Artificial intelligence is a cornerstone of the current transformation of industry, often referred to as Industry 4.0. Initial moves in this space involved collecting and collating data from plants and processes. The focus has turned to the industrial internet of things as extra instrumentation and communication systems are put in place, and traditional data-gathering systems such as historians – which collect all the real-time data produced at a plant – are combined with these new streams of data. Historians can also be populated with more accurate machine-generated data, such as from computer vision, for better process integration.
Given the volume and speed at which this data can be generated in an industrial process, it naturally falls to computing power to process it. AI techniques such as machine learning models are trained on these data sources and can report trends and outliers that may not have been apparent previously. To be effective, these AI models, once trained, need to be run as close to the shop floor as possible to reduce latency and data transmission costs. It is here that engineers gain the benefit of IIoT real-time infrastructure and AI modeling – a rich toolset they can engage with, asking questions and exploring options based on their expertise and direct knowledge of the plant.
AI in industrial use cases is not a single decision-making entity but a series of highly efficient tools embedded at the most appropriate points; for example, some ML models may be inside machinery to keep a process on track, only requiring human intervention at the extremes of tolerance – in the same way a mechanical thermostat, once set, will control heat. Other AI processes can sit inside simulation and modeling tools, with no-code/low-code interfaces allowing engineers to use their expertise in the industrial process to manipulate and query the models in ways that are adjusted to their workflow; they don’t require an IT function to create applications for them. With the live data provided by IIoT, engineers can make decisions based on the actual state of operations rather than relying on a deterministic model and spreadsheet calculations.
At a higher level, the use cases driving adoption are currently led by quality assurance, according to 451 Research’s Voice of the Enterprise: AI & Machine Learning Use Cases 2021 survey, driven in many cases by computer vision. Newer use cases expected to be prominent in two years’ time include assembly line creation and optimization as well as employee safety.
While the adoption of an AI-powered piece in a plant may have little impact on the engineers – simply being a natural upgrade in performance – the application of AI tooling will not be successful unless the people asked to use those tools trust and understand them. Because AI models require the knowledge and skills of frontline workers to continue to learn and improve, it is important that AI-powered recommendations can be explained, examined and challenged. It is just as important for workers who are experts in the specific industrial process to understand the results and to have the ability to communicate any adjustments to the learning process as the initial data training from the historian data and IIoT processes.
The goal should be to drive toward a safer, more efficient and flexible manufacturing plant where the current workforce’s abilities and experience are enhanced and supported by a range of AI apprentices and assistants. In turn, these AI tools are improved by, and hence amplify, the skills of the engineering workforce in a virtuous loop.