 I'm Rao Govindaraju, but I go by GS. And it is my pleasure today to introduce one of our colleagues, Muhammad Jahan Shahi. Muhammad got his PhD in about 2011 or so from University of Southern California. And then he stayed there as a postdoc, did some work as a research assistant professor. During this time, he also worked at Jet Propulsion Lab and finally joined our program in about 2014 as an assistant professor. So we are very thrilled to have him. He works in the broad area of structure. His teaching is mostly in our structure's courses. But his research is very interesting. I think he works in big data analytics. He works with sensors which collect disparate heterogeneous information, combines them using machine learning, deep learning, and other techniques to essentially lead to informed decision making. And in the application area is primarily into developed resilient communities. So with that, I'll let Muhammad come and tell his story. Thank you, GS, for the kind introduction. And thank you, everyone, for your attendance. It's an honor to be here. Typically, when we have a goal in our life, sometimes we say that if I get to this goal, I would be very happy. But the real life is that there's almost always a process until you reach your goal. And I always try to remind myself about this quote that says, if you do not enjoy the process, you are not a happy man. And I tried my best to enjoy the process of being at Purdue and the tenure process as much as possible. Of course, as you know, there are going to be lots of parameters that can affect that. Today, I'm going to share with you some of the major ones that helped me to enjoy as much as possible the tenure process. The first thing is the ecosystem. So I have been very fortunate to have lots of good people around me, supportive people. These are the names of my official mentors, late Professor Metisosen, Professor Mark Bowman, Professor Melba Crawford, and Professor Dulcy Abraham. Of course, GS has always been instrumental and very supportive. I've also had lots of other unofficial mentors and role models. And as you can see, they are from different departments. As I'm going to be talking about my research in the next few minutes, I should mention that all the hard work comes from my fantastic students, both in PhD and master's students, as well as undergraduate students. As you can see, I've been involved a lot with working with undergraduate students from different departments, ECE, civil engineering, computer science. Last but not least is the family. I have been, again, very blessed to have a very good supportive family, my siblings, parents, in-laws. These are the three people who have been with me during this journey in Midwest. And I'm very grateful specifically to my wife for her support and patience. Indeed, without her support, I wouldn't be where I am today. So with that, let me give you a little bit of background. I was born in Shiraz in Iran. It's a city with 3,000 years of history. I got my bachelor's in civil engineering. Then I moved to the capital of Iran, which is Tehran. And I got my master's in the structural engineering. Then I moved to California, Los Angeles. And as G.S. mentioned, I joined the University of Southern California. I got my master's degree in electrical engineering and PhD in civil engineering. Then I was a postdoc there for one year. Then I joined Caltech as a postdoc and worked in the NASA Jet Propulsion Lab for one year. And then they hired me as a research technologist in the robotics group with PhD in civil engineering. Anyways. And then after one year, 2014, I moved to Purdue as part of the big data cluster I have. So if you ask me what you do in two words at Purdue, my research is data analysis. We try to extract useful knowledge from sense data, variety of different data, primarily computer vision type of data like images, videos, point clouds. We also work with time histories and vibration signature of the structures. OK. Now, if you tell me, OK, we didn't see what engineering. Give us a little bit more understanding, a little bit of a scope of what exactly you are doing. I try to deal with the aging infrastructure. This is the grade for the report from ASC about the infrastructure, which is C minus. And you can see many of the infrastructure get D or D plus or D minus even. And the challenge is that insufficient inspection, because our current inspection method is manual, time consuming, subjective. Sometimes we inspect bridges once every two years, believe it or not. So the idea is that can we use artificial intelligence, data science, now that we can collect huge amount of data from different sensors and robotic platforms to make sense out of them. That opportunity is tremendous, particularly due to this recent 1.2 trillion infrastructure investment and jobs action acts, which is a once in a generation investment in civil infrastructure. And probably it will be benefiting the research in this area. So what I will be doing, I will just share some of the research topics that I've been doing. And some of them are ongoing at Purdue for the next few minutes. Again, just as a recap, this is using artificial intelligence and robotics. These are the two components of what I do. This project is related to condition assessment of nuclear power plant reactors. The reactor is under the water. So there would be robotic arm that collect videos. And someone looks at the videos at technician and provide a subjective report. But if we can develop algorithms that they can go through the whole video and provide a probabilistic report and tell you, just watch this part of the video. With this confidence, there's a crack with this thickness. And let's look at this part. So this technology is currently being rigorously evaluated by a couple of companies, including Westinghouse, to be incorporated into their inspection pipeline. Another project, which is quite interdisciplinary in collaboration with my colleagues Ed Delt at ECE and David Johnson at Industrial Engineering is about flood risk mitigation. Flood is the deadliest natural hazard. The idea is that there would be a website. You put your address. There it is, the coastal region of Louisiana. And you would get a map like this that provides the flood risk. So in the back end, in order to have this risk analysis, you have to have the information about the attributes of the buildings across a large region around you, like the type of buildings, number of stories, type of foundation, height of foundation. So what our students have been doing, they have developed algorithms based on deep learning and multitask learning where they can autonomously analyze Google Street View images and extract this information. So the information about 800,000 buildings are provided to the authorities in the coastal regions of Louisiana. And we are now extending this to the whole state of Louisiana. The other project is related to using robotic systems for inspection of bridges. And then here is an example of Bowen Lab. The idea is that if you have a UAV, you collect data, say, every six months from the bridge, can you identify and track the defects? Like in this case, you can see four different rounds of inspection. We have identified the same crack, quantified, as you can see, it's thickening. But the interesting thing is that these devices have limited computational capability. So in collaboration with our colleagues in computer science, Professor Elisa Bertino and funded by IBM Research Lab, we came up with an idea that if you have a very large deep neural network and you want to put it on an edge device as part of the vision of Internet of Things, how can you systematically prune the network? So in this case, for this very popular network, what we did, we were able to reduce the size and computation costs for 90% and yet get the same detection performance. This project is very close to my heart. And it comes kind of related to my teaching as well. As Gia said, I teach classical mechanics, statics, and structural analysis in civil engineering. In graduate level, I teach computer vision for civil engineering students. But I also, I was an advisor for EPICS and with my students, we have VIP team. So this is the idea and it came from EPIC and VIP. I would like to spend a little bit of time on this because it relates to teaching and research how they connect to each other. The idea is that if you put a set of inexpensive sensors, commercially available off-the-shelf sensors, which would be $300, like a GPS, an RGB depth sensor, an edge device, and you can transmit the data. And if you develop algorithms that can analyze the data and you put these devices at the back of the vehicle, I think this is U-Ting's vehicle here, she is in the front row. And the thing is, if you can collect lots of data, say in big cities like Chicago and say San Francisco, you would have thousands of these cars collecting data, sending the information through internet to a central system and you can analyze it. You can have updated information about the condition of the road. So to this end, I had this vision, I started with EPIC's student. The students, undergraduate students, they tried to put the sensors together, then we moved it to VIP, then they start really making it work. Then my graduate student working with her, supervising the effort, and she developed the deep learning algorithms. And now this became a major component of her PhD dissertation. So it just came from VIP and teaching with, and working with undergraduate students. As a follow up to this, actually, we're developing, well, it's not showing the video, that's all right. Probably I can, it's not a big deal, but. So we are developing virtual reality modules for Department of Transportation, where you can wear the headset and inspectors and frontline employees, they can interact with the immersive world to be trained to how to make decision about the condition of the road or bridges. So they would be better prepared for future work. This is a project, another interdisciplinary project. This was something that a little bit was far away from my expertise. I have had wave propagation courses when I was graduate student. I work with Monopoli colleagues, professors Fabio Sanperlotti in mechanical engineering. And the idea was that, can we use artificial intelligence to design new materials, new material that can have a specific characteristics. It was a very, very beautiful experience, probably. If you have questions, I can talk about it at a later time. And this is the last project that today I'm gonna be talking about it. I'm part of the Rethi Institute, the Resilient Extraterrestrial Habitat Institute, funded by NASA. And the idea is that if you have habitats, and say moon or Mars, and there would be a disruption, say there would be a meteorite impact when you have a sensor network, how can you detect, quantify, and localize damage so that you can dispatch robots to go and fix it in a timely manner? So with that in mind, to wrap up, my research is about data sensing, data analysis, and data fusion. And it has lots of applications within engineering and outside engineering. I forgot to mention something. I have been collaborating with almost all nine divisions in civil engineering faculties, either writing proposal, writing papers, presenting things with other faculty, outside civil engineering, in mechanical engineering, industrial engineering, electrical and computer engineering, and also outside college of engineering with computer science, social science, and flat and botany. With that in mind, I would like to thank you, and one more time mention that if you do not enjoy the process, you're not a happy man. Thank you. Thank you. All right, thank you, Mohammed. We have time for some questions. Hi, so about the crack detection in like nuclear power plants and the road infrastructure that you mentioned, what do you observe in terms of like improvements between the AI versus how it was done manually using people, because that was more subjective, right? So is it the same or are the algorithms better detecting these issues than humans were? Yeah, that's a very good question. Let me mention one thing that for the road, please keep in mind that the type of crack that we have is very different from nuclear power plants. Under the water, it's metallic surface, lots of scratches, tiny, tiny cracks, so the approaches are not the same at all. I hope you understand that. For the road one, we have been closely working with the Department of Transportation and particularly with both cities of Lafayette and Lafayette. I personally have met the mayors about this project, and we were planning to use when the prototype is ready to put it on the bus system in Lafayette to collect data and monitor as part of smart cities. So what they do in municipalities, they do based on surveys, or there is a website actually where you can go and report that I saw a pothole here and there and this is not convenient. When I was working with Epic team, we worked with the city of Lafayette to develop an app, and basically you could easily report it. So you can see it's very subjective. Every few years, they rent a $2 million van to collect the data, so they cannot afford that van. And there are lots of other issues here that I didn't mention about it. We have been talking about this specific project from different aspects. I closely have been working with social scientists and because this can bring equality because if you are in Beverly Hills, you make a call and they fix your pothole. If you are in a poor neighborhood, no matter how many times you call, they will not get back to you. If we would have such a system like this in future, as part of smart cities, then you would have the information and it's gonna be bring transparency to government. You see the impact, it's not just a matter of detecting this crack or that crack. The big picture is bringing equality to the society. And so basically we have done rigorous measurements between manual inspection and using these techniques for roads and they have published papers. Roads are, these techniques walk much better, obviously. As I said, because the other one is just manual, someone reports there is a damage here or not, or the garbage truck drivers do that. For nuclear power plants, that's another issue. So it seems that we're not replacing human, we're assisting the human. So instead of watching several hours of video, you just watch the places that the computer or artificial intelligence help you to narrow down. So you can use your time more efficient and effective. I hope I answered your question. Yes. Hi Mohamed, long time no see. Yeah. Okay, my question is pretty simple because you have done a lot of excellent work on data-based data analytics method. I'm being more like trying to, what are your, my research is more like a model based. What are your vision about the integration of data driven or model free and the model based method? We kind of feel that actually, you know that most of them have their other advantages. I believe for complicated situations, it will be definitely beneficial to integrate data into the model based method. And sometimes model based methods are more robust than purely data driven stuff. Yes, as you said, they have their own advantages and they say advantages. With this particular two projects, I'm glad that you asked this question. We try to integrate the physics into the problem. So for this one, it is physics constrained because when we were working with our colleagues and we realized that the machine learning cannot converge and you know machine learning, if you have data, it is good loosely talking for interpolation. If it's something happens outside the data training data, it may not work well. That's why you may need to have some physics constrained on top of it. So that's why we have to integrate the physics into the deep learning algorithm. So it is physics constrained. And particularly for the second project, as I'm talking to you in Heraklab, we have a dome and it's instrumented. And I believe that probably for that one also at the end of the day, we need to go to combine these methods. As you said, I agree there are two different words, but you can merge them and take advantage of both of them. You're welcome. And again, as you mentioned, depending on what's happening in the problem project that you were talking about, if you don't have enough data, you may need to use the model-based approach that you mentioned. If you have enough, you can probably use the other one or combine them. Just a follow-up question to Shashway's. But one example that you talked about are the nuclear reactors. So we had a deep neural network and you were able to reduce the complexity of edge computing needs down by significantly. Was that physics-based? No, no, no, no, no. That one was not physics-based. So basically, there are two things here. For crack detection on nuclear power plant, it was just based on visual inspection. And we were not incorporating any physics. In order to reduce the other project, which was we were trying to reduce the, if I'm not mistaken, if I understood it, the size of the large neural network, that one also, it was not based on the physics. So basically, what we came up is a mathematical approach that you train a neural network. You train it and you test it. Then you can rank the, say, the weights, get rid of 10% of it, retrain it. And if the performance is good, do it, do it, do it, until the performance goes down. So you should do it gradually. Otherwise, it breaks the whole network. So there was no physics involved. Go up. Any more questions? Thank you for your presentation. I just was curious to know if you have any suggestions to succeed in the tenure track besides being a happy man. I think that's it. You should enjoy the process. That's the answer. If you don't enjoy the process, you are not a happy man. And I firmly believe in this. And I keep repeating this to my students. And I keep repeating this to myself. And I think, as you know, being a faculty, it needs lots of things, as you mentioned, prioritization, decision making, communication, leadership, learning new things and so forth. I would say, and prioritization of the time and so forth, and balance in the life as the first speaker mentioned. So I think you have to have all of these. But I put them under the umbrella of enjoying the process and not be burned out with just working, working, working. It's very general, I understand. But we can talk about it in more detail. You're welcome. Any more questions? And one more thing that I would like to emphasize. Ecosystem, good people. That's very important. From the first day that I came to interview at Purdue, I felt that. The second day, we had some extra time. The gentleman, I, Hunter, from over there, when I went to get into his car, he put Persian music for me. He is not Persian. He took me an extra mile and took me to some areas and show me houses to buy. You know what I'm saying? His office is across my office. Every single day, before pandemic, whenever he was in office, really I'm talking about, not exaggerating, I was happy he's there. One of the other guys in the picture, Santiago, was next to my office. I don't lie to you. The first couple of months, I was so worried. I was thinking these guys probably put me in between themselves to figure out what I do. But the other days, whenever he was coming from the class, yes, I am here. But again, because of the pandemic, we don't see. So I believe supportive and good mentors that they would be really helping you. That's very, very, very important, I believe, ecosystem. All right. Well, so enjoy the process, good ecosystem. With all that deep learning comes deep wisdom. So thank you all. Thank you, GS. I just wanted to also take a moment to thank all the organizers here. I wanted to, Marcia, Maria, and Amy are here, as well as Ed Dunn and his team. So please join me in a round of applause for everyone. And a final reminder that the next event in this series is April 19th. That's going to be our last for this spring semester. We have three, again, outstanding colleagues that we wish to recognize at that event. So hope to see you all there. Thank you for joining us today.