example, can help to reduce the risk of equipment damage and downtime, as problems can be detected early, and immediate action can be taken to eliminate them. Without automation, machine designers and operators would have to create their own analysis and optimization solutions or use costly cloud solutions.
- Take advantage of AI “at the edge.” Instead of laboriously searching through data for patterns, find technology that approaches things differently – for example can the required algorithms can be integrated into the machine control to create the framework for real-time optimization – at the machine level (the edge). This involves monitoring production lines and machines with real-time sensors, which immediately collect the data and check it for anomalies.
- Focus on real-time data processing. While cloud-based AI solutions place enormous demands on infrastructure and IT, and the processing of data volumes is a tedious and time-consuming process, AI at the edge suits predictive maintenance and control of machines. It combines line control functions with real-time AI-based data processing. One advantage is it is possible to reliably identify unforeseen situations and react quickly, improve quality, maintenance cycles and machine lifecycles, and scale as needed. The processes gain intelligence on the basis of previous findings and improvements, and drive the holistic optimization of the manufacturing process.
Tim Foreman is European R&D manager at Omron. This article originally appeared on the Control Engineering Europe website. Edited by Chris Vavra, production editor, Control Engineering, CFE Media, email@example.com.