Good day, Rasananda Muduli. I'm Arjun. I've had a look through your background and your professional record. I'm interested in hearing more about your work and how you see yourself fitting in here with our faculty. Ready to start?

Please start by introducing your professional journey in academia.

Hello, Sir, Good afternoon. My name is Doctor Sanandli. Uh, I have completed my PhD from National Institute of Technology, Karnataka, uh, in 2025. Uh, before that my M tech, I have completed my Mtech from National Institute of Technology, Silchar, Assam. Uh, they're my specialization was power and energy system engineering.

Uh, in 2015 and my $12.10 standard I completed in state government board Odisha.

And my recent research work is application of reinforcement learning for power system and control system. Thank you. This is from myself for academic journey.

Thank you, Professor Muduli. I appreciate your summary. Let’s delve into your research expertise. Could you elaborate on how reinforcement learning is used in load frequency control for renewable-integrated power systems? Specifically, what makes it advantageous over traditional control methods?

Uh, during traditional control method uh mainly we used PID Pi controller or sliding mode controller which mainly required.

Which mainly required the system parameters. However, in power system design, the exact exact simulation model will not find. Uh. So what we did we in place of the conventional PID and sliding mode controller, uh, we use adaptive controller where reinforcement learning is help. Reinforcement learning helps.

To design the controller from the model whose parameters we don't know, it helps to train from trial and error and after a point where we are getting dissatisfactory result, we confirm that our controller design and we use that controller using in place of the PID controller so that without knowing the system parameters system model we can design the control.

That makes sense—the adaptability of reinforcement learning is clearly a powerful advantage in a system where parameters are uncertain or unavailable. How do you ensure stability and robustness in this trial-and-error-based training approach, given that power systems are critical infrastructures and cannot tolerate instability during the process?

Is a stability.

However, in our control mainly we worked on the controller part. In our controller we we know the we know what should be the behavior of the power system.

Stability. Yes, for a complex power system, we should take care of the stability.

What should be our result? So what reinforcement learning is doing? It is trial from the simulation model it trial and error trial input it is giving and it is sensing the output and it is seeing the output is whether it is good or bad. How much it is coincide? How much it is merging to the actual our intended result? If not, then this reinforcement learning is giving one penalty to the.

Agent. So that agent will change its, uh, input Agent will try to design the controller again. It will go, it will, will give the system again will sense the system after search iterative method at a point, we'll see that our system is going to uh, our controller is going to give a exact means, not exact exactly a similar, uh, output. Then only we'll assure that our system is proper.

Hence our controller design is correct.

Understood. So the reinforcement learning agent fine-tunes the controller iteratively by penalizing incorrect results and optimizing its parameters until it achieves outputs that align closely with your desired system behavior. That iterative refinement indeed sounds robust. Now, moving forward, could you describe how you incorporate hands-on laboratory work into your teaching when covering concepts like load frequency control or reinforcement learning? How do you bridge theory and practical implementation for students?

This is our research workshop for students purpose. We have not done this work but we can show 1 demo how our model is working for a I triple E2 area system. I triple E2 area LFC system will bring and for that system we know what what is the expected output. Everything we know then what we did in place of PIPID controller will replace that PID controller and our RL controller.

And whatever result we are getting from the RL controller and whatever result we are getting from the PID controller will compare that controller and we'll see that our controller is showing good result. And the main advantages of our control is we do not know the system parameter still we designed the controller so that demo we can show to our students.

Thank you, Professor. Your approach to demonstrating practical applications, such as replacing the PID controller with your reinforcement learning-based controller on an IEEE two-area LFC system, is a strong strategy to convey theoretical and practical insights effectively. How do you evaluate students in similar hands-on projects or research-oriented tasks? What methodologies or criteria do you use to ensure a fair measurement of both their understanding and application skills?

But the application skill set uh we designed our PID controller based on the reinforcement learning. That PID controller is not restricted to apply only the LFC system. Many many places we are using the PID controller not only LFC. Instead of LFC we can use same controller.

But design will be different, KPKIDKPKI and KD parameter will be changed and that controller we can use for other systems. So in that way we can.

Expand our design. Design the control, not only the LFC system. That way we can do the practical application.

Thank you for elaborating on the extensibility of your controller design beyond load frequency control systems. It's clear that the adaptability of your approach has broad potential applications. Returning to student evaluation, could you detail how you would assess their grasp of these concepts and practical applications? Do you focus more on their theoretical understanding, the accuracy of their results, or their problem-solving process?

The accuracy we can, we can see that whatever controller we are designing, uh, conventional way, we are designing from the border plot or from the root locus whenever we, whatever the, uh, requirement given based on that we are designing from Pi, uh, root locus or depot upload for in our control.

This is completely this controller we can use for the non linear system.

Where where designing the Pi and PID controller, conventional control need exact system model still at a particular operating point. We can design the Pi PID controller for our controller. Even the system parameters vary, system nonlinearity is there still the the required performance is given by our. Our controller so that way.

We can tell the main advantage. We can tell practicality without knowing the system model. We have system we can design the controller if if we know the what is expected output or design given.

Your explanation highlights the uniqueness and practicality of your approach, especially for nonlinear systems where traditional methods requiring exact system models might fall short. By focusing on outcomes and designing controllers without needing precise system parameters, you’re driving innovation in control design. For evaluating students in such a setting, would you primarily assess their ability to apply theoretical concepts to practical designs like this, or would you also heavily weight their demonstration of foundational control techniques such as Bode plots and root locus analysis?

So we can we that our controller and that conventional controller we can do without knowing the system model. I'm my my approach is not to restrict in the Pi and PID controller. My approach is we can design other controller using the reinforcement learning algorithm. Many reinforcement learning algorithm is there when system is given and system expected outcome is given. We can without knowing the system dynamics.

And design our controller there. We don't need the, uh, we, we don't need the body plotter, uh, root locus method. But by knowing the upper value under lower limit of our PID controller design or whatever controller we can design, we can analyze using the Bode plot. After that designing, we can analyze Bode plot. We can tell, uh, at what position should we operating when and whether our controller. Uh, operating the system within the.

A operating region that we can.

That we can do.

Understood. Balancing innovation with foundational principles will provide students with both the versatility of modern methods and a grounded understanding of classical techniques. Let us now discuss your teaching approach more broadly. How do you ensure that students, regardless of their varying levels of prior knowledge, fully grasp complex topics like reinforcement learning in control systems during your lectures?

It's a mainly.

Mainly our system, whatever we are, we are teaching our student at conventional method we are teaching and in that conventional method we are just teaching that board plot, root, locus, Nyquist plot and some now controllability, observability using that we are designing and Kalman filter also we are designing for the control purpose and the time domain we are designing and frequency domain and both time domain we are using you are designing the controller using.

Border plot and root locus all this method. However, the point is that when the main machine learning algorithm we we are interested to design when we have data means for a system already previous data system data is there and we want to design one controller. In that case, from the data we cannot design the PID controller. However, from the using the data we can use the machine learning model.

And that model we can use as a controller. And before that, using that model, we can analyze using the border plot after the designing. And we don't need the conventional way. So main, main advantages is that when we have historical data or some system model, we can design without, uh, knowing, uh, without requiring the system parameter. That is the main advantages. And for the student purpose, we want to. We want to give the teaching style in advance way not to did that convince.

We have only designed DPI DPID controller for sorry in border plot and root locus and time domain analysis for controller. We want that we should more we should more go more ahead using the technology which we we make our system more easy to design. So that way machine learning model will help us for advanced control control where non linearity more non linearity is high and.

System model we cannot design that way. We can use the.

Machine learning model.

Understood, Professor. You emphasize a well-rounded approach by introducing conventional methods like Bode plots, root locus, Nyquist plots, and Kalman filters, ensuring students have a solid foundation in classical control design. Simultaneously, you advocate advancing teaching methodologies to equip students with modern techniques like machine learning-based controllers for handling nonlinear systems and leveraging historical data effectively. Your vision to modernize teaching aligns well with preparing students for real-world challenges. How do you guide students in integrating these modern machine learning techniques into their projects or research work, especially when addressing real-world, complex systems?

Suppose we have system complex system for that system. The main point is we cannot exactly design the system. If you want to apply the conventional model, we should know the system model exactly system model how system parameters are behaving. But whenever we don't know the system model then only this machine learning.

And the adaptive control will help us to design the controller. And for student student understanding purpose, uh, we should add.

This uh, adaptive controller and machine learning based controller along with our conventional controller and we have to show how conventional controller is behaving and after same system for the same system when we are using adaptive controller using machine learning model, how system behavior is same? Both ways system behavior is same but in which case we are getting more uh more difficulty to design.

And we can see that an adaptive controller, most of the time adaptive controller design method will be easy. And when system model we don't know. When system model we know we can use the conventional controller or we can use easily conventional controller. The main point is that for the exact design when system model is done, we don't know we should go with the higher controller design using machine learning model.

Your approach for guiding students by comparative demonstration is excellent. Showing them both the challenges and effectiveness of conventional and machine learning-based adaptive controllers on the same system provides a comprehensive learning experience. This not only solidifies their understanding of classical methods but also illustrates the efficiency and necessity of modern techniques for complex, poorly-defined systems. Let us now touch on your research contributions. Could you elaborate on the significance of your recent research publications and how they advance the field of power electronics, power systems, or control systems?

Our main whatever reinforcement learning application we are doing that we are doing a application for controller design for different system. Does controller design we can use in power system For example in our case we used in load frequency control. However that for for example in power converter design when a bug bug converter to a designing bug converter we are designing voltage mode controller.

More controller there we need feedback controller where there will be a reference voltage and there will be actual voltage we are sensing or reference current or actual value we are providing and that error we should we should control using a Π controller so that our duty cycle will change very fast as as required. Because for the bug converter we need the constant disable output constant DC voltage. The DC voltage will get by using the switching.

Frequency by switching and that switching is go governed by the PWM generator and that PWM generator will design by this Pi PID controller from the error value. In that way we can use it in the power electronics one. This is one example in power system power system in many cases ABR automatic voltage regulation or load frequency control we need π controller or PID controller that controller we.

Use using the our model. In that case two example I have given from one power system. That way you can decide in control system. Nowadays we can see nonlinearities. When nonlinearities high and system model we can design correct system model. We cannot design that way. There also we can use the reinforcement learning control system controller.

Thank you for that detailed explanation, Professor. Your examples—from designing a voltage mode controller in a buck converter for DC voltage regulation to applications in automatic voltage regulation (AVR) and load frequency control in power systems—clearly demonstrate the versatility of your reinforcement learning-based controller design. Employing this method to address nonlinearities and unknown system parameters is a significant advancement in both power electronics and control systems. Your work highlights a practical bridge between theoretical research and real-world applications. Let me move on to another aspect of your academic contributions. How have you guided or plan to guide student projects or theses in research areas related to your expertise? Could you share an example or outline your strategy for mentoring students in advanced research?

For example, whatever the controller we are designing, suppose we are designing the big bug converter. In that bug converter what we need, we need at 12 Volt or 24 Volt this year, whatever our specification, that voltage we should be, our output should be that much voltage. But that voltage what we need, we need a controller for two design DPI PWM converter. Suppose we we want we designed everything in our model.

And we used Pi and PIPI, the controller we used to generate PWL. In place of that, suppose we know the historical data that this type of converter we have and we want to control using the computer only that PW instead of electrical designing, we want to give a signal that PWM signal from the our computer. To control the switch. In that case, what we will do, we will.

Collect the data previous data from the PID after PID block and before the PID block whatever the input and output data and using that output data will.

Model, we'll design the controller and we'll train our agent then that if you will use that agent that accuracy will be more high as compared to PIPID control. Because whatever Pi control we use for a particular condition we are used. But in our data that is for different, different various condition we collected in data and using data we could design the controller. That way our controller will be fast and.

Implemented.

Understood, Professor. Your approach to

Understood, Professor. Your approach to mentoring students revolves around exposing them to cutting-edge techniques, such as transitioning from electrical hardware-based PID controllers to data-driven, computer-controlled PWM generators. By leveraging historical data across varying conditions to train reinforcement learning agents, students not only achieve higher accuracy but also gain insight into adaptive, versatile control designs. This hands-on experience and exposure to advanced concepts undoubtedly prepare students for real-world engineering challenges and foster research innovation. As a final question in this regard: how do you mentor students to critically analyze and improve upon existing methodologies, encouraging them to contribute meaningful advancements to the field?

Students we need a small interested group where student is very much interested in learning in that case, what we'll do, research paper will read and whatever the.

Requirement, whatever the research gap, we'll collect that research gap and we'll give the student and we'll explain the student. This is the this is the area in that area. These are the research gap for to maintain that research gap, we should.

That will be the good mentoring, uh, student can understand because, uh, we have to make them understand what is the system and in this the system, what work we need to do. Then only they will do the work. In that way we can mentor the student. So research gap, finding the research gap is our duty. We have to give the student that research gap and student will work on that and will give the good project. And, uh, practical.

Explain depth what to do to the next step. That way they will proceed and they will design the problem and again we'll verify if it is correct. Then that way we can help uh.

Practical output.

Thank you, Professor. Your approach of identifying research gaps, guiding students to understand the system comprehensively, and providing them with clear, targeted directions ensures that they can effectively work on meaningful, impactful projects. This strategy not only empowers students to navigate advanced research but also fosters critical thinking and innovation. Let’s turn now to your teaching philosophy. Beyond technical skills, how do you ensure that students develop strong communication and teamwork abilities, which are vital for both academic and industry success?

Student and teacher there should be talk bilaterally not this teacher will talk or student will talk and student will not talk both in the class. They should interact each other that way we can.

Proceed.

Thank you, Professor. Encouraging bilateral communication and interaction in the classroom is a vital step in fostering strong communication and teamwork skills. This open dialogue ensures a more engaging and collaborative learning environment. To build on that, how do you create opportunities for students to work together, such as group projects or collaborative research, to further strengthen their teamwork skills?

We need both. We need collaboration work with other institute so that our research should not restrict what we know. Our research should be elaborated, it should be expanded what others are doing in this real world. So we need both collaborative and with that collaborative also we need small, small group of research.

And faculty or the university faculty and other students, we both we should collaborate and we should make a group so that we we will know to know what real world problem is going on. That way we can carry on our research.

Students.

Understood, Professor. Encouraging collaboration both within your institution and with external organizations or universities provides students with a broader perspective on real-world problems while fostering teamwork. Establishing small focus groups comprising faculty and students ensures a hands-on, guided path for research and practical contributions. Thank you for sharing your insights so far. To summarize your contributions, could you briefly outline your key achievements or milestones in academia that you believe distinguish you as a candidate for this role?

But academic purpose I completed my period in 2025 October 2025. After that, during my research work, I got.

Good Q and paper, 2Q1 paper, one Q2 paper and another SA paper. I published 4 paper I published that will be good research that showing my good in research and all paper ACI and Q1Q1 and Q2 paper. So that way that we can tell 1 achievement we we validate our research.

Using by publishing the papers that is 1 achievement we can tell.

Thank you, Professor. Your publication record, including two Q1 papers, one Q2 paper, and an SCI paper, underscores your strong research capabilities and dedication to advancing knowledge in your field. These achievements also validate the quality and impact of your work. It has been insightful learning about your expertise and teaching approach, Professor Muduli. Is there anything else you’d like to share or discuss regarding your qualifications or aspirations for this role?

I mean, I'm eagerly interested, uh, to, uh, get the best, best interview and I'm excited to go to BIT and see how BIT is working and, uh, what are the infrastructure BIT will provide to us and how we'll proceed in research, how we'll proceed in the outreach and how pH will help us that we want to know.

Thank you for sharing your enthusiasm, Professor. Based on the job description for this role, BIT places a strong emphasis on teaching excellence, student development, and impactful research, alongside fostering industry-institution collaboration. Your publication achievements, reinforcement learning expertise, and vision for modernizing teaching methodologies align well with these expectations. Given your excitement about the infrastructure and opportunities BIT offers, I believe you would find their focus on supporting high-value research initiatives and student development particularly rewarding. With your deep expertise and aspirations, you are poised to make significant contributions here. Is there anything else you’d like to discuss or share before we wrap up, Professor Muduli?

No, Sir, I understood the process of interview. There will be no vital question in my mind now.

But, uh, we'll be there. Indeed, during the face to face interview, I'll, I'll be understanding.

What will be the next step? Thank you.

Thank you, Professor Muduli, for your thoughtful insights and for sharing your academic journey with me. It’s been a pleasure speaking with you today. The next steps will be communicated to you by the recruitment team, and I wish you the best for your face-to-face interview and the journey ahead. Just to confirm, do you have any further questions, or can we conclude the interview here?

We can conclude, Sir, there will be no more question. We can conclude.