Transcript of interview with Dr. Sohail Nazari, global vice president of automation and digitalization - feed and biofuel, Andritz
Jackie Roembke, editor-in-chief, Feed Strategy and WATT Feed brands: Hi, everyone. Welcome to Feed Strategy Chat. I'm your host, Jackie Roembke, editor-in-chief of WATT Feed brands. This edition of Feed Strategy Chat is brought to you by FeedStrategy.com. FeedStrategy.com is your source for the latest news and leading-edge analysis of the global animal feed industry.
Today, we're joined by Dr. Sohail Nazari, global vice president of automation and digitalization for feed and biofuel with Andritz. He's here to talk about some of the trends happening on the automation front and how they can benefit feed manufacturing.
Hi, Sohail, how are you today?
Dr. Sohail Nazari, global vice president of automation & digitalization - feed and biofuel, Andritz:: I am fantastic. Thank you very much. I am excellent, thank you.
Roembke: Well, thank you for being here. Let's get right into it. What innovations in automation do you think will have the greatest impact in feed manufacturing in the next decade?
Nazari: There is not one specific technology that is going to revolutionize everything. It's a combination of basically different categories that come together, and then help us to grow into complete autonomous operation.
We have to (ask), what is the end game? What are we trying to achieve? Then start walking back: What are the technologies that are going to enable us to achieve an autonomous operation? Which is basically trying to get 24/7 production as seamless as possible, with no disruption, and increasing the availability, the reliability, the efficiency to help us to have less energy consumption. Safety is another very important part, is the end goal in the process industries, at least as we know it.
There are a lot of studies regarding all the management executives, what percentage are they trying to get to autonomous operations. We see the majority of the executives in the process industries would like to get to that — and that is what we call "Level 5 autonomy." Then we can talk about what technologies will allow us to get there.
Everybody talks about artificial intelligence, machine learning and algorithms, and then generative AI and what is the role in this process. Here, we can look at other industries and what they have achieved so far, and the trajectory that they have in the more traditionally advanced industries, like oil and gas, pulp and paper, and see where they are at. Then we can then extrapolate where the feed industry will be in the next 10 years. What (the other industries) can see is that the operations that are almost autonomous is that they have combined a lot of different technologies together.
To achieve (Level 5 autonomy), we are able to predict the behavior of the process, analyze, diagnose any conditions happening, and then do preventive measures to mitigate those conditions almost automatically. The technology that allows us to do that, they're not revolutionized algorithms — I want to make sure that your audience understand that. They're not something that was just invented yesterday and then changed everything in the world. These algorithms have existed for the last 20 or 30 years or these are all mathematical algorithms. But what happened is that now that the computation power has gone to the point that we can have millions and trillions of the calculations in less than a second.
Now we are able to use these algorithms to do something complex with them — that's the important part. The other part is that now that we have democratization of knowledge, coding and writing programs, that allowed a lot of people to get involved into bringing these algorithms and build applications with them that are useful for different industries.
This is what revolutionizing our industry, our process industry. The same goes for the feed industry.
Roembke: I recently came across, “digital twin modeling." Please examine what that means, and then also how that technology could potentially be used in feed mills.
Nazari: The term digital twin is a big hype right now, right? It's something everybody talks about, something that all the companies are trying to say, "oh, we have a digital twin, we are doing this ... " There is no technology company now that has not said, in one shape or form, that they they're doing something in this area.
The question is, what is a digital twin? It's a very general term, and everybody is interpreting it the way they want. I looked at this some time ago to find the origin of this term, where this started. And surprise, surprise, started in NASA and in aerospace around 11 years ago. Basically, NASA started — we've been doing simulation and modeling before. But the question is, can we bring the simulation and modeling in the whole entire life cycle of the project and then connect it into a real-time operation? And those two or three key definitions are making the word "digital twin."
Now, if you come to process industry, what does digital twin mean? That means you want to have a mathematical model. I want to have a model or a simulation model that represents the reality of my operation, one way or another, and then use that model in the entire lifecycle of the plant. In order to generate value out of that model, I want to connect that model into the real-time operation and use the information that is coming out of that model out into the operation that will become the digital twin.
Nowadays, maybe some people call CAD drawings, physical drawing of the plant, a digital twin — and it is not wrong. The question is, what is the applicability and the added value of changing the name of what we had into to digital twin? Now, if we come to the original definition I provided for your audience, the question becomes, how am I going to build that model? It becomes an important factor of how accurate that model can represent the reality of the operation.
Then there are so many different ways and methods to build a model. The first principle way means that, hey, we design these plants, we design this equipment, based on our knowledge of physics and chemistry, and stoichiometry, let's use that knowledge and put it inside the model so that we have a representative of how this plant, this equipment, should work. And that's one way of building models. The other way of building models is, let's gather information and the data and then build a mathematical model out of the inputs and outputs of this process. We call this data-driven modeling.
The trend that is going now is that there is no one true way or best way to move forward as with the majority of engineering and scientific problems. There is always a combination of different methodologies that gives you the best way — and that is something that now, more and more, is becoming present in our industry: let's combine first principle modeling technology and data driven all together so that we can have a representative of our process.
Some companies, including Andritz, have built [models] for a lot of different industries, like oil and gas, like pulp and paper, like mining, mineral processing. We have done it for the feed industry as well.
Now that you have this model, let's connect this model into the real-time operation and generate value out of it. That's an extra step. What value can we bring out of it? There are many, many different values. One is now that I know how the plant is running, I can start predicting conditions coming out of the equipment, or the throughput of the process, or the quality that is coming out of the process. For example, let's become very practical. Take the dryer, for example: We can have an accurate representation of the dryer operation so we are able to predict the quality coming out of our pelleting, drying or extrusion. Once we have that, what can we do? We can improve the efficiency, we can decrease the energy consumption and predict some abnormalities that happen inside the dryer, inside the conditioner, inside the extrusion, inside the pellet mill.
If you're predicting, you can prevent (problems).
The other one is that if you're going to add additional lines in to your operation, can I bring that into the virtual world? See how all of these are going to connect together, all the new operations with the previous one, all connected. And then see how much value we're bringing from our investment.
That's a big trend right now to do a lot of HACCP studies using the digital twin, and also provide operator training, build a simulator that can train the operators, the same way that the aerospace and airline industries are doing for the pilots. We can do that in the feed industry.
The last thing that I would add that might be interesting for your audience is that we are now able using the digital twin to train artificial intelligence agents, like for example, reinforcement learning algorithms, to actually learn how to operate the plant. And this is another use case of the digital twin, and we can talk about it more, but I will stop there.
Roembke: Very cool. I'm interested in picking up that conversation at a later day. Are there any certain considerations or obstacles that feed producers might come up against when exploring or adopting this sort of technology?
Nazari: Yes, there are and there are concerns, and these are something that we see a lot with our clients, from field operators, operation managers, then executives, each of them have their own and very valid concerns and limitations and restrictions of adopting technology.
I would generalize it into adopting technology, not specifically, is it a high-tech technology or is it a low-tech technology? It is a change, and change is always scary. Change is always hard to accept. It's human nature. So to manage that change becomes a very important aspect into bringing any digital technologies into operation. A lot of times, it gets overlooked because maybe the executives get all excited about this new technologies and its potential. ... I always tell our clients, "Look, if the field operator wants the technology to fail, they will make it happen. So as easy as that, it doesn't matter how much investment you put into your plant, everybody has to be aligned. Everybody has to accept it and be promoter (of the change). You need to have champions that really believe in it, and then try to drive that change and slowly show, step-by-step, the value that it brings to operation. That's one part managing the change.
The other part that we see in adopting new technologies is the business justification. A lot of executives, stakeholders and investors, they're asking and demanding for digitalization automation. But this automation and digitalization, the plan is not a one-step approach. It's not something that tomorrow you can have everything. It's a journey. It's a step-by-step approach. And if you look at it that way, that this is a step-by-step approach, it is the combination of different technologies coming into your plant, then you can look at what is my highest ROI? How can I start getting some value out of my investment that can allow me to have the next investment?
One thing I want all your audience to be aware of, if we tackle the problem as just one problem, I want to have it fully automated or digitalized plan, and then the bill for doing that it goes so high that it never happens. That's why doing it practically becomes very important. It's an obstacle for adopting these technologies because there's risk involved, and how you mitigate those risks is something that executives have to be aware of. And companies like us are able to build that and basically help to build that roadmap with them.
Roembke: Excellent. Now, you mentioned AI and generative AI, which is of significant interest, certainly in our industry. Can you look a little closer on the AI applications and the best integrations that would make sense in feed production, either today or in the future?
Nazari: Let's start fundamentally on this because we all talk about artificial intelligence and I purposely during our conversation, try to not overuse the term AI because if I asked you, what is artificial intelligence? I have asked many people about this, and really everybody has their own perception, but what is AI really? We all now know about generative AI, a specific thing about (real-time bidding) algorithms, that allows us to communicate, but if you really think about it, artificial intelligence is where you as an audience, your understanding of technology goes from.
I understand this to this is black magic, and that's where the AI happens. I can make that case for you, e.g. 15 years ago, 20 years ago, programming logic controller (PLC) programming was artificial intelligence. Basically, if you look at the definition of AI, is that when the machine can make the decision based on the understanding of the environment? Well, in (programmable logic controller) programming, we can do that, we can write the code, and then based on that, we can make decisions, understanding that the coder did that.
AI is just algorithms, mathematical algorithms. Now, you understand some of these mathematical algorithms, like for example, when it's a simple logic, and then you might find it very complicated when it becomes reinforcement learning. But the whole entire spectrum is the AI.
That's why it becomes very important. You know why it's important? Because when we want to adopt technologies, we have to understand, we are adopting technology because we want to improve something, not just because it's cool. This is important because it has to have a business justification. It has to have improvement with the minimum amount of effort. And I really want your audience to really think about this: How can I have the minimum amount of effort to get the maximum output?
That doesn't necessarily mean I have to find the most advanced algorithm. That's why we really emphasize on the journey, let's use the algorithms and the technologies where they belong, because there is no one solution or technology that will solve all the problems. If you look at it as blocks to build something, then there are fundamental blocks that you have there, like some well-established algorithms, that allows you to achieve very good results.
I'll give you an example. Like model predictive controllers. It's a class of controllers, that, if you think about them, based on the definition of AI, it is AI. But since it is a very well-known algorithm in the scientific community, in the control engineering community, nobody calls them AI anymore, because they're well known. Twenty years ago, the results that they were giving was very similar to some of this decision making and we though, "oh my God, this is amazing."
Let's use model predictive controllers that decrease the variability. Now, let's go one step further: Now that I have a very stable operation, I want to be making a new high-level decisions, can traditional algorithms allow me to do that? No, then let's go to the next step. Let's now use more complicated algorithms to allow us to get that then generative AI part that is becoming very mainstream, and everybody knows about this. Everybody is using ChatGPT now. The most important thing about generative AI is that it allows us to really understand human language and communicate with human language. This is a class of algorithms that generates a step by step the next board basically, based on the prompts that we are providing and based on the result is giving us.
It definitely will have a role in our feed industry. Why? Because then we can use that generative AI to ... if I can put all the information regarding the maintenance and regarding the operation into a repository, now I can start communicating with my machine to understand the best course of action. That's one way of using it.
The other way, when we are training the operators using the operator training simulators with generative AI, the operator can be instructed and can ask questions and then can be trained using the basically virtual instructor.
These are the things that we can do with the generative AI. But the point I want to make is that it's not only one technology called AI that is going to solve our all our problems. It is a step-by-step approach. Each of these algorithms, terms and technologies have their own place. And then we just have to use and choose the best one for specific problems, and then go to the next one.
This is an approach that works, this approach that we have provided, brought into all different industries. And we really want to bring into the feed industry as well.
Roembke: Excellent. Thank you so much for those insights. If you'd like to know more information about some of the technologies that Sohail has talked about today — and also what Andritz is doing on this front — please visit FeedMilloftheFuture.com. That's www.feedmillofthefuture.com.
Thank you so much, Sohail. And thanks to you for tuning in.
Nazari: Great, thank you very much for the opportunity. Appreciate it.