The Importance of Industrial Data Models for CPG, Food and Beverage Companies

Many companies are rushing to embrace Industrial Internet of Things (IIoT) technologies such as industrial analytics―but they find that the impact of these transformational technologies is limited by challenges with data. Entrenched data silos often exist throughout organizations, creating significant challenges to accessing data. Also, most companies’ data architectures have been built solely for IT systems, ignoring the importance and value of data generated in plants and factories. Furthermore, the distributed nature and disparate formats of plant data is challenging for IT teams not accustomed to working with the unique formats, volumes, and speeds that come with industrial operations and processes.

To achieve the benefits of Industrial Transformation―especially when it comes to IIoT analytics―every company needs a common data model that spans assets, processes, systems, and departments throughout the organizational structure. A company that fails to define a common data model for its IIoT projects is likely to end up stuck with analytics deployments that never deliver promised results and never scale―in the dreaded “pilot purgatory.” On the other hand, a firm that creates a standardized plan to use industrial data can gain new competitive advantages in its markets and may uncover new business models empowered by data.

A data model defines the relationships between disparate data entities within an organization and provides perspective to examine the data. Data models play an essential role in any IIoT solution and are especially important in analytics, because as everyone learns in Technology 101, "garbage in, garbage out." While modern analytics solutions are much more flexible in terms of data types and formats, they still lose efficacy when the data is poorly conditioned or lacks contextualization. At LNS Research, we’ve heard many stories of problems with identifying data locations when building a data model. In one case, a large industrial company was unable to tie sensor data to originating machines in as many as 60% of cases. Certainly, this reduces or eliminates the value of insights produced by analytics.

When creating a common data model, there are many important decisions with long-term impacts that companies need to make. For example, it must decide what to do with data that resides in different locations within the enterprise―that’s a critical decision. Some data is most useful when utilized as close to the source as possible: edge analytics for real-time machine monitoring is one example. Other data is most useful when transferred to a data lake or a cloud storage system for use in combination with other data. In all, a sophisticated data model supports a layered analytics approach. Even the decision about who to include on the data model team is an important one. LNS Research recommends that companies assemble a cross-functional team of operations engineers, IT personnel, data owners, and end users that can collaborate to build the use cases required as a first step towards a data model. Including both IT and operations technology (OT) personnel ensures that experts in each type of data will be available to evaluate each data source.

Deciding on technology for the enterprise data store is yet another important group of questions. Time-series data is particularly abundant in industrial companies; this has led to the rise of data historians and other specialized databases that are designed specifically for managing that kind of data. When combining time-series data from these specialized systems with other types of data to power analytics, particular challenges arise from trying to mix dissimilar data types in a single database. In some cases, a "data lake"―a storage repository that holds enormous amounts of raw data in its native format―has solved this problem. More typically, however, data lakes are used for web, image, and sensor data. The company can then incorporate these data lakes into a data model as another critical data source. In fact, combining all these different data sources and types are one of the drivers of advanced analytics that can align production and maintenance data in the context of financial and customer data.

A data model is not just a set of ideas about how the organization treats data. The company executes its data model across multiple platforms, including databases, data historians, data lakes, and in many cases, IIoT platforms―the latter is becoming especially relevant for companies with multiple distributed manufacturing sites. Once the company executes the data model and empowers IIoT analytics across various plants, it should feel an immediate impact on business operations, including better forecasts, reduced downtime, improved throughput, and much more.

Additionally, once the organization establishes a data model, it can greatly accelerate the implementation of industrial analytics―moving to predictive and prescriptive analytics rapidly, as there will be a high level of trust in the data and the analytics output. In other words, a well-defined data model can eliminate the risk of pilot purgatory, accelerate the deployment and scaling of IIoT tech like industrial analytics, and become a driver of the organization’s industrial transformation. Without a data model, all of this is at risk.

At the same time, a well-defined data model readies an industrial company for emerging technologies like artificial intelligence (AI) and machine learning (ML). With these technologies advancing rapidly, the long-term value of a data model (and the analytics it empowers) is likely to be in uses that no one has yet envisioned, but that will emerge as companies begin to deploy these technologies on top of their data models.

Learn more about GE Vernova’s solutions for Food & Beverage and CPG manufacturers.