Machine learning assisted optical diagnostics on a cylindrical atmospheric pressure surface dielectric barrier discharge

The present study explores combining machine learning (ML) algorithms with standard optical diagnostics (such as time–integrated emission spectroscopy and imaging) to accurately predict operating conditions and assess the emission uniformity of a cylindrical surface Dielectric Barrier Discharge (SDBD). It is demonstrated that ML can be complementary with these optical diagnostics and identify peculiarities associated with the discharge emission pattern at different high voltage waveforms (AC and pulsed) and amplitudes. By employing unsupervised (Principal Component Analysis (PCA)) and supervised (Multilayer Perceptron (MLP) neural networks) algorithms, the applied voltage waveform and amplitude are categorised and predicted based on correlations/differences identified within large amounts of corresponding data. PCA allowed us to effectively classify the voltage waveforms and amplitudes applied to the SDBD through a transformation of the spectroscopic/imaging data into principal components (PCs) and their projection to a two-dimensional PC space. Furthermore, an accurate prediction of the voltage amplitude is achieved using the MLP which is trained with PCA–preprocessed data. A particularly interesting aspect of this concept involves examining the uniformity of the emission pattern of the discharge. This is achieved by analysing spectroscopic data recorded at four different regions around the SDBD surface using the two ML–based techniques. These discoveries are instrumental in enhancing plasma–induced processes. They open up new avenues for real–time control, monitoring, and optimization of plasma–based applications across diverse fields such as flow control for the present SDBD.