An incremental deep learning model using flame imaging and condition monitoring to predict NOx emissions in oxy-biomass combustion

This paper presents the development of an integrated Incremental Deep Learning (IDL) and Incremental Multi-mode Condition Monitoring (IMCM) model for predicting NOx emissions in an oxy-biomass combustion process based solely on flame images. The integrated model (referred to as the IDL-IMCM model) combines the architectures of the previously established IDL and IMCM models and is capable of learning incrementally from both ‘seen’ and ‘unseen’ datasets. The model is tested and validated using flame datasets obtained from an Air/Oxy-fuel Combustion Test Facility. The test results show that the proposed IDL-IMCM model is capable of predicting NOx emissions for ‘seen’ and ‘unseen’ conditions with a mean absolute percentage error of less than 3%, even after three updates.