 Good afternoon everyone. Thank you. I hope it's a good afternoon. I'm Eckhart Groll. I serve as the head of mechanical engineering here at Purdue and it's my distinct pleasure today to introduce the moderator of our panel discussion, who will then introduce the panelists. So the moderator is Professor Hector Gomez. He's originally from Spain and got educated in Spain, but eventually just like many many others, including myself, came over to the US. He joined Purdue in 2016 as an associate professor in mechanical engineering and is one of our wonderful, still I will call him, young scholars and educators in our degree program. Here in his short career has already received numerous award for his academic excellence, but I wanted to point out one that is actually not on my brief bio, but I recently had to introduce him someplace else and I recall the Princess of Gerona Award from Spain, where actually the King of Spain is involved and where he helps attract young people into the area of engineering or STEM in general and it's really quite an accomplishment. It's actually an award that goes across the entire STEM discipline and he was selected as a young very outstanding mechanical or now mechanical engineers. He had actually educated as a civil engineer, I heard, but eventually he came over to the right area. So with that I will let Hector take it over and introduce the panelists over here. Hector, thank you very much for doing this. Is this working? Can you hear me? No? Okay. Okay, so good evening everyone. Thank you Professor Groll for your very nice introduction. So welcome everyone to the panel. Today we're going to be talking about the future of computational engineering. So computer simulations are nowadays used basically in the design process of every manufacturer object. They are also considered to be the third pillar of science together with theory of an observation and we will try to discuss today how computational engineering is going to change engineering, science and perhaps also medicine in the next few years. So we have an amazing panel to discuss that today. So if I wanted to do a proper introduction of every one, I think I would spend the entire hour. So I'm going to keep the introductions minimal. So we have our distinguished lecture speaker Professor Tom Hughes. He is the Peter O'Donnell Chair in Computational and Applied Mathematics and a professor in Aerospace Engineering and Engineering Mechanics at UT Austin. He is one of the most widely cited authors in engineering science and among many, many, many other honors. He is a member of the National Academy of Engineering and a member of the National Academy of Sciences. We have Professor Arisua de Cani. She is an associate professor in the Mechanical Engineering Department and her expertise is in complex flows, multi-phase flows and fluid mechanics in general. We have Professor Sikyan Kai. He is a professor in the Mathematics Department and he works on numerical analysis and computational mechanics. We have Professor Arun Prakash, associate professor in the Civil Engineering Department with expertise in Computational Mechanics and Structural Dynamics. Professor Vitaly Reis. He is an assistant professor of Biomedical Engineering and he directs the cardiovascular flow modeling laboratory. Professor Thomas Sikmun, professor in Mechanical Engineering. He is a leading expert in mechanics of engineering materials and we have Professor Ganesh Shubharaian. He is a professor in the Mechanical Engineering Department and his expertise is in Solid Mechanics and Computer Aided Design with applications in Microelectronics. We would like also to interact with the audience so if you have questions please let me know. I think you have also index cards where you can write your questions. Those will be brought to me so I can ask those questions at some point to the panel. So let's get started. I'm going to start with a question for our panelists. So I would like to start with a personal perspective from the panelist so that you can get to know them a little bit better. So they have been in the field of computational engineering for some time. They have seen how the field has changed, all the things that were done differently maybe a few years ago, but I want to somehow ask the opposite question which is how has computational engineering changed your life and your career over the years? So maybe we're going to start with our invited speaker, Professor Hughes. Okay. My life in computational engineering goes back a long way. It goes back over 50 years. So if I tell you all about it and how it changed my life we'd be here all night. It's really interesting because when I got started computer methods were rather a small part of engineering I'd say. And of course one looks back now and sees it pervasive throughout engineering. But it wasn't always that way and you might have thought that the path was easy and obvious but it really wasn't. There were a lot of battles and a lot of resistance to developments in computational engineering. But I would say one thing, every area in which computing has really penetrated it has in fairly short time in retrospect maybe 10 years or so, it has dramatically changed the field. Things that were experimentally based like elasticity where you would do photo elasticity experiments became analytical and completely done with computer technology. So it's revolutionized every area that I've had any contact with in engineering and it continues to do so. And now it's gone to other areas that are very exciting such as medicine. And that's been going on for quite some time too in different branches of medicine. But when we get to the future that's one of the big areas. I think computing methods are going to continue to grow. Computers are still getting more powerful. And the opportunity to apply them to more and more complicated problems and to gather data to support that is greater than ever. So I think it's just changed the world already. It's probably the biggest development in the second half of the 20th century. So I can go next. So I got into the field of fluid dynamics after I started my PhD before that, but I was undergrad. I wanted to be roboticist. But then after learning about fluid dynamics, I started actually with theoretical work. My first few publications were all theory. And I didn't know much about computational fluid dynamics until I got a course and learned more about it. But as I was doing more and more theory and learning the techniques out there, I realized that even though there are lots of nice things that can be done with theory, because then you get the scaling, you are able to come up with answers much quicker and understanding of the physical stuff problem, there are many other aspects where you cannot extend your theory to limits of large dimensionless parameters like Reynolds number, Weiss number, many other parameters. So that's how we started doing computational fluid dynamics. And one of the area that I have interest in is bio fluid dynamics. And at the time, many people started looking at, again, theoretical aspect of bio fluid mechanics. And we were one of the first looking at bio commotion and bio fluid dynamics from computational perspective. And now there are many, many groups who are looking at that on computationally. So that's how computational engineering or computational bio fluid mechanics has changed my own career. I was a grad student, I was very interested in computational sciences. So I was trained to do computational sciences, but I didn't really see the I didn't have the experience to see the impact of it. When I went to work in industry after grad school, I worked in a department that was doing finite element analysis on a daily basis. And what the company was trying to do, I work for IBM, what the company was trying to do was to use computational modeling, this is about 30 years ago, was trying to use computational modeling to reduce product development cycle times. So the whole idea was somehow reduce the product's development cycle times. And that has actually come down significantly. In other words, in terms of the expertise that is required to do that sort of simulation has now come down. Now we are seeing average undergraduate engineers work in companies with the aim of using computational tools to reduce product development cycle times significantly. And I think computational mechanics has been very successful in making that possible. I think Professor Rice went to say that for me, computational analysis and mechanics allowed me to bridge the gap between engineering and medicine. I'm an engineer by training, but I was interested in physiology and medicine. And computational mechanics allows us to apply a lot of physics and engineering to clinical problems, which would not be possible without computers. We would not be able to apply lots of physics to realistic anatomical geometries, complex flows, complex interplay of different physical phenomena. And that's the tool that bridges the gap between theory and clinical practice. Yeah, I think this is actually one of the things that we wanted to talk more about what is going to be the impact of computational engineering in medicine over the next few years. So we have seen that a lot of research labs in the world are using computational methods to develop new drugs, to study two methods for drug delivery, et cetera. And we have also seen the interaction between computational methods and imaging technology. And I wanted to bring this discussion to the panel. And I'm going to use a quote from Professor Hughes that I heard some time ago that goes like this. So he said, I see patient-specific computational modeling as a natural and inevitable extension of medical imaging. So I would like you to share your thoughts on that. I was just saying, I never said that. No, I'm sure I did say something like that. Anyway, it's really interesting if you think about what imaging has done. I don't think it's fully appreciated. We've just gotten used to it. You know, you get an MRI when you throw your shoulder out or something like that. But medical imaging is the window on human anatomy, physiology. Once you can see things, just like in every other science, like in biology with a microscope or astronomy with a telescope, once you can see things and create a geometry, you can build models of that. And you can bring to bear all the engineering and predictive technologies to the field of medicine. Medicine is a field that has been traditionally statistically based and diagnostic in its approach. But once you can do predictive tools, bring predictive tools to the party, then you can say things about the future. You can predict the outcome of an intervention. You can do virtual simulation of various interventions and compare them. You can diagnose on a computer. You can compute blood flow. You can compute many things that could only be measured experimentally, maybe in a cath lab or something like that. And also, imaging is much like computer technology. I think everybody's heard of Moore's Law in computer technology. It's advanced of the power of microprocessors. We have the same things going on in various imaging modalities. They're getting better. The resolution gets more and more refined as time goes on. So again, that's a platform. If you're researching in this area, that platform is rising you up, even if you're not doing much new yourself. But when you are doing things new with this more and more power that you have at your fingertips, then you can make enormous progress. And I think like other fields, like engineering fields, medicine will be revolutionized by computational medicine. It has started, but we've just seen the tip of the iceberg. I mean, this is a fantastic opportunity area. It will change the way we diagnose, treat, enormous effect on human health. Be prepared to have a complete mathematical model of yourself on computers. And Facebook will have access to it. Google will have access to it. So will the NSA. They'll be watching you. But there'll be some good coming out of that too. We hear more often like words like in silico medicine or virtual human, which basically now it's possible just as you mentioned earlier, because like we went from megaflop calculations in 1970 to now a petaflop that is being discussed and supercomputers that are built at data scales. So computations of that scale will allow us to basically use virtual humans for directory test for drug discovery test for diagnostic. And all of those would be possible because of this change, like in fields like aerospace industry or other fields, like we always use numerical or it's been a while we've been using numerical wind tunnels, for example, for low Mach number or high Mach number experiments or Boeing was able to reduce the number of tests from 1890 80s to 10 to 1990s and even less these but it's less commonly used by in the field of medicine by either biomedical industries or by pharmaceutical industries or physicians. And this is now possible. And that's why we hear more like something that was surprising to me. Maybe five years ago, we needed to try harder to convince industry while doing a computational work is valuable and can give them useful results. But a very recently, executive director of one of the large pharmaceutical companies were at Purdue, and he was talking about what we need in the future and his exact word that we need in cynical drug discovery or other aspects of medicine. And so we don't need any more to convince people these are useful. They know the value of computational engineering. And if I can add to that. So essentially, we were turning medicine into an exact science. It's more like an art with all due respect. And that's nothing better than that. But very often, your surgeons or surgeons or radiologists based their decisions on intuition experience, but they don't have objective quantitative measures. And by extending imaging with computations, we can actually provide objective measures, objective indices, physics based guidelines that we can provide them. And so, and also importantly, it will be personalized medicine. So we can actually compute biomechanical conditions for a particular subject. So it's a patient based medicine. It's important for the certification. It's important for tailoring a device or a drug to specific patient. Following up on this question. So we have in the panel experts on simulation of cardiovascular flows. We have also Professor Sigmund, who is an expert on bone mechanics. My question for you would be, what do you think it's going to be the next area of medicine in which computational engineering is going to have like very significant impact? One of the challenges that I see is that in a living system, we have to consider aspects that are different from an engineering system. So engineering systems are built and pretty much done today, right? They are delivered as they are in medical systems. There's a homeostasis and things change. They grow or are absorbed. I think there are good challenges out there in that direction for computations and they have to go over a longer time than we are potentially used to do a stress analysis of reliability analysis for a given amount of time. We still have challenges in that even in engineering to predict long-term aspects of structures often enough. So I think in the medicine role, this is I think a good, really good challenge to have and in playing this way backward, I think there's something for engineers to learn from that as well. If we can understand the physical principles behind such structures, for example, that can regenerate each other and live in a homeostasis instead of living in a fixed state once delivered. And that has computational resource challenges. This has challenges of non-locality in bone, for example. What is sensing of the image? It's not a local phenomenon on a crack tip of a microcrack. That information is relayed through piezoelectric effects potentially or through cell attachments through long distances. And so there are good challenges out there. So I would think that there are a few areas in the near future and one would be, as we already mentioned, optimization and design of biomedical devices targeted drug delivery. Personalization, customization of treatment for patients. That's one. Another would be understanding the disease, predicting disease progression, re-certification of patients. For example, in cardiovascular flow we know that vessels adapt to maintain optimal wall share stress values. So we can predict how the vessels will remodel in a specific patient and predict whether an intervention is needed or we can treat conservatively with medication, for example. And yes, in diagnostics, obtaining information that's not available from imaging alone by getting quantitative data. And then the last thing would be modeling different treatment scenarios. That's also a very powerful capabilities of computations. You cannot do that with imaging or with anything else really. You can image only existing conditions. With computational modeling you can model different post operative scenario and decide which treatment option is better, safer, less likely to cause complications. More areas of medicine where you expect significant progress. One of the things that I think will be important is sort of an interaction between computational mechanics and information. I mean, our bodies are information carriers. We can't do this without that. And that's an area where I think there are really good topics out there. So then perhaps following up on this, what do you think is the main limitation of the computational approach? What is limiting the computational approach from making a more significant impact on medicine than that it's routinely using clinical practice? Is uncertainty on the governing mechanisms or? The limitation is exactly on certain parameters or large number of parameters that are involved or not large number of physics that are involved in the process. And like in these days you see more and more of deep learning or machine learning combined with computational engineering or coming into help of classical physics based modeling and both together we'll be able to answer questions by either connecting the data obtained from observation and experiments to expansive computations and coming up with either reduced order model or propagating uncertainty in the parameters all the way through different levels of modeling. For example, you're dealing with multi-scale modeling where then it's important to propagate uncertainty from one layer to the other. So these are challenges and which now people in the field of computational engineering are looking into more. For example, perhaps this is going further away from your question to medicine but something that I observed last week where we are attending American Physical Society Division of Fluid Dynamics Conference where more than like 3,000 talks are presented. Like a few years ago there were not as many talks as now we see this this year on topics of deep learning in fluid dynamics calculation and I'm sure they're the same going on in medicine conferences and they evolve. Now the question I have related to that how can we make sure this is not used just as a black box without understanding the physics and just forgetting about the physics and looking at the problem as a black box. So now that you brought up this topic I think this was probably one of the questions that everyone in the audience was expecting which is how do you how do you think that machine learning and deep learning is going to impact computational engineering and maybe we can have the perspective of Professor Cai and Professor Prakash first. Yeah I think there have been several efforts I think recently where people have tried to do machine learning combined with some sort of finite element models but I think with limited success until now but we even we are working I think if you ask anybody these days everybody is trying to do machine learning to try to expand what their models can learn. One way I think we see things working out is you know infinite models there are lots of uncertainties which which the models you know you cannot really predict them how for example I have a you know structures background so I'm going to talk a little bit from the structural engineering perspective buildings and bridges. The level of uncertainty is so high that you know people have difficulty even trusting that the computer model results that sometimes come out of these these models but deep learning could maybe supplement you know the way we determine parameters we can we can learn certain patterns of how material properties materials under different circumstances behave things that we cannot do experiments on so I think there is some future but we have to proceed cautiously on deep learning and finite elements. Professor? Okay well first of all I'm not in engineering so I'm more serious. We start using machine learning to solve in the PDEs more or less we treat the DNN as new class of functions. The so far the experience is good of course still cannot compete with the finite we're still doing the one dimensional sense but potential seems like has a lot of potentials for example is any since the problem finite element cannot do well the machine learning probably give us the potential like approximate the boundary or interior layer sense extremely nicely of course for the deep learning right now fundamentals is not there we don't understand what is approximation property we don't understand how to solving the minimization problem effectively so all these issues but seems like we see a lot of potentials one sense I'm talking to a professor Hughes this morning as the one sense we didn't if we use the finite element we have a mesh and the DNN does not have mesh not only that they even have the nicer feature from the movie mesh so that's a reason approximate very well but of course for simple problems we still take days to train in that so it's not competitive now but there's a lot of potential as a and the other sentence of course I'm doing that is because the student pushing me to towards to this direction because they weren't looking for a job in the industry and if I can add one sentence I think it goes both ways it's not only that deep learning can benefit computational modeling and engineering but engineering and computational modeling can also benefit deep learning approach we can reduce the training data sets since we already have some knowledge about the system we can substantially reduce the amount of data that we need to train our neural net because we can impose some loads of physics and and then we can learn the system behavior much faster and more reliably professor Hughes maybe you want to kind of connect all these things to the question you asked again like what about the the future and I think whatever is done in a field like computational medicine whether it's data-driven or physics based modeling or a combination of the two whatever tools and analytical capabilities are developed somehow to get that to the clinic you have to go through clinicians and that's really a challenge because there are all sorts of things you can do and compute and clinicians will have no idea what to make of it in certain areas there are metrics that they use that they have actually utilized from something imaging catheterization some type of a test and so when you can compute that without that test they understand that but of course you could give them a lot more they don't understand that and they don't know what to do with it and you you can't ram it down anybody's throat anytime you you build a technology or a product you have to figure out how it's going to be consumed and take it you know across the finish line and that's going to be a big I think that's going to be a bigger challenge in most things because you're dealing with people that fundamentally do not understand what you're doing and on the other hand as a computational scientist you fundamentally don't understand what they know and what they're doing so somehow that gap has to be bridged and I think the key in in that area is for partnerships and collaborations between clinicians and research engineers and scientists to build technologies together that are utilizable and can be brought to brought to the clinic otherwise you know all the technology in the world doesn't do anything if people can't easily use it has to be easy and fast because they work on a different timescale so following up on this do you think it's going to fully achieve these do you think we're going to have to train completely train from scratch a new generation of engineers who have in their core skill set like a very strong interdisciplinary type of knowledge I think already you're seeing some things like that to a certain extent I mean there are computational medicine programs right now we're starting one ourselves in Texas and so you want people to be skilled at scientific computing you want to be skilled in mathematics and understand basic physics and engineering principles and all the technologies but you want that to have a strong application area in in medicine and so people are going to be trained in that I mean you can't be trained in everything and throughout your life you have to retrain yourself on new things because things keep changing but I don't know that everybody has to have that kind of background I mean you know problems in the real world of sir solid usually with teams of people it's not just one person sitting in a corner with a pad and a pencil you have teams of people and people with different skill sets can be brought to bear on these types of problems so you might be a civil engineer and you can work in medicine as you know you know and in fact you know if you think about the history of medicine probably the greatest contribution to health on earth was the development of clean water systems that was done by civil engineers so you can you can play a role but you have to partner you have to be a good collaborator you have to enjoy working with other people that's the real world you know okay so I like now that you all of you share an anecdote from your life so as you know many of the researchers who do computational work including me at some point they face some resistance against you know computational work and trust in the models and the computation so I would like you to share an anecdote when you have to face this resistance I've got a lot of resistance stories I was a member of the resistance you know I first started doing finite element research in the in the research and development laboratory general dynamics electric boat in Connecticut and I got excited about the idea of finite elements and and talked my boss into supporting activity in that area so within a year we had developed the funny anecdote is he started the finite element development group I this is before I even did my PhD so he put some guy in charge of the group and I was the group but one year later we had a 57,000 line code solving many many problems in the laboratory and there was an analysis group in there that was using very archaic technology and it became a real battle it was a war and the director of the laboratory said that after a while he would decide on the future direction of development and research whether it's going to be the finite element method or these older technology that I won't mention and he said he'd decide on a Friday afternoon and on Monday if you were not prepared to sign up for the direction you could hand in your resignation this was an exciting day and so this other group was led by you know the resistance he was really resisting finite element technology but the director of the lab said five o'clock then he went home he said we're going to be doing the finite element method in the future and so this guy who had fought tooth and nail was driving me crazy trying to obstruct everything I was doing on Monday was an enthusiastic finite element man okay maybe more anecdotes I'm sure there are many of them one little story not as exciting as Tom's story but one little story I have and I worked in industry I was doing a lot of failure and detection problems root cause identification problems and I worked at IBM there was one mainframe that failed in Rome and that went sent the company into panic they don't want your main frame ten million dollar mainframe to fail so we had to do some root cause analysis and a colleague a metallurgist colleague and I we always work together in groups as professor Hughes said and we thought it might be a problem with some specific process issue during fabrication of circuit boards so I had some finite element model which had some fancy contact nonlinearity etc etc and I showed it to my first-level manager who was a who has a PhD in mechanical engineering but in thermo sciences he was a believer and then I showed it to my second-level manager who also has a P who also had a PhD in mechanical engineering but in some experimental mechanics he didn't believe anything that I showed and he was very concerned about my going and presenting to this senior VP level person would be assigned to oversee this problem because it was costing the company millions of dollars a day to shut down the assembly line so we had to go present every day to until we to show that we had understood the problem and we have solved it so my second-level manager was very worried that I might be sacrificing the entire department by presenting some results which are not correct so fortunately it turned out okay because the third-level manager who didn't have any PhD but he was convinced and we went and presented to the senior reviewer and everything worked out okay so I think we come with biases I think my message is we come with our biases based on our training and somehow computational mechanisms seem to have more bias against and then experiment my PhD I was hearing from one of the more seniors that if you do experiments everyone believes in what you're doing but you don't believe in what you did and if you're doing computations no one believes in what you're doing but you you're the only one believing in what you think so for few weeks I was thinking maybe I shouldn't do computations maybe I should do experiments but as time passed I learned that well what is most important is the experimentally validated computational models that came up with and everyone believes in it and life is good so you know maybe following up on that on perhaps a little bit on the other end when I talk to people in the engineering community the broad engineering community be it academia or industry somehow the general belief is that everything in the computational methods area is kind of done and known and is already packed into the commercial software programs than anyone can use and you know my opinion is almost the opposite to that I would say because you know that there are a lot of I see a lot of at many different levels a lot of computations that you cannot really trust at all that they are boggles at many different levels not only numerical analysis but even physical principles so my opinion is quite the opposite I would say and I would like to know what what you think about that so you know I get this question from students sometimes that when I'm teaching a finite element course you know what software are you going to be teaching and my answer to them is the best software best finite element software that is which is going to be your pseudocode that you will code up in matlab so no finite element software I believe will give you you know the the absolute correct answer to a problem to a real problem but you know if when you get to code when you get to understand the the equations that you're solving and get to code them then you really have a better understanding of the limitations of your model so and maybe you know a quote that comes to mind from professor George box who's a statistician was that all all models are wrong but some are useful and basically what that says is that you've got to understand the limitations of your model and when you apply it you stay within those limitations and life will be good if you try to push the limits apply a model where it's not applicable which is what I feel you know has led to this mistrust of finite element models people trying to apply it to situations where they should not be like where they are not validated I don't know the much about the application engineering but the interim of a computation I do have a go to the lab quite often still one of the problem haven't been really solved is the accuracy control of the simulations once we have the simulations we even don't know how good is our simulation is so in particular for the very difficult problems that's we basically don't know much about it in the mathematic community or in the computational community there is a field that exists for 40 years we call the opposite or error estimation which can help for the simple problems but for steel for the hard problems our basic principle does not apply so this is the one I think that's for one reason is the 10 years ago or 12 years ago the DOE was talking about predictable computation which is basically how do we assess how accurate is our simulation so we can trust and we can use that for design or prediction one of the aspects that I think continues on is this question what physical reality are we projecting onto computation and I think we talk about elasticity that might be settled in the large domain but there are certainly conditions out there that have shown us that linear elasticity or local elasticity is not sufficient for each and every material that we income take a foam and that foam has a size dependence in the mechanical response and things similar things occur in plasticity and so we will have to be careful about on what physical reality we project forward and know these limitations on where our simulations are valid and worthy or not and that is a continuous battle that I think will not thankfully not end right and hopefully employ many of us long-term. The main purveyors of the line that everything's done are commercial software companies their marketing departments will tell everybody that they talk to that every problem is solved what you have to do is buy their software and if you believe that well you believe anything and if you if you want to get a good sense of what these people are like you should read Dilbert. Dilbert would give you a good insight into the marketing departments of various technology firms. Okay so maybe now we can take questions from the audience so questions from students. Hi good afternoon everyone I have several questions I don't know which one should I do maybe to professor Hux and maybe to all of you so I guess all of you have developed some good idea my question is how do you develop I mean something that is growing up in your head for several weeks several months or is something that just come out just all of a sudden so how is the process. How do you come up with great ideas. You know I think when when you're trying to you know develop a technology or you're trying to solve a problem you kind of carry it around with you and sometimes an idea just occurs to you you know all of a sudden you're taking a shower. My wife claims I take the longest showers in the world because I start thinking about things and I forget that I've washed my back 30 times you know but ideas just come at any time and it's I don't think this it's hard to describe processes like this they're not like linear processes there's nothing you can kind of really do you know you just have to be really committed to the problem it's the wheels are always turning if you are and then things occur to you in my experience. I had a good idea yesterday on the plane. On a plane anything is better than paying attention to what you're doing but that was a good idea. Any more answers for this question or? I think to me that's basically you keep trying and don't be afraid of failure so maybe the best thing that will come out of your your thoughts and research not all the ideas are gonna be good so you keep trying and failing and then trying again until it works. You have to work you have to work on things Einstein said what 98% perspiration and 2% inspiration or something like that I don't know which came first either. Professor Raisa I think you wanted to say something. Well also going to conferences reading papers staying current you need to know what's important what is relevant in the field what would be of interest to people you're collaborating with what is the problem that would be important to solve and as you keep thinking about that then eventually finally you will have this flashlight and you will get an idea. So we spoke a little bit I mean all of our lectures focused on problems that are very discreet that relate to using computers to model systems that are discreet but can we also throw some light on how do we use computational power to solve continuous systems and complex challenges like let's say we spoke about water so any light on that front my questions open to the whole time. Who wants to respond to that? Well I mean a continuous systems in some sense are the solutions are living in infinite dimensional spaces and the way you make things in infinite dimensional spaces tractable is you discretize and project them into finite dimensional large but finite dimensional spaces that's the finite element method finite difference method spectral method everything these are all analyses of continuous systems of equations but they're done in a discreet way you have to make the calculations finite to do it on a computer. I think we have another question in the back. So I appreciate the opinions and the thoughts from the panel my question is related to something we spoke about earlier about resistance to kind of simulations and models and things like that and I there's a lot of experience on the panel there and I just wanted to get your opinions of kind of because as engineers as scientists we are always seeking what is the answer right what is the best way to do something but more and more I understand that it's a lot more political than that especially when you go into industry and into you know working in groups because everybody has different opinions so what are your thoughts on how to best you know bridge that gap and help bring someone on to your side if you will and kind of how to work within those groups and how to bring people to your side if that makes sense try to convince them in a rational way and if you can't punch them in the nose. There is another question there. Firstly I would like to thank all the speakers for the wonderful panel discussion it was quite enlightening. One thing I wanted to get opinion by the panelists on is typically a lot of most of the stuff what we design as mechanical engineers is built for a like cyclic failure say like fatigue and probably like a few decades back we use empirical models or curve fits in fact to calibrate our models just from statistics and now we are actually using finite elements and other computational methods in conjunction with statistical data to characterize these models and all of a sudden we are at the phase where now a topic like machine learning with a whole probably the physics of the problem is like a total black box and how is this particular aspect of the problem going to evolve is there any way that we could like bridge everything together or is it one like something like a machine learning which is probably more efficient to predict but takes more time to train rather than some computational simulation which takes more time to get the results but we know the physics of the problem how are we going to address this yeah and I would like to open this question to all the panelists and so that I can get different views on thank you described as a balance of two things so itself is an optimization problem which you can optimize for to see like for what type of problems which which approach makes sense and I guess it depends on the how complex is the physics how how much data you have and like how you can how much of experimental data you have how computational expensive is your model and combination of those will set to the solution so there is no single solution for all problems but yes I think that we will see the fusion of well if you will imaging or experimental data and computational models you would have fusion of different modalities you would see high-resolution numerical models and low-resolution experimental data and combining them you can get the fidelity of experimental results or imaging data with resolution and accuracy of computational models so yeah it would be a combination of these different modalities to get better description of the problem yes another question testing thank you for the opinions and you've talked about some resistance from the other world now I want to have some advices from maybe aged minds like I want to talk about the resistance for your inner inner for your yourself because for us like PhD students we want some good publications if I become a professor I want to get promoted so I have those pedestrian stops that that is keeping getting my time but I also have something that I really want to do but so it's kind of something my real world obligations are pushing me back from what I really want to do so I want to just have some your your maybe you can share some experience like when you were young how do you deal with this situation I don't remember when I was young but you might be surprised to find out that these types of conflicts never go away you know I still have lots of things that I want to do but I can't do because I have many obligations to satisfy so it's nice if you can just ignore everything but the things you want to do but that's not life you have to somehow find the right balance or the best you can it's a psychological problem not a scientific problem okay so we have only a couple of more minutes so I would like maybe to conclude the panel by asking our panelist to give a very quick advice for all the graduate students in the room so if you had to give a quick advice to them what would that be do your homework so I probably would say that trying to understand sense trying to get more deep understanding then everything becomes easy yeah I second that make sure you understand the limitations of your model and then start simple and then go complex yeah I would say you need to know math because math is our language and you need to know your domain you need to know the area where you applying your knowledge you really need to know the systems that you are modeling or analyzing so it's an interdisciplinary thing you have to know your tools you have to know visit and you also have to know that I mean which is not easy but that's what what it takes to let's say grad studies of research find the topic that really fascinates you that helps a lot opinion is that you should value every experience that you can get and choose a wide variety of experiences because you will get your insights from something that's completely unrelated to your main topic just a wide variety of experiences my career I wandered around quite a bit started my thesis wasn't born by a mechanics and then I went to work for IBM and then academia doing computational modeling so I think every experience gives you insight that's very valuable later on okay so then I would like to thank you all for coming here and of course I would like to thank our panelists for their time and their thoughts which are very useful to all of us so thank you very much