HALcoll: Neural Network-based Optical Collimation and Alignment at the MMT Observatory
Modern observatories produce rich streams of telemetry from a variety of sources during the course of operation. The challenge has been how to most effectively harness the information in this telemetry to optimize observatory operations and efficiency. This poster describes the development and performance of a neural-network based tool called "HALcoll" that we use at the MMT Observatory to predict the optimal collimation and alignment of the secondary mirror based on the current state of telescope telemetry (e.g. various temperature readings and telescope pointing). The neural network is built and trained using TensorFlow and Keras along with archived telemetry that coincides with converged wavefront sensor data. The main advantage of this technique is that it does not require any a priori assumptions about how different portions of the telemetry relate to each other. This makes it much easier to incorporate all available telemetry into the model.
