6 minute read

Find out about the benefits of AI for engineers as well as the barriers to its adoption in the industrial environment

ARTIFICIAL INTELLIGENCE

Getting on board with AI technology

Advertisement

It is becoming a reality that technologies such as artificial intelligence (AI) are starting to change the traditional role of the control engineer. Suzanne Gill finds out more about the benefits it can offer engineers as well as the barriers to its adoption in the industrial environment.

According to Jos Martin, Automation EMEA at Mitsubishi Improving efficiencies senior engineering manager Electric Europe, agrees that AI will AI algorithms are starting to improve at MathWorks, the biggest affect the control engineering role. He the efficiency of the entire factory impact of AI on control said: “Control engineers will need to production line, reducing energy engineering in the coming years will change their daily task list. Their role consumption and waste, enabling be on the workers themselves. He said: will start to include much more data organisations to meet important “As demand for data science skills analysis activities. When users start corporate social responsibility targets grows and the tech skills gap widens, to implement more self-learning and as well as deliver cost-savings. everyday engineers and scientists, as self-optimising technology in processes Traditionally, to achieve good AI well as data scientists, will be expected a big part of the control engineering accuracy levels and easy training of to fill the gap, undergoing training objectives will change and this will models, the use of high-performance on how to design and deploy machine mean that engineering skillsets will computing systems such as GPUs, learning systems to become ‘citizen also need to change. I believe that the clusters and data centres that use 32-bit data scientists’. job profile will become more aligned floating-point math, have been vital.

To be able to make the most of AI in with that software engineering and However, developments in software their work, engineering professionals data engineering. tools now mean that AI inference will need to possess skills such as the “In around 10-15 years it is very models, which use a range of fixedability to deal with large datasets, likely that process optimisation point math, can enable engineers to and to build and train AI models will be handled entirely by AI capitalise on devices such as electronic and understand how to use new technologies and the ability to control units and other embedded development tools and software. programme PLCs will become much industrial applications that run on Companies need to support their less important. Even today we are lower power. workers to upskill and must be willing seeing PLC programs being generated AI is helping to improve the accuracy to invest in adequate training to make automatically by higher level systems of predictive maintenance applications this a reality.” in the simulation space and then – such as those for predicting the

Hartmut Pütz, president Factory downloaded into the PLC.” remaining useful life for an industrial site pump. However, one of the biggest barriers to its adoption in the industrial space is having enough high-quality data to properly train AI models. “Lots of failure data is needed to ensure the AI model is accurate, but it is expensive and inefficient to create data from real, physical equipment. Fortunately, improvements in software now make it easier to recreate data from critical failure conditions and anomalies by generating simulations representing failure behaviour and synthesising it to train a model,” said Martin. “We are seeing AI being used to transform design in everything from

ARTIFICIAL INTELLIGENCE

industrial plants to wind turbines to autonomous vehicles to aircraft,” he continued. “However, another barrier to adoption of AI for smart design is the complexity of multi-domain, AI-driven systems. To get around this, engineers are turning to model-based design tools that provide an end-to-end workflow to reduce complexity. These tools can simulate, integrate and continuously test systems, allowing designers to trial ideas in complete context, identify weaknesses in the data and spot flaws in component design before they become a problem.”

Reinforcement learning (RL) – a form of AI famous for beating human players in chess and Go – is also now being employed to improve engineering design. It works by learning to perform a task through repeated trial-anderror interactions within a dynamic environment. Martin predicts that very soon engineers will deploy RL agents into AI models to optimise performance, for example improving response times in an autonomous driving system.

Where and how?

An important question facing industry today is where and how to leverage AI and the data that drives it, to capture as much value as possible. Andrew McCloskey, chief technology officer, EVP of R&D at AVEVA, believes that this offers a huge opportunity for modern control engineers as when properly implemented AI will make them more effective than ever before, enabling them to implement huge savings for their companies. “AIenhanced predictive maintenance of industrial equipment can generate a 10% reduction in annual maintenance costs, up to 20% reductions in downtime and a 25% reduction in inspection costs, said McCloskey.

Predictive maintenance will leverage both supervised and unsupervised learning – the two primary methods of machine learning that essentially describe the ‘training’ required for artificial intelligence algorithms to ‘get smart’ and provide these savings. Supervised learning enables knowledge transfer from the control engineer in a very short time while unsupervised learning is able to automatically recognise disparities in data that may have significant consequences if left unchecked. An all-in-one solution that is said to make it easy for users to get started with AI-based image processing is now available from IDS in the form of the NXT ocean. Users only need their own application expertise and sample images to create a neural network.

With the help of the IDS NXT lighthouse cloud software, it is easy to train an AI classifier with image data. Because it is a web application, all functions and the infrastructure for creating the neural network are immediately available. There is no need to set up a development environment first.

The process involves three basic steps – upload sample images, label the images and then to start the fully automatic training. The generated network can then be executed directly onto IDS NXT industrial cameras, turning them into inference cameras which are able to apply the ‘knowledge’ acquired through deep learning to new data. This makes it possible to automatically solve tasks that would either not be possible with rule-based image processing, or would require great effort. Together, these algorithms develop high probability predictions that often are not intuitive or otherwise easily identified. “This frees up more time for the control engineer to take on even bigger challenges and drive a flow of continuous improvement, not just a singular event of improvement,” said McCloskey.

“For example, equipped with predictions of impending failures, it is no longer necessary to perform inspections and maintenance based on a pre-determined time schedule. Instead, we maximise the lifespan of equipment parts and replace them as and when necessary – in this case, just before the impending problem occurs. With tight capital and operating budgets, manufacturers are looking to ‘sweat’ existing assets, and this predictive maintenance approach translates into significant savings in inspection and maintenance costs while keeping unplanned downtime to a minimum.”

Conclusion

There can be no denying that the technology is making waves in the control engineering sector but we have learned that AI is not a magic bullet. While there are still barriers to adoption of the technology, it is vital that industry starts to engage with AI as the benefits are too great

Creating inference cameras without AI expertise

to ignore. !

This article is from: