Applying approximation models#
The use of approximation machine learning as part of the REPEAT VISION project significantly expands the scope of solvable problems in modeling complex systems:
The project can simultaneously take into account both experimental data and the patterns identified in them, as well as various physical models describing the processes occurring in the systems.
The use of ML models on the REPEAT VISION platform allows to accelerate modeling of complex processes by approximation of calculation results Finite elements method models, Finite volume method and other models. ML models allow to replace computationally expensive modeling with fast-to-call models of neural network.
The REPEAT VISION platform uses the neutral ONNX format to load and call models. ONNX is a general-purpose format that supports a variety of machine learning methods for both Python, and other tools and software.
Attention
To work with machine learning models, the user shall have basic skills in mathematical statistics, linear algebra and numerical methods.
Basic knowledge of Python programming language and skills in working with data analysis packages are also required, as well as knowledge of machine learning libraries, such as Scikit-learn, Pytorch or TensorFlow.
If you plan to use another tool or software to prepare approximation models, it shall be compatible with ONNX to ensure that your models can be used on the REPEAT VISION platform.
In order to gain basic skills in preparing approximation models and using them on the REPEAT VISION platform, please review the following tutorials.
The tutorial reviews a practical example of solving the equation of oscillations using physically informed neural networks (PINNs). Provides skills in building models to solve equations, saving them in ONNX format, and using the model in REPEAT VISION.
See also
More industry examples of preparing approximation models, as well as their use on the REPEAT VISION platform, are given in the corresponding how-to guide