Advancing AI/ML at the Advanced Photon Source

To address this, we have designed a fully integrated digital twin environment for simulation and debugging based on experimentally collected data. We have been able to show that incorporating time and physics-based biasing can improve long-term stability and robustness sufficiently for long-term use. Resulting ML libraries were interfaced with a Tcl/Tk production control system via an SDDS-compatible interface, which allowed for easy integration and immediate operational use. Experimental benchmarks have demonstrated new methods to be faster than existing ones in recovering full performance of the accelerator after a perturbation.