DIGITAL TWIN
NeurEco Virtuous Circle Accuracy
-
Accuracy (which is a direct consequence of parsimony) allows for the early detection of the smallest drift between the model and the real system
-
Early model update, on the basis of a short sequence of data, maintains the virtuous cycle of accuracy
An asset for continuous learning
A digital twin is a virtual representation of an object, process or system that spans its lifecycle. It is updated from real-time data, and uses simulation and machine learning to help decision-making.
Our parsimonious neural network technology is an asset for digital twin applications:
-
Predictive Maintenance: Early predication and detection of anomalies
-
Evaluation and update of the model can be embedded on small chips
-
Long term dynamic prediction
OUR SUCCESS STORIES
Direct radiating antennas: Satellites
Reinforcement Learning: Embedded Software
Direct radiating antennas: Satellites
Reinforcement Learning: Embedded Software
Context
Satellite companies are unable to accurately control their antennas using traditional methodologies
Problem
Maximize the beam efficiently and minimize the interference with other antennas beams
Objective
Create a small embedded model for real time control of the antenna beam
Challenge
Supervised learning is not working in this context
Solution
By utilizing a NeurEco reinforcement learning algorithm, the antennas produced results outperforming all other previously used methods
Results
NeurEco’s produced model beat all of their previous model allowing them to improved their overall product
Benefits
-
Increased product reliability resulting in improved customer retention and profitability
-
More time to focus on other product development
-
Client was so impressed by the improvement they decided to embed the algorithm in their antennas. This was not the original aim of the project
Digital twin of a gas network
Context
Existing simulation tools are not able to model the network (PSI) as the real state of the network (measurements) are not modeled correctly.
Objective
Create a digital twin using measured data on the entire network.
Solution
-
Create a NeurEco model for each network element (tube, pump, valve, etc).
-
Assemble these macro-elements using NeurEco to provide an overall system view.
-
Quickly solve the resulting nonlinear system at each time step, taking into account the measurements.
Results
An accurate digital twin of the gas network enabling advance training and predictive analytics.
Benefits
-
Improved OEE (overall equipment effectiveness) through reduced downtime and improved performance.
-
Lower maintenance costs by predicting maintenance issues before breakdowns occur.
-
More efficient supply and delivery chains resulting in increased profitability.
Predictive maintenance
Partner objective
Create a digital twin for the purpose of predictive maintenance of a system.
Why NeurEco
Unique dynamic solution for long-term prediction.
Benefits to the client
-
Client can schedule maintenance in an efficient and cost reducing manner allowing them to maximize the time the system is working and subsequent profitability.
-
The client can use the data to optimize the system operation.
Renewable energy power production
Partner objective
Produce a continuous predictive model of the power production of a solar or wind farm. The solution needed to accurately predict power output for up to 30 hours in the future.
Why NeurEco
-
NeurEco provides an unique dynamic solution which is ideal for long-term prediction with dynamic inputs.
-
The energy company could not accurately model predicted energy production using either measured data or historic data.
Result
NeurEco was able to produce a non biased prediction of estimated power production using the provided inputs.
Benefits to the client
Ability to accurately sell predicted power to the national grid and schedule maintenance at the most optimal times.
