I recently participated in a discussion with the Microsoft industry experiences team (David Starr and Diego Tamburini) to talk all things IoT, including the cultural changes data-informed decision making has on organizations.
The discussion was recorded as a podcast. This is Part 1. We will be bringing you part 2 soon.
Part 1 covers overcoming the friction involved in analyzing data and the importance of using brownfield data to build context.
Welcome to the Microsoft industry experience team podcast. I'm your host David Starr and in this series, you’ll hear from leaders across various industries discussing the impact of digital disruption and innovation, sharing how they've used Azure to transform their business.
Today we're going to be discussing the Internet of things. This is industrial and we're talking about real uses for it in industry including what types of IOT data exist and how they differ, aggregating IOTs data streams and connecting data for faster insights.
Joining us today on the show is Sameer Kalwani. Sameer is the founder and head of products at Element. Before joining Element, Sameer ran product for a company that delivered a full industrial IOT solution loop for over 250 water treatment plants in India. Those solutions provided sensing and gateways, Machine Learning processing, and analytics in the cloud, and even cloud-to-edge based controls. And that was 10 years ago back in 2008, so welcome to the show Sameer.
We're also joined by Diego Tamburini, the principal lead for Azure manufacturing in the Microsoft industry experiences team where he focuses on developing technical content to help manufacturing companies and software developers deliver their solutions on Azure at scale. He also champions partners who deliver manufacturing solutions using Azure. So welcome Diego.
David: I wonder if we could start with Sameer, and ask about your company and its mission?
Sameer: Element focuses on making industrial data incredibly easy to use, to unlock insights never before possible. I started Element back in 2015 focusing on how we bring data science and data science capabilities to rich, industrial data. And what we found was that there is a high level of friction between data scientists, analysts, and engineers who want to have wide scale data access to do analytics because getting the data into the right form and context to manipulate it is challenging. So, owner operators use our software and services to easily provide the data to anyone within the organization who needs to perform analytics and deliver new insights.
David: What about that friction? What is it you're describing when you use the word friction?
Sameer: One of the key things that's different about IOT data - and obviously there are many different types of IOT data - is that it's time series data. You're measuring things that are happening in the real world. It’s not software generated data. As an example, if I am looking at a pump in a brownfield facility that’s been around since World War II, how do you know what the sensor – or the piece of equipment – is that it’s connected to? There’re a lot of things that have happened with that one pump over 50 or 60 years. How has that pump evolved? Have people changed the naming convention? How has the pump changed in the various turnarounds that have happened? And then you think about that across the entire facility, whether an oil refinery or mining site, and the way operations have changed and evolved. The data is not easy to use. Then, if you bring in an engineer who's fresh out of school with their PhD, and knows R or Python and wants to able to work with that data, they have no idea about the context. They can't look at that pump versus another pump. Where is it in the process hazards? Who's worked on it? Where is it located? So, you want to be able to make sure that data is easy to use, but also has context to create a digital representation of the physical world, allowing anyone to get the insights they need when they need it.
Diego: What Element brings to the table is that it connects the dots, between all the raw sensor and device data, for actionable analytics. A lot of people have jumped onto the IOT bandwagon and collect data from their devices. And they are realizing that it's not a matter of just plugging all that data somehow into a black box that is going to start spitting out insights or predictions. There is a lot of sausage making. IOT is essentially a big data analytics problem. It is unique in that, like Sameer mentioned, it's time series data and it comes in different formats at different frequencies through different protocols, and you don't always get what you need. So, if you are trying to train a machine learning model or calculate a KPI, such as OEE (Overall Equipment Effectiveness), you may not have the data that you need to do that. So, there is a lot of sausage making again in the process, and a solution like Element helps customers do just that.
Sameer: You're highlighting a key piece here Diego. So, if I am working at Fitbit manufacturing Fitbit watches, I know my sensors. I have the same piece of equipment that is going to every single person. And because it’s a consumer device, if it fails, it's ok as I can always replace it. But when you’re talking about industrials, it's a very different animal. So, let’s say that you install a Raspberry Pi or an NVIDIA Jetson that's providing edge-based data collection. And that data is being sent to the gateways then to the cloud. Or you might be using your existing control systems policy control network and sending that data into the cloud. Question is, how do you build context about everything connected to that control system? And how do you get the data into a central spot? How do you augment that, not just with sensor data, but also with data about maintenance, supply chain, integrity, hazards, piping and instrumentation diagrams, etc.? And then provide your analysts with a schema-on-read capability similar to the one they already enjoy with web traffic data? Or submit queries and get back relevant data to generate insights by using tools like Azure Databricks or Azure SQL Data Warehouse? And then to display those insights within Power BI or Azure Machine Learning?
David: So, what I'm hearing is there needs to be a lot of thought upfront about the time series data structure and what it's going to report. But people have been recording data for a long time and we’ve got a pile of brownfield data there. How useful can that be?
Sameer: Element can connect directly to that brownfield data. Whether that’s stored in your data historian, or your EAM (Enterprise Asset Management) system where all the maintenance work orders and functional locations of your equipment exist. All that is rich context about your physical operations. It’s really misguided to not use that information because you don't want to look at just sensor data. You want to look at all the rich context that’s there from other non-sensor systems. And possibly even augment that with new sensing around IOT. Then put that together into a structure that’s easy to perform analytics. So, it is critical to bring in brownfield data, data from a greenfield site, and other data sets to add context to it. Context is key when you are talking about the physical world and performing analysis on this data.
David: And the brownfield data provides the context. So does the new, structured data provide exact sensor readings that you're able to then extrapolate back in time?
Sameer: Exactly. The analogy I always like to make is to think of the IOT device as a stethoscope. You're getting more fine-grained information that you're not able to get before. Facilities such as a water treatment plant, or an oil refinery, mining site, or power generation plant already have a great deal of control networks. So, they have control systems, like a DCS (Distributed Control System), that’s connected to the actual equipment and its using the sensing that’s there to determine what needs to happen. But what if you want to add non-control data, for example, adding IOT data to tanks, or at a distributed site where you might have a gas pipeline and want more IOT sensors there. You want to be able to pull all that information in, in addition to your control data, then get that to the cloud. Now you can do fleet wide reporting. And do what-if analysis. And predict failures or optimizations. All those things are possible when you are making decisions, not just with data, but by using data to enable decision making for the user. That way, the right things happen at the right time, reducing operational hazards significantly. That water company that you mentioned in my introduction? We reduced maintenance costs by 60% because of machine learning and analytics generating new insights. Insights that helped our field service workers do the right thing at the right time.
Part 2 of the podcast will explore rethinking analytics as a capability, not a maturity model, and the importance of design thinking as part of digital transformation.