EP 16: The Manufacturing Journey to Cloud Adoption w/ Arila Barnes

ABOUT THIS EPISODE

It’s a challenge to transition to the cloud under most circumstances, but the unique nature of manufacturing poses even more difficulties.

A recent guest on Unleash IT is Arila Barnes, Chief Cloud Architect at Standard Industries, a company with a 140-year history in industrialism.

Arila and host André Christ discussed, among many other things, how manufacturing can transition to the cloud without stopping the production line or while managing terabytes of data per week.

They also spoke about the benefits of adopting the cloud in manufacturing, challenges of cloud adoption, data storage concerns, the optimal team for cloud adoption, and cloud technologies in manufacturing in the near future

To hear this interview and many more like it, subscribe to the Unleash IT Podcast on Apple Podcasts, Spotify, or our website.

Listening on a desktop and can’t see the links? Just search for Unleash IT in your favorite podcast player.

Well, major challenge for manufacturing is the nature of the business right it runs twenty four seven. It's pretty important to keep the production a line running. So planning is went to introduced changes can be challenging to coordinate around plant downtime and to be able to distill the use cases that can justify this step of the taking the risk. So most manufacturers are pretty risk conversed when it comes to technologies that can have a potential disruption to the day to day operations. Welcome to unleash it, a podcast where we discuss the experiences and ideas behind what's working in enterprise architecture and digital transformation within the IT landscape. Unlock Your Business has digital capabilities. Transform your enterprise architecture. Unleash it. Let's get into the show. Welcome to this new episode of the unleash it podcast. Thank you to all our listeners to tuning in again and especially a big thank you to our guests today, a Reala Barnes, who is a chief cloud architect at standard industries. So I'm excited that I have the opportunity today to learn more about her role at at her company and also her experience in the field. So welcome, a rela. Please tell our listeners a bit about yourself, about your job at standard industries, as well as your prior experience at Ge Great. Thank you, Andre, for the opportunity. At Standard Industries. I lead the cloud strategy and engineering team. Our team focuses on the SI data platform and how to leverage mlai capabilities for standard and its operating companies. So Standard Industries is the holding company. It has several large manufacturers in the roofing industry, in the US GF focused on shingles, and in the EU B am, which is focused on clay tiles and concrete tiles. It also has a portfolio of real state venture investment and most recently has added slow roofs through its GF energy arm. So that's a good mix of manufacturing use cases, both from a pure plant manufacturing scenarios, all the challenges of supply chain around dram materials and emerging markets like the clean tech through stolar. Yeah, when I prepared for the podcast with you, I did a little research on the company and I think I've found a date. Is it true...

...that than industries was founded one of it and forty years ago? Is that? Is that about correct? So it has a long history, of course as a business. Yeah, it's interesting. So it's very old company, family owned over a hundred and forty years, has grown through acquisitions with the go to add a modern industrialism approach. What that means. It's really forward looking at technologies, including cloud to and machine learning to bring the mine keep pace with the changing factors in the environment and with changing challenges that presented to manufacturing. Like, for example, covid that was a big challenge across the world, especially in manufacturing, last year. I can imagine that that it has changed a lot of the processes and a lot of the like working habits. Are there any any other examples where, would say, even outside of Covid what are some of the requirements for modern company today in this field? So what has changed over the years? Where also it can make a difference for standard industries? It's actually beginning. So last you like, the last couple of years, standard industry has decided to make an investment in bringing industrial iot platforms and capabilities to its plant operations. What is challenging in standard industry, space perche in particular, is the nature of the processes. So the processing is in the equipment of making concrete, making us foult singles or concrete tiles or clay tiles is in some cases over a mile long. So you get getting into really dusty environment that with not easy visibility. Like you cannot have visibility from end to end on the entire process. There's a lot of variables around the raw material mix and a lot of variables between sites, like there's no true manufacturing plants that are like so that it's been and it's also very traditional industry. Right. Towel roofs and concrete roofs have been around in certain cases for modern fifty hundred years old and in even a fout shingles in the United States can last fifty years on a roof. So what has changed right as this multiple disruptors. One is aggregation of businesses and consolidation of business.

So how do you leverage best practices between those different points? Another challenge is like how do you continue to respond to demand? So it's a business that's doing very well and the more natural disasters they are, the more that it has a direct impact on the demand for roofs. For example, the hurricane season in the United States drives a lot of demand for roofs, for new roofs. Okay, yeah, that gives a good background, like the starting point. I understood when we prepared this podcast that the clouds and the migration to the cloud plays a big role in this. So where would you say in this journey are you today? Are you at, and what's the pace this is going at this adoption? Yes, so where cloud comes into the picture is enabling visibility across the fleet of plants. Right. So say, if how can you putting the data of all your plants in one place to enable standardization around key performance indicators to also be able to provide data for enabling new type of analytics? In our case we focus on quality control around the shingles. So that has been a very challenging problem of figuring out the defects on a black asphalts shingle, with black minerals, all of it. So it's not as easy as recognizing a picture of a dog or cat for most visualization algorithms out there. It's actually involves a lot of coordination, a lot of trial and error, a lot of partnership with universities and cross organizational partnership with the subject matter experts at gf from process, from to distinguish what are the defects and what a defect means on asphalt shingle and how to leverage how to make sure that can grow into a product, by leveraging cloud, by leveraging machine learning, by leveraging IUT platform, and where all the pieces fall into place. How the pieces fall into place. What would you say are the top benefits and values for adopting the cloud in manufacturing, like in some of the use cases you mentioned? Is it to have access to more computing power, to better machine learning, to a central storage, to more elasticity when there's demands? So and then you can scale up and scale down. So what parts of the cloud benefits are most relevant for your use case?...

It's access to data, in its access to be able to share insights. That's what main benefits of the cloud. Of course, managing compute as it's needed. So unlike an on premp situation where they needs to be a lot of planning in a lot of investment upfront of the computer resources, cloud gives the advantage of being able to scale that as needed based on demand. So in our case, what leveraging the cloud has allowed us is to have a very stable and scalable approach to migrating plants on their own schedule, one at a time in the cloud, without oversizing the system. So you know, say we have twenty plants. Traditionally you need to get a data sentesized for twenty plants and by the time the migration plan is completed that data symptom might be obsolete, already needing an upgrade in hardware. That's something that the cloud solves, both in the initial setup and also ongoing maintenance and the reliability of the system as the adoption of new technologies and as the migration of existing components is moved in that direction in the cloud. Can you give us an example of our GIG give some ideas of what's what type of services you're using in the cloud today? Is it, I mean, I understand its storage of the data to to make that accessible, but are you using like specific services provided by the cloud provide us? By the way? Okay, can you share if you are on one cloud provider or on multiples? So get maybe provide a bit of an insights into what parts of the clouds are you using and what services are most important to you today? Yes, so we currently leveraging the Google cloud or wherever we also interesting infrastructures code so we can be cloud agnostic in the future. By would you. As I mentioned, we build the data platform it standard industry, leveraging Google capabilities through their manage services. So, for example, Bi Query and cloud storage for storing both road data and transformed data. That is transformed to enable various reporting capabilities. And we leverage machine learning, how to ML and vision, ai document ai capabilities through our cloud partners to analyze some of that data, and also a...

...text analytics capabilities to normalize some of the discrepancies in naming of dags, naming of columns. Each plant has its own variations of what they call, how they describe a temperature, humidity or some other parameter at the plant level. Okay, okay, yeah, that's that's interesting. And this it took two major use cases when it comes to manufacturing. One is focused on the digital twin. What that means is like how what kind of data do I need to have, like a digital representation of my process or my asset, so that I can make better decisions of maintaining that asset or configuring that asset for the future. Same thing for processes and those differ. In some cases that data can be based on physics models. Take that into consideration. In some cases it's more like statistical analysis techniques that bring to the surface of what is important to be addressed from an operations perspective. Okay, what do you find the specific challenges or pain points you encountered when you decided to move to the cloud and and adopt more of the services? Can you share some examples where it was more difficult to actually get this done or where you were facing hurdles in this adoption? Well, major challenge for manufacturing is the nature of the business right. It runs two seven. It's pretty important to keep the production line running. So planning is went. To introduce changes can be challenging, to coordinate around planned downtime and to be able to distill the use cases that can justify this step of the taking the risk. So most manufacturers are pretty risk conversed when it comes to technologies that can have a potential disruption to the day to day operations. So navigating the stakeholders concern, in navigating the use cases that in focus on the use cases that bring the most value is the challenge and has been the challenge, not just that standard. From my experience, same challenges existed edge renewable energy in regard to hydroplants or wind farms, Solo arms, any time that this is expectation for continuous production contention with changes...

...of the digital system and the digital and the processes and workflows that those changes introduce at plant level as well. So in our case, to move from to be able to take advantage of the of the cloud, what needed to happen first is update the network to handle the additional traffic and also to provide the necessary security layer isolation between production networks and the network that communicates to the cloud. So that was an investment that needed to happen to be able to fully take advantage of getting the data near real time into the cloud and have that eventually the bidirectional flow with information so that the operators can be enabled not just to monitor what's going on in the clouds scenario, but also to be able to take action based on those inside. Can you explain a bit more this fact what you said? You cannot actually stop a plant from working because they're producing seven how do you do that then? So, like do you need to agree then on downtime windows and how do you cut them? Cut them short so or how do you keep them as short as possible? So how do you how do you have you actually then introduced new topics? Is it that you try to have like always very short time windows where you are down, or do you put a lot of the changes into into one big big updates and then go from there? So what's the best strategy? Actually, they are you using? As it's I kind of figure out the path of list list resistance. For example, if there is a process that already sends the data with it's full back up or for some other purposes, to the cloud is, how can we level reach that process and also work on what are the incremental changes needed? So we were able to get the data first and work independently from the plant operations to get the pieces in the clouds to that confugue figured and ready. So we started with a one point data valid dated the pipelines are working correctly and then expanded to the entire data set. So that has worked very well. Another strategy that works well is identify a proof of concept and, of course, in that case, schedule around plant downtime to get it set up and a third strategy is to establish that works well for sensor data. Additional sense of data is to have a plow set up to that is independent of the production systems, at least in the...

...initial phase, and once that is working, to roll it out in a moint integrated in the integrated fashion. So all this integration, but we also taking advantage is of Iot Platform. So we're not solving all of those pieces ourselves. We rely on on partners that have deep expertise in the industrial iot platform. That makes sure that those concerns addressed. Of like being able to collect data at plant level, make sure that data is buffered for local usage in case connectivity is broken between the cloud and the independ the plant be able to forward relevant information to the cloud and be able to think the data in between. So if person needs to needs to react in real time, they get the information right there at the plant level. If this more time, for example, if we just want to see how did we do yesterday or the week before, and ransom forecasting or some analytics, predictive analytics on top of that, that's when the cloud can't work independently and various and now it is the various challenges of the data itself. For example, at a hydro plant, the sense of data flows at such high frequency the that to be effective at modeling and analyzing that data and building a digital between the text into consideration the specifics of the hydro turbines, that was not a physical use case to bring all of that data, or most of that data to clouds. So that was more like a balance of like how much do you to at onprent side. So you take out of its kind of having a mini, mini data center inside the hydroplant for the analytics and basically the leverage the cloud from an asset performance management to see across hydro plants and standardize around the hydro hydro model and basically report more on the on the results. It's kind of coordinating this local load of analytics versus what can be done at that cloud level for wind farms and solar farms. The challenges. So it's not so frequently updated like hydro every second. It's more like you can have more like ten minutes. Is Okay. However, it's very lots of details like it in...

...the magnitude of eight hundred or plus tags about an equipment, you know, betting from different parameters, round vibration, temperature, wind direction angles, like all kinds of information that is of interest to the to the manufacturer about to be able to analyze how does the equipment behave and react to in the field and be able to proactively monitor haze behaving up to space? Is this variance from expected engineering expectations? So in that case was like all these data that fleets become pretty big, like in wind, just like thirty fivezero plus wind turbines, which is different scale than a few hundred hydro plants. And when it gets to those solar space you can explode into the millions of solar arrays that need to be monitored, and that's from an industrial perspective, and if you get to individual devices, you can get into the billions. That's what makes it so interesting and challenging. Yeah, and I think I understand why you said in the beginning it's important first of all to have good network connection, because the data you generating, the more the more senses you put out, the more data you generating. And then obviously you need to think about how you get that to a central place. What I wonder is how much is it a concern to actually store all the data. So I'm thinking a bit about like retention of the data. How long do you want to keep all the all those masses and until you like aggregate and more? Is that a topic in terms of like also cost sensitivity, like storing large amounts of the sense of data, or isn't? Is that a topic for you at the moment? Yeah, in some cases it is like that was like a full steam turbines and gas turbines, a General Electric the volume of sensor data is huge, so that that's like, you know, in the terabytes and petabytes, reaching like you know, on the weekly basis. What what is different? That's not the always the case. So, for example, its standard industry, the data is much smaller. But then it's like the quality of the data can be a challenge and and then it's also we might be missing data, right. So it's kind of figuring out what instrument instrumentation needs to be added in addition to what they are already to solve a particular problem.

So it's not just like in some cases we have a lots of data, in some cases there is no data and well, like the data is really messy and the ideas like how do you like this whole challenge of what's the best approach of addressing that? For Analytical Purposes, any data between six months to two years, that's a best practice to have at least that much data available to be able to leverage a machine learning type of model based on that data, and the rest of the data you would put, like, on long term storage, which is then cheaper. Yeah, so that another one is like if you if this requirements to keep the data a bit beyond two years, then there's that kind of management into place. Cold storage versus hot storage, and that's another advantage of cloud is that you can, based on how often you are accessing data, you can take advantage of those different scenarios of storing the data. Cold storage is when the plan is to access the data maybe once a year, versus hot is it's ready to to be queried as needed. And there is also the in between, like be able to quit every quarter versus every year, and then there's the whole complexity of data engineering. So maybe I'm over simplifying, but all of those details, because of the scales, like hundreds and thousands of dags in certain cases describing in detail a particular equipment of particular process. Again, there was discrepancies what that information model is. In certain cases they standards to address the information models, in other cases they are not. So that's another thing we focusing on is identifying the common information model that we can standardized and normalize against as we move the move our partners into the digital space. Okay, I mean it sounds like a lot of very interesting architectural questions. I mean, and you need to solve like not not only on on like what is like the best way to transition into the clouds, to adopt plans, but also how do you get the data efficiency efficiently into to one place? How do you standardize it? How do you deal with data quality topics? How do you do that today? Do you have like a team of architects coming up with those topics? House that organized? How big is the team which is like taking care of that? Maybe you can give give our listeners a bit of an idea how this is organized today. Yeah, I mean it definitely to take advantage of technology. I recommend investing in data engineering and those are the people that understand the day, the source of the data and the destination and all...

...the transformations in between. So typical technique is like a load extract load, etl extract transform load. It's like load transform extract pattern. So to be able to preserve both, like the raw data as is, which usually it's what might end up in cold storage, to someplace, but also to the various levels of transformation for different purposes. Maybe one type of transformation is to take care of privacy concerns BII data, personally identifiable information, and now that's not so such a challenge for most industrial cases, but in some some equipment it also has like, for example, or geolocation information that can be used as identify. Maybe to come back just requick to the question I had. I mean, like it sounds like you have a lot of topics going on in parallel. So so I'm trying to get a better sense of how big the team is. How many people are like working on cloud architecture, how many people are working in in like broader it or data engineering? So maybe you can give us a feeling for what size, how many, how many people are working on this today? Yeah, I would say like it is important to invest in a data engineering team. So that depends on the of the size of the use case and where you are in the transformation journey. For example, if you just like it, starting to for proof of concert since May might be a team of three people. You know, when you reach the size of managing the fleets of gas turbines, that can be like, you know, team of hundreds of data engineers to continuously improve the data pipelines and run those in the cloud. The other skill set that's very important to invest especially on the cloud side, is the develops. So you have a you have a team which does all the infrastructure as a code and prepares that. So is that so? This like the DEVEOP steam? So the focus of the develop steam is to properly configure the infrastructure itself, in regard to Compute, in regard to storage, in regard to security controls, like who can have access to what data went, and by having all of that in code, it makes it easy to have a repeatable patterns. One thing. It's easy to provide audits and layout security concerns on top of that, and it also protects the business from a lock in with a particular cloud venom. So those are all like it's totally worth it to have that investment. The other one is the data engineering so that is the people that work and wrangle all those data, because it's not perfect. Like I said, that's...

...the challenge of industrial data. It's not perfect, it's dirty, it's like has different quality standards. Sometimes it's complies with Eyeso Code, sometimes it doesn't, depending of sometimes there's a proprietary naming conventions stand to be able to align all of that to a stand a destination of who they like, common information model, that standard that is normalized quality at the right level of quality as needed and also feeling in missing data in certain cases or generating aggregated data as needed so that is ready for additional qualies versus cutting to crunch compute all the date all the time, all of these tradeoffs as like you know, what is what data makes sense to bring in in patches? How large those batches? How often do they happen? What is more like of a streaming decision? Right, a lot of people take advantage of Kafka in Google Cloud. That's POPs up. So it left each pops up as the main pattern of coordinating all those moving pieces and all those different sources so that we can keep it saying, keep it saying the system grows and to be able to not have like a linear more like take advantage of automation as well, so you don't need to like, you know, three engineers, thirty engineers, three hundred engineers, like they is short, but you have scale and efficiency over time. Yeah, yes, scale to reach scale and efficiency over time, yes, but definitely those are the skills to invest data engineering and they're bobs, and also at the field level of we dealing with a lot of devices, a lot of servers, a lot of sensors that deployed in the field. So IOT field engineers. So those are the people that can understand the domain of the manufacturer and also understand what are the challenges involved in working with the device specific or plant specific setup and how to keep that in Stinc with what's in the cloud. So another technology that helps with that is containerization. So we leverage COUGARNETI's and GKA and those and those kinds of technologies to make sure that, through containers, we can have that consistency and automation enablement for delivering of analytics, delivering of applications. Again, to be able to have it secure, to have it portable and have it manageable within the constraints of budget constraints, tank constraints and form factor constraints of certain devices, you know,...

...like we want to in some case want to have an analytic that can run on a Rasperry Pie. In other cases it's an entire server, I raq, deployed locally and, depending on the judge, can be an entire cluster in the cloud. Okay, yeah, look, I read that this was like really, really fascinating and deep insights into what are the use cases? How do you solve it with the cloud technologies? Really great to see and understand some of the challenges you actually have in specifically the the industry you're working and so I want to say a big thank you for allowing us that inside today. If you probably as an as one of the questions I'd like to ask if, if you if you think I had in three five years, what do you envision are the biggest changes through cloud technologies in your industry? What would you say like a couple of years forward? What is what have we solved in your industry and made made so much better? Can you give some ideas there? Yes, I see a maturity. The landscape of Industrial Iot platforms maturing and also I see a trend among the IOT platforms is specializing and specific industrial use cases or industrial verticals, for example. Some iot platforms a better student for process oriented challenges versus equipment oriented challenges. Some platforms a better of a managing fleet of plants, others a better IOT platforms are better managing fleet of devices. So they're becoming more specialized. Saying yes and that becomes that is a function of the partnership. Like I said, it was very hard to interrupt the manufacturing business. So how do you build that partnership and how do you grow the technology through that partnership? So that's one trend I see. Another one is like more investment in the analytics. Now that we have the data, now that it's connected, how do we take advantage in analytics? And it can be simple analytics, it can be more complex analytics, like I mentioned, vision analytics to distinguished defects on black on Black Bose, and also taking advantage of virtual reality, which right now it's still more like in a proof of concept kind of like all the experimental stages. Can I trouble so should an equipment problem remotely through virtual reality, or can I train employees new employees on that equipment, on that process. I see that getting a little bit more mainstreament integrated as the network's improved, as the businesses are becoming more comfortable with leveraging the cloud versus just having a lockdown...

...environment at the plantlet more, I see those rowing and leveraging moutonomous kind of just like the self driving cars. How do you leverage some autonomous equipment for moving my Tudios at the yard level and elsewhere? So coordinating all of that and making sure all of that is closin as system is a very interesting challenge and that's the kind of puzzles I like to solve and make sure the pieces that day are two to keep it moving forward. Yeah, this sounds like a very fascinating challenge going forwards and a lot of a lot of innovations still to come. So thank you a lot again for forgiving giving me the opportunity to get more insights in into how the cloud can actually help there, how modern architecture can help there and how, probably this this area will further continue in the future. So thank you, Riya, for being for being our guests on unleash it today and, yeah, wish you wish you all all of success further in your role at the company. You've been listening to unleash. I T to ensure that you never miss an episode. Subscribe to the show in your favorite podcast player. If you'd like to learn more about enterprise architecture and tools to help unleash your businesses digital capabilities, visit lean ix dotnet. Thank you so much for listening. Until next time,.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (24)