good afternoon ladies and gentlemen welcome to today's backs asia's technical webinar series presented by conserve it and hosted by bags asia my name is patsy again and i'll be umc for this session we are delighted to welcome everyone back and hope that you have obtained insights from the past few days with hearing from many of the region's leading solution providers in asia's built environment before we get to the session proper i would like to encourage everyone to send in any questions relating to this session via the q a tab on the right hand side of your screen the speaker will try to address them towards the end of the presentation and during the q a segment today's this session's topic is titled reduce energy consumption of your hvac systems using smart machine learning ai based chiller plant control and optimization systems to begin we have a question for our audience which you will now see in the poll section on the right hand side of your screen again i welcome you to have a look consider this question and kindly register your vote we are happy to welcome chirayu shah back today to share how this session can actually help you improve um improve and reduce energy consumption in your chiller plan chirayu shah is the vice president of operations at conserve it with over 19 years of experience in the automation and controls industry he started his career in india in 2002 working on special purpose projects with multiple government organizations in various fields of research shah has considerable experience in technical product and application development product distribution and support as well as business development in smart buildings age to cloud iot solutions building automation and analytics as well as energy and efficiency industries without further delays here are you will like to turn on the camera and microphone thank you betsy for the introduction welcome chirayu before i hand over the session to you please allow me to take a quick look on the poll results and i can see that uh based on the questions uh what is the kilowatt autonomous efficiency for your plant room we have a vote of 71.4 for 0.5 to 0.6 when you are ready i will hand over the session to you and to kickstart thank you thank you thank you so much patsy and thank you everyone you know for helping us uh answer the poll uh i mean like we we are aware that you know singapore is one of the advanced markets globally and uh and the chile plans are very well optimized in this region and generally they trend very well as as we talk about energy consumption uh but when we uh you know use a smart technology like what i'm going to present today uh using uh you know machine learning as well as ai we can improve the kilogram efficiency by even more so thank you thank you pepsi and i would like to thank ib ew as well for the opportunity to you know allow me to present today uh as you can see my screen now uh i will i'll be talking about how we can use uh you know machine learning and ai techniques to further reduce energy consumption of the hvac systems so this is the brief agenda of our presentation today we'll obviously talk about the background and the common challenges that we all face today in this industry then the solution approach that we can have using ai or the artificial intelligence model then how do we use plan pro which is our smart machine learning chiller plan controls and optimization system how how do we use that and fit it into this overall solution uh what are the different artificial intelligence based machine learning methods that are used in our solution and at the end i will also share a success story at a local building in southeast asia so that will give you more context of how plant pro actually works in this setting so you know just a brief background i'm sure most of you are aware that buildings are responsible for almost a quarter of the carbon emissions globally currently and that is growing but that number is growing every day uh everything is now you know running of electricity uh we initially it was just the hvac and the lighting now you know there is televisions there is lighting all smart devices everything in a building requires energy and requires electricity so buildings are one of the highest you know consumers of energy as well as they emit the most amount of carbon emissions uh so uh um so to nullify that uh you know almost 200 countries have ratified the the un paris agreement where everyone has committed to reducing the carbon output as soon as possible and bring the global warming to below two degrees of its current levels and different companies have adopted different rates of how they will achieve this but significantly you know everyone wants to get there by 2030 in that range so how do we get to that uh you know target that we have what is our pathway to get to net zero emissions from buildings because they are one of uh the largest emissions uh you know contributor obviously we need to improve the energy productivity so you need to optimize you know how our systems work and use less energy for providing you know the cooling we also need to then reduce the demand using optimum optimal load management uh then this is something that is happening more and more uh is introduction of on-site renewables so renewable sources are being added to the existing uh you know supply of energy sources so either it's an on-site generation with or without storage uh that could be battery storage or thermal energy storage uh and you know using tv or or like geothermal which is the chillers and at the same time doing peak demand management so you know i'm sure most of you also have to pay high peak demand charges uh when there is a lot of uh requirement from the grid so how we can manage that so that's you know like the premise uh now when we break that down into buildings uh we sort of know that the hvac systems you know consume almost half of the total energy of a building and out of that almost 80 percent of the energy is consumed by the chiller plant so obviously you know making the chila plants more efficient and the expect systems more efficient goes a long way into saving the energy reducing the energy efficiency and you know getting us closer to the net emissions uh so going on from the background you know what are the common challenges currently that we face in the industry today in the building management industry today uh you know the number one is inefficient operation of hvac equipment uh the second is the amount of system down times that can happen which again you know causes the systems of nine we need to allow for more redundancy and you know we need to get people in to do service ad hoc instead of having like you know a data driven maintenance strategy then obviously there is a lot of you know there is a lack of automation in control uh and by this i just don't mean uh you know just having a fully automated system this is also having a fully optimized system making sure that we are running the equipment uh that is required that is the most optimal piece of equipment uh that can run at that point of time to provide us required outputs uh and uh you know essentially people do use bms systems and the bms systems i would i would admit and agree in in singapore are pretty smart uh because this because the standards set by bca are also pretty high so getting around you know 9 kilowatts per ton of 0.6 or thereabouts is is common in singapore which is not very common in any particular world but a vms can only take use you know somewhere but then if you need to get even more savings out of that you definitely need something which is a bit more smarter uh and on top of that you know uh when we do introduce renewable sources uh there is even more issues where you know uh like we do not have enough data of renewable sources how they are used how renewable sources as well as electrical sources we have today energy sources how they combine well together then if we are using solar as a part of the renewable sources we could have intermittent usage of solar availability as well as storage systems and you know we could have different fuels like gas electricity and other things with varying you know prices that you know it costs you more in the afternoon it costs you less in the evening and the nights so we have to take all of those systems into account and all of that into our solution approach so essentially the key question is how are we going to deliver benefit to the end user using real-time data and predictive modeling capabilities and on this slide you know as you will see it will come up for you very shortly in the presentation is the solution approach that you know we propose to use using ai so obviously at the bottom is going to be your central chiller plant that is going to supply the chili water to the buildings and the water returns back on top of that uh we uh proposed to put in a smart machine learning children solution uh which would be a product called plant pro and then uh there is going to be an optimization platform on top of that which is the model predictive control and an ai based prediction engine in there obviously you know there's going to be different inputs measurements weather series and on the output we will see calculations of the npc which sends back the controls to the server and controller and that obviously is going to send back controls to the central cooling system so a very high level architecture of the solution approach using ai so now how do we fit that into the product range that we have today which is plan so plan pro is a smart machine learning chiller plan controls an optimization system and it's an acronym for plant room performance reliability and optimization plan pro is a complete hardware and software solution that enables optimization of control of your chiller plant as well as enabling up to 30 to 40 percent of annual energy savings plancro can integrate with your bms systems on-site and can work with or without dns systems in singapore you're generally going to have a really good bms system so what we will do is we will use control as the brains of the system and the bms can act as the arms and the legs the plantar solution is a full solution which is the edge controller the dashboards as well as the reporting and when we say the dashboard the dashboards include the machine learning capability the control algorithms the alarming the trending everything is built into the software application uh plantro uses the state-of-the-art technology so plan pro has ai machine learning backed data models which are used to see how the chiller plant is operating and we are then using that information to make well-informed real-time decisions as well as modifying the models based on the data learned or the data collected we are just not relying the factory data that is provided by the expect manufacturers we also take into consideration the real-time data collected and make decisions real time all the data is not sent off site into a cloud server or or you know another system that does the modeling and sends information back everything is done on site inside the building premises on the controller that sits and the machine learning happens live all the data also stays inside the controller so the end user or the building manager owns the data it's not send off to an offsite uh server for security purposes then plantro also uses ai to automatically generate control algorithms so based on the information input into the system as well as the data collected we automatically generate different control algorithms uh that would best suit the plant configuration and the plant system this has been an ongoing development for over 10 years so now the product is very mature and very robust and we've got really really good results globally we even have multiple chiller manufacturers who oem this technology and they sell it as their own system as their own central plant optimization system then they sell their own chills plantro will also provide you automated alerts when the operations deviate from the target efficiency level as well as from the target performance level and obviously we can use that uh to do predictive maintenance as well as predictive work uh to ensure that the system does not fall down and there is no costly repairs needed to be done at a later stage uh so i mean like you know you've got the premise for plant room uh now you know i mean i'm sure everyone is thinking what exactly is it work uh and what does it do for me so on this slide uh our so the next slide comes up you will see that plancro uh the functionality of plum pro can be defined into six different steps the first step is chiller plant performance monitoring what this means is that we are going to create a full digital twin of your chiller plant inside the plant pro software model so we will be setting up how many chillers you have how many pumps you have how many cooling towers you have how they are associated with each other is it dedicated pumping one to one is it headed pumping uh and so and so you know other vsds involved and then uh also the design parameters from the manufacturer that are provided are added into control to create the digital models on top of the digital twins that are created then those models are then used to then compare the performance of the actual hvac equipment against the real-time performance of the so we will see what the real-time performance of the chillers is pharmacist and the plant group is so what is the real-time kilowatt spectrum for my children's what is the real-time kilowatts per ton for my pumps what is the real-time kilowatts pattern for my cooling towers and for my whole plan all chiller plants and at the same time because we have the modeling done and the result will set up inside the plant software uh we will then predict what the best performance for those equipments should have been at the same point of time then the best performance is then compared against the real-time performance uh and that is uh then you know going to give us a gap or it's going to tell us that yep now everything is fine so if my real-time kilowatts are done comes up to be say 0.6 uh and my best performance or my pro kilowatts pattern that number calculates that the system should be able to be achieved at this particular point of time comes out to be 0.56 then obviously we are missing out on you know 0.04 in terms of my performance in terms of my efficiency and i'm using more energy to produce the same tonnage of cooling that i need to do now that could be because of a few different reasons it either could be just that you know the control algorithms need to be tweaked dynamically to suit the current conditions or there is actually a physical issue in the system that is causing the system not to perform as efficiently as possible so it could be the heat transfer is not happening correctly on the chillers or you know there was a service technician who went out to site he logged the bms oh sorry logged the vsd at a fixed speed or it turned it off there could be various things you know the machine is running low on gas and so so all of that so when there is a difference and it's a substantial difference then plancros analytics and fdd engine automatically starts looking into the gaps what could be the issues and then not only tell you and give you insights but also guide you towards the root cause of what the issues could be so that information can be used to then you know make sure that the chiller plant room is always maintained predictively uh rather than having to send someone to site when something breaks down uh on top of that because we have all that information even the chiller plan controls and automation can be done in the most efficient way and the optimization so you know obviously all the standard strategies you know chill water resets condenser water resets you know smart staging and sequencing you know smart flow optimization all of those things cooling tower optimization condenser order optimization everything is done and we can do it in a way where a we use the machine learning models to push you know the set points and the limits of the system uh without damaging the hvac equipment of course all of this is done within the realms of safety so you know as i said before plan pro is an acronym for plant room performance reliability and optimization so optimization is done only after we achieve the performance and the reliability we are not doing optimization at the expense of reliability and on top of that the system allows you to continuously commission and fine-tune your your chiller plant operations uh at every second every minute so it gives you the live visibility of what is happening and the system tells you if something's wrong even like highlighting and and error in the heat balance or even sensor calibration issues so all of these things are then brought out uh in advance uh just going in a bit more detail of how the platform system works uh essentially there is a lot of data that we measure uh so for a chiller that is about 15 to 18 points then for each pump there is another six to eight points and for cooling towers the same and then there are common points across the chiller plant system on the common headers as well as you know the decoupler lines or the bypass lines uh plantro is a system for your chiller plant only we are not looking at the airside optimization with plantro we got another system called airpro which looks at that but today we are only focusing on chiller plant optimization so we look at everything on the water site we measure all the key data points and then we calculate the efficiency of every machine as well as concealer devices everything is available to you on the user interfaces or through on demand and schedule reporting but then that information is taken and you know verified against the design data and the best case efficiency modeling that is done using the machine learning and the total plant efficiency is then taken into account and compared with the design information using that gap we are then diagnosing the issues uh which may be causing the gaps and then either we are you know sending people out or personal out to fix the equipment or you know do the service and maintenance as required or if it's a controls and automation issue then everything is adjusted automatically to make sure the plant is working in its most optimized state at that point of time to automation now again what else is the value proposition program so plant pro works on any type of chiller it it doesn't need to be just a water cooled or air cooled absorption like you can even have a mix of chillers in your same plant room it doesn't matter you can even have any compressor type within your plant room even a mix of compressor type between your plant room so you can have two centrifugal chillers and a screw chiller absolutely no issues plan pro works very well in a retrofit environment as well not just in a new environment so if you had an enquiry or if you had a system where you know you had a decking machine a train machine and a carrier machine absolutely fine no issues even mixing and matching multiple manufacturers is completely fine because the planter system can learn the performance of any machine or any pump absolutely fine developer system u is based on the proven niagara framework from trillium so it can integrate with typically any bms and any equipment on site as well as clan pro provides high level interface mapping with various chillers as i already mentioned that plan pro is already oem by three different chiller manufacturers already so we have a lot of inside knowledge of working with chile manufacturers and how hvac equipment should work and the automatic mapping is already provided for a lot of the standard manufacturers again i would like to highlight that plant row is installed and deployed inside the building premises within the customer's secure id network it is fully independent of cloud so all the cyber security risks are covered the platform controller sits inside your building it doesn't then no data is ever sent out the only way to login or connect to the system from outside the network is using uh the recommended security with a vpn solution uh or through the customer type you know requirements so it's completely safe and it sits inside your building uh on this slide you can see a possible architecture uh for control system plan pro controller will sit inside your building then there will be a secure vpn channel or connection which will then be talking to your existing vms systems because typically i know in singapore vms systems are going to be connected to all the equipments they will have all the sensors and everything required so we can just have one connection into the dms system and get read and write all the data from the existing bms so we can reuse all the information that is available in the front room so we are not adding to any more you know infrastructure issues uh so you are only paying for the controller and the software technology nothing else now we will go into some of the machine learning techniques and methodologies within plan pro so clan pro uses various machine learning techniques to learn the performance curves of equipments uh the most common one that is used is multivariate polynomial constraint least squares regression for children efficiency which essentially means that you know it is a very carefully designed mathematical model uh which is then going to accurately learn the performance of the machine we are going to embed specific requirements within the model uh and we can come over specific data gaps because we can extrapolate and correlate things used on the hvac and thermodynamic equations we are going to learn the model and fit it into the closest data set as possible so when the models are built they are actually compared with the real-time performance and make sure that the the theoretical models or the machine learning models fit as close as possible to the real-time models and just not making the models for the sake of making a model which can be so far away and we use all of this to make sure that the hvac equipment specifically the chillers then uh you know work as efficiently as possible and we are able to accurately predict the efficiency of chiller plants uh to with the least amount of error in our modeling so just an example of how the machine learning uh is approached uh inside plant row so essentially on the very left you can see uh in the screen that we are plotting the actual power versus the predicted power the optimization algorithm then minimizes the difference between the actual and the predicted power those are then turned into specific mathematical models where we predict the power as a function of cooling load leaving chill water temperature and entering condenser water temperature essentially there we use the entering condenser temperature and living chill water temperature as a delta and that is then converted into the cop curves or the kilowatts per term curves of the chillers at various conditions so in this chart on the right what you can see is the performance of the same chiller at various conditions as the delta t changes and as the load changes the performance of the chiller changes is there so we can model that we can predict that and hence we can ensure that you know the killer plant is operating in its most optimal stage under any condition so plantro as i mentioned before you know we do use mathematical optimization we don't use brute force or you know any of those other methods which are not uh very nice and they actually put a lot of strain on the system and they are more trial and error machine learning models rather than an actual learning model so plan pro uses advanced mathematical optimization algorithms primarily for two reasons one is for machine learning uh and second is for child plant optimization so by optimization we mean we will find the optimal set points the target nodes and the channel combinations we can predict all of that based on the machine learning models that are there which allow us to do that again one of the core plant pro optimization algorithms is a large scale non-linear programming solver uh which is then used uh using a sequential convex programming interior point method uh by using these technologies we are able to find the optimum point or the optimum operation for a wide range of optimization problems including non-linear and convex or non-convex problems for inequality and equality constraints so by using all of this we have developed a feature called smart sequencing inside platform by using all the all the machine learning uh nai strategies that i mentioned about uh then we use what we call smart sequencing and smart sequencing does basic things sorry that's the three main things which is an optimum staging so it carries out the prediction of the cooling load to ensure that the chillers are enabled and disabled as needed and that the cooling demand is always met then we do the optimum sequencing so selecting the most efficient combination of killers that brings the lowest energy consumption and by optimum load balancing so once the optimum sequence is selected we will then again push those killers into their most optimum sweet spots so typically what happens in any chiller plant is that you know if you have three chillers running they are all going to balance out two similar loads so say if the building requires uh you know them to balance out at 65 load then all the three chillers will operate at roughly 65 load and what you will get inside the building is also 65 load that you require of the total capacity what plan pro does is that you know eventually in the building you will still get the 65 load that you need but if it is running if it needs to run three chillers to provide that combination and that happens to be the most efficient combination using and including the pump power and the cooling power energy then it's going to see uh you know which is the sweet spot for my individual chile and it will then push those chillers into their sweet spots so for example at the bottom of the on my bottom left you can see there are three different chillers and you can see that the optimal points for chiller 1 is about 300 kilowatts the optimum point for chiller two is about thousand kilowatts and the optimum point for children three is about 550 kilowatts so what plant oh then does is it make sure that we will try and push the chillers into their sweet spots in a way that when the water is mixing from all the sweet sugars we are providing the most efficient combination uh and we are providing the requirements inside the building but we are increasing the efficiency as well as the kilowatts per ton of the entire plant room by making sure my equipment is running at its most efficiency first and uh on the next slide that comes up you will be able to see smart sequencing in action on a real site so even on a 35 degree day so 35 degrees celsius you would agree is is not uh you know it's not a cold day where we can play with you know the system uh that means that's a warm day and we need to make sure that you know we are doing the best for the plant room we are getting a plant cop of almost 7 which is about 0.55 kilowatts per ton for the building and we are doing that by making sure that you know i'm pushing my children into the sweet spot so for example my chiller 2 i am asking it to run at 11.8 degrees set point uh which is a capacity of almost 70 percent and my children three i'm asking it to provide nine and a half degrees celsius uh which is a cooling capacity of just over 100 and by having that by mixing the waters i am still meeting my set point in the plant room which is of 11 degrees and i am not compromising on comfort but i am getting the best efficiency out of the system another strategy with machine learning that we use within plant row and again they are not just the only strategies but in this particular presentation i am only highlighting a couple of them to show you how advanced the system is is what we call a smart flow optimization so inside the smart flow optimization we do variable primary flow but we also do the most efficient condenser water flow object to reduce the overall power usage in the chill water plant this is based off scientific methodology based on data-driven machine learning and mathematical models essentially rule of thumbs do not work inside machine learning systems typically the bms systems are going to use dual outcomes or you know use priority to see okay this happens so i can generalize that i can use this strategy unfortunately in real life you know every plant is different and we are not able to do that and we use uh you know in our optimization the fact that lower condenser water flow reduces the pump power usage but in return it is actually going to increase the chiller power usage so we make sure that the trade-off is done in the best way by using predictive optimization and ensuring that my plan kilowatts pattern is the lowest and not just ensuring that my chiller load pattern is lowest or my from kilowatts per value so as an example here the different conditions of load and water temperatures the optimization algorithm finds different optimal flow for the same machine at different conditions so at 30 load keeping my chill water temperature and condensed order temperature is the same my optimal flow is calculated to be 50 liters a second on the example on the left but for the same set of equipment for the same configuration if the load increases to 75 percent then my actual flow becomes 750 second so this again is just to show that you know we cannot just have a set and forget philosophy it has to be a learning philosophy where you know things are learned over time and we are able to predict the performance and we go into the optimized set points as and when required so you know there is a very common question that people ask us why can't the standard control strategy be derived for all sites you know it's simple it's just math unfortunately no because different chillers and pumps uh the algorithm the optimization algorithm will find different set points at different conditions so now taking the last example but by changing the amount of you know by changing the pumps and the chillers so now we can see that it's they're both having the 50 load the same temperature two different sets but uh the set two has an optimized flow of 85 liters of seconds where the pumps were rated 14 kilowatts and in the set one where the pumps are 182 kilowatts the optimum flow now becomes 60 liters so again you know we need a dynamic self-learning optimization algorithm system uh to give us the maximum energy savings uh and cost savings again just you know showing the smart flow in in action on a live site and showing you how the machine learning helps to model the performance in real time uh and this one just shows us the results in a real-world site where smart flow was you know put in for trials uh and you can see that we are able to get 3.62 savings only from the smart flow optimization technique every other efficiency technique was disabled uh to see how the smart flow technique affects the savings and the real-time results are showing that we are able to provide over 3.62 savings just by using this one technique so now when we have multiple machine learning techniques we can keep adding the performance and we can see that you know how the energy efficiency can improve drastically uh so moving on from that what i will do is i will very quickly take you through a case study uh for a building in southeast asia so that you know you get to see how platform functions in the southeast asia whether uh not just you know i mean like i'm from australia so not in the australian weather their conditions are very different you know in a hot and humid climate how plant pro functions so uh as a high level uh we can say that you know the data uh uh in this case study is for a six month period running from june 2020 to january 2021 18.8 energy savings were reported uh as well as almost 45 hours of manpower savings per month were reported so here we are now talking about uh savings as a total cost of ownership not just energy savings but also the manpower savings that you are able to save using a system like pro using the predictive analytics and data driven maintenance from blindfold so what is the setup of this system within uh plan pro in this site so this one was specifically three watercool chillers uh having four headed chill water pumps four headed condenser pumps and they had a very unique uh condenser side system where they had three radiators with 12 vsd fans on each radiator so this was like you know having like a micro cooling tower sort of a space a very unique system having multiple radiators you know i'm sorry multiple vsd fans on the radiators instead of pulling camera and i just highlighted this just to show that nonpro is very adaptable very flexible and we can take on you know various uh configurations as well now the deployment on site was very simple you know uh on the next slide when you will see that on the left is the actual bms panel that was then replaced uh you know by uh the plant pro hardware where uh just a plant pro edge controller along with uh remote remote connectivity kit and a power supply was put in and then just connected into the vms system via a local lan port very simple connections we are not even touching the bms system vms system stays as it is and absolutely uh working fine then uh once plan pro is set up uh you know the clan pro dashboards are able to give us the detailed plan condition at any point of time also allow you to shut down or manually operate the plan from the screen if you did not want to have the plant running auto for servicing or you know whatever reasons moving on to the results of plant row in this specific site so as i said before in the summary uh plant pro was able to save 18 energy savings on this site this is the actual chiller plant energy consumption under plant growth and all of this is done using a proper ip mvp energy study and by doing that you know we are able to uh compare the actual as well as the medicare energy savings and you know in a short period of time we are able to see there is a massive gap in energy savings and gives us 18.8 the overall energy consumption in red declined in comparison to the cooling load after the plant system was deployed so what that says is that the cooling load or the turns the output of the system never changed only the kilowatts used to provide the same pulling damage was improved hence showing that there is an improvement in kilowatts per ton uh so it's not that we just reduce the amount of cooling hence we are using less energy we were still providing the same amount of cooling that is required but the energy required to produce that cooling was drastically reduced and this resulted in a 27 improvement in cop or kilowatts per hour uh when the system was on the bms control to compare to when the system was put under plant pro control so that is where all the energy savings come into picture also at the same point uh there were significant manpower savings uh which led to the lower total cost of ownership for the system so typically before plan pro the building managers used to uh you know use up almost 50 hours of bms contractors work or neighbor to make sure that you know system is always tuned and you know make changes as required adjust things to suit whereas after plantro they only required the bms company to come in and tune check the system for almost two and a half to three hours so that is almost you know uh at the current energy or the labor rates that was almost thirteen thousand dollars of savings in six months uh just in manpower series we are not even talking the 18 energy savings in the plant energy consumption that is even higher so this is the extra savings and this was the client feedback uh after planter was deployed initially they were very hesitant to install a machine learning and optimization system because you know they thought it's like a black box we don't know what is happening and you know it would be hard for us to you know intervene but since they have installed plant rope they are very very happy with the performance the optimization strategies as well as the energy and the manpower savings they get out of it and essentially the client has provided feedback that they their engineers now just need to periodically monitor and adjust operations only when required they can now focus on the critical bms and site issues and the customer is very willing to recommend provide other sites in the region so i think that was the presentation of you know how we can reduce energy consumption and improve the efficiency of kilo plants using machine learning and ai techniques i can now open the floor to anyone for any questions and answers that you might have uh or happy to take yeah any any feedback thank you to ryu for the very insightful sharing if you could please stop screen sharing we will start the q a session for those in the audience please feel free to submit your questions via the q a section on the right hand side of your screen just let me take a look sure are you from your sharing itself uh i do have some questions where i'm actually interested to have a deeper understanding is the chiller plan automation using smart machine learning ai that is yes absolutely so the system that we use uh definitely uses uh you know artificial intelligence techniques as well as machine learning techniques and benefits of plant pro which is a product that we provide uh definitely uses you know machine learning and hence we are able to get the type of results that i just presented in the case study brilliant how can any of the existing are bms actually uh upgrade to smart machine learning ai would it be possible no absolutely absolutely again as i said you know in in my in my presentation uh plantro is provided on its own own edge controller so all it needs is a controller will go inside your building and it will communicate with your existing bms's over us over a single ethernet or serial communication and we can read and write all the information from the dns so we don't need to upgrade your entire system just an extra machine learning optimization product is installed and it communicates with the bms and uses all the existing infrastructure good to know good to know okay i have a question here from the audience how is the baseline derived when the plan is already on plant pro optimization is the baseline obtained by calculation so uh so the so the baseline is basically derived uh using the actual energy consumption by the building from an independent energy meter and not from plant prodata just to make sure that you know the base line is fully independent uh so we use the actual energy consumption uh and the actual meter data for the building and we take that on and then build the baseline using ipmvp methodology so ipmvp is a protocol that is used very widely in the industry which has a set of rules and guidelines and using that you know there can be no doubts or calculations on how the baseline or the predicted savings are calculated so we don't use internal control data do that we could do it we could absolutely use it but just to avoid questions on the legitimacy of the baseline we use independent raw meter data if that is available from the client amazing we do have a another question do you have solutions for vrf systems uh yes we have solutions for vrf systems uh it's just not planned pro but we have other solutions uh that we can and yeah uh mr boon i mean like maybe we can reach out to you after uh the presentation privately and we can have a chat about your requirements then whatever did any challenges face when connecting to oems sure are you no absolutely we have never had issues when connecting to oems because plantro is based on a very open technology very open framework so we can easily communicate with any bms or any oem systems we use standard bacnet ford personally all the standard protocols so there has really never been an issue and again the fact that you know plan pro is used by various oems already so we do not have any issues you know at all working with oems or other manufacturers next up how do you eliminate other factors affecting energy consumption example the weather occupancy or even a change in operation etc oh absolutely so so we actually consider exactly the weather the occupancy all of those as a part of our baseline modeling as well and based on the conditions uh the machine learning and the modeling the ei and pc does consider you know the weather if we predict uh the weather for the next hour we see how the how the cooling water trends are happening as well so we use all those information to make sure that here we are always on track and dynamically update the algorithms as required so it's not a set and forget solution it is continuously looking at the input parameters and continuously updating the output solutions amazing okay how do you actually convince um owners to let the solution control the chillers in real time though uh essentially as i said you know so with dantro uh as the name of the product is performance reliability and then optimization so everything we do is within the scope and the constraints so if the building manager or the facility manager has certain constraints the plant pro system will definitely respect those confront and we will only control within those constraints once we can demonstrate and do that only then we will ask is it possible to you know expand the constraints even more and then the savings are achieved as we expand the constraints more but everything is done within the safe you know safety limits advice by the oems or the manufacturers then what is being predicted in the predictive control why prediction is needed at all this is a question from dr mann okay so essentially i mean the predictive essentially prediction is required because you know absolutely we cannot see into the future if you could change the future that was great so prediction is required to predict what the load of the building could be in the future using the current variables and the weather trends and we use those predictions to then see what control algorithms and what strategies i need to meet that future demand that we are predicting i see interesting okay brilliant um we do have many questions coming up maybe just let me have a look um how do you eliminate other factors affecting energy consumption i think we took this question already yes oh how about this question how is the hardware attached to the bms uh the the hardware is attached to vms using an ethernet cable or a serial backnet or modbus connection so it's simple one one connection only okay thank you very much uh we do have more questions from the audience to ryu but unfortunately time doesn't permit us to continue for those whose questions have yet to be answered please feel free to contact chirayu and conserve it at the digital booth directly before we close this session can i invite chirayu to say a couple of last words for this viewing right now absolutely thank you thank you so much patsy and i would also like to thank the audience i think you have been fantastic and i can see from the amount of questions coming up that you know you found the session very interactive and you know very interesting uh i mean like i do apologize for the shortage of time we have but i will go through all the questions you know that you have put up and i'll try and reach out to all of you you know in the next couple of days and maybe we can set up a future meeting uh with you guys or i'll try and answer your questions you know in the chat box as as possible uh and yeah i mean like please feel free to reach out to conserve it uh on the virtual booth uh we are one of the exhibitors at the bex show and you will also be able to find my information uh as one of the exhibitors or participants in the show as well so please feel free to reach out to me there as well but i'll definitely make it a point to reach out to all of you over the next couple of days thank you thank you so much everyone and thank you patsy thank you very much and with that on behalf of backs asia i would like to thank chirayu conserved and all the attendees for joining us and sharing your insights into the topic do take note that the recording of this session will be available on demand after the show you will be redirected back to the back asia platform after this session i encourage everyone to continue to explore and chat with the various expertise at their booths you can also find out about the other live conference and technical webinars to taking place thank you once again for taking time out of your busy schedule we look forward to seeing you in more of our business coming soon stay healthy and goodbye for now [Music] you