Lecture 38 : Part 2

Students welcome back to the part 2 of lecture number 38. So, we have been talking about different kind of databases and I told you a materials and the three types of databases available traditionally material behavior which is giving you mechanical physical functional properties and then you have also databases from thermo chemistry or thermodynamic data sets which can give you specific heats information and enthalpy or calorimetric informations and then you have a crystallography data set which can provide you crystal structure you know lattice parameters atomic location etcetera, etcetera. So, this three databases are the ones which are available which you can access obviously by not free of cost by paying some money, but this are available this has been developed over the time scale. So, let us now discuss how we can go about it you know I will go back to Michael Asby it is not because I like him, but it is because whatever he has done is phenomenal when you talk about materials behavior you have to discuss different types of materials metallics, ceramic, polymeric, glasses, elastomers and combining this can gives you hybrids or components right. So, that means it is a massive number of you know materials to collecting data or creating a database of different properties of different materials it is not a very easy job it is rather a very difficult job. So, we always undermine the material world material world is very big and when you talk about different alloys god knows may be there are 10,000 alloys possible out of which we are using probably 2,000 alloys in our real life steel itself has different types then you have ceramics and polymers, composites, glasses, elastomers right. The specs which I am working have working wearing has glasses as well as the frame is a metallic right. So, you can see that different combinations of materials are used. So, now, the question is if that is the case how do you understand the behavior of these materials well obviously Michael Asby proposed material selection charts and that is the way better to represent data and this is visualization. So, it will allow us visualization other properties and we always like to see things in front of our eyes we do not want to see things in our mind right that is why visualization of these material data are very important and it can show different types of materials lots of different type material we can actually even drill down to specifics and we can always find out balance center properties this actually rough ideal for the first rough cut. Now, you can say that, but they are good for making selections as you know the Michael Asby's work is. So, good in those days and this is consider one of the systematic efforts to address the challenge that we face when you talk about data or developing data sets and this is done by Michael Asby and co-workers long back 20 years 25 years ago they have shown that by merging phenomenological relationship from the established properties with discrete data on specific materials characteristics one can begin to develop patterns of classifications of the materials the hidden patterns can be found out. So, I remember when I was a student I have done this course on material selections in mechanical design and that is how I was exposed to this concept called this material selection charts and then I was amazed how it was thought about it how data are obtained from different sets are different parts of the world and the selection charts are made. So, the visualization provides you lot of idea about how to balance the properties how to make selections this is something which you must have heard about it right. So, what are different properties we talk about it many mechanical means strength modulus ductility toughness physical means density melting temperature electrical means conductivity resistivity many more we can also talk about the acoustic the appearance texture color processability if the material is not ductile you cannot process this by deformation processing moldability we all talked about all these things at the beginning of this course and last but not the least cost is also a factor when you talk about material properties right because you are design and discovering a new material for actual applications the cost is a factor. So, where from you get the data there are four sources of these material properties first is a text book you can also have data books manufacturer literature and internet sites that is very important these are the different sources of data you can get it right. So, now let us discuss one by one out of these text books they are good for general information some of the text book will provide you tables of properties like the text book or mechanical behavior of materials by George G Dieter will provide you tables of some properties right, but you know they are not good for detail specification and properties they may give you some ranges of properties. So, therefore, text books are good for general information data books this is the one of the quickest source of detailed information to give an idea Smithell metals reference book you know this was edited by W. F. Gell and T. C. Tottmayer long back now you can get eight edition also in two thousand twenty twenty two or something. So, this is one of the quickest sources of detailed information I remember when a student who is to use this Smithel handbook very often find different data or properties of materials and they usually contain grades and specification as well as properties they are small but perfectly filled form pocket books that is why they are easy to navigate. So, this you understand whatever we are talking a new subject of informatics people have or our forefathers have discussed about data books we always used to think about it, but nowadays nano pass maintain data book. So, nicely this is one of the problem of using mobile phones and computers then the third one text book data books third one is manufacturers literature they are not very useful because they are variable in quality and usefulness manufacturers will advertise their own products. So, they will focus on certain things often only cover their products usually do not compare materials. So, therefore, can be biased, but you know they can be good for final selection before ordering that is what it is otherwise they are not very useful please understand that they are not very useful in the sense that they are kind of you know advertise that is why then come the internet which we all are so much you know glued to every day and night internet can be a real minefield lots of poorly presented information you do not know which one to take which one to filter out Google searches brings up lot of rubbish yes that is not my sentence that sentence of many of these scholars also many times you will see that the searches will provide you authentic informations you need to dig deeper to get authentic informations and many cases it is hard to find even technical informations. So, therefore, best is to use non-commercial sites and Google will take you those non-commercial sites at the end after you probably some 3-4 pages you will find those non-commercial sites, but we always like to see first 10 hits first 15 hits in Google search that is why internet is not a good way to get information now look at what Michael Asby's book can provide well Michael Asby has given this material behavior mechanical physical functional property in terms of charts and that is his noble idea because charts I am going to discuss about that in quite a extensive manner in the next 10 minutes or 5 minutes or so 5-10 minutes this is this is the you know chart between young modulus of density of different materials and if you look at he has used classification algorithm right he has classified the different materials like foams woods and wood products polymers rubbers porous ceramics metals and alloys ceramics components. So, you can actually have a nice correlation of these properties a different materials having the young modulus and different densities now if you look at carefully here that is what sub classes are given look at the foam these are actually rigid foams this is cork flexible polymers within the foams sub categories are given similarly within polymers even different sub categories are given like silicon elastomers polyurethanes many others. So, he actually methodically collected data authentic data from different sources and classified them put in terms of charts. So, this charts actually source of information one can actually gather information material property information from this charts very easily nowadays you can get in terms of CD or so or CD ROM or may be in pen drive you can get this data and use it. So, therefore creation of data looks like is very, very important for material informatics and this is authentic source of the material selection charts authentic source of the data. So, please keep in mind that once if you get time go through this book by Michael Ashby material selection in say material selection in mechanical design another chart called strength density charge look at it density is basically in this unit meter cube strength on the right on the y axis ceramics are the strongest one that is why they are at the highest composites are mixed of ceramics metals or polymers that is why in between them then you have woods forms all these things on the right I am showing you best combination of data. Now, you can also see the different lines are drawn these are basically correlation analysis which will do the next lecture how to do a correlation analysis of these kind of data sets and these correlation analysis are considered as a guidelines for like minimum mass designs. So, as you understand when you talk about design you are having a set of properties given to you and then you have to select a material or you have to design a material or you have to make a new material or discover a new material whatever that is only possible when you have data. So, these actually charts are providing you lot of data and they are not only important in terms of providing you correlations, but also in you know providing you classifications. So, classifications of the different kinds of you know models we talked about it when you discuss machine learning remember at the beginning and of machine learning classification is one of the unsupervised learning techniques, but here you look at it he has classified this if you look at this chart is much better he has classified this you know woods, foams, elastomers, polymers, metals, composites, ceramics very nicely. And this is a set of materials in a materials we use in day to day life and if I want to have a strong material which are low density that is what is for space applications. So, I should be searching density between 1 and 2 or 1 and 3 right I should not search densities of 10 or 15. So, therefore, then I have a limited choice window and then if I want to have a high strain like more than 100 GPS then I have only set of metals, ceramics, composites only that is what is the main idea of this charts that will allow to select materials based on this set of properties. So, let me move ahead on this, this is on functional property like thermal conductivity versus electrical resistivity as you can see here thermocouple activator metals are normally very high as there is electrical resistivity is also very low we know that. So, they come here as a one elliptical area then you can also have ceramics and composites polymers elastomers. So, polymer elastomers they are normally used as a insulators some ceramics also. So, they have very high resistivity 10 to the power 20 10 to the power 15 to 20 right, but composites can be having combinations of higher you know high thermal conductivity not high, but medium type thermal conductivity with reasonably good resistivity. This is what is subclasses you can see as in the metals from my class behind steam is showing copper alloys has the highest electrical thermal conductivity after the aluminum zinc tungsten magnesium steels titanium alloys has a much lower thermal conductivity that is why for any thermal application we do not use titanium alloys we use mostly copper aluminum alloys although titanium alloys may have higher melting temperatures, but copper is probably the best one because it has higher melting temperature as well as higher thermal conductivity and it is it can be used for a application where you also need to take care of the electrical conductivity. So, that is why copper alloys are the best candidate for electrical thermal applications you may say that we know yes we know about it, but it also talks about it if you want to find out some alternatives of copper what all you can choose next alternative should be aluminum that is what is done in air conditioners if you have seen air conditioners the you know inside in the air conditioner kind of pipes they use are basically made of either copper or aluminum similarly, wires can be either copper or aluminum electrical wires. So, these are all experience based, but the theory or the data science also provides a strong confidence on that. So, that is the one which Michael Asby's charts can provide. So, I am just showing you a tip of the ice bar his book is going to provide you details of that and these charts are made after lot of struggle in getting the data from different parts of the world that tooks lot of time probably for the team to do that. Now, let us go to the next set of data right what is called thermodynamic data sets thermodynamics data set means we want we know that important function in thermodynamics gives free energy which is nothing, but H minus TS enthalpy is enthalpy T is the temperature and S is entropy. So, we can always write down this is C p d t and this one is C p by d t right. So, what you need to know is the C p heat capacity also you need to know latent heat that is what calorimetric data this is called calorimetric data. So, most of the thermodynamic data set is will have that and then you can build free energy data sets free energy things right. So, this is taken from thermocal and you know thermocal provides you lot of databases I am just listed few, but the actual thermocal products will provide you many databases steel iron databases nickel alloys aluminum magnesium copper base alloys titanium molten salt defectory alloys lumen metals like platinum palladium gold silver silicon base alloys semiconductors oxides slags high entropy alloys holder alloys cemented carbides zirconium alloys high temperature materials ultra high temperature materials permanganates minerals general alloys and substances nuclear materials to list few there are many more actually. So, these databases available you can buy them actually you cannot use it free of cost because these are expensive because that is how they make business. So, as you know with these databases you can use a CalPHAR approach to design or get phase diagram informations right. So, what do you do that and CalPHAR approach you have thermodynamic functions G H S P you can see that thermodynamic functions and then you can you can actually have generate this thermodynamic function from the stored databases you can actually adjust various free parameters of the thermodynamic functions and using different models models will have adjustable parameters and this adjustable parameters this model actually can be obtained from quantum mechanics experiments or estimates also we have talked about it these aspects thus require some kind of optimization and you know once you optimize thermodynamic properties then you can use for your calculations. So, remember using only thermodynamic data may not allow you to get you know good informations from your systems many times you need to do lot of optimizations and that is something which we talked about it when you discuss about different length scale issues. So, finally, you get phase diagrams phase equilibrium informations and you can make a graphical representations of that that we material scientist know very well that graphical representations of this phase equilibrium phase diagram informations are there. So, CalPHAR approach does that as you know free energy is a function of these plus excess free energy that is why you have lot of free energy parameters for the these first two terms you cannot do much only third term you can do lot of adjustments. So, and then you can get a you know assess the data optimize it to get a good free energy estimates and this once you get that and you can use this free energy information for thermal calculations or calculation for phase diagrams you can do diffusion induced phase transformation calculations or you know you can do even both that is what done is prisma now a days these are all built in softwares by the way by thermocal and you can you can actually get this softwares when you buy the databases from them. So, remember other than thermocal you also pandat which is also a new company not new they also selling databases and all this software also do that. So, now third one is called crystallographic databases and this is a comprehensive literature coverage from one twenty eight thousand original public this is taken directly from peers and crystallographic home page I am just copied from there I do not want to distort the information what they have given it. So, they can it can provide comprehensive literature world literature over one lakh twenty eight thousand original publications and it can provide you phase informations like crystal structure chemical structure systems unique formula it can also provide you various kind of crystallographic data like atomic coordinates crystal structures their lattice parameters y cups notations and many more you can go through this literature and now a days we can get this you many of us has this data sets available in our libraries. So, therefore, crystallography data sets actually are very unique and they provide lot of insights into the different kind of crystal structures which forms and this data set actually every time updated because of new discoveries or new concepts which are coming up all the time always. So, this is the home page of the crystal impact dot com and it is by P. A. Villars and Karin Sensual they are editors you can buy even this hard copies and they provide three lakh ninety five thousand that means three point nine five lakhs entries almost close to four hundred thousand entries and this is released recently two thousand twenty three twenty four. So, you can imagine it is a huge database one can use the database to do lot of calculations crystallography do that and this is one such snapshot directly from this home page or this is their product catalogs you can see a complex chemistry can you see that this complex chemistry they are analyzing it. So, you can analyze all kinds of chemistry look at crystal structure atomic location everything information is possible whatever you can get it well that is about the available databases. So, can you create new databases answer is yes that is why I taught you NLP the natural language processing. So, students you have learned lot of techniques of you know databases different types of databases you can also can create your own databases right by using what is known as a technique called NLP. So, that is what is called use of NLP in database creation. So, NLP can be used to create new databases sometimes you can create your own database and use it right that is what is called NLP and you know as you know machine learning is a versatile tool it can do all kinds of things it can do image recognition it can do email spam it can also do online fraud deductions. Similarly, it can also do speech recognitions or many other things which is including the natural language processing what is called NLP it makes your task very easy by the way. So, how it is making easy well if you want to do experiments and generate data and use it analysis that will be time consuming that will take lot of resources on the other hand if you can generate data using you know different kinds of you can develop a model and then you can predict it then it will be fast very fast time efficient resource efficient and that is the new the hour because we want to accelerate the design and discover your new materials and that is possible once you have machine learning based models possible. But you know underlying fact of having machine learning based model is data and without data you cannot do modeling right that is why data to database what we have talked about it in the earlier part of this lecture. So, how can you generate data? So, we call it design to device we can take scientific literature what are available in the globe in terms of publication in terms of reports in terms of books or many other things that is part of scientific literature extract the information and compile it once you extract information you have to do data preprocessing also because there may be some missing data wrong data some outliers then you compile it once you compile it you get data set right from that you can predict you can validate whatever you want to do. So, these two from the generation extraction of the data from scientific literature to extraction to compile that is can be done by NLP right that can allow you to create new databases. So, what is that you can have articles scientific articles you can have you know papers you can retrieve information from that then you can also have databases as we have talked about it. So, from there also you can find out relationship extractions and finally, you can do data preprocessing and technical evaluations right that is what is done. So, how this extraction of data are done that is what is done by NLP right. So, NLP uses different phases you can use morphological processing, syntactic analysis, semantic analysis, discourse integration and pragmatic analysis to go back to my earlier lecture on NLP. We can do all these kinds of thing, but most important thing what you can do is what is called a name entity tagging because if you want to create database let us say you want to create database on mechanical properties of steel. So, that is your tag and we can this NET identify allows you to identify categorize and tag name entities in data. So, it if you create this name entity it will go and search that in the literature identify it extract it and classify it. So, purpose of NETB is to automatically extract the structural information from the unstructured text and that allows machines to understand categorize entities in a meaningful manner for various applications right. You can create in terms of graphs, you can create in terms of you know plots or you that is that is the idea actually and how does it work it provides you NEP tools NER tools or rather name entity tagging tools like name, country, company or whatever for any person you ask and it creates this tag words or phases basically that will be then searched in the literature and once these tags are found you can extract information that is something which is widely done. So, there are different kinds of name entity tagging like let us say material word embedding you can have this kind of software word to VEC, fast text, ELMO. Annotated material text it can be plain text synthesis of paragraphs or materials like text mined curie or nil temperature curie or nil temperature it can mine it this is found from ELISA. So, there are many ways actually one can do that as so once you do that you can develop a query search on the downloadable papers and then you can create you know what is known as a database and that database can be used for all kinds of things analysis like you know you can do classification, you can do recognitions I do not want to go up details of that. So, this is one such examples so you can have you know corpus of 2.5 million abstracts 4 million abstracts and using this NER you can actually pick this poly word from polymers and create you know lot of materials obtain lot of data that is what I was trying to see this is one example. So, one can take many more examples and show that but that is not the purpose of this lecture what I am trying to say is that if you want to create your own databases you can use NER and technique and generate databases. So, let me stop it here thank you and we will get back to lecture number 39 with more analysis on data mining that is what is the next part of lecture data mining data analytics where we talk about it how data mining can be used to predict the different kinds of trains and along with we talk about statistical correlations what is known as a data analytics. Thank you.