Our projects
We analyze data
We build predictive and classification models
We deploy our computing engines in the cloud
We deploy our solutions in industrial environment
We turn our models into mobile apps
We create dashboards integrated with your business environment
Contact us,
We will find a solution dedicated to you!
Predictive models

Our models answer questions like:

  • What will be the heat demand in Warsaw district heating system in the next 5 days? How high will it be in specific areas of the network?
  • What will be the volume of gas that specific clients will inject or withdraw in an underground gas storage facility?
  • What will be the electricity demand on a given day?
  • What are the odds of developing drug-resistant epilepsy for a child treated since first abnormal EEG?

Our models are usually a part of bigger systems for which we develop dedicated Graphical User Interfaces (GUIs).

 

Predictive Maintenance / Anomaly Detection
District heating network pipeline
failure prediction
  • District heating network pipelines are prone to failures that are costly to repair and may result in interruption of heat supply and therefore generate additional financial losses.
  • Machine learning-based system created by Transition Technologies S.A. determines probability of pipeline failure based on historical data (previous failures) and information about the element such as length, diameter, construction type, producer, date of commissioning etc.
  • The solution proved to be twice as accurate as decision of experienced team of experts and about 12 times better than the reference method which randomly selects pipelines for modernization.
On-line diagnostics of a coal mill
    • The abnormal operation conditions and malfunctions of coal mills negatively affect the boiler operation and furthermore can be the cause of emergency boiler shutdown. In coal mills the coal is pulverized and mixed with air necessary for combustion in the boiler, hence abnormal operating conditions can decrease boiler efficiency as well as increase CO2 and NOx emissions. Not detected at the right time, mill faults are the cause of numerous issues such as failing to maintain steam parameters, reduction of generated power or flame instability in the furnace. The last may eventually lead to flame loss which is one of the most dangerous situations in a power plant.
    • The algorithm is based on 2 key parameters of a coal mill:
      – electric current supplying the coal mill engine,
      – outlet temperature.
      The models have been developed using machine learning techniques.
    • Pilot implementation has been done in Rybnik power plant.

Read more >>

Optimization
SILO

SILO is an immune inspired Artificial Intelligence system. Immune system is unique as it has memory and is improving through learning and experience (self-adapting). These features are highly desirable in solutions which perform on-line industrial process optimization and control. Thanks to them SILO learns the process and its input-output relations and calculates setpoints or setpoints corrections for controllers that are automatically passed to the control system. Optimization strategies used are very similar to those used by our immune system fighting off viruses.

SILO II system performs an on-line optimization of MIMO (Multi Input Multi Output) industrial processes. It is responsible for:

  • Maximization of income that is related with output product
  • Minimization of costs which are related with fuel costs and penalties for air pollution.

The SILO II economical calculation module helps to define optimization goals, that assure the best economical profits. In the case of combustion processes in power boilers, SILO II increases process efficiency, reduces NOX, CO and SO2 emissions, reduces LOI (Loss of Ignition) and decreases unit heat rate.

The system usually controls around 100 control inputs in order to optimize technological and economic indicators of a process (about 10-20 at the same time).

Read more >>

Read the article >>

Production line configuration optimization
  • Within Horizon 2020 Programme, we took part in the creation of algorithm optimizing configuration of a production line. Designing a production line needs answering questions like: What should be its configuration? Should curved conveyors be used? How many conveyors should be installed in order to minimize purchase expenses and maximize profits?
  • Our solution mimics natural selection algorithm and genetic heredity. Just like in nature, where only best adapted individuals can survive, goal functions that give the best results are maintained and reproduced, while functions with unwanted features are eliminated.
  • New generation of goal functions is determined based on the results from production line simulator. The simulator calculates quantity of products, number of times production line stops and total energy consumed.
  • The goal function is optimized to maximize production line efficiency i.e. to increase productivity while minimizing unit costs.
SOE – Storage Operation Expert
  • SOE is a software solution for optimizing the operation of underground gas storage plants. The main goal of the software is to maximize plant capabilities and to minimize energy usage, while ensuring compliance with operational safety standards.
  • SOE enables the operator not only to choose the optimal configuration of the plant components and caverns, but also to forecast the plant’s flow potential for a longer time-horizon (e.g. 7 days).
  • The plant model implemented here is based on real operational data and fulfills geological boundary conditions defined by expert rock mechanical reports. Within the project we developed models of withdrawal and injection flows.

Read more >>

Read the article (1) >>

Read the article (2) >>

Statistical analysis
Pulverized fuel burners analysis
  • We have helped EDF Polska in analysis of data obtained from experiments performed in Rybnik power plant. The study’s purpose was to determine whether power unit load and coal mill operating parameters influence combustion process in a boiler. Collected data consisted of 4 series of main tests with additional 8 secondary per each main test.
  • After getting familiar with the experiments methodology and data, we proposed a method of the analysis, implemented it and interpreted the results. We used ANOVA (analysis of variance) method which proved that combustion process is dependent on both power unit load and parameters of coal mills feeding the analyzed burners.
  • The study results were published as a research article (see link below).

Read the article >>

The above typical data analysis cases do not exhaust all various projects that we realized for industry.

We are always eager to take on new challenging and non-standard projects.