Manual Contextualization Avoided with Metadata

• A data value means nothing without its context.
• Until now data must be manually contextualized in every software where it is used; time-consuming, costly, and error prone.
• The recommendation is to deploy an industrial data fabric that automatically relays contextual metadata with the data.
• The result is richer displays, easier cross-referencing of additional information, and richer interaction with generative AI.

Imagine operators having data values for which they can get complete contextual information such as status, engineering unit, location in the plant, description, and scale etc. to make sense of the data quickly and correctly to run the plant better. Not just from one, but from all plant subsystems.

Imagine system integration without having to manually configure data type, engineering unit, plant location, description, and scale. And without configuring logic to determine status. For every single piece of information. This makes it easier, faster, and lower cost to deploy new software apps to solve plant problems. Engineers love a challenge, a problem to solve, but it should be the big plant problems, not manual data contextualization.

Imagine future user-interface to software including industrial AI apps where you can ask natural language questions or give instructions including focusing details and aspects like plant location, timeframe, description, or status. And get natural language answers that includes explanatory details like engineering unit, plant location, timeframe, description, and status – numerically and graphically on a scale. And, automatically converted to the engineering unit you asked for. This requires a contextual industrial data fabric you should start building today.

What is required of modern automation systems to bring these benefits today and lay the groundwork for realistic use of LLM and generative AI in future plant operations. What are the recommendations? Here are my personal thoughts:

Data Context

Until now, data in software is mostly presented without relevant context. This is true for safety dashboard software, sustainability and energy management software, reliability and maintenance software, as well as production and quality software. Making a good decision therefore requires additional manual information lookup (other software, web, and documents) and requests (from colleagues). The vision is that of all personnel receiving data in-context. The enabler is metadata. Metadata is ‘data about the data’ which either directly helps decision making or enables apps to automatically cross reference other data sources to aid in the decision making by providing additional detail. The result of fast, correct, decisions is improved plant safety, sustainability, reliability, and production.

Metadata matters

Metadata

Metadata provides context to the data. For instance, a sensor for flow measurement provides a value which is the measurement data itself, like ‘10.13’. But that number means nothing on its own, without context for the receiving application. That is, data taken out of context has no meaning. Metadata gives the data-value context such as the engineering unit is ‘m3/h’. The tag name is ‘51-FT-201’. Timestamp that the value was sampled on ‘15 of August 2024 at 11:28 and 53 seconds’ – which software use to establish sequence of events and for root cause analysis to distinguish values which are cause from values which are effect. Other metadata is the measurement status such as validity ‘Good’ or ‘Bad’ and if it is limited ‘High’ or ‘Low’ – which software and humans use to determine if the data-value can be used for control, safety, machinery protection, or algorithm training etc. The plant location takes a machine-readable form similar to a URL such as plant/area/unit/equipment but can be presented graphically like a tree or in natural language such as ‘the Austin plant, the utilities area, in the boilers process unit, and it sits on pump-02’ and the description tells this measurement is the ‘feed water flow for pump #2’ – which helps humans and AI apps “connect the dots” by automatically cross-referencing other data sources. Its scale is ‘0 – 100’ so software can display it as a needle gauge or a trend chart etc. And there are many more metadata items like data type, permissions, and update period used among the software apps.

Simple data sources like PLC, HMI, and SCADA software do not have metadata but high-end data sources like control systems (DCS), safety systems (SIS), and machinery protection systems (MPS) already have contextual metadata internally, some of it dynamically updated in real-time, some of it static in the configuration database. Historians also hold metadata internally. However, the historian system which plants have today only communicates the simple value for each data item, without the contextual metadata like engineering unit, location in the plant, and description etc. So when a problem arises this makes it hard for plant personnel to make sense of the data. So investigation and decisions take time. Maybe days or weeks. Part of the reason why metadata from automation and control systems has not been integrated into the historian is because the ‘connectors’ from automation and control systems were not designed to communicate the metadata along with the data, so the metadata for each data item must be manually recreated (contextualization) in the historian system and again in software receiving data from the historian to once again get the context for the data values. Manually recreating contextual metadata for many data points in the historian and other software would be impractical so it is rarely done which in turn limits software apps and their benefits. In future Ethernet-APL devices metadata will originate from the field instruments; instrumentation protocols like HART-IP makes it possible.

Metadata gives the data context which apps use for correct function and to provide a better user experience such as richer display to guide user. Metadata will be particularly useful in the future for GenAI to interpret natural language queries from plant personnel and to generate richer natural language responses. The recommendation is to use a modern industrial data fabric storing and communicating data with its contextual metadata. Metadata will be passed along from the data source system, so metadata need not be configured again in the systems that receives the data. The context for the data which previously was missing will be provided by metadata. This makes it easier for plant personnel to make sense of the data, to conclude investigation of issues and make their decisions faster. Industrial AI apps receive metadata for context to provide new capabilities.

System Integration

With an industrial data fabric the metadata will be preserved from one automation and control system to the next, and thus the wonderful automation we have talked about for so long can become a reality. Imagine clicking on a tag, and:

• It will open the process flow diagram and shows where that tag is located and provides an installation photo.
• It will cross-reference the instrument index to open the product data sheet for the transmitter, type certifications, and the product manual opens.
• It will open the P&ID and shows where on the diagram it is, the MSDS of the process fluid and cross-references the pipe index for pipe and flange data opens.
• It will open the instrument loop diagram to show how the instrument is wired and cross-references cable index for cable data and DCS IO index for IO card data.
• It will open the device records to bring up device-specific information like material certificates, calibration history, and service history.
• It will open the instrument specification form.
• It will open the process model where simulation and testing can be done.

These lookups are today done manually, or document links configured manually and then maintained manually – which is not practical to keep-up for the life of the plant. Therefore, setting contextual metadata only once, at source, and then preserving contextual metadata from source to finish will be critical, both among systems from the same vendor and third-party systems. The recommendation is to use OPC-UA between third-party systems. Metadata will also be an enabler for future GenAI apps that rely on metadata to “connect the dots” between various sources of information to provide more exhaustive answers.

Action Plan: Rich Data

Digital transformation requires plant automation. As existing plants roll out a ‘second layer of automation’ for Industry 4.0 / IIoT / Monitoring & Optimization along the lines of the NAMUR Open Architecture (NOA), the recommendation is to use a data infrastructure, an industrial data fabric, that supports contextual metadata. This includes both data storage and communication. Companies must invest more in plant automation, assigning a greater part of the technology budget to plant automation; the I&C team.

The recommendation is to write metadata interface and storage into the engineering specs for automation systems and software.

For new plant projects the recommendation is to deploy a DCS and second layer of automation with an industrial data fabric, that supports contextual metadata.

And remember, always ask vendor for product data sheet to make sure the software is proven and pay close attention to software screen captures in it to see if it does what is promised without expensive customization. Well, that’s my personal opinion. If you are interested in digital transformation in the process industries click “Follow” by my photo to not miss future updates. Click “Like” if you found this useful to you and to make sure you keep receiving updates in your feed and “Share” it with others if you think it would be useful to them. Save the link in case you need to refer in the future.