The analysis of charge exchange recombination spectra represents a very challenging problem due to the presence of many overlapping spectral lines. Conventional approaches are based on iterative least-squares optimization and suffer from the two difficulties of low speed and the need for a good initial approximation to the solution. This latter problem necessitates considerable human supervision of the analysis procedure. The authors show how neural network techniques allow charge exchange data to be analysed very rapidly, to give an approximate solution without the need for supervision. The network approach is well suited to the fast intershot analysis of large volumes of data, and can readily be implemented in dedicated hardware for real-time applications. The neural network can also be used to provide the initial guess for the standard least-squares algorithm when high accuracy is required.
Performance prediction of radio frequency based negative ion source using fusion neural network model