The Advanced Sources and Detectors project is building an advanced multi-pulse linear induction accelerator capable of generating a 1.4 kA electron beam at energies up to 24 MeV. The accelerator, named Scorpius after the brightest known x-ray source in the sky, will be unique in its use of solid-state pulsed power (SSPP) to generate the voltage pulse for the injector and accelerating gaps throughout the accelerator, giving Scorpius unique control of the pulse shape by independently triggering 45 individual stages stacked in each of nearly 1,000 line replaceable units (LRUs). To take full advantage of the SSPP flexibility, automated optimization of the pulse shape to a desired waveform is currently under development. To demonstrate this capability, nonlinear surrogate circuit models of the SSPP have been developed using the hybrid transmission line/modified nodal analysis code, CASTLE, that include parasitics and a dummy load to generate reflections. Data-efficient Bayesian optimizations calling CASTLE directly for each iteration are compared with results from a convolutional neural network or other machine learning model trained on data generated by CASTLE, and the effect of the number of stages on pulse flattening is discussed.

The photocathodes used as electron sources in high-performance electron accelerators today are largely one of only a handful of materials. While there has been an increased understanding of the properties of the electron beams produced by these cathodes, there has been little change in the overall selection of materials used at accelerator facilities. In fact, nearly all of the photocathodes in use today originated in the photomultiplier tube or night vision goggle industries, where efforts were aimed at discovering new materials by employing trial-and-error based iterative experimental approaches. Our work in the field of photocathode discovery was initially directed towards improving the brightness of electron beams used in FELs and was the first data-driven approach to screening for high brightness photocathode materials. Through screening over 74,000 semiconducting materials, a list of novel candidate materials was generated. Our current work is focused on two other areas of interest for photocathodes: very high average current photocathodes and spin-polarized electrons. We will apply active learning techniques to reduce the amount of computationally expensive calculations that need to be performed in order to discover more new materials for photocathodes.