From years of hard lessons, manufacturers attempting digital transformation know they must contend with the troublesome nature of plant data, which flows in a torrent without consistent structure or sequence. It is now clear that to use plant data strategically, three levels of technology are essential:
- The first is pipes that get data flowing smoothly from the myriad sources—sensors, PLCs, MES/ERP, historians, operator entry and more. The need for good pipes has been long understood, and with edge, connectivity, aggregation and cloud being rapidly adopted, data is now flowing fast. With private 4G LTE, 5G, hybrid cloud, cloud edge and other big steps in rapid connectivity, innovation in this area continues.
- The second is analysis—business intelligence, data science, and AI and machine learning. Significant time, money and skills have been devoted to this level, but scale has been a terrible challenge. Efforts stall because plant data is so varied and difficult. The deepest challenge in this area is one of perception and misunderstanding, and it’s a misunderstanding that sadly has been perpetuated by the technology industry itself. In comparison to the pace of manufacturing transformation, many technology companies have plowed through integrating data from most business activities: sales, marketing, finance, and other business process data, such as data from ERP and PLM. But OT data—the real untapped gold in manufacturing companies—is not IT ready. As powerful as AI/ML and BI tools might be, they can’t be applied to OT data in its raw form.
- And then there’s a layer that is too often overlooked, right in the middle, between pipes and analysis. This level is sense-making. Sense-making turns data into information that can be analyzed and understood.
In early stages of digital transformation, as Big Data technologies matured and Internet of Things investments took hold, manufacturing enthusiasm for IoT led to massive investment in pipes. A focus on AI followed, accompanied by a widespread belief that pipes and AI combined would radically benefit plants. But as companies increasingly engaged with huge amounts of plant data, the need for sense-making became evident.
In sectors like advertising and finance where Big Data got its start, sense-making may be as simple as cleaning data, changing formats and standardizing columns. The sense-making work has been done by data scientists (who themselves consider it lower-level drudge work), outsourced or done through a proliferating collection of DIY data manipulation tools. This area of the IT market is now exploding, because IT tools work really well on IT-ready data.
But in sectors with especially complex data environments—including the most complex, manufacturing—sense-making is the most difficult problem in digital transformation. Companies work hard to develop AI/ML capability, they train and hire their best controls engineers and data scientists, and almost invariably, when it comes to any sort of scale, they hit a dead-end. They may succeed in solving one problem, but not the dozens of interrelated problems that face a manufacturing plant every day. AI/ML models are built, one at a time, but all on different data sets. BI systems to help plants make decisions run on different data, and it’s impossible to cohere the workflows, other than through very painful grafting. Data sources change constantly in the background. Models drift, the solutions don’t scale.
Engineers and data scientists build pipeline models to target specific plant problems. They design a data model to fix a production problem, only to later find that in failing to account for related problems, the model is insufficient. So they build another model to counter that problem. But the models and data pipelines are incompatible and can’t be linked. There’s a gap that makes it hard for plant managers to use data sensibly.
“No, We Don’t Have a Better Tool”
In discussing this landscape with manufacturers, we’re often asked, “So Sight Machine has a better tool, right?”
“No,” we answer, “we don’t have a better tool.” Sight Machine isn’t a tool for building data pipelines. It is the data pipeline—Pipeline as a Service—an end-to-end offering that also includes everything from pipes to BI, ML and AI, but also works with and bolts into existing technologies.
The heart of the offering is the sense-making layer—a streaming Pipeline as a Service that continuously transforms and analyzes all plant data. The product works with data from any manufacturing environment in process and discrete industries, and plants small and large. It complements and accelerates existing investments, and provides OT and IT leaders with a single layer of trusted information.
As technologies mature, there is a move from bespoke to standardized, from tools to products. It has taken longer in manufacturing than it did in advertising because of the nature of the underlying data: unwieldy, unmanageable, physical world data.
How Sight Machine’s Sense-Making Works
Sight Machine captures manufacturing reality with a Data Foundation composed of standardized data structures. Standardized structures are generated by standardized models, frameworks we call Common Data Models. These models are NOT models of specific machines (scaling that would be impossible), but rather are standardized schema to make all aspects of manufacturing relatable. These Common Data Models include the Production Model, capturing the data from each cycle or unit of work; the Parts Model, capturing everything that went into a unit of output along with its resulting quality; and other models including KPI, Line,
Factory, Downtime, Defect and more. These Common Data Models capture a living, continuously evolving snapshot of all the key elements of manufacturing.
As a first step, we pre-process and stream data from all plant floor sources—controls and automation, software systems and tools, databases and data lakes. Our Pipeline as a Service then transforms and maps the restructured data
into Common Data Models. Now the data is contextualized and ready for analysis within Sight Machine’s products or with third-party tools.
All this happens continuously and in real-time, so that companies can fix production problems as they emerge. The Data Foundation is the single source of truth, capturing the production process in digital form, from which to manage productivity, quality and availability, as well as to streamline supply chains and enhance sustainability.
Our Pipeline as a Service is browser-based, highly configurable and transparent. It can manage time-stamp variation, incorporate late data, and automatically apply countless contextual rules.
The Pipeline includes hundreds of techniques to map, model, and distill information so that raw plant data continuously streams into the Common Data Models. It is built around operators and pipelines. Operators are algorithms or functions that transform raw manufacturing data into normalized formats. Pipelines are a series of operators arranged in sequence. It isn’t uncommon for a Pipeline to include 50 or more operators needed to transform incoming raw data into the plant’s Data Foundation.
Think of the operators as LEGO blocks that can be stacked and rearranged as needed. Sight Machine has already built the operators most commonly needed to transform raw manufacturing data into more useful analytics-ready formats. Manufacturing teams can add their own custom operators via extensions to the Pipeline as a Service platform.
Operational technology and IT teams use different tools but they have the same goal: to turn vast reams of data into one foundational data set. There’s factory data, then there’s the ability to use it sensibly. It’s one thing to know what happened in production, but another thing to know what’s happening in real time.
Sight Machine’s platform allows you to make sense of your data in real time so that you can increase your plant’s efficiency, productivity and profitability.