It’s a challenging time for Artificial Intelligence (AI) as the wider public have dramatically become aware and been able to access the technology. Broad public consciousness has mushroomed as tools like ChatGPT became mainstream in just a few months. However, this hasn’t been all good news for the technology as there have been many more negative press stories about its potential misuse than its power for good. The break-out of AI has stirred negative associations from fiction and gaming about rogue machines and computers wreaking havoc. It is important the engineering community is aware and communicates the benefits of the technology, differentiating the industrial applications where we can massively improve productivity, reduce energy consumption, and become more competitive.
In the industrial sector, I believe the wider awareness and acceptance of AI will accelerate its uptake within automation, design, and manufacturers. However, will we have to use terms like Machine Learning to avoid some of the more lurid headlines and perceptions?
AI for productivity
Within Festo, AI has been a focus for a long time. We have developed the Festo Automation Experience [AX] artificial intelligence tool which enables engineers to use machine learning algorithms to achieve high added value from the data produced by their systems. It was designed to address three key areas: preventative maintenance, energy, and quality optimisation, so customers can ultimately increase productivity and reduce costs although new additional applications are evolving rapidly. These include a robot vision aid called Grip AI and a stand-alone actuator package Cylinder AI.
Festo AX keeps “humans in the loop” by integrating the users’ valuable application knowledge and findings, with our algorithms in what is called Reinforcement Learning Software. The user inputs helps Festo AX ‘learn’ more about the state of the assets providing continuous learning and thereby improving the algorithm predictions. This means anomalies are defined with greater reliability, finer classifications and alerts raised more appropriately. Human feedback on anomalies detected within the data trains the software, translating raw data into systemised models where patterns are established, and the causes and solutions are correctly categorised and communicated.
Having learned what is important, or not, notifications and alerts are raised to the appropriate people at the best time and in the preferred way. This can be via high level reports, maintenance interventions, text messages or direct input to smart maintenance tools.
AI simplifies the design process
It’s my view that the future of Machine Learning and Industrial Automation are intrinsically linked. We will see more powerful software tools in the not-too-distant-future that speed-up and support designers and programmers with AI optimised machine designs. I’m sure that these will be easier than ever before to use, for example, by being programmed more intuitively than today, using plain speech or text.
Complex machine models will be created more quickly and error-free from standardised and structured data models. Intrinsic to this is the role of Digital Twins for the components and sub-systems. These incorporate all the information about components’, their physical attributes, performance dynamics and operation. Having this information in a standardised and structured format enables it to be read digitally and means the user doesn’t need to ‘hard program’ all this data. The system ‘reads’ its constituent parts and ‘understands’ their operation, interpreting anomalies in their performance and supplying valuable information in plain text enabling improved predictive maintenance, energy consumption etc. This means the digital system configuration is always up to date and isn’t reliant on updating documentation or programs during the machine lifecycle.
From a sustainability perspective engineers can already see exactly how much CO2 it takes to manufacture a machine, right down to individual components, as well as how much CO2 is consumed during its lifetime. This has huge benefits in terms of sustainability, giving engineers visibility of each machine’s complete carbon footprint, right from conception to end of life, and all before anything has physically been built. When individual machines’ Digital Twins are combined, complete plant operations can be mapped, monitored, and optimised.
I anticipate that designing machines using AI will work in a similar vein to creative AI programmes like Stable Diffusion Art, whereby the user defines the task - what they want to achieve, and AI APIs suggest alternative solutions. The engineer can either directly review the solutions or use a secondary AI package to challenge the designs and ensure they are feasible (as Chat GPT already does). We can already see AI tools being used for software programming based on the clear text definition of the task, the proposed machine design and any constraints and frameworks, such as safety.
Realising the potential of AI
As with the introduction of any new technology or way of working, there are certainly risks and challenges to be expected. For example, there is the question of IP ownership which needs to be addressed. In-built bias also needs to be closely monitored and we need to ensure that logic errors aren’t built in, reinforced, or repeated as AI solutions become increasingly convincing. Staying up to date with the latest AI developments is essential if we are to realise its potential and limits. Only then can we work with our customers, enabling them to gain confidence in the full benefits of the technology. It’s a new era for AI that we are positively embracing at Festo and look forward to our customers reaping the rewards.