In the project, we used deep neural network learning for the reduced order embedding for representing the dynamics of the flow system.
We used reduced order embedding for training and predicting the flow in future flow time and obtained 2 orders of magnitude reduction in computational run times. We developed technology for reduced order modeling of complex processes, which generates a large amount of data.
The technology developed during the project can be used for the prediction or generation of video frames using small computational costs.
The project cost 200,000 PLN and was financed by IDUB PW. The fund was allocated for designing a fluid mixer with cross activation using machine learning techniques.
Starting the design process, we used machine learning algorithms to create a model of the behavior of a fluid mixer. This model is trained on a dataset of known input and output parameters, such as flow rate, fluid viscosity, mixer geometry, and mixing efficiency. Once we obtained the model, we used it to explore the design space and optimize the mixer's performance. We can use techniques such as Bayesian optimization or genetic algorithms to search for the best combination of input parameters, such as cross-activation frequency, mixer blade angle, and flow rate.
By utilizing machine learning to assist in the design process, we can efficiently explore a vast design space and quickly identify optimal solutions. This approach can lead to the development of highly efficient fluid mixers that can be tailored to specific industrial applications.
The technology developed during the project has a wide spectrum of applications in areas where small and accurate representations of large data for future state predictions are needed. The most popular applications are: Wind turbine industries Computer vision-based technologies