Combustion Regime Monitoring by Flame Imaging and Machine Learning

Абдуракипов, С. С.; Gobyzov, O. A.; Tokarev, M. P.; Дулин, В. М.

doi:10.3103/s875669901805014x

Cited by 17 publications

(4 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One of the applications in which these networks perform particularly well is the detection of anomalies in the time series of different signals. 32–36 In the study, we utilised three types of DRNNs. They are as follows: LSTM recurrent neural network, one-dimensional (1D) convolutional neural network LSTM (CNN-LSTM) and 1D convolutional LSTM (ConvLSTM).…”

Section: Methodsmentioning

confidence: 99%

“…Many of them work very well in the process of recognition of time series and images, including flame. For example, Abdurakipov et al 11 studied the possibility of monitoring combustion regimes using flame images of a gas burner. Lyu et al 12 applied a deep belief network for the detection of abnormal conditions in the experimental combustion system.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Combustion process monitoring based on flame intensity time series

Omiotek

Smolarz

2020

Proceedings of the Institution of Mechanical Engineers, Part I:

View full text Add to dashboard Cite

The pulverised coal and its mixture with biomass are one of the most popular fuels in industrial energy. To ensure, on one hand, minimal greenhouse gas emission and, on the other hand, maximum energy production efficiency, it is necessary to monitor the combustion process of these fuels. One way to do this is to monitor the flame intensity. This is an optical, non-invasive solution, and information on the status of the process is obtained with minimal delay. The article proposes a method for identifying undesired combustion states for which the excess air coefficient is greater or smaller than the value ensuring total combustion. Three deep recurrent neural network architectures for classifying the flame intensity time series were explored. The best results were obtained using the convolutional long short-term memory model, which offered the accuracy of 86.5%–99.8%, depending on the current thermal power. The prediction time of a single data sequence was about 0.6 ms. High accuracy and low time consumption of the proposed method create an opportunity for its use in industrial combustion systems of pulverised coal and its mixture with biomass.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Combustion process monitoring based on flame intensity time series

Omiotek

Smolarz

2020

Proceedings of the Institution of Mechanical Engineers, Part I:

View full text Add to dashboard Cite

show abstract

“…Various architectures of such networks have been developed, and many of them achieve excellent performance in image recognition tasks, including flame images. Examples include deep convolutional neural networks [ 2 , 3 , 4 , 5 , 6 ], deep belief networks [ 7 , 8 ], deep convolutional auto-encoder [ 9 ], deep convolutional auto-encoder connected with the principal component analysis and the hidden Markov model [ 10 ], deep, fully connected neural networks [ 11 ], deep convolutional selective autoencoder [ 12 ] and various architectures followed by a symbolic time series analysis [ 13 , 14 ]. Each of the architectures mentioned above plays a vital role in a specific application area.…”

Section: Introductionmentioning

confidence: 99%

Flame Image Processing and Classification Using a Pre-Trained VGG16 Model in Combustion Diagnosis

Omiotek

Kotyra

2021

Sensors

View full text Add to dashboard Cite

Nowadays, despite a negative impact on the natural environment, coal combustion is still a significant energy source. One way to minimize the adverse side effects is sophisticated combustion technologies, such as, e.g., staged combustion, co-combustion with biomass, and oxy-combustion. Maintaining the combustion process at its optimal state, considering the emission of harmful substances, safe operation, and costs requires immediate information about the process. Flame image is a primary source of data which proper processing make keeping the combustion at desired conditions, possible. The paper presents a method combining flame image processing with a deep convolutional neural network (DCNN) that ensures high accuracy of identifying undesired combustion states. The method is based on the adaptive selection of the gamma correction coefficient (G) in the flame segmentation process. It uses the empirically determined relationship between the G coefficient and the average intensity of the R image component. The pre-trained VGG16 model for classification was used. It provided accuracy in detecting particular combustion states on the ranging from 82 to 98%. High accuracy and fast processing time make the proposed method possible to apply in the real systems.

show abstract

“…Wang et al [15] utilized a CNN model to extract flame features for predicting burning states of power plant furnace. Abdurakipov et al [16] established a CNN model based on labeled flame images, which is used to predict the combustion regimes of a laboratory-scale swirling gas burner. Even though various progress has been made through CNN-based models, one non-trivial problem in these models is that a larger amount of labeled data is needed for training.…”

Section: Introductionmentioning

confidence: 99%