Probably the best compass for finding the right data analytics solution is if the user defines a specific operations/production problem, and uses its details and requirements to guide their search. For example, when plant managers at two of Evonik's specialty chemicals plants in Mobile, Ala., recently found that 75% of their corrective maintenance costs were going to emergency pump and compressor maintenance, they decided to implement condition-based monitoring, predictive maintenance and root-cause analytics to avoid device failure and improve uptime. The crosslinkers and active oxygen facilities make paints and coatings ingredients, hydrogen peroxide and other chemicals, and possessed information from Emerson EMS and GoPlant software. However, their production data was admittedly messy, often plugged into Excel, and hard to sort through, so 50-70% of Evonik's initial data science labor went to information cleanup instead of optimization.
Because creating integrated views of time-series and transactional data was painful and not scalable for other plants, Arpan Seth, senior data scientist and process engineer at Evonik, recently adopted Element Unify software from Element Analytics to produce cleaner, contextualized asset data models for digitalized operations. The software is designed to break through operations technology (OT) and information technology (IT) data silos, bring OT and IT teams together, and bridge the gap in the Industrial Internet of Things (IIoT) that makes 95% of industrial data unusable because it's fragmented and disconnected, according to Andy Bane, CEO of Element Analytics.
Map tags to build models
Element Unify snaps into Evonik's existing Internet as a service (IaaS) and cloud-computing environment; connects to Emerson EMS and other sources; imports metadata from AspenInfoPlus.21 and maps tags to preconfigured templates; adds a master equipment list from SAP; and integrates electronic operator logs, work orders and vibration metadata. The asset data model created in Element Unify provided the context and relationship between Evonik's time-series and transactional data residing in Microsoft SQL Server, which let users connect to Microsoft Power BI and create five dashboards for condition-based monitoring, predictive maintenance and root-cause analysis. The dashboards deliver a 360° view of pumps and compressors at the plants and IT-OT data management for Evonik's data scientists (Figure 1).
In just eight weeks, Evonik reports that Element Unify let it deploy two asset data models for 464 pieces of rotating equipment at both plants. It took 80% less time to build the asset models, and 40% less labor to deploy their analytics. As a result, the operations teams can see pump or compressors operating statuses, amperage, vibration history, operator logs, work orders and other parameters on the Power BI dashboards; predict equipment performance and failures; and make more informed, predictive maintenance decisions. The managers estimate Evonik could potentially save $550,000 over five years by preventing four pump failures per year based on one analytic at one plant. In the future, the teams and managers add they can rapidly scale their new analytics to other plants, and quickly test and validate artificial intelligence (AI) and machine learning (ML) solutions.
"We help users bring together OT and IT data using modern software. We don't do analytics ourselves, but we make it possible with middleware software that creates a context layer, and provide it as a cloud-native, easily consumable, software as a service (SaaS) product that focuses on metadata," says Bane. "Because metadata describes identifying information associated with devices, such as numbers and locations, it can be used to develop hierarchies and asset or enterprise models, and redeploy them in production processes. For instance, we can take data from an OSIsoft PI System, use the tag stream metadata that describes its pumps, add time-sensitive events like previous maintenance and failures, and help users create open-design data models."
Bring databases into development environment
Cleaning and contextualizing data is precisely the problem that Element Analytics tackles, but developing an efficient solution has been circuitous. "Traditional manual entry and spreadsheets were about chasing data, but now we have Power BI, data scientists and artificial intelligence, so users have to spend 80% of their time wrangling data," says Bane. "Most analytics packages want to get time-series data from operations, which is usually compressed and stored in historians, and integrate analytics at the tag level. The problem is this method doesn't scale, so its data lacks or loses context. This was the problem Evonik had when it wanted to do predictive maintenance by viewing critical equipment, figuring out best practices, and sharing them with its high-performance teams.
"However, historians don't have object models inside, and the process industry doesn't have an object model that can combine P&IDs and tags that identify which label goes with which device and its data. The only historian with an object model is OSIsoft's Asset Framework, but now developers on the IT and consumer sides are bringing together relational databases and production schemas. These include Snowflake's cloud-based data platform, which lets users view their sensors, pumps, tags, schemas and everything else at the same time to do more advanced analytics. Combining databases and schemas resolves the former context problem by centralizing metadata, so models can be evergreen and reflect operational changes when they happen."
Bane adds that Element Unify software provides a low-code and no-code development environment, templates and a graph model, so it can bring in data from design and engineering systems, and create multi-domain models that let each user look at the information, parameters and operating conditions they want. "Instead of the old way of extracting, transforming and loading data, Element Unify makes the process faster and more efficient by getting data off spreadsheets and into enterprise software more quickly. We extract and load data first, and then transform it. We also build the tag map and attribute it, so we can represent all the instruments on a pump or other device, and make what some users are calling an 'operations digital twin' that can map to all its data and show it in a graphical model."
