Data without constraints
Purely data-driven models can learn powerful correlations, but may behave unreliably when production moves beyond familiar conditions.
Technology
AI you can trust. Grounded in science, proven in practice.
jora.twin combines machine learning, physical process knowledge and operator expertise to create robust models for real-time industrial optimization.
From signals to process intelligence
Plant data becomes a dynamic model of the complete production line, guided by physics and process expertise.
PLANT SIGNALS
One dynamic line model
Data + Physics + Process knowledge
Feedstock
Processing
Quality
Why one approach is not enough
Data, physics and expertise each solve only part of the problem.
Physics without adaptation
First-principles models provide structure, but can be difficult to calibrate and maintain for every line, material and operating regime.
Expertise without scale
Operators and engineers understand context and exceptions, but that knowledge is difficult to apply consistently across every shift and decision.
Hybrid intelligence
jora.tech unites data, physics and process knowledge to build AI models for smarter extrusion.
The three sources of intelligence reinforce each other: data captures actual line behavior, physics constrains what is plausible, and process expertise defines the targets, exceptions and operating context that matter.
Data
Historical and live signals reveal dynamic dependencies across sensors, actuators, environmental conditions and quality measurements.
Physics
Known relationships, conservation principles and causal structure provide guardrails and improve behavior across the complete process chain.
Process knowledge
Engineers and operators define regimes, constraints, quality targets and practical response options for the specific line.
Dynamics across the line
Modeling the entire production line
Production quality is the result of a time-dependent chain. Material moves, energy is added and removed, equipment stages interact, and disturbances take time to propagate. jora.twin represents these delays and dependencies in one dynamic model instead of treating each machine as an isolated optimization problem.
Time-aligned modeling
Align signals with material movement and capture the delays and interactions between process stages.
Operating context
Distinguish operating regimes to account for different products, recipes and line states.
Prediction & validation
Forecast process behavior and product properties over a useful horizon, then continuously compare predictions with observed behavior after deployment.
A material change propagates through feedstock, processing and product quality with delays; jora.twin aligns these signals in one dynamic model.
FOLLOW THE MATERIAL
Feedstock
Input change
Processing
Delayed response
Product quality
Downstream effect
One time-aligned dynamic model
Signals are aligned with material movement and process delays.
From foresight to response
Turning predictions into actions
A useful forecast is only the first step. jora.twin evaluates possible responses against production targets and operating constraints, then identifies the set-point changes most likely to improve the predicted outcome.
Explore the application modules →Conceptual forecast comparison: possible set point responses are evaluated against a target band and operating constraints.
EVALUATE POSSIBLE RESPONSES
Target range
Observed
Now
Forecast horizon
Select a response within operating constraints
Conceptual illustration of response selection; not measured production data.
Platform architecture
Designed for industrial reality
jora.twin was designed from the outset as a modular, scalable software platform. Its containerized, microservice-based architecture supports rapid deployment, whether on-premises or in the cloud. The platform connects to existing IT and OT systems to minimize integration effort and build on the capabilities already in place.
The deployment layer contains the data, model and application layers. Plant data feeds models; models deliver forecasts and optimized responses to applications.
Deployment layer
On premises or in the cloud
Application layer
Interfaces & operational decisions
Model layer
Forecasts & optimization
Data layer
Plant signals & quality data
A modular architecture connects plant data, models and operational applications within the chosen deployment environment.
04
Deployment layer
Fit the architecture to plant requirements, network zones, latency and governance instead of forcing a single deployment pattern.
03
Application layer
Deliver results through jora.twin interfaces, existing dashboards, operator workflows or the control layer.
02
Model layer
Train, validate, version and monitor the dynamic models used for forecasts, anomaly detection and optimization.
01
Data layer
Connect historians, databases, quality systems and live plant signals while preserving time alignment and operating context.
Research to production
Science and validation
The underlying methods were developed through multi-year research and industrial collaboration, including work supported by the Austrian Research Promotion Agency (FFG). The focus is not a laboratory-only model, but an approach that can be validated against real production behavior and operated within clear boundaries.
Technical questions
Technical FAQs
How much historical data is required?
This depends on sampling, process dynamics, product coverage, target quality data and how much relevant variation the history contains. In practice the required data range can span anywhere between days and months of data. The pre-check assesses this directly for a specific line.
What happens when the process changes?
The model performance is constantly monitored. Once the gap between prediction and actual process exceeds a defined threshold, a retraining may be required. Our software provides a guided wizard for model training and ensures that this is done in a controlled and traceable manner.
Can jora.twin run on premises?
Yes, and this is usually the preferred way for most enterprises. However, a cloud implementation is also possible without any problems – the modular containerized architecture of jora.twin enables an easy deployment on a wide range of hosting solutions.
How is control safety handled?
Automated actions are limited to validated targets, constraints, rate limits and operating boundaries, with the autonomy level agreed for the application.
Let’s look at your line
Bring your process question
Talk to us about the target, constraints and data behind the production decision you want to improve.
