Intelligent Control Systems

Until recently, I would’ve bet that PLC and DCS logic would be the last frontier untouched by LLMs. Now, I’m not so sure. A new research paper from ABB, “Spec2Control: Automating PLC/DCS Control-Logic Engineering from Natural Language Requirements with LLMs – A Multi-Plant Evaluation”, takes a major step forward. The authors demonstrate how Large Language Models can generate IEC 61131-3 compliant control code directly from natural-language specifications. Things like: “Open valve V-203 if tank level > 80% and pump P-401 is off.” Across four industrial plants, the system achieved: → 86–91% first-pass functional accuracy → 55% reduction in engineering hours for repetitive logic → 40% faster acceptance testing with human validation in the loop The models didn’t just translate text. They reasoned about control logic, detected missing conditions, and flagged unsafe interlocks. Spec2Control hints at a future where engineers design through intent, not syntax. Where control narratives, standards, and logic are part of a single intelligent workflow. And where “AI-assisted control engineering” becomes a practical reality, not a conference buzzword. The question isn’t if this will reshape control engineering, but how soon it will become standard practice. What do you think? Will AI-generated control logic become trusted across regulated industries like chemicals and energy, or will safety and accountability concerns keep it on the sidelines? #industry40 #ai #manufacturing #automation #plc

As the demand for smarter, more connected systems continues to rise, PLCs are evolving beyond their traditional boundaries. What was once considered a rigid, low-level controller is now starting to behave more like a modern computer—bridging the gap between industrial automation and full-stack development. I experienced this first hand recently as I had a project where I needed to pull data from a third party system. The catch? The data was only accessible via a REST API. Instead of routing everything through a middleware PC, I implemented an HTTP GET request directly from the PLC. The response came back in JSON format, which I parsed on the controller to populate target parameters in real time—no external hardware or conversion layer needed. Today’s PLCs are capable of much more than deterministic scan cycles and I/O control. A lot of PLCs are adopting items we see in a regular software development setting: - HTTP requests can now be sent and received directly from many brands of controllers - JSON parsing is becoming supported across several PLC platforms - RESTful APIs can be integrated to communicate with cloud services or MES/ERP systems through PLCs - Secure communication over protocols like MQTT and OPC UA is becoming more common - File handling, string manipulation, and even structured object handling are part of the toolbox - Some platforms support object-oriented programming and event-driven architectures Why does this matter? Because the modern factory is no longer isolated—it’s part of a broader ecosystem. Smart manufacturing, Industry 4.0, and IIoT demand seamless data flow between machines, systems, and people. As system engineers, we’re entering an exciting time where the roles of industrial control and software development are blending. This shift opens up new possibilities, but it also means we must continue expanding our skill sets beyond traditional methods of PLC programming. P.S. the controller I used for those HTTP requests mentioned earlier was an AutomationDirect BRX Model PLC. #IndustrialAutomation #PLCs #IIoT #Industry40 #AutomationEngineering #SmartManufacturing #PLCProgramming #OTmeetsIT #ControlSystems #JSON #APIs #EdgeComputing

Day 45/365 23 years ago while in medical school, and passionate about decoding the mind and brain, my curiosity led me to study neuroscience and later control engineering and computer science and I was fortunate to become a student of the Late Prof. Caro Lucas. He was one of the gurus of control engineering globally, a student of Prof. Zadeh, the father of fuzzy systems. After a couple of semesters, Caro asked me who are you, and I told him my story we became friends and eventually started working on intelligent control systems, specifically bio-inspired control systems, combining neuroscience and AI. With our unique approach and inspired by the emotional learning process in the mammalian brain, we invented the first controllers that made decisions based on emotions and named it BELBIC (stands for Brain Emotional Learning Based Intelligent Controller). Based on the computational model inspired by the neural structure and function of the amygdala and orbitofrontal cortex in the human brain, BELBIC is designed to process sensory inputs and emotional signals to generate appropriate control outputs. BELBIC has shown improved performance compared to traditional control methods, especially in handling uncertainties and disturbances. It's also way faster and computationally more efficient and is particularly effective in dealing with complex, non-linear systems and environments with high levels of uncertainty or disturbance. BELBIC can adapt to changing conditions and learn from past experiences, making it more flexible than many traditional control systems. It can process multiple inputs simultaneously, including both sensory and emotional signals, leading to more sophisticated decision-making. For the last 20 years, BELBIC has been applied in various fields, including robotics, industrial control systems, and autonomous vehicles. Here are some major categories among more than 400 applications: Industrial Control Systems Robotics Automotive Industry Power Systems Aerospace Consumer Electronics Medical Devices Financial Systems Environmental Control Transportation BELBIC's development and widespread application illustrate the rapid progress and far-reaching impact of AI technologies. As an early example of integrating emotional learning into control systems, BELBIC represents a stepping stone towards more sophisticated AI that can process and respond to complex, multifaceted inputs. This trajectory points toward future bio-inspired AI systems with increasingly human-like decision-making capabilities, potentially leading to advancements in areas such as natural language processing, adaptive learning systems, and even components of artificial general intelligence (AGI) and something more powerful, Artificial general emotional intelligence (AGEI). To be continued on day 46/365

When models are wrong or dynamics change, most controllers struggle — we need controllers that learn online 🧠. We propose a certainty-equivalent MPC scheme that adapts to online system changes and provides strong robustness and performance guarantees ✅. 📄 Preprint: https://lnkd.in/ervSdn2n 💻 Code: https://lnkd.in/e6aFCDjX We combine a certainty-equivalent MPC with a least-mean-square (LMS) adaptation. The proposed approach relies on a novel MPC formulation that is easy to implement and provides strong stability and robustness guarantees. Theoretical guarantees provide suitable bounds on the tracking error with respect to output references and on state-constraint violations, given (unknown) time-varying parameters, measurement noise, process noise, and nonlinear dynamics. The figure below shows how the adaptive MPC navigates through obstacles while self-correcting the dynamics, whereas an approach without adaptation simply becomes unstable. I want to thank my former colleagues at ICS - Intelligent Control Systems, in particular Melanie Zeilinger, for fruitful discussions during my time at ETH Zürich. Imperial College London; Imperial Mechanical Engineering #MPC #AdaptiveControl #LearningBasedControl #Robotics #AutonomousSystems

I'm very pleased to share with you my project to modeling and control the Li-ion battery charging using Adaptive Neuro Fuzzy Inference Controller (ANFIS) developed for DC-DC charger. By incorporating inputs and desired output data pairs in the ANFIS toolbox of MATLAB software, a successful prediction model can be constructed with the lowest possible error margin, this technique is adapted for the control of Battery charging profile as an intelligent control technique. Approximate knowledge reasoning and uncertainties could be modelled by fuzzy logic, but it lacks learning rules whereas the neural network has learning capabilities which strengthen the adaptive learning rules. On the other hand, the neural network lacks representation of knowledge as compared to FL. Neuro Fuzzy systems (NFS) combines the main features of both NN and FL, ANFIS is the combination of the two soft-computing methods: ANN and FIS. ANFIS uses the NN learning algorithm to generate a Tagaki-Sugeno type Fuzzy Inference System that approaches a nonlinear system with a variety of linear systems. Fuzzy rules and Membership functions (MFs) are obtained by training the system using experimental data sets. In order to determine the parameters of the adaptive system, back propagation or hybrid learning methods are used in the learning process, however MATLAB / Simulink with its comprehensive and powerful control library. This combination makes ANFIS a robust and effective technique. At the same time, the adaptive capability of ANFIS is increased by a trial-and-error process where expert knowledge is not mandatory. In a fuzzy system, rules are generated with human knowledge and manually, where ANFIS generates sufficient rules with the reference of input and output data considering the benefits of ANN. However, ANFIS chooses the finest combination between these criteria to get the maximum output with a minimum error during training operation, It is generally used for complex and nonlinear systems in various fields. Used ANFIS approach to control with large uncertainties and highly nonlinearity, designed an ANFIS based energy management system. ANFIS is utilized as a modern controller by researchers because of its enormous advantages, and it has demonstrated its ingenious performance in different sectors. ANFIS controller outperforms the traditional controller with respect to time efficiency and optimization of membership functions (MFs). The use of ANFIS is broader in modern control systems. However, ANFIS makes the system easier in terms of parameter choice and MF optimization and is time efficient as ANFIS utilizes training data rather than human expert knowledge. Literature shows that ANFIS controllers have been employed to control systems with battery-integrated renewable energy resources better. One study showed that effective battery power management is possible with an ANFIS controller.

Engineering automation is no longer theory. And I am shocked! Last week I posted about the nonlinear control of a subsystem from the machine in my PhD thesis (3D Servo Press from Institut für Produktionstechnik und Umformmaschinen (PtU)). Time required: 45 minutes This week I automated the control of the entire machine. Time required: 3 hours. From scratch! Honestly, I’m shocked by how much engineering work is already automatable. For context: the control of the 3D servo press is not simple. We are talking about an overactuated system with 5 decoupled drives and 3 degrees of freedom. This requires model-based control - Kinematic modeling of the press gear - Splitting the control problem into subproblems - Computing Jacobian matrices between drives and output - Designing nonlinear robotics control laws A lot of complexity. The spindle drives q control the dead centers t The eccentric drives phi control the 3D ram pose x If you don’t understand the control structure at first glance: no problem! I didn’t either when I first worked on it years ago (Sorry for all the dumb questions back then, Florian!) Here is the interesting part: Based on the equations describing the press model and the control concept, Claude Code rebuilt the entire control system in MathWorks Simulink from scratch. It: - created Matlab functions for each subsystem - generated Simulink blocks - inserted the Matlab functions into the blocks - connected everything into a working control system Result: - Two control blocks - Five plant blocks - Nonlinear MIMO control - Fully generated without a single manual click in Simulink! Three years ago, this work resulted in a paper in one of the most respected control engineering journals (Journal of Process Control from International Federation of Automatic Control). Today, with AI, this is basically a weekend side project. And that leads to an important conclusion: Engineers with deep domain knowledge who use AI to accelerate their work will be indispensable in future companies. Not because AI replaces engineers. But because AI multiplies engineers. Get rid of operational work, more system thinking! Many people I talk to still believe AI will not have a major impact on engineering. Often because they tried free models once and concluded it’s just a toy. "Too prone to errors, too simplistic!" But if you have not used state-of-the-art agentic coding tools, connected via well-designed MCP servers to engineering tools and combined with high-performance reasoning models and skill.md files, then your mental model of what AI can do in engineering is simply outdated! We are moving towards a world where engineers describe systems and machines build it. And this will fundamentally change engineering work. The real question is no longer whether AI will change engineering. The question is: Who will be the engineers that learn to work with it first? Viktor Arne | Vlad Larichev | Atilla Akdere | Rick Bouter | Dr. Pascalis Trentsios