Neural Network Simulation of Burner Operation Parameters

Instruments and Experimental Techniques · Springer Nature · 2026-09-01 · DOI: 10.1134/S0040601526700230

Аннотация

Abstract— The article presents the results of studying the applicability of deep learning methods for predicting the operation parameters of a liquid fuel fired burner with steam-assisted atomization and a gas generation chamber, which is intended for combusting low grade liquid hydrocarbon fuels and waste based on a flame visual analysis. The scientific novelty of the study is that with steam-assisted atomization it is possible to obtain visual flame patterns differing from those obtained using conventional methods, for interpreting of which a dedicated approach is required. A modified architecture of the dense convolutional neural network DenseNet-121 is proposed with weights pretrained on the ImageNet data set, which is adapted to solving the regression problem of predicting the device process operation parameters (fuel and superheated steam flowrates). The DenseNet-121 architectureс with dense interlayer connections is able to efficiently draw fine visual features from the flame images, which are critical for establishing the functional dependences. The regression subnetwork includes fully connected layers with the rectified linear unit (ReLU) activation and Dropout regularization. Validation on an independent test sample of experimental data has shown high accuracy of the model: for the fuel flowrate, the mean absolute error (MAE) is 3.8%; the root mean square error (RMSE) is 9%, and the determination coefficient R 2 = 0.957; for steam flowrate the MAE is 2.6%; RMSE is 4%, and R 2 = 0.994. The achieved results ( R 2 > 0.95 for both the parameters) confirm that the model is able to catch complex interrelations between the flame visual characteristics and burner operation parameters. The proposed solution is aimed at developing an intellectual tool for prompt contactless monitoring and optimization of liquid fuel combustion processes in burners with steam-assisted atomization and a gas generation chamber.

Читать у издателя →

Похожие статьи