INDUSTRY & R&D
Sensor data modeling at scale: use every sensor, keep the explanation
Sensor data modeling at scale with TIMi uses a very large number of sensor variables at the same time, for research and technical analysis, without limiting the work to a small subset of measures for lack of processing capacity.
Thousands of variables
Start from all the sensors, not from a pre-filtered shortlist.
One standard server
No cluster and no dedicated cloud infrastructure.
Readable at scale
Modeler keeps the variables that matter, so the result stays explainable.
HOW IT WORKS
How sensor data modeling at scale works.
Three steps, in one engine, on a standard server with no cluster.
1. Prepare the data
Anatella gathers and synchronizes the flows from a large number of sensors, often heterogeneous in frequency and format, then cleans and structures them into variables.
2. Build the model
Modeler builds models that start from a very large number of variables and keep automatically the ones that matter, without a manual reduction of the sensors considered.
3. Result and explain
The model produces its result (score, classification or estimate, depending on the goal) together with the sensor variables that weigh the most in it.
Anatella
ModelerWHAT CHANGES
Why scale changes the method.
The usual answer to too many sensors is to drop most of them. It does not have to be.
Do not pre-filter the sensors
Many tools force a reduction of variables, through sampling or a manual selection of the sensors judged relevant. This risks discarding a useful signal before testing it. Here the model can start from all the sensors available, and the result tells which ones are really influential.
A technical capability, not a business use case
It is a cross-cutting capability that supports several business uses, such as predictive maintenance, asset degradation modeling or industrial process analysis.
Batch or real time
Both are possible depending on the deployment architecture, according to the latency constraints of the project.
EXPLAINABILITY
More sensors should not mean less understanding.
A model that uses thousands of sensor variables without explaining how is a black box by construction. A white-box model keeps the influence of each variable readable even at this scale: Modeler automatically keeps a small set of variables out of thousands, and the selection is documented.
For technical teams
You keep understanding why a result is produced, as the number of sensors grows.
For R&D teams
The variables that carry the signal point to where to look in the physical system.
For model owners
Modeler’s built-in Drift Report shows when a model loses accuracy, so you update it when needed instead of on a fixed calendar.
DATA
What data a sensor model uses.
Anatella connects to more than 80 sources, so this data can be gathered without prior custom development.
Sensor streams
Potentially heterogeneous in frequency and format.
A clear modeling goal
What the model must predict, classify or estimate.
Enough history
To cover the variability of the conditions met.
Sensor metadata
Position, type of measure, unit.
PROOF
Sensor data at scale, in production.
Three deployments with very different sensor volumes.
More than 1,400 sensors in a single analysis
Telemetry from over 1,400 simultaneous sensors during accelerated aging tests, used to isolate the variables behind battery degradation.
10 billion rows on a standard server
Ten machines, each generating over a billion rows, analyzed to find rare failure signals.
500 sensors and 4,000 derived variables
Three years of data from 5,000 wells and 50 GB of raw data, used to detect pump failures early.
FAQ
Common questions.
Is there a limit to the number of sensors?
TIMi does not impose a fixed cap on the number of variables in a model. In practice the limit is the capacity of the hardware used, which can be sized to the needs of the project.
Is this a business application or a technical capability?
It is a cross-cutting technical capability, which supports several business uses such as predictive maintenance, asset degradation modeling or industrial process analysis.
Do we need to pre-select the most relevant sensors?
No, this is precisely what this capability avoids: the model can use all the sensors available and identify, through the explainability of the result, which ones are really influential.
Does it work in real time or only in batch?
Both are possible depending on the deployment architecture, according to the latency constraints of the project.
RELATED USE CASES
Keep exploring.
Asset Degradation Modeling
Model how the condition of an asset evolves with use, and estimate its remaining life.
Predictive Maintenance
Anticipate equipment failures from the data you already collect, with the reasons.
Automated Defect Classification
Classify defects during production, in real time, with the measures behind each call.
SEE FOR YOURSELF
Test it on your own data.
Run TIMi on your own data with a free trial, or talk to us about your case.
