Multi-mode combustion process monitoring on a pulverised fuel combustion test facility based on flame imaging and random weight network techniques

Combustion systems need to be operated under a range of different conditions to meet fluctuating energy demands. Reliable monitoring of the combustion process is crucial for combustion control and optimisation under such variable conditions. In this paper, a monitoring method for variable combustion conditions is proposed by combining digital imaging, PCA-RWN (Principal Component Analysis and Random Weight Network) techniques. Based on flame images acquired using a digital imaging system, the mean intensity values of RGB (Red, Green, and Blue) image components and texture descriptors computed based on the grey-level co-occurrence matrix are used as the colour and texture features of flame images. These features are treated as the input variables of the proposed PCA-RWN model for multi-mode process monitoring. In the proposed model, the PCA is used to extract the principal component features of input vectors. By establishing the RWN model for an appropriate principal component subspace, the computing load of recognising combustion operation conditions is significantly reduced. In addition, Hotelling's T2 and SPE (Squared Prediction Error) statistics of the corresponding operation conditions are calculated to identify the abnormalities of the combustion. The proposed approach is evaluated using flame image datasets obtained on the PACT 250 kW Air/Oxy-fuel Combustion Test Facility (PACT 250 kW Air/Oxy-fuel CTF). Variable operation conditions were achieved by changing the primary air and SA/TA (Secondary Air to Territory Air) splits. The results demonstrate that, for the operation conditions examined, the condition recognition success rate of the proposed PCA-RWN model is over 91%, which outperforms other machine learning classifiers with a reduced training time. The results also show that the abnormal conditions exhibit different oscillation frequencies from the normal conditions, and the T2 and SPE statistics are capable of detecting such abnormalities.