Scale By The Bay 2020: Dirk Slama, AIoT: Why now? And How To?

Scale By The Bay 2020: Dirk Slama, AIoT: Why now? And How To?

Recording: Scale By The Bay 2020: Dirk Slama, AIoT: Why now? And How To?

[Music] thanks a lot and hello everybody this is uh dirk here um yeah so i would like to um take a look at uh ai and iot why now and how to um so um as devin said this is not new right so we've been doing iot we've been doing ai um depending on who you ask for years or decades yeah um and um if you talk to iot folks they will always say like of course i mean ai is something that's kind of like a part of iot and we do predictive maintenance and preventive maintenance and all these kind of things and if you talk to ai people they will tell you like yeah of course we can process sensor data et cetera and then make use of it but what we're really now seeing is that um these two concepts um are more and more um to be seen in in context and um especially in larger organizations um you will often have a problem that they are to a certain extent siloed so you will have maybe a center of excellence for ai you will have the center of excellence for iot you will have on the operations side uh people focusing on the one topic or the other and the question really is um how do you bring all of this together how do you enable um iot with edge and valve computing and then ai with intelligent assets on the edge and also swarm intelligence in the back end and what we see here is that um really the first step is that um you have um a lot of deep ai knowledge experts who can tell you everything about the different algorithms and their use cases etc and then on the other hand we have experts who know everything about certain asset categories they know everything about manufacturing these assets they know everything about operating these assets and bringing these two worlds together is not easy um i have brought with me a couple of examples um just to talk about where we are there at the moment i mean um obviously everybody has been talking about uh awesome striving uh the last couple of years and it is happening i i think um we're still at this point where we have realized okay going from 99.5 to 100 percent is a bigger step than we all had hoped and yes there are regulatory hurdles etc and everybody's taking a slightly different take on it how much of this are you rolling out now in the public and which steps i've brought with me here one example um bamler and bosh uh rolling out uh autumn's driving in the public now for example for autonomous parking as a first step um this is not so much i guess um a technical hurdle we're taking here but really um getting this out into the real world and then obviously um the whole discussion around uh autonomous shuttles and that's i think the first good example for the difference between asset intelligence making the asset itself more intelligent and enabling awesomeness driving versus managing a swarm or a fleet of assets for example a fleet of shuttles and then solving the problem of matching demand of people who want to use these shuttles through the available shuttles in the backend so two different types of a iot um another example is um the smart connected lawn mower yeah which is obviously an i a sorry an iot solution yeah but also becoming more and more intelligent not only to optimize the way how uh the the lawn is mode but for example also to prevent that um the more is not uh interfering with any uh wildlife uh in the green etc so that's also a good example another example um you can see here a smart connected tightening tool this is not you know the type of tightening tool that you and i would buy in a walmart or so uh this is actually a high-end industrial tightening tools that's used by manufacturing um customers who um depending on the type of factory we have customers who perform 100 million tightening processes per year and each of them in a a mission critical tightening process so um we've made this type of tool not only connected but also intelligent that for each tightening process you can basically control torque and angle and have individual tightening programs that are specific to the type of material type of screw you're using the coating on the screw etc and then in addition to the asset intelligence to control the individual typing process we also have swarm intelligence that basically captures data from uh all the tools in one fleet of tools in a factory to basically derive knowledge about uh potential uh problem areas and automatically make proposals for solving them i brought another example here with me which is actually not the bosch example it's the uh large hadron collider at cern in geneva and it's uh potentially one of the biggest aiot applications i have come across so far it's basically controlling a um 27 kilometer long particle collider and then deploying a million sensors in a collision chamber to basically detect and analyze particle collisions for example the higgs boson so all of these are different types of aiot products or solutions and this is where actually one important differentiation starts that we have learned you really have to understand what is it that you're building is it something that's highly productized is it a smart connected product um or is it actually um a a once-off um solution and some some some piece of critical infrastructure like for example a traffic network uh operated by a traffic authority etc where kind of uh different kpis are important so for products it's typically about usability and um revenue models whereas for solutions it's typically more about [Music] efficiency and um yeah basically making um these uh different types of sensors and um products that you're deploying the solution um operate as efficiently as possible so how do we basically build these these kind of products and or solutions we have developed um a best practices framework from this um with the aiot user group it's an open group of iot practitioners and we have learned that you really need to bring these two different worlds together so you have the agile cloud world yeah and then you typically have the not so agile very safety uh concerned world of um building uh embedded software uh hardware etc that's uh deployed on the asset or most of the assets and uh that's um really um the first challenge that you need to overcome this is where we've developed um as part of this framework um recommendations how to basically bring these two worlds together we have learned also that you can't use a single answer for a product organization that needs to work this way versus a solution organization so we're making a differentiation here and have different uh blueprints um developed for this on the organizational level and and we've learned that this is really one of the key challenges that you have to overcome um starting with the ai perspective the ai people it's usually a world on their own they have their own processes going about understanding the business problem understanding um the data that they have or can get yeah preparing the data creating models and then deploying them either in the back end or on the asset itself um that's one key challenge that we're currently seeing so uh obviously there's def ops out there and i'm assuming if you're working for a hip startup in silicon valley this is what you do day and night if you are working for the it department of a large manufacturing company that's still struggling to get um stuff to the cloud and still uh this is running a lot of things um on premise then um you might not be completely there and now in addition to um these um you know devops um related challenges the next thing um that's coming is that suddenly in addition to managing um your code and your applications you also have to to manage uh ai and your ai models in the same way and so ml ops is kicking and then you learn that um all of this uh in terms of deployment um in an alt solution uh is also not uh only um you know deployment in the cloud but it's potentially um deployment on the asset so suddenly over the air updates um become a huge deal and need to be integrated um in your devops cycle and again that's the world on its own not only from a you know deploying updates uh to a potentially heterogeneous um asset with potentially um multiple target platforms uh multiple bus systems etc but also from a security point of view from safety point of view managing update campaigns uh and so on as well uh it's it's really the challenge to bring all of these things um together and then from a iot project point of view as i mentioned um ideally you really want to run this as agile as possible but there will be some parts um where agile simply um does not work so procurement and sourcing yeah this is something where at least i think nobody has found the silver bullet how to combine agile and sourcing right so especially when um you have to source complex hardware yeah that's involving potential custom manufacturing um potential uh variants of the software because you have to support different chipsets for different regions uh that work on different frequencies and so on yeah you suddenly are living in a world that requires a lot more long-term um planning ahead than than you would typically have in a pure agile organization right the same applies to the safety aspects of this right so uh that's why um we we have developed some some blueprints for uh how a uh per se agile um product organization works for for a iot and taking the non-agile aspects of all of this into consideration as well um when you look at this beyond software and beyond the hardware that's that's running the ait software and the sensors you realize that there's also a whole other world if you look at these quadrants we have the digital world the products that you roll out and the way how you produce it using some of the things that are just outlined but then you also have the physical world right so how do you make sure let's say for example you're building your next generation uh forklifts and you want to uh build um additional sensors for location tracking and maybe proximity warning or something like this in your forklift then this is not only about everything we discussed up until now yeah but this is also about then um extending your physical product making sure that the additional sensors that you're building in for example can be connected to the powertrain can be connected to any communication buses you have already on the physical asset et cetera so this is why um these things really have to um work hand in hand and um when you come back to the uh ait organizational model that i was talking about earlier on uh there's for example one important organizational boundary which we call asset preparation so in the uh airt model this is the work stream at the very bottom that has a dotted line between the main organization and um that's basically related to the physical asset design so your digital uh product engineering and also then of course to the manufacturing itself right so i mean not only um the need for for for using more and more advanced uh manufacturing ways uh of setting this up but also uh suddenly challenges for example how um do you validate certain intelligent functions um of the products that you are building in your manufacturing center and so the manufacturing for example in the past you know kind of like the the height of uh quality assurance manufacturing was that you closed the car door and then there was this person standing next to the car listening to the noise that the closing car door makes and if it's not a satisfying flop then you know you have a quality issue right but how do you basically uh do qa from a manufacturing point of view for um all of the intelligence that that you're building and so on yes so um this is something uh that's happening yeah and and has successfully happened uh but i really think we've seen only the beginning of it um let me talk a little bit about uh what bosch is doing in the space so bosch for those of you who don't know bosch very well we are a multinational engineering and manufacturing company uh in automotive and white goods and energy and building technology and so on we have about uh 400 000 people working in our engineering organizations and also in our factories and um we have i mean as you saw earlier on um a lot of products out there already who are applying this but um i still think this is just the beginning of the journey and um in these different areas that need to play together we have quite a lot of activities so in terms of products smart connected products i was giving you some examples from mobility industry energy building and residential and so on um the digital product engineering um this is something that we see i mean this has been strategic obviously already for the last decade but now with you know digital twin all these uh strategic new initiatives coming in we really need uh to further uh advance this so um we have um a holistic plm initiative here um then of course uh factory of the future we have um dedicated business units so for example our uh rexroth unit that's a multi-billion business supplying components for the factory of the future and then when it comes to iot products we have for example our global ai units we have an embedded systems unit we have an industrial engineering unit we have an iot unit we have sensor attack that's a mass manufacturer of micro electronic mechanical sensors mem sensors and so on we have another unit that's more specializing on the initial phases of smaller scale custom outdoor development et cetera and all of this basically plays together to realize uh this vision of iot and to come to the uh other question as we we've looked a lot at the how-to uh so now let's also look at the why now uh i mean obviously there is no lack of analyst forecasts that promise to you that this is a gazillion dollar market opportunity i decided to spare you these type of slides today um but um generally speaking yeah um there is so much activity in the space that i think everybody uh who's involved in their uh greece uh yes this this is happening on the other hand i think you can also see that this is still very early stages yeah so we're not living in a back to future world yet where you know everything is smart and intelligent and you know has overboard type of uh capabilities so we really think uh this is just the beginning and this is gonna keep us busy for the next 25 years um if you look back the last 25 years what what what happened at internet speeds yeah um if you combine internet speeds with manufacturing speed which by its nature is orders of magnitude slower than internet speed and you take the average of this um you can still see that these are um yeah basically much longer lead times as soon as physical products are involved that require physical product design manufacturing line setup and so on then you know just putting an ndp together in a cloud and be done with it over the weekend so this is uh why we think um yeah this this is a long game but as i said uh we already see that this is significant kindly starting to transform many asset-centric industries and yes there is a steep learning curve and that's why what we see the time really is now to start building up your skills and yes you will most likely already have iot skills you will most likely have ai skills you will already have backstreet of the future skills but the trick is really um to orchestrate all of this to make this work together right and uh that's what we see um is um the time to do now so with this um i hope i i was able to share an interesting um perspective with you uh answer uh the questions as promised even at the beginning of the presentation and we would now have uh some time for questions from you [Music] you