Data contextualisation: a brief history, key aspects, and best practices
Contextualising data is a hot topic in the industrial sector. Although it’s not a new concept, it’s gaining attention as industries face growing challenges that can only be solved through contextualisation.
Contextualisation: A Bit of History
Before diving into what contextualisation is, let’s take a brief look at the past.
The first contextualisation approach was introduced by historian softwares several decades ago, mainly through the use of asset frameworks to provide context to data from an equipment standpoint. And that was the extent of it.
The Era of Big Data and Unstructured Data
When we started Optimistik in 2015, it was clear from our industrial experience that contextualisation was the key to addressing the challenges of data in process manufacturing. At that time, the hype was around Big Data and unstructured data. Many thought that contextualisation was unnecessary, but we worked hard to build standard data models (physical meaning, traceability, genealogy, events, assets, etc.) in our solution to contextualise data from our customers. Those who followed the hype lost a lot of time and money, and are now considering how to structure data for their use cases.
The Era of AI
Then came Artificial Intelligence with all its promises. At Optimistik, we were convinced from day one that Machine Learning and Data Science could be used to help the industry become more competitive and sustainable. Many companies have been offering solutions or services around AI in the past decade, but the number of use cases implemented in industrial production processes remains limited. We have identified two obstacles:
First, compiling clean and relevant datasets to build an AI model.
If an AI model is successfully developed, implementing it on the shop floor is not straightforward. This involves establishing a clean data flow, performing proper data preprocessing, running the model code in an industrial environment, and deploying and managing it at scale.
Once again, contextualisation is a key enabler for removing these obstacles.
So, what is contextualisation?
In simple terms, contextualisation refers to embedding industrial knowledge into digital infrastructures so that raw data can become information. To be more precise, it involves integrating new data that will provide context for the data being manipulated. Here are a few examples:
On raw numerical data: what is the unit, the origin, the physical meaning of the data, where and when its as been collected…
Production Traceability: when and where a specific production took place, what are its caracteristics, details of the productions phases…
Production Genealogy: for raw material or when intermediates products are being uses, in which production they were used…
Assets: what are their physical caracteristics, the data attached, their localisation…
Assets traceability: what are the equipment cycles, the consumable parts cycles…
Events context: what happened, where, when…
But it doesn’t end there. Contextualisation makes sense when you can automatically combine this data to generate the proper information, such as:
How much energy did I use to dry each batch of powder?
What is the variability of my filtration time between two cycles?
How do my different reactors behave compared to each other?
Do I see differences between batches produced on different reactors?
Are every approach equals in term of contextualisation ?
To be clear, not all approaches are equal when it comes to implementing contextualisation. There are several factors you should take into account when considering such an implementation.
Contextualisation must be at the core of industrial digital infrastructure
To be relevant, contextualisation must be at the core of your digital infrastructure. It bridges technology (databases, data treatment capabilities) and your business. This is intricately done as use cases drive the requirements for data models and infrastructure performance. Additionally, business applications must be fully integrated with contextualisation to offer a fluid and comprehensive user experience.
Contextualisation solution must provide standard yet adaptable data and context models
If you find yourself needing to rethink how to handle data within a specific context and build the data model and processing from scratch, it may indicate a dead end. Your solution should offer predefined data models, context models, and combination functions, alongside optimized data structures, storage technologies, and treatments. The data and context models should be configurable to suit your specific use cases. The goal is to accelerate the deployment of your contextualisation framework by using standard, pre-defined templates, while also capturing your business specificities through proper setup.
Contextualisation is a matter for business experts
Do not ask the IT or OT department to set up data contextualisation, as it requires the expertise of your business experts. To achieve contextualisation set up by business experts, you need a solution that provides predefined data and context models that are business-centric, as previously mentioned. You also require context setup tools that are designed for business experts, not IT or data experts. A no-code or low-code approach with rich documentation is essential for the adoption of contextualisation by business experts.
Contextualisation must be agile, flexible, and traceable
As plant processes and equipment continuously evolve, the contextualisation framework needs to adapt accordingly. Its structure should be able to handle modifications while keeping historical data in a usable form. Additionally, it should keep track of every modification to ensure coherence and compliance.
Contextualisation must support your data governance
Roles and responsibilities are essential to maintain coherence in your data approach within your corporate organisation and remain aligned with your data governance strategy and rules. At the same time, you need to delegate data contextualisation to the business experts who know the data best. This requires a contextualisation solution that offers fine-grained role and access management and a proper audit trail.
Contextualisation solution must be open
Another important aspect is openness. A contextualization solution should not only be able to collect data, but also to share data with other corporate systems. Particularly it should enable you to provision Data models and Context model from other corporate system.