Application of machine learning methods in designing combustion chambers of gas turbine engines

The article considers the application of a recurrent neural network with long short-term memory (LSTM) and a gradient boosting algorithm for determining the key geometric dimensions of the diffuser of the combustion chamber of an air-craft gas turbine engine. Numerical modeling of physical processes in the diffuser is performed based on the finite ele-ment method, and total pressure losses are subsequently calculated. A database is compiled containing various geo-metric configurations of the diffuser model alongside the corresponding total pressure loss values. The configuration with the lowest total pressure loss is selected as the reference. The performance of the gradient boosting method is compared with that of the LSTM neural network, based on the total pressure loss data obtained from numerical model-ing of the diffuser across a range of geometric configurations. The gradient boosting approach yielded an error of 1.64%, whereas the LSTM network demonstrated an error of 7.28%.This approach enables the creation of a design database for diffuser configurations, facilitates the use of simulation data to train neural networks, and allows for sub-sequent training on alternative designs. The results can be applied in the design and optimization of combustion cham-bers in aircraft engines.