Siemens, FFT & Databricks: Operationalizing Industrial AI from edge to cloud

Summary

Siemens and Databricks address IT/OT convergence challenges by combining Siemens Industrial Edge with the Databricks Data and AI platform to stream contextualized shopfloor data to the cloud in near real time.
The edge-to-cloud architecture allows AI models to be trained centrally on the Databricks Data and AI platform and then deployed back to factory-floor edge devices, enabling closed-loop industrial AI at global scale.
Real-world use cases such as Audi's quality monitoring and predictive maintenance demonstrate how the partnership bridges talent shortages, localization requirements, and operational flexibility challenges in manufacturing.

Siemens, FFT & Databricks: Operationalizing Industrial AI from edge to cloud

Industrial AI often stalls due to fragmented data, disconnected systems, and lack of scalable infrastructure to manage data integration and local AI execution. In this video, Siemens and Databricks show how an edge-to-cloud architecture enables an AI-ready data foundation by streaming contextualized shopfloor data to the cloud in near-real time, with Siemens Industrial Edge apps and FFT DataBridge app, training models centrally, and deploying them back to the edge—unlocking scalable, production-grade AI across global operations.
Talk By: Francisco Javier Franco Espinoza, Global Ecosystem Manager - Edge & Cloud, Siemens AG ;

Chapters

FAQs

What is IT/OT convergence and why is it difficult to achieve in manufacturing?

IT/OT convergence refers to integrating information technology systems with operational technology systems on the factory floor, which have historically operated in separate silos with different standards, teams, and tools. Challenges include organizational divides, retiring workforces with specialized technical knowledge, talent shortages, and the need to rebalance manufacturing capacity across geographies.

How does the Siemens and Databricks edge-to-cloud architecture work?

Siemens Industrial Edge apps and FFT DataBridge stream contextualized shopfloor data to the Databricks Data and AI platform in near real time, where models are trained centrally and then deployed back to edge devices for local AI execution. This creates a closed-loop system where insights from the factory floor continuously improve the AI models running on it.

What use case does Audi demonstrate in this video?

Audi's use case involves quality monitoring and predictive maintenance on the factory floor, leveraging the edge-to-cloud architecture to detect issues before they impact production. This illustrates how the Siemens-Databricks partnership can be applied to real production environments at a global automotive manufacturer.

What brownfield connectivity protocols does the architecture support?

The architecture supports OPC UA, Modbus S7, and MQTT, which are common industrial protocols found in existing brownfield factory equipment. This allows organizations to connect legacy manufacturing systems to the modern edge-to-cloud data pipeline without replacing their installed equipment.

Full transcript

[00:07] Good afternoon everybody. My name is Herman Lombard. Uh so I look after global ecosystems at data bricks. Uh it's wonderful having you all at uh at the summit. It's my first summit as an employee. So I'm going to be sharing the stage with uh Francisco uh or Paco for
[00:21] short. uh we've been working together for the last couple of months uh to get us to this point and I just wanted to recognize Justin Fenton uh sitting hopefully you can see me uh pretty key in in making all of this happen. Um so
[00:33] I'm really uh thrilled to be presenting on this topic today. Uh but before I get into it I just wanted to kind of just mention again like most of these announcements this is a forward-looking statement. So um so still a lot to come
[00:46] in this space. It's our first um partnership and venture with Seammens, but uh just note it's forward-looking statements that we'll be making uh based on the partnership that we've established. Um secondly, um please
[00:59] complete the surveys. It's really important for us to get feedback from you as
[01:12] and that you want to hear more of. So please take some time to complete your surveys um at the end of this. So the topic for today is IToT convergence. It's a topic that I'm personally passionate about. Um I joined
[01:25] from Deote. I was a partner in the Canadian practice uh for a few years before joining data bricks. Um and one of the the kind of roads that led me to join data bricks was the struggles that I had working with industrial customers
[01:38] in bringing it and OT together. Um it's a real challenge as I as I'm sure you're aware. Um it's both an organizational challenge, historical challenge and a tools challenge. So everybody kind of um
[01:50] you know across the divide of it still in most of our customers today are not really um operating in a harmonized and integrated way and that's what we're trying to solve partly with this partnership. Um and as you can see on
[02:03] screen there's a number of challenges that we need to address as we converge it and OT. So um a lot of demographic change uh especially now with a lot of kind of people reconsidering where to
[02:15] place manufacturing capacity and where to locate production facilities. Uh talent crunch is very real uh both in terms of our organizations and customers that want to grow but then also in terms of our customers um wanting to um you
[02:29] know um just sustain what they do today as we have a retiring workforce and losing a lot of technical skills. Second one is localization. So uh so we have again re rebalancing of the portfolios
[02:41] of where production happens but then at the same time um figuring out how to produce locally maybe in new jurisdictions but also keep in mind the global aspects of demand and then finally flexibility. So I'm a chemical
[02:54] engineer started my career in manufacturing. So my manufacturing plant back in the day was it was all about flexibility. How do we use the installed base of assets that we have and how do we make sure that we can optimize that
[03:06] and be flexible enough to uh to bring agility into the production line but then also have agility when we have shortage of supply or new increasing demand. So uh so IT and OT integration continues
[03:19] to be a challenge. Um and it's kind of the value of death is kind of what I call it is often a lot of um um successful data and or II initiatives basically needs input from both of these
[03:32] functions and because they don't historically are best friends because they don't historically talk and because they historically optimize for different KPIs in the business um that's why we see a lot of digital transformation
[03:44] initiatives fail historically. I I also see across our customer base that they haven't necessarily done uh either not done the work in terms of creating the necessary context for the operational data um or B uh they've done that but
[03:58] they've grown so quickly through acquisition that they've inherited a lot of technology and a lot of data and standards from other places and we're seeing a lot of consolidation and integration happening across the sector
[04:11] and then finally you know where we come in uh together in our partnership with Seammens now is saying that how do we make sure that we can create a trusted foundation of data sets uh that's contextualized within the context of
[04:23] your operation and then we able to bring that into the cloud for more sophisticated workloads. Uh this is a partnership that I'm personally extremely passionate and excited about. Seammens has been on the top of my list
[04:35] of partners to to announce this with. So um so we're really excited for for what we can do together. Um so some of the challenges that we think we'll be we'll be solving with this uh which is all the
[04:47] classic challenges I'm sure you as an audience face in your businesses today or if you're providers I see some of our friends uh there from ibite you know um we we face these challenges and we know that they're real um and I think these
[05:00] types of partnership goes a long way to make it a lot simpler for our customers to be able to bring that contextualized data across u you know interoperate across a very diverse stack of technologies that they might have across
[05:13] the factories bring that um make contextualize it harmonize it and then bring it into data bricks. So a lot of the pre-processing will happen in the industrial edge software of seammens and then depending on the use case that data
[05:26] can then now simply find its way to data bricks uh with a little bit more detail that PCO will share with you. So thanks again for the attention. We're going to do a bit of Q&A maybe at the end for a few minutes, but we also we're going to
[05:39] be sticking around for a little while to maybe answer some questions. So, Paka, over to you. >> Thank you, Herman. >> Thank you. >> All right. So now I want to talk about how Semens
[05:51] is answering um to these challenges and our answer uh to how customers can break these data silos and make the data AI ready is a platform called Simmons
[06:04] Industrial Edge. Um, Simmons Industrial Edge is a decentralized industrial-grade edge computing platform for solving these challenges that we've been talking about. Um, and implementing intelligent use cases that start with industrial
[06:18] connectivity, connecting to vendor agnostic uh, industrial assets. Um, doing IToT integration at scale um, integrating to enterprise systems to databases to the cloud. Um, but it goes
[06:31] beyond that. It's also about doing local analytics um executable digital twins right in the process um custom applications low code applications on premises um local AI execution and
[06:45] management. So the breadth of things that you can do um with a platform um goes beyond just uh connectivity and integration which is what makes it very powerful. Um it is not only a platform,
[06:59] not only a runtime for application, it is also an ecosystem of industrial applications from semens and third parties um third party partners some of which are sitting right now in the
[07:11] audience like our uh valued partners from Hibite which together um form a collection of let's say Lego blocks that you can put together to implement these kind of use
[07:23] cases at scale. Right? Uh the platform is also centrally managed which gives it that that scalability. You know once you define your standard architecture with industrial you can deploy it globally
[07:36] across your um uh facilities and it is a platform that's made for automation engineers um so that you can empower your um factory employees to implement
[07:48] and manage your um intelligent use cases in the shop floor. Now I want to talk about some examples that our global customers like Audi are
[08:00] implementing. So in the case of Audi they are standardizing um on the architecture on Simmons industrial edge to do things like um visual inspection for uh weld um welding splatter. So they
[08:15] are essentially uh taking images of the welding spots in the car um and detecting if there are defects on that on that weld. Uh previously this was um a manual uh a manual inspection. Now
[08:29] this is automated. So they are able to reduce the cost while increasing the quality of their products. Another example uh in this case now time series based is Impro. a seaman's
[08:41] partner and customer that develops technologies for production. Um they were having issues with um uh failures um on planned downtimes of pneumatic
[08:53] plants. Um and they leverage industrial edge in a standardized architecture across different production lines to uh deploy an AI model that predicts the failure of uh these uh pneumatic clamps
[09:06] and be able to then proactively reduce the downtime of uh of those uh components. So these are just a couple of examples of what customers can do or are doing
[09:18] right now um in real production environments with Simmons Industrial Edge and the ecosystem of applications and partners that we have around it. Um but edge computing is not the whole
[09:30] story. For many of the use cases that I mentioned, including the examples from uh from the customers, you also need scalable edge computing and services. And this is why we are
[09:43] partnering with datab bricks and um uh also with our joint partner FFT production system to bring direct connectivity from semens industrial edge to data bricks to make this whole
[09:55] process uh much easier for our customers. So essentially what you can do with this um collaboration is you will be able to
[10:09] load AI ready data directly from Simmons industrial edge to data bricks and the way to do this is you would you would install your distributed edge devices across your factories. you would install
[10:22] um the industrial edge connectors that would connect to the industrial assets from different vendors uh in multiple protocols. It doesn't matter if it's um seammens S7. Uh it can also be you know
[10:34] Modbus Ethernet IP or you know a Mitsubishi protocol or something standard like OPCUA etc etc etc. Um and once you have connected your assets, you can use our industrial edge uh
[10:48] industrial information hub application to begin to harmonize contextualize that data um and make it AI ready. From this point on um data uh the FFT data bridge application will take that AI ready data
[11:02] and load it directly into data bricks using the serial serial bus interface. Uh in this way uh you have your AI ready data in data bricks from from the get-go um in a managed in in a in a data bricks
[11:15] governed environment and from this point on you can use it uh to train machine learning models. You can use it to feed cloud agents with this kind with this structured data. You can do global
[11:27] analytics um because you can ingest data from different production facilities um and so on and so on. Now the type of use cases that you can
[11:40] do with all these it's not just about you know sending the data to the cloud um but industrial edge gives you the possibility to for example train AI models on data bricks and deploy them to
[11:53] the edge to be managed on premises to be monitored on premises and to be executed on premises. Why would you want to execute an AI model on premises? Well, if you have um a a closed loop
[12:07] application that's talking back to your machine and if you want the outcome of the AI model uh to be able to have an immediate effect in your machine like detect a defect and remove that product immediately um you need a very low in in
[12:21] u in many cases you need a very low latency to do this. So you will need to execute those models on premises uh for example with GPU acceleration. So um you would you would uh need local infrastructure to be able to to deploy
[12:34] that right um and um not only that but you normally do not want a let's say your data scientist to be pushing new production critical AI
[12:47] models directly from the cloud to production but you want to give a local repository to your factory employees to be able to manage uh those models and to be able to say when a particular model
[13:01] goes uh to production and be able to react if that model is not um is is not performing as expected. So with all of this infrastructure um you can you know
[13:13] go beyond IoT integration and do um advanced industrial analytics at the edge and in the cloud um you can do AI supported quality and and and predictive maintenance. You can do closed loop AI
[13:26] and you can do manufacturing agents whether they're running on the edge or in the cloud. So it's all about different types of smart manufacturing use cases.
[13:38] So what we have discussed right now is um you know the benefits that you get from this collaboration is that you can build a unified data foundation from edge to cloud with high quality data
[13:50] from the get-go in a datab bricks governed environment. You can go beyond data integration and optimize your operations with smart manufacturing use cases that span the edge in the cloud. And you can accelerate the time to value
[14:04] um from AI. You can deploy intelligent um uh AI models at the edge in closed loop talking back to your industrial assets for example you know physical AI
[14:17] use cases. Now, if you would like to learn more, I'd be happy to have a conversation uh today or tomorrow during the summit. You can reach out to to our inbox uh to have a conversation later and you can uh
[14:31] check out these resources to uh learn more about our uh collaboration and the broader industrial ecosystem. Thank you very much. >> Thanks.
[14:44] We'll be taking some Q&A. >> So, happy to take any questions if anybody has. Yeah. So >> hi >> um can you go back to your slides where you was actually your whole architecture where you had data bricks where you
[14:58] activated the final layer. Yeah, here sorry how I saw that you have a SAT distributed for PLC. So different modalities um coming through data bridge
[15:10] how actually you are stitching to them like together how you marrying this all this modality together which lay is responsible for that >> sorry I I I didn't hear um uh the first
[15:22] part of your question how are we connecting to what to the industrial assets >> all all this modality like SHDD distributed control system PLC I see different modalities >> so all of these industrial assets um communicate with different protocols So
[15:36] if we're talking about brownfield assets, you'll find things like you know mudbos like semens S7 for semen PLC's um you know Rockwell PLC's will talk Ethernet IP and you know every vendor
[15:49] will have its own protocol um newer industrial assets may rely on on protocols like OPCUA. So this um in the industrial ecosystem has um specialized
[16:03] connectors that talk all of these protocols. So you can connect to all of these assets. >> But are you merging them together >> one single insight from the from all of them like let's say I won't see one
[16:16] picture eight pictures I won't see one painting. So once you get the data from all of these different assets, you would use an application like um industrial information hop or alternatives from our
[16:29] ecosystem like hibite to um uh create a semantic model that's um uh kind of like unifying all of this data. You can harmonize because the data might be coming in for example in different
[16:41] units. um you may want to kind of like rename those variables uh because in the brown field you will find that some of these um so-called tax may come in a
[16:54] format that's or with a name that's unintelligible and unusable. So that's why you always need some sort of integration layer um to harmonize and contextualize that data and from this point on you can reuse it at the edge um
[17:09] you know at the machine edge at the factory edge in the cloud and so on. So in this case in this particular architecture uh we would be using the Simmons industrial information hub application to uh put this together.
[17:21] >> Okay, got it. Thank you. So um so are you guys looking for something
[17:34] like MQTT as your data broker or you said Hibbyte is it's the integration layer. >> So Hibite is um a recently onboarded partner to our ecosystem. Um, so you can
[17:48] definitely use H Highvite on Industrial Edge or connect to H Highite. Um, >> and they're using MQTT as the broker or >> so they support UNS and I'm pretty sure they support MQTT, but they're right
[18:01] there. So um, uh, how do the our colleagues from HY would be able to tell you exactly, you know, which kind of like protocols they they manage. Uh but talking about the broader industrial
[18:13] ecosystem uh we definitely have MQTT connectivity. So if you have let's say an external MQTT broker at the factory level that you want to connect to um through MQTT um you can you can connect
[18:26] there as well and then from that broker to data bricks if that's the architecture that you want to use.
[18:40] um like >> sorry um SACD or this you mention or do you guys have a real time kind of workflow which actually gives a real fines let's say you have a um I know maybe it's not health domain but
[18:53] somebody has like a cardio episode how how this system going to work if you want a real time inside but sent to the doctors >> um sorry could you repeat the question >> so let's say we have a kg machine sensor
[19:07] on the raw data layer and it collects the AKG information real time. Is this does this system allow you to kind of send insight that there's something going to good bad going to happen to the
[19:19] doctors from the final layer which is data bricks to the doctors? >> Um well as assuming um we can connect to that um machine it's a it's a um medical
[19:31] machine that you mentioned. >> Yeah. Yeah. Like a KG machine. >> Got it. So depending on the protocol that that machine talks um I mean you can always deploy a custom connector right but let's assume we can connect to the machine um you can send that data to
[19:44] data bricks and I'm pretty sure um in data bricks you have some sort of service to notify whoever needs to be notified that that's there's an anomaly um ongoing >> but it's real time possible that's what I'm saying really
[19:56] >> um >> close to we call this connectivity near real time so in in automation when we say real time um we mean something very specific. So all of these industrial um
[20:09] well sorry industrial assets such as PLC's they talk real time meaning they guarantee that they will execute the logic and produce the result in a certain amount of time which is normally
[20:21] milliseconds. So it's normally immediate and deterministic. That's what we call real time. Um when we connect to the cloud often what we say it's it's near real time because there's no determinism. um to to the cloud, right?
[20:34] So you're you're talking maybe about, you know, a few seconds of uh latency. If that's sufficient for your application, you can do that. If you need low latency, you often would need to execute that workload right next to
[20:47] the process on premises. >> Yeah. And I'll say that um so near real time is what most of our customers currently use, but as per the announcement this morning, we've now got real time lat. So we bring the latency
[21:00] down even on on some of those workloads as well. >> Okay. >> All right everyone. >> Well, thanks a lot everybody and you know happy to get in touch with you. >> Thanks. Enjoy the show.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.