Building Operational Intelligence

Summary

Typical industrial automation methods in the digital age excel at monitoring field conditions and executing prescribed logic to perform real-time control. Human operators are still involved to some degree in most automation implementations, and they are integral decision makers in many cases. But maximum benefit—in terms of performance, efficiency, and quality—is achieved when the automation system effectively informs operators, who in turn make decisions to improve performance.
 
Gartner defines this type of “decision intelligence” as “a practical domain framing a wide range of decision-making techniques bringing multiple traditional and advanced disciplines together to design, model, align, execute, monitor and tune decision models and processes.” (Reference 1).
 
Unfortunately, most automation systems in service today are limited with regards to decision intelligence. However, by combining aspects of modern artificial intelligence (AI), industrial internet of things (IIoT) technologies, and human intelligence, a digital decision-making framework is enabled. Using innovative hardware/software architectures and this approach makes it possible to build true decision intelligence into manufacturing and processing applications of all types, leading to operational intelligence, and creating an artificial intelligence of things (AIoT).
 

Human and machine

Humans can be quite good at making good decisions based on limited or superficial data, even during rapidly changing circumstances. However, they are not usually able to process or act on large datasets without assistance. And unfortunately, humans can also make poor decisions based on biases and other improper determinations.
 
Computing systems can rigorously follow rules and can be used to analyze massive datasets to expose patterns leading to insights. But while powerful, most traditional computing systems used for automation systems simply do as they are told.
 
What about artificial intelligence (AI)? Some would call typical logic-solving digital systems a form of basic rule-based AI. Developing the rules—also known as decision logic, algorithms, and procedures—requires engineers and operations experts to define the operating requirements so programmers can create corresponding code. Unfortunately, the resulting rules typically become frozen in time during commissioning. They are largely static, requiring specialized personnel to improve or expand them.
 

More effective AI

Modern AI is far more capable, compared with basic AI using only logic-solving systems. Certainly, logic-solving and decision trees are still a part of modern AI. However, instead of only being designed by humans, modern AI rules are dynamic, developed by machine learning (ML) which discovers patterns and rules within massive datasets to support reinforcement learning.
 
While modern AI is applicable to most any type of computing, that does not mean it is a plug-and-play solution. And for traditional industrial automation industries, modern AI has been applied sparingly and slowly up to this point.
 
Remembering that both humans and machines maintain certain strengths—and weaknesses—begs the question of how to apply modern AI most effectively for industrial automation applications. The answer is found by creating AI solutions which include tools enabling operational experts to shape decision logic and deploy it into production. This type of operational intelligence enables production systems of all types to operate efficiently and constantly adapt.
 

Implementing operational intelligence

A more advanced architecture is necessary to preserve local real-time control and integrate it with higher-level computing. This architecture enables the merging of traditional hard sciences and human experience with AI and ML, executing in the field and in the cloud, to deliver AIoT (Figure 1).
 

Growing an AIoT application

One proof of concept has been for greenhouse control in the agricultural industry, but the ideas are applicable to many other types of control.
 
Optimal greenhouse performance is measured in terms of increased produced quantity and quality, while minimizing consumption of energy and other resources. Greenhouses represent an interesting test bed involving significant physics and science. There are many intertwined variables and control functions, such as lighting, heating/cooling, humidity, irrigation, and others. Even plant health can be assessed, and some systems provide plant spacing adjustments.
 
Local μcontrollers—typically in the form of industrialized Arduinos or Raspberry Pis, but also traditional PLCs—provide real-time control, hardwired I/O and fieldbus connectivity, and robust form factors.
 
The μcontrollers transmit field sensor data to the macro-control AI program. Human farmers also supply crop data to the AI, and the AI can use sources like weather forecasts. ML models may also include advanced real-time information about plant weight, transpiration rate, photosynthetic rate, and more.  In turn, after detailed analysis, the AI provides setpoints and optimal strategies to the μcontrollers, as well as crop handling instructions to the farmers (Figure 2).

Power to the operators

Another example of applying AIoT to industry is at a large biomass production facility in Vietnam. The system was originally designed and built with traditional PLC/HMI/SCADA technology, and provided baseline functionality. However, the facility owners were looking for a way to improve operations without getting into complex and expensive engineering cycles.
 
To start, they tapped the PLCs to provide source data to an edge-located computer running AIoT software. This data, supplied by the field sensors and other operational parameters, enabled operators to better visualize the process operation and perform analyses, sometimes with the help of cloud-based methods. Therefore, the operators who work with the equipment every day in a hands-on manner are able to consider, implement and update improved control logic and algorithms for optimizing operations.
 
One of the tools they use is an Excel-like language which is easier for operations personnel to work with compared to traditional industrial languages like ladder logic (Figure 3). This low-code language is very expressive, and enables to develop strategies ranging from simple to sophisticated. They can leverage all types of signals and information, even the results of other blackbox ML learning models.

Digital decision benefits

AIoT concepts are disrupting traditional industrial automation practices by forging a better partnership between human intelligence and AI. Classic digital decision-making techniques are well understood and can perform reliably, but they are relatively static and require specialized personnel to create, maintain, and update. Modern AIoT techniques are often the best method to provide complete operational intelligence by merging the best capabilities of humans and technology.
 
_{All figures courtesy of Koidra}
 
Reference:

1. Gartner, Decision Intelligence