Smart Experiment Control at FAIR Phase 0

As a part of the international accelerator facility FAIR project in Germany, a study was conducted into the viability of implementing machine learning (ML) based algorithms for realignment of the HADES' forward Straw Tracking System (STS). All models tested were forward feeding regression models, utilizing a custom loss function which calculated the minimum residual distance between the model's adjusted hit location and the provided track. Output layers and loss functions were modified depending on if the model's purpose was to test translational realignment, track parameter correction or both simultaneously. The models were trained and tested on Monte Carlo simulated data where translational misalignments were introduced into the two modules of the STS and aligned, misaligned and generated tracks were used. It was found that the models designed for translational realignment could learn and correct for numerous simultaneous shifts in both modules of the STS, but could not correct for misalignments not present in their training datasets. The models could learn how to adjust the location to better fit hits to misaligned tracks, but were unable to correct simultaneously for both misaligned track and detector parameters. In this poster presentation, I will present the status of the current work and discuss the plan for future work.