uh thank you very much bye and for the invitation and offering this opportunity for me to come here to share this uh the excellent people I was really impressed by the atmosphere and the environment so daily filters I I I should be here early so as a color said I I have been working on this many different aspects of the automation systems and hope to use this opportunity to share with you my personal Reflections and experiences also some challenges that I I look for some collaboration with academic partners and before I start the presentation I like to emphasize this is just some academic dialogue is not representing the ABB some official statement in these areas don't over interpret the messages from this presentation so see this as my my like justice interactions with academic people and Inspire future research together and the topic is control over the cloud and fog for automation for advanced digitalization or Industries it's not a bit fancy uh before I start the presentation I like to introduce the company first maybe for some of you uh AVB is not very well known name because we don't do the consumer business we only deliver all of our products services to the to the to the the companies so ABB Visa is electrification and automation company we have the full main business areas first is electrification it's a low and medium voltage par Electronics power supply for factories for buildings for the city infrastructure for example the evening charting system and we have the second one is process automation it includes the hardware the sensors actuators controllers and control system the DCS for all the type of the process industryfacturers and then we have the robotics sorry motion that is the the motors or different size of Motors can be very small can be super big motors and also the drive system behind the motor and then the last one is the Robotics and the Machine automation they produce the robots maybe this is the most uh yeah uh well-known product from ABB because it is shown everywhere and it is used in the automotive industry basically and also expanding to the service scenarios like the hospitals even in the future in the restaurants you will see more and more robotics in this business areas Abbas ranked number one or number two in general we say about that leading player in the automation world we are providing um almost everything about automation and this is the agenda of this presentation today first I share with you the vision about Cloud fog automation it's still a very fancy uh concept it is more about using the cloud and focal Computing for deploying the future new generation of the automation systems and then the look at where we are today especially about the 5G technology and for this for this for this vision and then how much can we do from the control side it's not just the communication and Computing challenge if we do something from the control models we believe it can make things much easier and then the future perspectives so even though we have done something the beliefs the challenges uh even uh bigger than what we understood so we do need this cross-disciplinary research we need the expertise from all the three subjects communication Computing and control to solve the challenges and I also like to organize the the contributions from my colleagues my students and collaborators in the recent years in this area yeah first it's about close-up automation uh for this picture maybe the right side is easier to understand the other scenarios of this automation systems it can mean the factory can be in the infrastructure can be the energy a power plant and even in some like Offshore platforms for all your gas and the vessels the big vessels the onboard automation systems for the vessels if you look at the abstract uh the Technologies behind this automation systems it they always look like this you have a lot of high-level workstations software the DCs then you have the controllers they are connectives or different type of networks they are distributed decision making system then through the controllers it controls it gets the data from the sensors from the input devices they make decisions then give back the feedback command control command to the actuators and depending on the applications there are different type of controllers and sensors actuators but in general the the architecture looks like this it is a layer structure uh we call that the pyramid automation structure on the top level it is the Erp system enterprise resource planning it's a high level software managed like the orders from the buyers and Supply the scratching and and then in the middle it is the manufacturing executing system or some collaborative production management system depending on there can be different names but the task of this layer is the same it's dispatch the orders into some production plan then give this plan to the lower level control systems then you have the supervisor control this is a bit higher level of the control task it will execute the production plan for the mes typically it controls one uh production line one process then go one level lower that is regulatory control this level is much more comfortable than the supervisor control in this level it manages the very fast control loops uh actually this Loop cycle time can be also very diverse depending on the different applications it can be from 100 milliseconds to 100 microseconds so this is let's say the most uh challenging uh level of the control task it needs real time you need a very precise very reliable and deterministic behavior of the communication computer and control then the bottom is the process it can be a chemical process physical process some hybrid even some bell process notice that that pyramid structure is the result from the industries 3.0 the third revolution of Industries it's automation technology but what is happening now we have the fourth revolution of Industries so here I gave us a few game changers I call that game changers because they are really changing the things and also the way that we do the other things the first is international things we know so the communication Technologies is is expanding from the human communication to this machine type Communications it is connecting almost everything everywhere you can get the data from from from almost everything in our life this is also happening in the in the industrial world the factories more and more sensors are deployed everywhere you can get the information from all the machines clearly so this is a about connectivity then second is about Computing and the computation power is growing so quickly it's driven by the Moore's law and the processors and the storage of the information is going is developing so fast and but today most of the factory control systems are still using the old generation hardware for computing it is deep embedded devices of course it has a good deal stick performance a whole real time and also it is certified in many cases for different safety regulations about how can we utilize this super powerful computer infrastructure this is a very important topic that we should consider and then the third one is artificial intelligence okay okay it's intelligence [Music] this is another new thing really introduced by this new generation digital Technologies we see that the intelligence can be involved everywhere in the whole life cycle of the automation systems during the planning phase design phase and commissioning and the maintenance everywhere it is really delivering the value of the of the new driver infrastructure so you get the data from the iot right you you have more powerful Computing so what you do what you deliver to the customers that is optimization more vulnerable Energy Efficiency all uh higher quality from the product and thus this is the value so the AI we could say the AI stuff applications powered by the AI will be the like most attractive values that that is supposed to be delivered to the customers and uh another important thing is for all the three things it has gathered or a tactics so much uh investment and also the best talents I believe they will become self-fulfilling that means if everyone believes that will happen it's really really will happen so the it is just a matter of time I believe and then this is uh the the the high level abstraction about the the trend uh towards this cloud and for computing for automation we use this fancy term close all Automation and this is the the layer structure the pyramid automation structure so this is not just the logical structure but also the physical structure if you go to the factories you will see the the iOS controllers and networks they are really organized installed in such a way layered and what is happening already now it is Cloud automation the high level applications for example Erp mes and partly the the supervisory control are being deployed more than one two-time Cloud platform or Edge Computing platform and of course the industries are realistic and conservative this start from this outside from this less critical applications if anything goes wrong if the latency is too long if anything if even there's some security attacks it wouldn't would not damage the whole process it will not cause the the the dangerous impact to to human life or to environments so that's why we start from outside approaching the critical part but this there is also already some industrial initiatives driving this kind of change for example the number is one the number or open architecture is really promoting this concept so no way and also the OPC UA Foundation is also promoting to use some unified communication protocols to collect the data everywhere and use that for this high level monitoring and optimization of the processes but in this case anyway the the most critical part here the regulatory control the lower level i o operations this is making they are still in the traditional hardware and software architecture this is due to the security and safety concerns but the new way of the more aggressive reason is called fog automation here we do really want to change this also to the small fog based infrastructure here I use the term fog to reflect the difference and between the cloud and Edge and because the fog can be more independent to the to the cloud you can in principle deploy the fog fully in your facility uh you can you can have a very strong security and safety boundary in between you can deploy the dedicated hardware for this deminitarized Zoom that is a the boundary between the public domain and this private sensitive domain you can do more strong protection and also when you do some optimization based on the data collected by the iot by by this uh Cloud platform you will change something right you need to give the feedback to the process then you will pass through this very strong security boundary so so that you can give the feedback in some automatic way so the idea is not to do the human uh operation to do this Opera to do this optimization so otherwise there's you cannot do this fine grain optimization feedback and here there are also some some industrial initiative driving this change so now for the the old part the open process automation forum this is a big Consortium with the involvement from the I.T World nearly all the famous I.T companies Computing communication companies involved and also the automation companies involved very much the launch players are all there of course this will take her take a longer time because the challenge is super big right safety security when you ask this question today we don't have a clear answer so this is also some prioritize the research topics in my day-to-day agenda so if you have to put some gear on The Middle on the cloud Edge and some year on the cloud cloud Edge fog what would it be when will you see this in production even even here give me the boundary now when if you have to say you know when we talk 60 for instance yes they say 20 30. so when we I mean we don't see the cloud automation completely in this way you have it here right now right so would we see that before 2030 and is the cloud for automation what yeah I I if this is just a really personal uh dream I think this would happen after uh between 10 and 20 years and in general Industries uh are slower than the Consumer domains right and also uh they are very diverse so the the way the best way of penetrating to the industry is to find the phone number first use case which uh uh where the the the wireless or the cloud candle is AI technology will be some enabler full novel no variables and also the service robots so this could be the some area of these technologies will become uh like practice earlier than the other domains the industry has a very wide range many different segments we have very different uh pace of adoption the new technologies and the actor in the loophole takes area there's already some open robotics initiative and some four robotics initiative about this yeah in general we say maybe 10 or 20 years to see a broad adoption of this this kind of architecture then most specifically the objectives the first one is about the virtualized Computing as I said today the industrial devices are using the very deep embedded systems and to use with some customized software it's a very hard real-time operating systems and but actually now the the network infrastructure is providing very powerful Computing capacities like like the 5D infrastructure like the uh while the Broadband networks they all have this Computing capabilities and also remotely in the data center is super powerful but to really utilize those Computing infrastructure we need virtualization so that the upper layer developers will not care too much about the low layer stuff it can provide the same interface and also with with guaranteed performance providing the interface apis is easier but currently the performance especially the latency availability and the security quality of service they are more challenging but this is really really needed without this we cannot really use this those new new things in the industrial world and another one is the generalistic and converged Networks today there are super strong uh technology pushing from the Telecom Industries like 5G and also from the Wi-Fi 6 of S7 and to really use those new generation communication infrastructure we need to solve the end to end determinism first you need to find a proper way to integrate the interfaces today if you look at the machines Motors or robot arms evolve convertibles they are all using some wire interface typically industrial internet we are these kind of protocols they're not there they are not that so how to integrate so there are many uh tricky things to be to be done it's not just plug and play similarly from the controller side if you virtualize the control logic yeah control engine and run that in some data center or cloud computing infrastructure you also need to solve a lot of interoperability issues both of these protocols because controllers that they are using these protocols and more importantly also more chatting is the messages need to pursue all those virtualization layers as I showed just now uh so many layers the messages are transmitted from from the physical media then going up until the application on the top and really use the message so you need to really solve this virtualization issue and also guarantee the determinism sometimes we call this deterministic virtualization this is a very uh important topic then let's look at where we are today especially uh from the 5G perspective in the recent years we have worked a lot with our strategic Partners on the 5G and also with our uh engage over customers because 5G is really promoted uh everywhere it promised to give very very attractive benefits it is not just replacing the cables by the wireless infrastructure but also enabling new things for example if you have a very high uh height of the infrastructure if you have to wide ire are in rural area you have noticed uh while the infrastructure where the networks and if you have mobility of course it doesn't make sense to have a cable behind the robot and if you have a rotating part typically like the wind turban with some some Greening machines so cabling is very difficult and some hot environment or corrosive environment the cables can be damaged very easily and if you don't have the access to the to the land this is also the case especially in North America the land is owned privately it's very difficult to to get the land to even make a cabling under the under the the ground and in some Modern fashion buildings it is hard to change anything to add a cable other device or remove a device the wireless is more Natural Choice and also medic industrial facilities we need the temporary development the devices are just used for short while it should be removed immediately after the task is finished and another special case is that high voltage electricity stuff in some Power Systems in factories in in the energy facilities there is very big potential between the controller and the control Target you cannot use the metallic cables for communication today's only choice is the optical favor but you know the material with the optic fibers is much shorter than the solid state semiconductors so this also gives some opportunity for the virus to contribute there if we can manage to solve the challenges in the latency availability basically the lifetime of this wireless devices is comparable to the to the polytonics solid stage devices yeah all these cases give us a stronger motivation to try with the wireless communication another aspect important aspect is the Computing so 5G is not just communication this is also Computing it's natively providing this Edge Computing capacity and here the general idea is in the factory residitions it can yeah do a lot of computing uh process you can you can you can offload the computation load and to the base station and also you can get a very close uh distance between the the the producer and user of the data and this is a really a very very Natural Choice full full level for the high Mobility applications to use a 5G as a Computing resource and what is the primary concern today if I may talk about this 5G communication Computing for the for the for the for the Indus domain especially for the control applications uh I use this dinosaur curve to describe the challenge it's about determinism latency and reliability so if you describe the performance over latency you need always give the the probability otherwise it is mediumless so here I I use this curve from some old technology developed by IBB many years ago maybe 15 years ago uh the white cell technology we measure the performance in in such a way first in the x-axis that is the latency of the messages delivered from the the sender and receiver and the y-axis is the probability of this latency distribution it can be the PDF curve it can also be some CDF Community function but anyway what do we want to see is this low probability events how many packets will experience a longer latency even though that can be very low probability but it is really of our concern uh for example in this example if we set the deadline to be 20 milliseconds here okay we can get uh 10 to the power minus 8 probability that's pretty good yeah in some cases this is really really required uh in some cases it can be minus six here if we just require minus six then we can expect we can promise 10 milliseconds this is the meeting uh of this curve but actually most of the consumer oriented communication or Computing technology mainly provide this type of red curve it has a very big tail so the effort in the recent years including the 5G and the 6G in the future it's trying to push this red curve to the left and to the bottom well of course the cost increase quickly you want to reduce especially if you want to cut the tail that really cause a lot of challenges and you even you need to change the design philosophy uh vulnerable introduces redundancy that is the most practical way to cut the tail but you know the consumer technologies always try to to minimize the redundancy that's an opposite in the design philosophy then let's look at where we are today this is the some experiment uh evaluation that we did uh yeah almost one year ago in our lab together with our family partner notice that that was not using the the latest uh the best profile the 5G for the short latency it was using the embb enhanced mobile broadband it is more oriented for the for the consumer vacation for the video streaming applications it's not for the real-time control yet and but that is the first like mature product release of the value technology in the market and we did this long-term evaluation using a real uh test pad of the automation we use a controller to produce this cyclic traffic and transmit over the 5G then our automation device received a message and give the feedback so this is a closed loop control your case and and we also measure this latest level every packet for each Direction very precisely for every packet and also we developed some dedicated Hardware software to do this because to really know phone number there's a low probability issues we need to have a super reliable measurement tools otherwise you cannot say your numbers are trusted or not we managed to do this and also I believe this is really a significant progress from the 5D area if you look at this curve here this is the uh this Downstream this is from the base station to the device this is Upstream is from device to the to the base solution direction for One Direction if you set that line to be 50 millisecond you could expect to get a at least 10 to the power minus 6. uh reliability that is very very impressive and if we be more conservative we can resolve some margin for the for the engineering implementation we could say this is already capable of doing the control closing the control Loop uh at some 200 millisecond cycle time it is it is not enough to to to to cover all the control use cases but it can already do a lot of course this is done in your lab it is nearly the perfect environment for wireless communication uh there's no Mobility there's no big obstacles there's no long distance but but this is already a very good starting point uh as a reference of course more field tests and with more harsh environment is necessary to draw the solid conclusion but this is already a big big progress in this area uh we are continue working on this computer Computing part there's no reason to share today hope hope next year next time I can have some results from the completion side but you know family is also very very good Computing a very good server inside and then uh other side without changing the control models we can do the control Loop like 200 milliseconds cycle time it is attractive but let's see how much can we do from the control side to help to make the same easier so this this is what we called latency aware control over the cloud and fog first we made a a real life testbed because this control stuff is difficult to really do do this bad simulation because the wireless networks is nearly impossible to be precisely modeled by simulation you can do some high level simulation or the metal Behavior but the physical layer stuff is too complicated you you can never guess the precisely enough results if you do the adjuster simulation and then in our case uh we have this love setup of the 5G and Wi-Fi 6 and also a reference network using the industrial internet and we have this fully virtualized Computing environment it has an industrial PC and servers running some mainstream Edge cloud computing framework which is called openstack it is also used everywhere for doubling the Amazon the Microsoft Edge or cloud services which is the mainstream in principle we can make this as big as it is Center using the same Hardware infrastructure however architecture then also we have built different type of the control Target uh first use case is a mobile robot uh we use this open source Rows 2 environment running some navigation application on top of that another use case is robots one is the safety coordination because the wireless communication has a very very uh attractive feature for the for the mobile robot and in the mobile robot safety is the most important pattern actually it's filled up in some Hardware automation you have a lot of big robots uh big advs carrying the containers which can be 30 or 50 tons super big fully unmanned operation so if there's any clear in between there's where there will be a big problem you have to stop everything and you have to fix the problem manually a lot of economic loss and another deal case is is a television title operation of the robot this is also very very very very valuable use case for the robots uh for example remote soldering or telemeda thing they all need this kind of tele operation and for the Boost cases we need to really concern the natural performance because they are safety critical they are time critical we want to know what's the impact a for the network is not perfect if the other latency if there's a packet loss if there's an old teacher with negativity whatever happened in the control applications and then another one is even more time critical which is the bond beam this is quite popular education case for for the for the online control subject students and in this system the controller will know the uh the angle with a beam and also the angle of the of the ball it will control the rotating part of the motor to adjust the angle the beam so that the ball will roll on the beam and reach the target position that you want and for this process we require basically four millisecond level second time otherwise you could not really close the control group and for all these cases we deployed applications in the same infrastructure this was not possible before if you look at the the segments of these use cases this is robotics this is the robotics this is another motion control or process control elements then you often have different software architectures so different Hardware platform and here we converge them into the same virtualized Computing infrastructure that's definitely uh one of the values uh imagine in the future in the factory if all these Technologies are mature enough I just buy one big server or data center or you can even Outsource this to some third-party service providers it becomes a control at the service right this is another fancy fancy concept so there is not just the technology motivation but also the business motivation this can create a new business model if you can you can use this virtualizing fractions to support very diverse applications uh this is a test bed and today I mainly introduce uh what we do with the ball beam process [Music] on this platform uh we first use this layer the structure for the server side for the controller side uh you do this virtualization based on the therapy and and the KVM virtualization is the kernel virtualization machine and then openstack to manage all the Computing resources it can allocate how many CPUs how much memory you have to each application then on top of that we deploy the soft controller it is software based control ending which can do this technique did correction and determaking then send back the country command and here we uh pick special tricks in the control module we call that the digital wire control uh we model the process use the basic very International method internal model control and but in this model we consider not only the process itself we also consider the networks as a part of the process especially the latency introduced by the network as I showed the 5G can introduce 10 milliseconds to 50 milliseconds latency into the communication and that process is so fast it requires 4 milliseconds cycle time so if you just use the wireless network in the same way as you use the ethernet you can never close the loop because it didn't say of the network it becomes the major portion of the latency then after this we consider more those Network latency and into the control model then we managed to to do the to do the control to close the loop this is the final uh demonstration after all this tuning the ball can can can slap between the two set points in there for a few seconds 20 or 10 seconds then we change to another Ascend pointer with the ball partition um actually it is difficult to see the difference by your eyes if you compare the different networks 5G Wi-Fi and and ethernet no no big difference uh if we also measured the the the quantitative accuracy of the control yeah you can see the slightly some difference but as I said if you don't do this this is aware control if you just use use the 5D or Wi-Fi in the same way as the internet not in our closed the loop so the ball is always shaking oscillating on the beam so this basically suggests to the latest award control is effective strategy for this specific case of course it is not generalized yet but it is a good indication there's a good opportunity to to really really improve if you remember the 5G has 50 milliseconds one way then round trip is 100 milliseconds then you can you are supposed to do the control with a longer cycle time than 100 200 milliseconds but here you can really do that for this a few SEC a few millisecond level later uh cycle time that's a big difference right basically we can improve the performance by by by 10 times and then uh what is it next this is just uh some uh the current progress in the filter if we want to generalize make this uh really into some serious business call for organization what is needed first from the technology push it from this direction uh Computing and combination topics we need more real-time more bonded latency of the networks and also the the the software in the server in the com in the computing and we also need orchestration the orchestration is the resource allocation prioritization scheduling although all these different tasks especially when you when you share the infrastructure with multiple applications but supposed to be shared with everyone right even you have the private 5G in factories it is also supposed to be used by many applications in in the factory it can be the normal I.T traffic it can
be the like the radio streams from the cameras can also be this real time control traffics they have very different uh patterns and traffic and also requirements of the performance so how to do this proper orchestration and this specific considerations that I would like to do to emphasize is first redundancy this was not very well considered today in the consumer Technologies as I said in the factory is if you look at the critical uh factories or control systems so almost everything is doubled or even tripled sensors networks the cables switches controllers prcs and actuators everything is basically two copies running the same algorithm so if anyone fails the other one will take over seamlessly so in principle we hope the the the controls critical control systems will never fully uh stop working for the whole life cycle it can be 20 years it can be even energy industry it can be 50 years so even during the maintenance it's do the maintenance for this uh setup another database is still running so that is a redundancy and also failover as I said if anything fails it can be the failure of the hardware it can be failure of the communication can be the some errors in the Computing the software whatever you must be able to detect this failure and also pick over activate the Redundant uh hologram software make sure the whole system is still up and running then the configuration today the the communication and Computing stuff in the factory is already is already too complicated for managing for configuration and if you introduce these things you know the shared infrastructure which can support so many different things it comes uh really a really really nightmare for for the operators in the factory so how can we ease the configuration basically make it easy the best is make self configuration and here the AI will really play a role so how should we make the the infrastructure Lure by itself what is needed maybe you just give a very high level description of about the intent sometimes it's called intent based networking or intent based Computing so so that the the system the human structure can can adapt confirm the optimal configurations as also dynamically based on the change of the environment change of the traffic load then save the under security I will come back to this after another column then from the control side as I said this is more application pulling I believe this type of effort is even more important and also more urgent because these applications will already deliver the values of the new generation infrastructure so without that without this new Cloud 5D stuff the factories can still work right basically has been there for tens of years without 5G without cloud so to really convince the users the factories customers to to invest in this new generation we must deliver a new values and to add some some urgent topic we have shown this benefits we don't have to wait wait until the 6G is there using the existing 5G this is Magic product it's available off the shelf right we already can do a lot of control so can we generalize this we verify we become treated it is effective using the pawn beam but can we do that as a general solution and other processes and we need some spherical work and also experimental work we also need to expand this uh to tolerate not just the latency but also the teacher of the latency this is a different concept right latency from here to here But the teacher means yeah in average you will have like 30 millisecond latency but sometimes you have 10 milliseconds sometimes you have 100 milliseconds this is the teacher sometimes the detail is more killing uh in the in the control perspective and also in the end we hope we can get some quality of service aware control quality of service covers even more uh aspect uh latency reliability Jitter and also time time synchronization and some some some some general availability long-term availability and so on and then as I said the value creation so after we have this uh uh showcases which can demonstrate the feasibility so how can we deliver more values of this and here I just gave a few examples which I believe are valuable from the user's perspective phone number the hybrid data driven and the model based control and today the control system is mostly based on the first principles that's why we call that model based control and it is not really benefit benefited by the large amount of sensors iot devices and the data and if and also in practice a lot of processes are not easy to be modeled using the purely first principles a lot of non-linear high order effects which cannot be captured very low in this first principle so then the data driven methods and neural networks some learning especially the physics informed learning so it's not purely data driven it should based on the basic principles so that you can have the bounded performance you will not mix the very stupid mistakes all right but here to really get the benefit from the data from this digital approach we need more computation power we need a lot very large storage of data and also big enough boundaries to get the data to train the model and to to run the model and the second is uh optimizing based based control today in foreign mainly on the high level of the control room supervisory control maybe using the optimization to produce the the set point of the low level control groups well actually one of the reasons is computation power it's too complex to be executed in the low level as I said sometimes require a few milliseconds cycle time so the optimization engine takes much longer time sometimes it can be second level right depending on how many steps you want to predict but if you leave how this super powerful data center that could change the game if we can do the optimization with a millisecond level second time that could really really uh make more intelligence into the low level because as if you go to closer to the devices you can do final drawing control you can do final grain optimization as it breaks basic principles and then the camera at the sensor more and more cameras are being used today for the surveillance right monitoring the activities maybe for security purpose can we use the cameras as a generic sensor in the processes and this was not the case before actually one of the reasons is the camera screws produce big data right aiming issue is super big much bigger than the temperature than the 10 series data and also it requires a lot of intelligence to really get the meanings get the values from these camera images but if we have this black competition power if you have enough bandwidth yeah we do love to use more cameras in the processes it's more like the human intelligence introduced into the process so there's another another very attractive valuable use case then augmented operator in the loose and human workers are already overloaded to be honest even they have a lot of globals they have a lot of TCS software that are supporting The Operators to operate the factories but the operators are still overload it because more and more complex complexity of the whole system is introduced before one operator just need to take care of one machine today you need to supervise the whole process and also during the the the the the most arrested pandemic years this becomes really really the challenge so how how can we operate the factories of production facility uh with with less human integration with less uh tumor effort especially for the decision-making part and this is really really valuable valuable thing then uh the event triggered control this is another aspect from the control perspective this is not earth if it needed but it is very much meaningful after we have this loud follow infrastructure in either past the the networks the controllers are customized for the for the for the process and if you have some room unused capacity of the network and the computer this is just there nobody else can use it it's uh it's wasted of course but you don't have the way to to use utilize it but today after we have this virtualization after we have the infrastructure shared by multiple users multiple applications it becomes possible if we can save some traffic same some batteries from communication and Computing so this Savory courses saved boundaries is valuable for all the users it can be used by more applications or it can provide a better experience for the other applications if they are less critical that's control tasks so that becomes irrelevant if you haven't triggered the communication and compute the end control of course we can save a lot of resources so those benefits can be used by many other applications this will really like reduce the overall cost for the for the factory owners this can help the adoption of this new infrastructure of course we need to have the other use cases first this can be the secondary certain level and then the functional safety and and security again so I I like to emphasize this on the top actually security and safety they are not solely in the communication Computing or control it is really system uh level challenge and for example uh for the for the for the communication part uh let me see function safety from the communication perspective it's about deliver a message especially some alarming message maybe emergency stop message over the network we need to guarantee first this message should be delivered before the the deadline the deadline is calculated derived based on the safety process for example if you want to stop a mobile robot so if from you sending the message until the robot is really stopped there is some time breaking time it is not zero so when we make the machine when we make the robot we calculate this time we can estimate this time then your communication and Computing the whole process needs to guarantee you must deliver this message before this deadline so this is really required by the safety regulation and the second is you must make sure there's no error beat a single beat error in this message so this requirement is very critical sometimes it requires 10 to the power minus 13 minus 14 level beat error rate so you should tell me basically if there's any error in this message you should notify me there's the error maybe you cannot crack at it but basically you should detect it so you should very very provide a very very strong error detection coding in the message so this is the the practice today and sometimes it's called black Channel approach so that means the safety communication layer doesn't care what is uh that works we assume the network is not trusted it has packet loss it has error beats in the message we just take care of that ourselves from the top level well this introduced a lot to overhead and also makes this I even not possible today to meet all those requirements even we have improved the wireless very much then here I propose to have some great Channel approach green Channel means we need to really know uh the channel and from the control perspective we know uh this channel is is different from another Channel this one is uh 60 um it's a Wi-Fi seven it's not ethernet so let's keep this in mind let me do the control logic this is from the control side and so from the communication side a full double if the 6D Network knows that oh this packet is for safety message let's skip over all the other overheads let's apply the very strong error detection coding in this message and also we support that using as dedicated physical layer parameters you don't have to develop a totally new physically about you give this special care of this saved message and transmit that which is the minimum overhead and you do the automatic retransmission automatic or stronger Arrow extraction and so on in the physical layer in the low layer of the of the of the the wireless communication this I believe largely improve the performance but it is it is not there yet then another another aspect about cyber security before it was also like some black box uh approach uh the cyber security managers or engineers in the factories they have a very long list of the security policies for example you are not allowed to use cloud in this process you are not allowed to use a wireless communication you are not allowed to use a single Factor authentication for novel these are their policies so they follow the policies they do the assessment of the security and they do a lot of uh encryption decryption using the password but for me that's not enough and we need to know the process for novel imagine if the attacker has tracked all those uh encryption decryption mechanisms for example using some social engineering you lose your password this was really the the root cause of the most famous accidents security accident happened in the recent years so the VPN password was stolen that somewhere but some phishing email on just buying some face-to-face contact to the to the to the persons as it got the password then all your security mechanisms becomes useless but they really invade to the to the critical control system and then they send some fake command damage your system and here the gray box approach is exactly addressing that kind of issue if we know for example this machine the temperature cannot be higher than 100 degree if you send a send Point Which is higher than 100 degree it should be like stopped and also you will take a security alarm this is simple from the control perspective right so able to control Engineers if the physical knowledge if the process knowledge is engaged in the development of the the security process security solution this can really solve the problem this is what I call the greenbox approach so all both of them you can see may need the knowledge from the both side it's it's definitely cross-disciplinary research and that's also uh my my personal uh preference so in the in the company as kalena said we have the vertical coverage of the subjects as one person I I was like I forced myself to learn the things from all these subjects because sometimes of course when you formulate a product team you can have the person from control from communication from security from software engineering but I think to really have some scenery of the knowledge especially for this disruptive ideas you need to have something really close in the disciplinaries in in your brain so I do love to to to to learn this broad subjects and also put the knowledge together yeah yes this is the last slide I managed hey thank you
2023-03-21