Although industrial factories and processing plants have long been automated, it remains vital for human decision-making to be involved in operations, sometimes to a great extent. Automation in and of itself is very effective, but it can deliver the best performance, efficiency, and quality when it is coordinated to inform operators so they can make decisions, and even impact the logical control.
Much of today’s industrial automation is carried out by traditional controllers with mostly fixed programming rules. Operators can start and stop systems and change setpoints, but the automation systems are largely static and inflexible once commissioned. This is unfortunate, because conditions change over time, and given the right information, hands-on operators can create improved insights and make decisions, which they would like to apply to the automation platform.
To nurture this type of improved “operational intelligence,” it is now practical to apply modern artificial intelligence (AI) techniques, in conjunction with the industrial internet of things (IIoT) technologies, to upgrade the capabilities of traditional industrial automation. With a digital decision-making framework built from innovative hardware and software architectures, automation platforms can take advantage of user intelligence, traditional hard sciences, and modern technologies to create artificial intelligence of things (AIoT) systems (Figure 1).
AI and Humans Together
Humans are good at making decisions based on limited data, while computing systems can analyze massive datasets and meticulously execute rules. However, humans are subject to biases, and traditional computing can only do what it is told.
Modern AI systems are a form of computing that dynamically updates its rules based on machine learning (ML) techniques, building on massive datasets. Therefore, AI is much more capable than standard logic-solving systems, and it can also evolve. But in many cases, especially those involving the control of real-world industrial equipment, a pure AI solution would not be ideal.
The answer is to implement AI solutions that provide tools and analytics to help operational experts shape AI rules and deploy the resulting information.
An industrial AIoT system must include local micro-control (μcontrol) and higher-level supervisory macro-control.
Installed in the field near the sensors and actuators, μcontrollers can be traditional programmable logic controllers (PLCs), or newer options based on Arduinos, Raspberry Pi, or other hardware. The μcontrollers are necessary to provide near-real-time data gathering and control.
Because μcontrollers lack the processing power needed for AI, it is necessary to have an on-premises or cloud-based computing solution running the macro-control functions. These functions include analytics, advanced visualization for the operators, and the development of AI algorithms. The resulting setpoints and AI rules are transmitted to the lower level μcontrollers to execute the functionality in the field.
Traditional automation methods are capable, using PLCs in conjunction with human-machine interfaces and supervisory control and data acquisition (SCADA) systems. But they can only help users implement a certain level of operational intelligence.
True industrial AIoT functionality is how end-users, like a biomass production facility in Vietnam, can continuously improve operations using their in-house experts, without undertaking costly and complex engineering projects. The site was originally automated in a basic fashion using PLCs, but the operations team had many ideas for improving production.
The team used an AIoT server to gather data from existing PLCs and external data sources (some originating in the cloud). One of the tools they used with this data to improve logic control is an accessible Excel-like language (Figure 2), which enabled them to perform analysis and computations on the data, and also visualize the results and the process operation. With these capabilities at hand, they could adjust setpoints and even implement more advanced strategies for optimizing operation, and examine and refine the results.
Web- and mobile-accessible data dashboards let users interact with the system, and the AI can detect anomalies, incorporate a variety of external data, and help the users migrate advanced functionality from the PLCs to the macro-controllers for more intelligent operation.
The concept of creating AIoT systems built on μcontrollers and macro-controllers is disrupting the traditional industrial automation model by providing a more advantageous partnership of human intelligence and AI to deliver operational intelligence.
Ketut Putra is an IoT and Automation Engineer at Koidra Inc (Seattle, WA). For more information, visit here .