Digital Twin Principles
Digital Twin is a software component that mimics behavior of an arbitrary physical system.
Unlike physical systems that are closed, have hidden states, give partial, noisy and delayed information, the twin:
- is open and 100% transparent
- provides all actual and historical data through API in unified model
- suppresses measurement noise, model errors
- estimates hidden state information from historical states and actual observations of the original physical system
- data is available even during infrastructure outages and before the real data has arrived (prediction)
- acts actively - autonomously issues alerts and events when something interesting happens
Digital Twin Implementation
360º - we put the physical system into its full context – that includes its historical, present, and future states, as well as of information about all interacting systems.
It is not realistic to have 100% information available at the beginning. We integrate information in iterations, where the digital twin becomes more precise and accurate after each iteration.
Digital Twin for Manufacturing
Digital twin in manufacturing provides these outcomes:
- predicts future development - probability distribution of machine defect (e.g. predictive maintenance)
- detects anomalies - detection of exceptional behavior
- fires alerts and watch for important events - hitting tolerance limits
- root cause analysis - what impacts quality or caused machine defect
- online dashboards and reports for managers, operators and QA
Common implementation obstacles in manufacturing:
- machine heterogeneity - different vendor, age and level of wear
- hidden states - some states are not physically possible to measure
- unstructured, disconnected data in different format - data has different data models, is not correlated, is stored in data silos
Digital Twin for Health Care
The latest research says that 80% of the diseases can be prevented. Imagine that you are going to a doctor who will make you a unique "digital twin" - your unique digital image that will based on your health history, special tests and ongoing measurements and will give an objective assessment of your health.
More over, the digital twin will be able to track, detect anomalies, predict trends in your health and let you or the doctor to act in advance. Imagine that the system is constantly learning from the latest research and experience that has helped people with a similar diagnosis anywhere in the world.
Digital Twin for Living
Every building - house, office or whole city part - is a physical system with complex internal dynamics and many evolving variables. Typically, only a small subset of this information is visible and the building is de facto a black box. If enough variables with sufficient sampling frequency gets available, you can create digital model that behaves similarly to your building. If you then connect the model with the building you get the physical system and its digital twin synchronized.
Now imagine that you connect data from all other interacting systems such as people, outside whether, forest nearby... You can go further and start wisely influencing the complete ecosystem through actuators and strategic planning to achieve environment that is green, healthy, energetically and financially sustainable .
Visual inspection in manufacturing process
- Identification of product defects: scratches, cracks, sink marks, welding lines (bad melt flow), bad coloring, bad gloss, bad filing, flush, non intrusive 3D welding inspection
- Object distance, angle, object counting
- Product layout and component placement and assembly check
- Process decision support for robotic arms and handlers
Real time object detection
- Real time detection of persons, objects, animals
- Safety hazards and security risks early warning
Aerial images analysis:
- Analysis of satellite, plane, and drone images
- Change analysis, tracking and future development forecasts
- Utility networks issues tracking
- Vegetation composition, health and density analysis
- Digital Elevation Model from radar topography
AI Guided Product Design and Optimization
Industrial design and optimisation process starts in CAD tooling where CAD model is used to build a physical prototype or simulate behavior using CFD. Both is costly and time consuming. CAD => physical prototype/CFD loop is repeated until desired performance is achieved.
Our AI driven approach predicts product behavior from geometry and material to:
- Avoid costly and time consuming physical prototyping and CFD computations
- Uncover new dependencies and rules that can be applied by engineers
- Make recommendations on how to modify product/machine parameters to achieve better results
We turn the latest AI research to production. We build digital twins for manufacturing, health care and day to day life. We want to create solutions that make the world a better place.
Artificial Intelligence shall serve people.
Artificial Intelligence shall be aligned with human values.
Artificial Intelligence shall make world a better place for everybody.
cooperation in applied research
We closely co-operate with top-class universities and organisations that support innovation and applied research. We track the latest technologies, explore them and leverage them.
It is hard to turn any software solution from prototype to production. Turning any AI algorithm from lab solution and ideal data to practical solution on real data is even harder.
Our core team is made up of top architects, researchers, AI developers, mathematicians, and subject matter experts. People with open mind, not afraid of innovation, risking and learning from other colleagues, and of their own mistakes.
Learning and development is our key
Main technology stack
Programming languges - Python, Java, Scala
Python libraries - Pandas, Numpy, Scikit-learn, SciPy, Seaborn, Pyplot, PySpark, OpenCV
Big data - Hadoop ecosystem (Apache/Cloudera, HDFS, Parquet, Avro, Ambari, Zepellin, Hive)
Deep learning - Tensorflow, Keras, H2O
Data streaming - Spark, Storm, Kafka, Spark ML, MapReduce
Research platforms - Python, R, Matlab, ILog CPLEX
Data streaming - Kafka, Fluentd, Mosquitto
NoSQL storage - Influx, OpenTSDB, Elastic, Redis
AWS Lambda, S3
PCA, SVD, auto encoder, clustering, correlation analysis
Stochastic processes and bayesian inference
Bayesian data assimilation, ensemble kalman filtering, particle filters
SVM, SVR, gradient boosting, XGBoosting
Neurodynamic programming, stochastic machines
Convex optimization, quadratic optimization
U Nikolajky 1097/3, Smíchov, Prague 5, 150 00
Apply for a full time/part time job
Become part of our DNAI team