Pioneers Talk #2: Meeting EAF Performance Targets Using Data Analytics
Welcome to today's Pioneers Talk: Meet EAF performance targets using Data Analytics. My name is Christoph Stangl and I will be your moderator today in my studio and my today's guest is Manuel Sattler, data analyst for Primetals Technologies. thanks for inviting me today nice to have you here it's a pleasure so uh within the next about 30 minutes manuel and i will discuss um that often the little things make a big difference when talking about the performance of your eaf plant and why this is the case we will have a look at the solution that determines improvements for eaf performance based on operation practice using data analytics methods and we will have a look or share insights how you can transfer generate information automatically to your automation solution system and why this is important as well if you have any questions to manuel satter suttler please just add a comment and uh input the question to this live stream on the desktop you can do it on the left hand side and on a mobile phone you should have an input field on the bottom of your smartphone later on we'll have one two sessions in between where we will uh read out some of your questions and i will ask manuel to answer them um so this is a power in stock interview so let's start pioneering well before we really start with discussion model i would like to show a very quick result of an online survey i did maybe we can show this result of the online survey to our participants you can see i asked the question uh if our participants already use data analytics methods uh in the sleep blend and as we can see about like uh more than sixty percent sixty-two percent said yes we'll use data analytics methods so if you're maybe willing to share for what kind of applications we would be glad um another like 12 percent said okay not yet but they plan to use data analytics methods in future 30 say they are not quite sure if they use it another 30 percent uh don't not interested in data solutions right now hopefully we can convince you with this interview that this makes a lot of sense for your ef performance improvement so thanks for sharing now manuel um you have developed a solution um that improves exactly the performance of an eaf based on uh operation practice using data mutex methods and you presented this solution uh in the another conference named parents connect 21 uh spring of this year um and the interest was quite big and so i thought it might be a good idea to take this a topic for our today's interview session so that we can have a deeper dive into the topic and that participants can really have a more interactive discussion with you and ask you questions um so to start so you told me just prior to this interview that usually when you have an ef just starting ef you just uh input all the rules and recipes for the operation and then it's like operating very statically and stable and and this is sub optimal but why is this the case maybe you can elaborate depends on this i think you're now talking about the steel making practice yeah so so when we when we have this when we go to the plant we make the startup and everything um we have to meet the performance values so for these performance values we make the optimal uh steel making practice uh which is based on on the on the materials we have at that moment so we have scrap dolo lime and all that things and they can change over time so um if these things change because you have more scrap qualities your supplier for your electrodes change you have different qualities of lime or toro you also need to tweak a little bit the steel making practice to stay at the optimum of ref and from our experience this might not always happen because i'm the one responsible for it are you know loaded with with other work for example i'm dealing with with quality of strep uh quality of the seal of the final steel and such such things so it's it's just it's just a minor part which can make a a good difference on the outcome yeah okay very interesting and i think you also told me once i read that that it's really a matter of i know minutes or seconds like scheduling the the the sequence of the baskets how they were actually i don't know preheated or inputted into the eaf so why is this crucial i was surprised that these kind of minor time frames really make a difference yes that's um we got some sample data and and it was really interesting and then we found that um for example if you have the first basket just a few minutes longer and also the second basket a few minutes longer so it was actually first basket was one and a half minutes long and the second basket was half a minute longer so in total two minutes and we could save it at the third basket we could save four minutes so it's for us it was kind of kind of surprising that the real good timing uh really makes it makes a difference in in the tap to tap time or the power on time depending on what you're focusing on well so thanks a lot for sharing this these insights i mean i'm not an af expert yeah but i think most of our participants are um so to your experience i mean this this this this kind of sensitivity that these details are so important are like the operators of es usually uh aware of this or you see some some know-how which i feel fades out now over time because maybe the more senior operators just just fade off also i think the operators are we are are aware of that so i think everybody is aware of that the good timing of the baskets um is is really crucial for for for for the success of the heat but you know every heat is kind of different um so so you have you have different scrap recipes and all that things and and so it's always uh it's also for the operator pretty difficult to to find the perfect spot when to charge and when not to charge and and you know timing is crucial even for for gas oxygen and and and the slag builders and all that stuff so yeah timing is crucial very good and and mana is this something you experience which is specific to eaf operation or also for other aggregates so usually for other aggregates you have most of the time you know what is coming in so if you have for example an ef route most of the time and level furnace follows so you have a sample of the ef and you totally know what's going into the elf so there's not much surprise on the other hand if you put scrap into an ef um there might be some surprises so so it's it's very difficult to to get a real model there which is always working perfect and and you know it's that's one of the things and yeah we thought we want to solve this this problem very very good so thanks molly for this first year in the first insights um well if you have uh joined our live stream a bit later on recently my name is christoph schrange um marketing manager at parameters technology head over today's live stream pioneers talk interview with me is manushatla datalust also in parameters technologies and we discuss how to improve ef performance and to better meet targets using data analytics methods um you have any questions during this stream uh please just input them as a comment uh as it used to it when you just make a post a comment to a regular post on social media later on we will try to read out some of the questions and manuel will answer them um normally we've just discussed about this like uh problem or this challenge or this importance that the or the potential what could be uh how to improve the performance of an eaf and now you have developed um a solution called eaf heat cloning you also presented already at this conference in april this year maybe you can just describe a bit more how this solution looks like and how it works yeah sure and i think we also have a bit of diagram maybe we just can display this diagram please also work with this image yes you sure the diagram you can see that that at the beginning we we are clustering this crap so not much rock and science we we cluster the scrap recording to quality and scrap weight so nothing more so we get this cluster where we know that that we have scrap qualities and in the kind of range of of the uh of the tons of scrap which which are going in so it's you know it's just a clustering that we know what what is going in to the 2df and then we make for each cluster we then find the reference heat and we think reference heat is one of the most important steps because it defines the outcome of our search for example if you want to increase productivity of your ef this reference heat should also reflect that productivity approach so it should be a highly productive heat or on the other hand if you want to improve for example costs or something like that the reference heat should be very you know cost efficient heat and the other steps also three and four are usually there to to really check if this heat is repeatable so step three um we look at the aggregated data so one would say level two data or reporting data something like that so we find these potential heats where where these gold markets or these similar heats uh are in and from that potential heat uh or from that potential pool of heat we really go into the time series of data so so we check uh really when was when did the operator push the carbon button or when was uh the the gas or oxygen activated and and all that things and we compare it and and with similarities we find this we find this last cluster and and uh then we can see okay this reference heat was produced 10 times 15 times and it's realistic that we can reproduce that and so from that we extract the best practice and at the end it's more or less just a table which you can put on the on to the to the operator and say okay let's try it and do that and that and check the outcome and should be better very interesting so uh this looks like very comprehensive to say and very specific what they have presented us thanks for sharing the diagram um now as far as i understand it's like it's all about of course getting the data to be able of course then to apply the solution right uh and to experience how easy it is to get access to the data from sleep producers because i know this is quite a sensitive topic so it's easier difficult what is your experience yeah you know it's it's um you always has to prove what they're doing with the data so so if you can prove that the that the zebra juice has advantage it's normally pretty easy but when you say okay i want data because just to have data and nobody knows what is done with the data it really gets difficult and and so you know we have to transparent approach that we really can check what is done with the data step for step so everybody can follow it and and yeah that's very good and and to experience uh i mean i know that there are different kind of like concepts how to to collect data or ssd producer and your experience is more often that i only have data in a data lake without any structured or like we're getting there from a data house or people with the data mart structure or other other other types of of data interfaces yeah so so usually we don't care how we get the data yeah so if we connect a data lake or to a sql database where everything is really nice structured um it's it's you know we have a very flexible approach we need to get somehow the data then we transform our data ourselves or customer can already do it if if you can say okay this is this is what he's interested in um and and if he thinks there are some some data signals which are very important he can spare it out and yeah that's that's the thing so we're pretty we're flexible and and uh the only thing we really want or or needs um is is that the data you know the quality should be okay and the second thing is um it should be really comprehensive so most of the data should be in there so that the results can be used and would you actually then for example support sd producers um to make sure that you have like the proper data points plus uh to validate them somehow i think that they have the proper setup okay now this is fine to be a good basis for the analysis of the tutorial you're sure you have to talk sure there's no way around that i think okay perfect and the solution presented uh they have these five steps is this normally one solution or is it a combination of solutions that execute the steps there are some mineral steps for example defining the reference heat but the search steps are more or less automated so so um you know we can tweak it a little bit to say okay for example um electro consumption is more important so we want to focus it a little bit so we can put some weights on there and and so on so we can tweak a little bit around on that but that is more the last the search steps are automated okay super it all sounds good one point i would like to come back because um at the beginning we said it and we pointed it also out in your description that that this eaf heat cloning uh solution as far as i just got it from your is is really based on operation practice so it's not like uh out of box or standard or i don't know some some advice which which comes maybe from a powder like us it's really based on operation pictures uh maybe you can just describe again how this works and why this is important to your opinion okay so so um the thing is you know we know how to build plants and and our customers are good in operating plants so we don't want to have some some some kind of model in that direction which is predicting something with a you know if you have a currency of of 95 or 99 it's pretty good but um that's that's not what we want because you have to five percent of failure rate so um we think that our customer is really good in operating plans and so we just look into the data and and check what happened in the past and and so what we extract from there is it's not something which we fantasize about it but it's just a little help for for our customers to see what they have done in the past and and what they have done good in the past and if they can maybe transform it into the future so that's that's the approach we have very good um so what i'm also interested in um is is that um look maybe we have some comments not so far so if if you want to add any comments uh please just uh use the chat chat function of of our live interview session i mean we received two comments to our chat in between i will try to look them up a little bit later on yeah um so maybe to proceed let me quickly try to summarize manual but i understood so far i i'm a metals expert the human metals expert and our participants are so what i understood so far so i understood that operating eaf is whenever it's stable or a static process no that's team making practice okay with which you know um you're still level two and c making practice is is somehow in the background and it says the operator when you push the buttons it just says to push the buttons okay and the operator can decide otherwise okay okay and and the thing is that steel making practice um is often not not adapted but this means usually operators do not do it they just stick to the extended operation mode yeah because it it it worked in the past yeah so why shouldn't it work now yeah and um so this means that uh it's it's it's rather fine that operated but there miss actually a potential to improve the performance to improve productivity or improve profit or quality whatever yeah you know in in my feeling uh ef must have some kind of targets for example if steel prices are good you need to produce as much as possible to get to get the profit out of it so you want to focus on productivity so that's what we do then with the reference heat on the other side if the steel prices are not so good you you want to produce as cost efficient as possible to get at least some some profit out of it and you don't care about how much you produce because steel braces are bad and you just want to be as cost efficient as possible to still have some profit very and and that's what we try to achieve with our reference heat and we can also with our tool we can easily switch that so we can extract uh this smp's seal making practices within one or one and a half days and then we can say hey here here's our here's your new new plan and how to reduce and you can switch to target okay so it might add some flexibility on top okay very very good and um you also state that yeah or just a question i mean if you're now a very very advanced senior operator of an ef and really have like a really bunch of years experience uh why isn't such a person not able now to to to to look at the performance of the data and analyze it why do they need this kind of tool yeah i'm not sure if everybody needs this kind of tool but you see this seasoned operator will also you know retire at one point and or or has has new operators and all the things who needs to learn over the time and and operators in my experience and they just get better over time and over experience and so we can we can um so if if this one operator you talked about is is really good and we will find his heat in the data and we will make it a best practice so so we will extract his his knowledge on how to process on certain steel grades or or anything like that we extract it out data and and deliver it to to all the operators and not just one okay i see this means does this tool then also serve as a solution to collect ensure and grow corporate know-how that's a good question probably yes yeah yeah if you think of that yes sure because um if there are good heats for example one one operator i don't know retires and you change in you change the target of your ef and then he was for example good at cost efficiency and the last three years you uh focus on productivity and and you find this data from five years ago something yeah and you can renew it and put it back in into its operation again cool cool because i i think to my impression that this is really a big topic currently i'm not only seeing producers i think all industries have this problem that um experienced people retire and and it's difficult to find anyway highly skilled people uh to to to replace them um so uh i think it's an interesting side aspect so uh we have received some comments maybe to try to to go through them um so one is asking for example okay can one apply this this heat clone this heat cloning as a technology also for fox finger shaft type furnaces actually unknowable disease but uh yeah fox finger shaft um it's it's it's the it's the um how do you say the predecessor of of the qf quantum uh i see i see so is this applicable there now this is this long heat cloning so i know that the finger shaft is is the that it is there but i'm not sure how much hot heal it has oh you see if if if the thing with hot teal um it influences the heat yeah the outcome of the heat is pretty pretty good and and i think if if if there is a constant hot deal um i think we we could apply that there sure okay cool so if you're interested please contact miles out there of course later on just another question i mean maybe it's a bit tricky about that let's try uh do you have any any uh data you would like to share i mean it's up to you um how much the efficiency now could be increased by a heat cloning app i mean it's like five percent two percent twenty two hundred it's not possible yeah um the thing is um we unfortunately didn't manage it to put into practice now so we we got some data from from some real data and we we made our studies on that and one of the reasons we are doing this today is maybe to find some interested ef operator who is working together with me or us um to to see how much it really takes or how much you can really get from that out and and yeah to to to prove that yeah okay very good so um i think it's an important point uh um so if you're really interested now to to utilize this technology to him up for the performance of your uh furnace contact manuscript on the one hand i think the easiest way is only way to contain by linkedin because this is the good thing all of us now attend this livestream for sure on linkedin because this is the only option to attend um um or of course you can leave a comment to this livestream or the recording if you watch the recording later on because this will stay remain on linkedin and if you're interested please leave a comment and i'm able to make sure to connect you to the manuscriptlet and then later on right um there's another question very interesting uh stating like currently electrodes are extremely um extensive or expensive maybe most likelier can the methodology you just present a trusted process to minimize the electrode consumption i think yeah consumption it depends on the data you know um usually electro's awaited a weight i don't know depending on on the operational practice um once a week once a day something like that so it's pretty hard to really say which heat had which electro consumption um that that's that's that's i would say it's i wouldn't um um how to say it i wouldn't say it's it's impossible but with our method it's it's probably not that feasible i think okay because we're really checking into the heats and if if you just have an average consumption of of electric consumption it's you know i'm i'm not sure if it's work but i don't think that it will work yeah okay very good also thanks for sharing this kind of limitations maybe although what is the best case maybe for the application there's another question just asking is there an automation algorithm behind the solution or have all the steps of the heat cloning concept to be done manually um so it's really hard to have automation concept for the reference heat because um you know as i said it has to it depends on the target of df but mo all the steps so so you can also have this this this clustering done automatically everything and we also have the search algorithms automated and also the the extract of the data so so you really get very nice curves so um we just finished with that um three weeks ago and it really works good um but the reference heat um it's one of the things where you still need to need a human brain and an expert who says okay this this this is the path we want to go yeah isn't this any very often the case to say that okay if we have big data what is obviously the case in this application you need data analytics methods so just determine patterns get inside just for an example if if you look at level one data for three years you have i don't know if you have all two seconds one row um you have millions of rows you you have to automate it because otherwise it's not possible yeah that's also one of the things we have this two two steps because um um comparing the time series of you know ten thousands of sheets on on on time series data um it's it's uh you need a you need a pc cluster and and big pc machine and so so we have this two search approach to get the potentials and then then the time series so i think there's also no way to use excel for that purpose or is it i think excel ends at 1.2 million rows okay and still it's a lot but there's something efficient for this application very cool um there's another question um which type of data analysis uh is is okay or is it used for this technology is a statistical neural networks or any other so so the thing is um neural networks always make a model they predict something and that's not the path we wanted to go because if you have a model which is predicting you always have this error and and the in in in situations which they don't know they're predicting sometimes absurd things and and so we have statistical models which which are just based onto the past and we extracted out of that and then we have some some machine learning algorithms in there but they are very very small i would say okay very cool um so if you have further questions please uh keep on typing them into the chat we won't be able to answer all the questions there are quite some of them thanks for your interest but mother and i will make sure that the questions we will not answer now in our live stream we will answer later on by just making comments to your comments and we stay reconnected by linkedin so it's easy to follow up for us uh one maybe last question in this live stream uh what are the prerequisites for the solutions he states that the participant we might not have the data in one location for example it might be next or other data in the data history so is is this possible then to connect to your tool as i said if it's a date delay this is excel something like that we can deal with so so we have to put it in one place at some point yeah but but you know that's not the problem if we have different data sources and we just put it in one place and then we continue to work with that okay cool cool so do you you do also provide them this kind of a little database yeah we will put it into the data warehouse um no data warehouse yeah and and and then we we have put our solution on top of the data house and and so so that's that's the way so we have to get it somehow into our data data yeah okay so i think we're now really close to the end of this live stream thanks a lot for joining this first live stream pioneers talk interview i'm very glad that mano sattler was with us status as epi metals technologies are very glad about the insights you gave us about the solution we showed about ef heat cloning if you are not interested uh applying this solution to your furnace or would like to have more details please just either leave a comment to this live stream if you are seeing our recording of the live stream please also leave a comment reviewing them regularly or please contact manusata directly via linkedin uh if you like our full month please also be free to leave a comment on the prime stock interview format or our studio uh our next upcoming pioneers talk will be on the twenty on the second of december this year entitled from automation to digitalization and if you want to be sure not to miss any upcoming pioneers talk just follow our hashtag pioneerstalk on linkedin or follow myself and i will make sure that you will get notification will most likely be getting away automatically thanks again manuel thanks thanks a lot for being here yes it was a pleasure for me and thanks for visiting and goodbye bye you
2021-11-29 01:44
