Podcast Episode 47 min
The Forgotten Foundation: PID control and Process Automation with Prof. Margret Bauer
We talk a lot about data platforms, AI, and digital twins on this blog. But all of that sits on top of a process automation layer that most people take for granted — and that layer is quietly eroding.
In This Episode
- Margret Bauer's PhD (2002-2005) developed mathematical tools to diagnose the root cause of plant-wide oscillations in chemical processes; the resulting code was purchased by ABB and incorporated into a commercial software product.
- Margret's central, underappreciated thesis: nearly all industrial processes, and even the vast majority of rockets and robots, run on PID (proportional-integral-derivative) control, not the flashier techniques taught in most university control-engineering courses, and PID is far harder to implement well than its reputation suggests, mostly due to real physical actuator (valve) limitations.
- Margret describes a genuinely "missing layer" almost nobody documents: how controllers actually connect to each other and to sensors, distinct from and not shown on P&ID diagrams, which themselves are rarely accurate after decades of undocumented plant modifications.
- Many companies eliminated their dedicated process-control departments in the 1990s and early 2000s once automation worked reliably, and academia largely stopped teaching process control as its own subject, leaving today's brownfield plants running on expertise nobody is actively replenishing.
- Margret's diagnosis of why process automation feels genuinely stuck: vendors lack the process knowledge needed to improve systems, users no longer maintain the internal experts who'd hold that knowledge, and system integrators in between have no incentive to change anything either, since they're paid for their time.
- A master's student's reinforcement-learning project improved simulated penicillin batch-process throughput by 25%, using the well-known academic "Tennessee Eastman Problem" reference model, a genuinely impressive result that nonetheless remains confined to simulation and isn't scalable to real plants.
Meet Margret Bauer: An Electrical Engineer Who Fell in Love With Control
00:00 – 02:06Margret Bauer is an electrical engineer by training, not a chemical engineer, who studied in Germany and completed a PhD, officially titled "data analytics" but really process-data analysis, between 2002 and 2005, working with early time-series data historians like PI and IP21.
Her PhD focused on plant-wide disturbances: in theory, a chemical plant's key variables should stay smooth and constant, but in practice, everything oscillates and fluctuates, sometimes intentionally, often bothering operators, and because everything in a process plant is physically connected, one oscillating disturbance can show up across many measurements at once. Margret developed mathematical tools to trace such oscillations back to their root cause, work she considers genuinely successful: the resulting code was purchased by ABB and incorporated directly into a commercial software product.
“In my PhD, I tried to find the root cause of such an oscillation. And I developed some mathematical tools and I felt it was super successful because the code was bought by a company called ABB.”
Prof. Margret Bauer · 02:06
From BP Chemicals to Eastman: Applying the Code, Then Joining ABB
02:06 – 06:13Margret applied her oscillation-diagnosis code at BP Chemicals in the UK, where she did her PhD under supervisor Nina Thornhill, later the ABB Professor of Process Automation at Imperial College and now succeeded there by Mehmet Mercangöz. She also applied it at Eastman Chemical Company in Tennessee, home to a legendary Advanced Controls Group led by Jim Downs, a chemical engineer she worked alongside for a stretch of his decades-long career there.
From there, moving into ABB's Corporate Research felt like a natural next step, driven above all by a genuine love of control engineering itself.
“I applied my code to chemical companies, to BP Chemicals in the UK... to Eastman Chemical Company... I was in this group for a while and applied the codes there. So it just felt natural to then move on to work for ABB.”
Prof. Margret Bauer · 04:52
IT/OT Integration Before It Had a Name: Connecting SAP to ABB's 800xA in 2007
06:13 – 08:36Margret expected to do process control work at ABB, but instead found herself doing early IT/OT integration, connecting SAP to ABB's own 800xA automation system, around 2007, genuinely cutting-edge work at the time. Installing SAP alone reportedly took three days, and 800xA needed a dedicated computer of its own.
Working with a preconfigured "paper plant" test environment, her team successfully got data flowing from 800xA to SAP and back, and developed early asset-management use case scenarios that this kind of connection could enable.
“I thought I'm going to do process control. But I didn't, I did IT/OT integration. So I tried to connect SAP with ABB's 800XA automation system... this was cutting edge, like trying to get the two to speak.”
Prof. Margret Bauer · 06:13
The Missing Layer Nobody Talks About: How Controllers Actually Connect
08:36 – 11:40Margret is candid that she found this work somewhat less interesting, and that her real underlying concern was different: coming from academia into corporate research, she felt she didn't actually understand the process underneath the data she was analyzing. Her PhD supervisor suggested taking chemical engineering courses; ambitious in her 20s and determined to finish her PhD in a record two and a half years, she declined, a decision she now regards as a mistake.
That gap in process knowledge shaped her entire career direction. She'd study P&IDs (piping and instrumentation diagrams) to understand how a plant was connected well enough to diagnose plant-wide disturbances, but even that proved insufficient, because P&IDs don't show how sensors are actually connected through the control layer itself, what she now calls the missing layer that almost nobody ever talks about, poorly documented if at all, often buried in extra code somewhere inside the automation system.
“I felt I didn't know enough about the underlying process... this is what I think is the missing layer. No one ever, ever talks about it. And this is what I call the process control layer.”
Prof. Margret Bauer · 08:36
Why P&IDs Are Never Quite Accurate, and Why Academics Get Better Access Than Vendors
11:40 – 13:13Margret explains P&IDs concretely: every reactor, distillation column, and tank gets represented, connected by the pipes that give the diagram its name, all in a standardized, readable format process engineers use when designing a plant. It's genuinely useful, but never an exact, real-time representation.
She also points out a real irony in access: as an academic, and especially as "just" a PhD student, she got far more access to P&IDs than ABB itself ever could as a vendor, since companies were noticeably more guarded sharing proprietary documents with a supplier than with a researcher. But even that access revealed the documents were often inaccurate: plants are typically 20 to 40 years old and have been modified repeatedly over that span, with nobody bothering to keep the P&ID updated to match.
“As a vendor, as ABB, it would be almost impossible to get access to the P&IDs. As an academic, you have more access... very soon you find out that these P&IDs are not accurate. They're not a true representation of the plant.”
Prof. Margret Bauer · 11:40
"The Best Strategy Ever With the Worst Name": Why PID Control Runs the World
13:13 – 16:27Margret's central thesis is that PID (proportional-integral-derivative) control is a brilliant, deeply underappreciated strategy, made worse by a name constantly confused with the completely unrelated P&ID. She explains it intuitively as covering the only three aspects of time relevant to holding a variable steady: proportional, the present measurement; integral, the past; and derivative, the trend heading into the future.
She only recently discovered something she finds genuinely striking: nearly every industrial process, and even the vast majority of rockets and robots, ultimately runs on PID control, not the flashier state-space and Kalman-filter techniques she was formally taught at university, which barely touched PID despite it quietly running the world. PID is also far harder to implement well than people assume, largely because of real physical actuator limits, specifically valves, which can only be fully open, fully closed, or somewhere in between, a far less glamorous constraint than the freely moving robots everyone loves.
“It's the best strategy ever with the worst name and it's called PID control... 99.9% of all the rockets that have flown into space work on PID control. All the robots, the underlying control strategy is PID... It's completely underestimated.”
Prof. Margret Bauer · 13:13
"Big Valve, Little Valve": The Undocumented Control Strategies Nobody Talks About
16:27 – 19:19Margret extends the valve-limitation point with a vivid example: a control signal asking a valve to be 150 percent open is simply nonsensical, the valve has nowhere further to go. That constraint led to genuinely clever workarounds, like a strategy variously called mid-range control or, more memorably, "big valve, little valve": running two parallel flow paths, a large valve handling big flow demands and a small valve deliberately held around 50 percent open so it can react quickly to small disturbances, coordinated together through PID, since a single large valve alone can't deliver fine-grained flow precision.
Her key point is that strategies like this are genuinely widespread across industry, essentially undocumented anywhere formal, and without knowing they exist, it's simply impossible to make real sense of the resulting process data.
“The problem is that we not only have one valve that we often want to interact with, but maybe two valves... this is called... big valve, little valve control strategy... these are the strategies that no one ever talks about.”
Prof. Margret Bauer · 16:27
From ABB Corporate Research to Professor: A Well-Worn Path
19:19 – 22:56Margret always wanted to both teach and do research, and started her professorship in South Africa, where she'd moved during parental leave (her husband is South African), before eventually returning to Germany. She jokes that working at ABB Corporate Research is practically a direct pipeline into becoming a process-automation professor, naming roughly 20 former colleagues now teaching across Germany, Switzerland, Finland, and the UK.
She explains the German academic system's split between research-focused universities and teaching-focused Universities of Applied Sciences, and her deliberate choice of the applied-focused institution in Hamburg over the more prestigious, research-heavy RWTH Aachen, since process automation is fundamentally an applied subject and genuinely difficult to publish in academically. Her actual research runs primarily through master's students placed directly inside companies, many of whom already worked as operators before returning to study, giving her real hands-on process perspective she never had herself, including a memorable student working at a bakery producing roughly 12,000 bread rolls an hour.
“If you want to become a professor of process automation, you go and work for ABB Corporate Research because I have so many peers... it's such a beautiful place to be. So many clever people in one building.”
Prof. Margret Bauer · 19:19
Skepticism of Asset-Naming Hype, and Why Data Analytics Needs Real Process Knowledge
22:56 – 27:42Margret is openly skeptical of the heavy focus in the German automation community on asset-organization and naming standards, arguing it's really about administrative tidying rather than genuinely automating or managing a plant, with essentially no real business case for retrofitting a 20-year-old brownfield plant purely to reorganize how its assets are named. She sees more genuine value in industry-specific standards like MTP (Modular Type Packaging), which makes sense in pharma's genuinely modular plants but doesn't translate to chemical plants, which aren't modular and don't get reconfigured module-to-module day to day.
Her real passion remains data analytics, and she sees real progress in making process data genuinely accessible, pointing to her work with a company called Seek, whose platform her master's students get direct access to as part of coursework, an arrangement she says other academics genuinely envy. Her core teaching philosophy is that data analytics only really makes sense to students who are process engineers first: understanding both the operator's perspective and the underlying process is, in her view, the absolute key to making real sense of the data itself.
“It's all about process knowledge. I think understanding the operator's perspective, understanding the process is the absolute key, particularly if you want to make sense of the data.”
Prof. Margret Bauer · 24:22
Why Companies Fired Their Control Departments, and Why That Was a Mistake
27:42 – 31:44Margret traces the field's historical arc through colleague Tore Hägglund's work at Lund University in Sweden: process control was a genuine hype topic through the 1970s to the 1990s, with exciting new developments like model predictive control (MPC). But once automation worked reliably and got installed, many companies concluded they no longer needed dedicated control departments and eliminated them, often on the advice of management consultants questioning why 20 people needed to sit around watching a smoothly running process.
Exceptions exist, Eastman Chemical still maintains a large control group, and Sasol in South Africa runs the highest number of MPC applications in the world at its Secunda site, but the deeper tragedy, in Margret's view, is that academia largely stopped properly teaching process control as its own subject, so almost no new control engineers are entering the field even as the world's brownfield plants keep running on decades-old expertise nobody is actively replenishing.
“The tragedy is in academia, it's not been taught properly. Process control is not taught as a subject. So there's no new process control engineers coming... we are a handful in the world, really focusing on it.”
Prof. Margret Bauer · 27:42
Autonomy's Real Bottleneck Isn't Technology, It's Keeping Operators Engaged and Retained
31:44 – 35:03Margret connects this directly to today's autonomy hype: a genuinely well-automated plant can become almost unbearably boring for an operator, one of her students described literally reading a book for an entire eight-hour shift with nothing to do until the next shift arrived. Some sites deliberately assign make-work tasks, walking the floor, checking a pump, taking a manual sensor reading, purely to keep operators mentally connected to the process.
Margret doesn't believe full automation will ever be fully achievable, both because unpredictable exceptions can't all be anticipated and because every individual process is genuinely unique. The real challenge, in her view, is retention: finding someone willing to do the operator job at all is relatively easy, but finding someone good who's willing to stay long-term is nearly impossible, especially given real safety and chemical-exposure risks that come with working that close to a process, making shift design, team structure, and genuine people management essential skills the broader, technically minded automation community still hasn't fully absorbed.
“I personally think we'll never be able to fully automate everything... finding someone who's good and is willing to stay is next to impossible... you have to think about how to make that role attractive.”
Prof. Margret Bauer · 31:44
Process Automation Is Stuck: Nobody Has Both the Access and the Incentive to Improve It
35:03 – 36:52Margret's diagnosis of the field's current state is genuinely bleak: real improvement requires deep process knowledge that vendors simply don't have access to, since users no longer maintain the control departments that would normally hold and share that expertise. Users, meanwhile, lack the remaining internal experts needed to drive improvement themselves, and the system integrators sitting in between have no real incentive either, since they're paid for their time and structurally can't unilaterally change a client's system.
She points to NAMUR, the German user association, noting that building up real process-control capability now is especially unrealistic given the broader economic downturn hitting the chemical industry. Her blunt summary: nobody in the current ecosystem, vendor, user, or integrator, has both the access and the genuine incentive needed to move the field forward.
“No one can actually improve automation because the vendors don't have access to the process. Only with process knowledge will you be able to really improve. But the users are not going to give them that process knowledge... So we're kind of completely stuck.”
Prof. Margret Bauer · 35:24
A 25% Throughput Win With Reinforcement Learning, But Only in Simulation
36:52 – 40:08One of Margret's master's students built a genuinely successful reinforcement-learning project optimizing throughput for a simulated penicillin batch-production process, achieving a 25 percent improvement. It worked because the project used a well-understood academic reference model, the "Tennessee Eastman Problem," a famous 1990s paper and open dataset created by her former Eastman colleague Jim Downs, still widely used today with freely available MATLAB and C code for anyone to experiment with.
Her important caveat is that despite the genuinely impressive result, it remains confined entirely to a simulated environment, specific to that one particular process, and isn't scalable or repeatable to real-world brownfield plants.
“He could improve the throughput by 25%, which is magic. It's amazing... But it's all in a simulated environment. And it's for this one particular process and it's just not scalable or repeatable or anything.”
Prof. Margret Bauer · 37:12
Understanding, Not Dismissing, the Frustration With How Slowly the Process Industry Moves
40:08 – 42:18Margret has real sympathy for outside frustration at how slowly the process industry seems to modernize compared to consumer technology, phones, apps, even the ease of recording a podcast, all unimaginably easier than a decade earlier. But she gently pushes back too: the process industry was actually an early leader in interconnected sensor communication, well before the modern internet existed.
She shares her own lived frustration from her ABB days, working on a genuinely compelling "Google search for the plant" project: search for "pump," and instantly retrieve every pump in the plant along with its maintenance manual, historical data, and live vibration-sensor readings, all unified in one place. It worked and was genuinely valuable, but ultimately stalled because nobody was willing to fund the real, ongoing engineering effort required to build and sustain it.
“We came up with this Google search using all the data... you Google search and you find this one pump and you get all the information... you had everything together in one spot. And that'd be so amazing. But it's going to be a lot of work and no one's going to pay for it.”
Prof. Margret Bauer · 40:28
Wallpaper Made of Math: Margret's Ongoing Side Project to Raise Baby Mathematicians
42:18 – 46:32In a delightful, lighthearted closing story, Margret describes wanting her children to grow up mathematically gifted. Her first attempt was playing classical music nonstop to her pregnant belly for nine months, which, amusingly, did produce two musically talented kids.
Her next idea came from an episode of the In Control podcast, hosted by Alberto Padoan, about mathematician Sofia Kovalevskaya, who reportedly first learned mathematics as a child in Russia from literal mathematical papers used as makeshift wallpaper in her family's home, too poor to afford real wallpaper. Inspired, Margret set out to design an actual wallpaper covered in control-engineering equations for a nursery. She's drafted a first version, but friends, an architect and a PhD in French literature, weren't impressed by the aesthetics, so it remains an unfinished, ongoing side project, one she openly invites anyone listening to steal and finish themselves.
“I've got the equations, the control engineering equations together. I started drawing them and then I ran it past my friends who are architects and PhD in French literature, and they didn't like the idea.”
Prof. Margret Bauer · 44:34
Full episode transcript raw feed ▸
Speaker 4 (00:01) Hi David! We're back for another IT/OT Insider. Now before we introduce our next guest David, uh you and I have been busy for the last weeks with our third group on the ITOT Academy. Super cool. Speaker 2 (00:18) And actually I like it even more this time because yeah, this one is with all the feedback we got from the first two groups. So it's even better. It was already really good, obviously. Speaker 4 (00:30) no, I have to I really like to see all the interaction of having people from IT, OT together, but also vendors coming maybe more from an IT background to people with a more OT background and just getting those discussions that in real life don't always happen. ah It's very interesting also for us. Speaker 2 (00:49) So in any case, uh here to announce that our uh fourth group starts in May. uh We have early bird discounts available. So for those interested, go to itot.academy for all information. And this kind of concludes the commercial block in this podcast. Speaker 4 (01:11) We need to make a living as well, so. That's true, that's true. um like I said, we have a new guest today from a very different field than what we're used to, David. ah Here is Margret Bauer. She is a professor in process automation. Now, I think for most people working in IT and digitalization, process automation is probably like something that's been solved already. Why are we even talking about it? Now, I'll let... Margret explained that in the coming 40 minutes. Speaker 1 (01:47) Absolutely. Speaker 4 (01:47) uh Margret. Speaker 1 (01:51) Yeah. Speaker 2 (01:53) Thank you for joining us. let's just, as we always do, introduce yourself. Where do you come from? How does someone become a professor in process automation? And then we can start from there. Speaker 1 (02:06) Yeah. So I actually am an electrical engineer by training. So I wish I were a chemical engineer like Willem, but I'm not. I did electrical engineering. I couldn't use much of my degree actually. It's all very mathematical. And I did this in Germany. And then I did a PhD in something that was called data analytics, but quite a while back. in 2000, started in 2002 and finished in 2005. And I looked at process data at the time. And at that time, it was actually really difficult getting the data from the plant. So I'm talking time series data. I'm talking em pressure, temperatures over time that you would get from your automation system. And we'd get it from the data historian. So often you'd have a pie historian at the time or an IP 21. And then we try to make sense with the data. I particularly looked at plant-wide disturbances. So in a chemical plant or in most process plants, you want everything to be nice and constant. So you want your temperature to be at a certain point and everything's smooth. And that is the theory because practice, if you walk into any site, anywhere I've been to, everything's oscillating and fluctuating. And sometimes it's not a problem. Sometimes it's intended. but very often it's something that bothers the operators. And because in chemical plants and in other process industry plants, everything is connected. Things travel through the plant and one affects the other. And then everything suddenly, if you have a disturbance that's oscillating, all your measurements are oscillating. And in my PhD, I tried to find the root cause of such an oscillation. And um I got some math, I developed some mathematical tools and I felt it was super successful because the code was bought by a company called ABB and they incorporated these codes into um a software product. Speaker 2 (04:12) Very small niche player in the automation. uh Speaker 1 (04:19) them? I suppose the, I mean, this is so funny because we are in this specialist community where we know these companies, but if you talk to anyone in the street, they're no Siemens because they're household appliances, but ABB, unless you live in the south of Germany or in Switzerland, where you have lots of red logos um on the highways, on the side of the highways, you probably haven't heard about ABB. So yeah, so I have to introduce it. Speaker 4 (04:44) in process automation for the people who don't know it. Definitely not a small startup that started in 2000. Speaker 1 (04:52) So they hired, so I applied my em code to chemical companies, to BP Chemicals in the UK. This is where I did my PhD. did it with my professor who's called Nina Thornhill. She then moved on to Imperial College and she was the ABB professor of process automation at Imperial College. She's now retired and there's a fantastic colleague who's taking her position. His name is Mehmet Mekangoz and he's continuing her great work. um So I applied this to BP Chemicals and to Eastman Chemical Company, which sits in Tennessee. It's a fantastic, they've got a fantastic controls group there. They call them, I think the Advanced Controls Group, and a person called Jim Downs. He did this for like 40, 50 years. He's incredibly good um and a super chemical engineer. And I was in this group for a while and applied the codes there. So um it just... natural to then move on to work for ABB with in corporate research. And because I went into this area because I love control. loved control as a control engineering, not control, not like being in control, supposed to, but I loved control. Yes. Speaker 2 (06:13) Sorry, wrong podcast. Speaker 1 (06:16) uh So I loved control engineering, I uh don't know, Willem, you probably had as an undergrad student. Most engineers, think, yeah, most engineering students, mechanical, electrical, um chemical, they have control systems as an undergrad topic. And I loved it. And when I started with ABB and process automation and in uh controls related group, I thought I'm going to do process control. But I didn't, I did ITOT integration. So I tried to connect SAP with ABB's 800XA automation system in, and I did this, this was in 2007 maybe around that time. So we were like, this was cutting edge, like trying to get the two to speak. I had some fantastic, amazing colleagues who helped with that because as an electric engineer and doing this kind of thing, I wasn't too comfortable. Just remember, I think it took three days to install SAP on a computer. then installing 800XA was another mission and we had a special computer due for that. um Speaker 2 (07:32) three years to install and three days to install and three years to configure typically. Speaker 1 (07:36) Yeah. that's fast. Speaker 2 (07:39) Oh, sorry, sorry, sorry. Speaker 1 (07:41) We had this paper plant that was like uh a pre-configured plant which we worked with. And then we managed to send some data from the HMXA to the SAP and back. And we came up with some scenarios where this could be useful, some asset management scenarios. And em yeah. m Speaker 4 (08:01) I don't know why we're even talking to it. mean, clearly the problem has been solved already 20 years ago. I mean, we've connected it. Let's move on. Speaker 2 (08:12) damn, just today I announced that I was going full time for the ITOT insider. Is this fine? Well, yes, do we bother? Yeah, that's interesting. um But then for some reason, I assume Margret that the ITOT side of things that it wasn't was less interesting for you. Speaker 1 (08:36) I felt it was a little bit less interesting and I felt that I didn't, my main concern was that I felt I didn't know about the process industry. So I'd come m from academia and I moved into corporate research and I felt I'm looking at this data, I don't understand the process underneath. And my PhD supervisor, she said, Margret, you should do, um you should take some courses in chemical engineering so you can understand it better. And I said, No, I was in my 20s, so I was super ambitious. I said, I want to finish my PhD in two and a half years. I want to set a record. And I did. Why would I take courses? This is going to slow me down. So I stupid at the time, um not taking the patients and not having the patience and doing these courses. So I felt I didn't know enough about the underlying process um that we worked on. I didn't understand it. And this is the key, actually, because I tried to make sense of the process data. And I looked, obviously I looked at the P&IDs, the piping and instrumentation diagrams that shows you how everything is connected. And this is important when you have a plant-wide disturbance that travels through the process and you have to know how things are connected. But it didn't make all that much sense. And then on top of this, the P&ID doesn't show you how the connection is between the different sensors through the control layer. Yeah, yeah. this is what I think is the missing layer. No one ever, ever talks about it. And this is what I call the process control layer. And the connection of the controllers lives in it. So how is everything connected? And it's poorly documented, if at all. You can see some of it in the P&ID, but often it lives in some code, in some extra code on the automation system. Speaker 2 (10:33) maybe for the, I say our listeners who come more from the IT side, so a P&ID, and I also used many, many of them in my previous jobs, uh is some kind of a, it's a plan, it's a, how do we explain it? It's actually a physical representation of your vessels and your lines and where your sensors are located, but it's more the physical layout, right? Speaker 1 (10:58) Yes, so it's not an exact representation, but you have every key equipment, every reactor, every distillation column, every tank. You will have that represented. And then you will have, this is why it's called P &ID, piping. You have the pipes that connect the equipment represented there. And it's in a standardized format, so people can read it. There are standards that describe it. um So it's pretty much... It's pretty readable and it's what the process engineers look at um when they design the process. m Speaker 4 (11:40) how do I isolate? Yeah. is available. Speaker 1 (11:44) Yeah, it's usually very proprietary. So when I did my investigations, it's like as an academic, you have better access to it because you're not the supplier. So as a vendor, as ABB, it would be almost impossible to get access to the P&IDs. As an academic, you have more access because they're thinking, this is Margret and she's a professor and who cares? She's a PhD student. That's even better. Hahaha. And they would share the P&IDs with me after some, often just in paper form, not in electronic form, um because people are very cagey about their P&IDs. And very soon you find out that these P&IDs are not accurate. They're not a true representation of the plant, because the plants are 20, 30, 40 years old, and they have been changed and altered. And no one's kept up with, no one bothered to implement this in the P&ID. And then on top of this, you have your controller structure. So you have this little controller. don't know how to explain that for the IT, IoT cloud. Speaker 4 (12:52) Well, I assume most of our listeners do have some idea that there's process automation. ah But of course, there's logic in there. If the temperature goes up, close the valve. Stuff like that needs to be programmed somewhere. And you will not find that program description in your P&ID diagram. Speaker 1 (13:13) Yes, you will find some of it. So there's what people don't, I think, fail to often to realize is there's one control strategy. It has a terrible name. It's the best strategy ever with the worst name and it's called PID control. And it's so awful. The name is so terrible also because it's so easy to confuse this with P&ID You think, you know, it's the same letters, only the one has an I. It stands for proportional, integral and derivative. And if you think that we want to keep a variable constant over time, so we always have this time, we've got time series data with time axis. There's only three aspects of time. It's the present, the past and the future. So the P part concerns with the present, because we're looking at the proportional part, and this is proportional to the current measurement. You have the integral part, which looks at the past. And you have the derivative, which looks at the future. shows you in which direction we're heading. So there's only three parts to it. And as I say, it's a magical tool, this control strategy. It's completely neglected. People think it's a solved topic. um But the word runs on PID. I thought it's all, I only found this out in the last year or so, because I always thought it's the process industry that only does P and ID. And I'm sure, All these robot robotics guy and the space exploration guys, et cetera. I'm sure they have more advanced algorithms, but they don't like 99.9 % of all the rockets that have flown into space work on PID control. um All the robots, the underlying control strategy is PID. And it is, people say that how complicated can it be? And it turns out it's actually really complicated and people have neglected it because when I heard control engineering as a course at the university, I was taught about state space and some very fancy mass and yapunof function and Kalman filters, etc. And then was like one half election PID saying, well, by the way, there's also PID and and not explaining. I think I don't know, maybe my professors didn't understand that the world turns around PID. It's run by PID technology runs on PID control and Yeah, it's completely underestimated. It's very difficult to implement, by the way. So the difficulty is that we have physical limitations and that is mostly the actuator. So what we actually do is, and this is why we're not as fancy as robotics. So everyone loves a robot, right? And it moves around. What we have is we've got a valve and the valve will open and the valve will close. And it's just a little... And it opens and closes. It is not fancy to look at. This is how we work. And this is what we do. We open and close valves in the process industry. Speaker 4 (16:10) I've never heard of the job description that way, but it's actually quite close to reality. What do you do all day in a plant? Speaker 2 (16:17) Open and close valves. and our products are used all over the world. Your job is, if you choose to accept, open and close files. Speaker 1 (16:27) Well, this is pretty much the process automation job that we do. Yes, but the point is the all these pumps and all the, yeah, those pumps too, yes, but most of the time we control with valves and. um The point is that these valves are limited. So they have, they can be fully closed or completely open. So if I computer control signal that tell my valve, please be 150 % open, the valve will say, are you nuts? Can't be, I can't do 150%. Speaker 4 (16:56) for deep bottlenecking exercises. don't open the valve, 150 % problem solved. Speaker 2 (17:01) Well, and then you talk about open and closed, but I've already walked through Valve set. So you also have the ones who take minutes to open and minutes to close. Speaker 1 (17:19) That's a great, that takes me to my next point, which is, so the problem is that we not only have one valve that we often want to interact with, but maybe two valves. For example, if you have a very big valve, you cannot control the flow accurately with a big valve. So what people do, a control strategy that's called big valve, little valve. So you have two parallel flows, um one is controlled with a big valve and one with a little valve. And um then you have to find a control strategy that adjusts the flow accurately at this very high volume with the two valves using PID, but finding a strategy that's on top of that. And that's basically a controlling the one gives the set point for the other valve and the one is always supposed to be 50%. So the small valve strategy is to be always 50 % open so it can react to small disturbances and the other valve will take care of the big one because it takes so long to open and to close. And so this is called, oh gosh, it's called mid-range control. And my colleague at a chemical company will call it big valve, little valve control strategy. There's different names for it. But this is one of the many control strategies that people use in industry. And these are the strategies that no one ever talks about. that are not documented and without with, if you don't know about this strategy, you can't make sense of the data. Speaker 2 (18:46) It's interesting because we're here, we're really talking about the low layer control as something which in we in our ITOT world, don't come across that part of the world, but it's extremely important. Now, having said that, you were still not a professor because you were working on your PhD, you were working with several companies. And then at what time does one make the link on I wanna teach and... research on these topics? Speaker 1 (19:19) Yeah, so I always wanted to teach and I always wanted to do research. And I started actually teaching in South Africa because I had my kids and my husband's from South Africa. So I moved there to my parental leave. I actually became a professor there first. um And I am now a professor in Germany, eventually moved back to Germany. And um I jokingly say, if you want to become a professor of process automation, you go and work for ABB Corporate Research because I have so many peers. I can give you a list of, I think, 20 names easily of former colleagues who are now at different universities across Germany and Switzerland and Finland um in the UK. um As I said, the colleague who took over from my supervisor, Mehmet in that period, he used to be at ABB Corporate Research in Switzerland. So we have ABB Corporate Research is a very, very special place, especially if you love research. It's the best place to be if you're working at this intersection and if you have an academic career in mind. It's such a beautiful place to be. So many clever people in one building. And yeah, so this is how you can enter academia. There many paths. I have to say, I don't know if you're familiar with the German academic system. So we have two different types of universities. We have the universities that do research and then we have the teaching colleges. And there's a bit of a hierarchy. When I came back from South Africa, I applied to a couple of positions. I got two and one was at a very big, very research focused university, RWTH Aachen. And one was at a teaching college here in Hamburg and I chose the one in Hamburg. because I think if you do process automation, it's an applied subject. I'm at the University of Applied Sciences, we call this the University of Applied Sciences. And I don't see how I can supervise many PhD students in the subject because it's very difficult to publish in our area. So I basically do my research through master's students who are placed in companies. Many of my students have worked, have... worked before they came to studies, many worked as operators and I learned from these students probably more than they do. Well, I'd like to think they learn more from me, but I do learn from them a lot because I've never worked as an operator. em I don't know what it's like standing there and operating like a metals factory or I have one student who works, who does a bakery. he... Speaker 2 (22:03) we love those. Speaker 1 (22:04) I love that. And it's so glad it's a good bakery because I'm buying that bread. It would have been awful if it had been a bad bakery, but it's really nice bread that he's doing. And he does per hour about 12,000 bread rolls an hour on his own. So it's pretty, pretty awesome. So I learned from my students now, which is great. Speaker 2 (22:27) about PID now a bit. What are other process automation concepts people, or at least our listeners, so the ones who are typically not active in the process automation world, should know about? Speaker 1 (22:45) That's tough one. Speaker 2 (22:50) Yes, we are known for our very difficult and in-depth questions. Speaker 1 (22:56) Yeah, so there's a lot of noise in the, particularly in the German community, automation community about uh different standards on how to arrange assets. And it is not really about, I find this process somewhat problematic because it's not really about managing or automating the plant. It's more about managing the assets. So about uh organizing them, giving them good names, making sure everything's kept up. I never find that very useful when it comes to Brownfield because no one's going to start organizing their assets in a plant that's been there for 20 years. It's just not worth everyone's time. There's no business case. Speaker 2 (23:42) I you're now referring to asset administration shell as an example. Speaker 1 (23:46) Yeah. of standards. I have just supervised a student on an MTP standard, modular type packaging. And that for me makes sense in the pharmaceutical industry. um So that's very industry specific. don't see an application in the chemical industry because our plants aren't modular. You can't, it's a module. I mean, even my... Speaker 4 (24:19) Yes. Speaker 1 (24:22) And you wouldn't really connect one module to another module and today and tomorrow you've connected in a different way. Yeah. So I work a lot in the data analytics. I still, I love data, I have to admit. So I love looking at data. I love making sense of the data. So... Um, the, the, work you're doing is it, getting the, getting, having access to the data, um, is excellent. So I think this is something where we see a lot of progress. I work with a company called seek. I haven't noticed them on the podcast. Yes. Have you, you haven't had anyone from, from seek. Speaker 2 (25:01) We haven't got people from Seek, but they are highly welcome to join us. We obviously, there are really no name in the industry. Speaker 1 (25:07) Yeah, so I've got some super, two super nice colleagues I work with because we have my students have access to the Seek platform as part of their degree. So I do a course on data analytics and there we use the Seek platform and the students can play with it. I think it's very unique because some of my colleagues are jealous that I have my students have access to it and that is don't. um and we do the data analytics, but it makes sense if you're a process engineer. So I teach predominantly process engineers and it makes sense for them. At the end of the day, it's all about process knowledge. I think understanding the operator's perspective, understanding the process is the absolute key, particularly if you want to make sense of the data, I think. Speaker 2 (25:59) But I saw on your LinkedIn feed somewhere a post where you mentioned process control engineers are a dying breed or something like that. Speaker 1 (26:09) Why? So it's because we automated the automation. That's the reason. So basically it was a hype topic. So I also, I was a professor at Lund University in Sweden for three and a half years. And um yeah, I'm still affiliated with them and I work with the colleagues there very closely. They're the best people working process control. And one of my colleagues is called Tura Heglund. this is where I can point out his book. So if you want to know about process control. can't see this. There's a book. I'm sure you can put this in the notes later. Thank you. the show notes. Speaker 4 (26:46) more about process automation. Speaker 1 (26:50) Yes, it's very applied, doesn't have any mathematical equations in there. It is just about talking about process control and why it's important. he's now retired, he's in his, well, he's still active. I shouldn't say he's retired because he's still very busy doing research. He's in his seventies and he's been observing the community for a long time. So he said it was an absolute hype topics in the 1970s, 80s and 90s, even early 90s. So people got very excited about it. Lots of new developments, MPC, model predictive control, another strategy got into the area and it was super exciting. And then what happened was that the process automation worked and people installed it. And then the companies got rid of their automation department because they said, well, you've automated everything. Speaker 3 (27:40) That was a check mark. Speaker 1 (27:42) The one exception is Eastman Chemical Company, where they still to this day have this very large control groups. um In South Africa, there's a company called Sasol, spelled S-A-S-O-L. In German, we say Sa-Sol. And in South Africa, we say Sasol. And um it's a big em chemical uh company. And em they have the highest number of MPC applications in the world in a place called Secunda close to the Kruger Park. And em they have good controls people, but the majority of the companies got rid of their control departments entirely. Some business, some management consultant, like some McKinsey or BCG's or whatever came in and said, well, why you don't need these guys? Because the process runs smoothly. Why do you need 20 people sitting there? So they get rid of these groups in the 1990s and early 2000s. And the tragedy is in academia, it's not been taught properly. Process control is not taught as a subject. So there's no new process control engineers coming. And now these brownfield plants, they're running and um they're running well, but this is an ongoing, people don't understand that it is an ongoing subject. You can't just optimize your plant and improve it. And so as a result, we see very, very few people doing process control. And I can tell you the academics that work on the subjects and we are a handful in the world, really focusing on it. We are an absolute minority and we are very passionate about it, but it's a hard battle. Yeah. Speaker 2 (29:21) it's definitely an extremely, extremely interesting topic. My master thesis was in nonlinear model predictive control. ah So, and the fun thing is then, well, actually I never applied. That's typically in a master thesis, you go a bit to the edge and then you come to the conclusion that let's just take a couple of steps back in reality when you enter and, yeah, not just spreadsheets, but. definitely is a extremely, extremely interesting topic. um But it also has a downside in my opinion on SO obviously you have less people who are interested in these control strategies, although they are indeed extremely important, but you need to understand the physical, the actual process, et cetera, et cetera. There is also actually a second downside on the side of the people who, the operators or technicians who work with the process, because what I've witnessed for example, in our past BASF, was that the more you automate, more stable your process becomes, the less upsets you have. Also, gives us a positive point from an economical perspective, but a negative point from a personal perspective, that these operating technicians, they are less exposed to upsets. So you come to this point where first you have a good PID control and then you have a decent SCADA system, DCS system, then you have a m model predictive control layer on top of that. And then all of a sudden, operators are faced with upsets they don't understand anymore. And then we start talking about operator training simulators, for example. So we're adding another layer on top of that, where we're doing some kind of a flight simulator, but then to train operators, for example, to do a startup or a shutdown or product change or whatever. And it's very interesting. But I don't know how you look at it from an academic perspective, but it's a... there's less variability in the data. it's a very interesting thing, because you also can't let's turn off the model predictive control layer to make our process less stable again. Speaker 1 (31:44) Yes. Yes. I mean, this all ties in with this like the latest hype term, is autonomy. We want autonomous plants. And but it comes to exactly down to this. And when I speak to my students who used to work as operators, they said, well, it's the I'm not going to work as an operator. That's why they come to see us because they want to study further, because it's the most boring job. If you have a if you have a beautifully automated plant. One of my students said she just took a book with. It's like took a book, pressed the button, sat next to the machine and then waited. And then eight hours later, it's like the next shift came over, shifted. And yeah, she read her book. So it's a fine balance. And then I have other students who reported that they get tasks that are... I mean, just to entertain them, just to keep them connected to the process. You you walk around, you have a look, you take a reading from a sensor, you look at a pump and check the pump and you sort of like sort of little bit, yeah, making them work and making them be connected to the process. So it's a fine line. This is what I love about your process because you say all the time it's about people, because it's about people at the end of the day. And I don't think we'll, I personally think we'll never be able to fully automate everything. we'll always have a human there because of these exceptions. And you cannot take everything into account, one. And the other thing is all the processes are so like individuals. Every single process is different and you're not going to automate that. um So we always have a person and we have to think about it very carefully how we're going to train the operators, how to deal with the operators. I think one key is that it's very easy to find someone who's going to work as an operator, but finding someone who's good and is willing to stay is next to impossible. so, yes, so I mean, they go and they move on, they would want to do something different. So you have to think about how to make it attractive, how to make that role attractive, how to challenge people, because it's a very important job. It's, mean, because you're working close, you have to make sure that it's safe. I mean, often it's not because you're working close to chemicals that are not good for you em as a person. And you can be exposed to them if you're working so close to them. So you have to think about all of this. em A lot of them is with people's management, um working in shifts, working in teams, setting up these teams. And people have learned, I think, a little bit, but not enough yet. um about the people aspect and that's part of our community. We are all nerdy tech people. I mean, not us, not the three of us, not the listeners. But there are some nerdy tech people out there. Speaker 2 (34:52) My wife always says David I refuse to see you as a nerd you are a geek that sounds better. said whatever works for you. Speaker 1 (35:03) Well, I have the benefit of being German because Germans don't have a sense of you though in any event. So I keep saying, well, I'm not a nerd. I'm just German. Speaker 4 (35:11) Works No, so basically you have the the automation going so far. It's also redefining the job of an operator actually and redefining what is a good operator if I understand you Yes. Speaker 1 (35:24) And I think this is where we have to work on and then also find a place for developing new technologies because currently I think it's not getting developed. Process automation is stuck. And the reason for this is that no one can actually improve automation because the vendors don't have access to the process. Only with process knowledge will you be able to really improve. But the users... are not going to give them that process knowledge. The users don't have control departments anymore. They don't have the experts to improve anything. We're struggling. I don't know if you know about Namur, um this user association. It's becoming more and more difficult, especially now with the economic downturn. mean, the chemical industry is struggling. You're not going to build up a group now of process control engineers in this climate. But they should, because then they could develop it further. They could work with the standards. They could make sure it's important to them. But this is not what we're seeing. We're seeing fewer and fewer people being involved from the automation user side. And then you have the system integrators sitting in between, and they don't have an interest either because they're selling at their time. also, they can't change the system. So no one can actually improve anything at the moment. So we're of completely stuck. It feels. Speaker 4 (36:52) Margret, of course, I'm going to come in with the solution for all your problems. You need to use AI. It's going to solve everything. mean, it just needs to be more innovative, use AI and unleash it on the fields. Speaker 1 (37:12) So one of my other master's students worked on a reinforcement learning strategy for a batch process. So he optimized um the throughput of the batch process for penicillin production using AI. And that worked really well because he had a model of the plant. uh This is a model that's out in academia. So in academia, we have a number of processes that are well understood. So one is the Eastman chemical, sorry, it's called the Tennessee Eastman problem. So you can look it up. It's a paper from the 1990s, a TEP, and it's done by Jim Downs, who I worked with. And people use it all the time to study the behavior in these exceptional situations. So... Everyone should have a look at this because there's MATLAB code, there's C code available for free for academics online to develop things. And so the other process we have is this penelucinin process. And my student developed a really good algorithm using AI, using reinforcement learning. And he could improve the throughput by 25%, which is magic. It's amazing. That's fantastic. But it's all in a simulated environment. And it's for this one particular process and it's just not scalable or repeatable or anything. Speaker 2 (38:33) it's a great point you're making. that's actually, ties back to the fact that uh companies are so close about their process models. Because we have nearly no access to actual data and actual models, it's extremely difficult to come up with new strategies. uh When I did my non-linear MPC thesis, I also tried running it on actual process. and it didn't calculate at all, too much, much, much, too hard to make the calculations. And then I reverted the back also to uh some model uh I got somewhere with some easy data and all of a sudden it worked. But the problem is that I also knew back then that those, that way of calculating uh the Optima would never work on actual data. And here we have the same problem and indeed with AI and I love the question, um Willem, that's one of the things we have been talking about quite a lot in the last month is the understanding of the physical twin, the physical assets, the actual process. Well, that is what makes or breaks real AI in a process slash manufacturing context. And all the rest is Yeah, it's nice to have some marginal improvements, but nothing fundamental, I'm afraid. uh Speaker 4 (40:08) what makes operations operations also. It's like that links to the physical reality and every physical reality is different. Even if you have the same plans constructed just next to it, they will be different. They will be different. You can be 100 % sure. You might not know why though. Speaker 1 (40:28) can understand the frustration though, because we are in our, I mean, our world is changing. We are in this consumer world where we look, mean, look at our phones, look at our devices, look at this, this, this podcast we're doing now, how easy it is. I mean, this wasn't possible 10 years ago. So, I mean, we can understand the frustration from seeing how the world around changes and not seeing this change in the process industry and on the, on the side. And I think. one has to relax a bit because the process industry were the first ones that communicated really well. I we had all these sensors speaking to each other before we had the internet. I mean, it's sort of, I can understand this. I I was equally, when I was working for ABB, was equally frustrated. because we worked on this project where we had to Google search of the plant. I mean, we came up with this Google search using all the data, using this IT integration and getting all the data. And then you just search for pump and whoop, it comes up and everything you want to know, all the pumps are listed. And then you Google search and you find this one pump and they get all the information you get the maintenance manual and you get all the past data and you get the vibration sensor data. You had everything together in one spot. And that'd be so amazing. But it's going to be a a lot of work and no one's going to pay for it. Speaker 3 (41:47) Yeah, that's it. Speaker 2 (41:53) It's an interesting discussion and it's also a bit more philosophical one and we can continue that for a very long time. But I have one final question I'm actually dying to ask you because I also read that you were creating an art project with control engineering equations. That kind of triggers me. Speaker 1 (42:18) yeah, I would love to have something here behind me, Yeah. oh let me pitch this to you. Because I started and then okay, let me let me explain. basically, when I had my kids, I wanted them to be mathematicians. Because I mean, what's the point of having children if they're not good at maths? Speaker 4 (42:36) So I read some exactly that's the whole point. Speaker 1 (42:39) I wanted a mini, I wanted to do minimies. And then so I read somewhere that your kids are going to be very clever if you play classical music to them. So I I bombarded my belly with classical music nonstop for nine months. And what do I have now? I've got two kids have very good at music. So then I, um, we, we, did, um, I participated in this, in this, uh, it's called in control podcasts. It's if you want to know more about control, the in control podcast is a really good, um, starting point. And, um, the, um, The host is called Alberto Padoan and we did an episode on Sofia Kovalevskaia. He was one of the first mathematician dealing with control-related topics. There's the Koshy-Kovalevskaia theorem uh named after her. And she's an incredible, uh clever mathematician slash control engineer. And so she learned control because she had a wallpaper. because she grew up in Russia and they couldn't afford wallpaper. So they had a wallpaper made out of mathematical sheets. So all the leftover... So then I thought, okay, clearly the playing classical music doesn't work. But what is going to work is designing a wallpaper made of mathematical equations. It's going to look really beautiful. And then we put it up in the nursery and you can have a color scheme, choosing different colors. And then all these babies will grow up with these mathematical equations. They become really good mathematicians. So, so I started. Speaker 4 (44:27) room, this is the analytics room, this is analytical room, algebra room. Speaker 1 (44:34) So I've got the equations, the control engineering equations together. I started drawing them and then I ran it past my friends who are architects and PhD in French literature and et cetera. And they didn't like the idea and I didn't think my drawings are very nice. Speaker 2 (44:52) don't like the idea because they just don't, no, no, no, no, the idea is brilliant. Speaker 1 (44:57) Well, they didn't like my drawings as such, so I'm currently working on them. So I've done a first draft and it didn't hit the spot as much as I liked. So still working. If anyone wants to steal this idea, absolutely welcome. still going to, I I love drawing. So that's my favourite pastime is to draw. so I haven't given up on it. I'm still busy drawing, but it's taking a little longer than I thought it would. Speaker 2 (45:24) but yeah, we will eventually get to a paper. request towards or ask towards our audience if there are people who are really into art and drawing, why don't you reach out to Margret. Speaker 4 (45:37) there's a big overlap between mathematics and art anyway you'll find also with also with music Speaker 2 (45:45) I fully agree. I always loved the look of creations, but now we're going in a wrong direction. With that, Margret, thank you so much for joining this podcast. I very, very much enjoyed it. I hope you will be able to finish the art project and when you do, let us know. We'd love to be one of your first. rooms, we'll put it in control rooms. Yes, you can put it as your background. Speaker 1 (46:15) yeah, absolutely. Speaker 4 (46:16) change the poster, you know. Yeah. Speaker 2 (46:20) Okay. Or I can change my my roadmap from 1985. Speaker 1 (46:27) Yeah, it's the wrong projection. mean, Africa is much too small on that map. So you have to change it in any event. Speaker 2 (46:32) Yeah, yeah, yeah, is, is, is, but it's 1985. There are still some countries which don't exist anymore currently. Anyways, again, thank you very much also to our listeners. Thank you for tuning in. And as always, if you enjoy our content, make sure to subscribe on the itotinsighted.com or take a look to our academy, itot.academy. Willem, Margret. Thank you and see you next time. Goodbye. Thank you, bye. Bye bye.