What is Visual Intelligence?
Visual data: an uncharted resource
Visual data coming from all kind of commoditized capture devices (smartphones, cameras, drones, satellites, lidar systems) is today generated at a massive rate and in overwhelming quantities. It is rich in information but unstructured, hard to handle and interpret. It is today a widely uncharted resource.
The future of operation and maintenance on industrial sites is based on visual data: automating inspection tasks, digitizing the real world, creating predictive models, paving the way to full automation. Mastering visual data is at the heart of all large enterprises’ digital transformation.
Massive amounts of data, many sources
Since digital cameras, all industrial players have started to make large use of visual data. They include it in their reports as visual support to describe complex scenes. Because “one picture is worth a thousand words.” Once stored on a local machine or worse, burnt on a CD, that data becomes simply lost.
Organize visual data, make it easily browsable and share it across the organization through the cloud is just step 1 of visual data management. But it already carries a huge value.
Mastering visual data, a trigger for digital transformation
Then, with AI comes the possibility to make sense of visual data. The machine is capable of indexing it automatically regarding a context, a localization, a time. The database becomes structured. It can be mobilized to answer specific questions about the current or past state of industrial assets. It becomes possible to capitalize on visual data aggregation.
Visual data is not the only type available to assess or probe industrial assets. IoT for instance is a powerful source of insight. as well as all types of OT reports. Visual data is technically the most complex. However, to be complete, the software stack should be able to ingest all other type of rich data coming from the field.
There is a challenge in finding a common sense in heterogeneous data. Data that have different formats, different natures, different time sampling. Data that does not “speak the same language.” This is possible by creating a model: a representation of a physical thing associated with rules that describe its behavior (like physical rules for instance). Now, any inflow of data is used to update the current state of the model, in respect to the rules. This is a way to reconcile all data types: even thought they do not speak the same language, “they all speak to the model.”
Toward smart industrial operations
Once you have agglomerated sufficient data over time, you know the history. With enough history, you can predict the future (in a certain measure). The models become predictive. They can be used to run simulations and optimize the complete operation and maintenance chain.
This is how a visual data management stack should be the cornerstone of a large enterprise’s digital transformation.