To content

Setting the operating points

The operating points (tunes) of a storage ring describe the number of so-called betatron oscillations of the stored particle beam per revolution. They are crucial tuning  parameters for stable operation and must therefore be constantly monitored and readjusted if necessary. The tunes are set with the aid of beam-focusing quadrupole magnets. For this purpose, 3 horizontally and 4 vertically focusing quadrupole circuits are available, each of which is arranged in the arcs of the DELTA storage ring. For automatic control, a classic Proportional Integral Differential (PID) feedback system has been used for many years.

Simulated operating point calculations (Qx, Qy) with superconducting wiggler magnet (SAW) switched on and off.
Structure of the neural network for a ML-based tunes control.
"Necktie" stability diagram for a triplet unit cell (insert) of the DELTA storage ring. The operating point (horizontal tune) per cell is shown in colour as a function of the quadrupole strengths (QF, QD).

In a test scenario, this system was replaced with an ML-based control mechanism. For this purpose, quadrupole strengths (actuator data) were systematically and randomly varied, and the corresponding tune shifts (sensor data) were measured. These actuator/sensor measurement pairs serve as the dataset for supervised training of classical, "shallow" (not deep) feed-forward neural networks (see picture left). During training, the networks "learn" the correlation between changes in quadrupole strengths and shifts in operating points, and can subsequently be used for automatic correction. This approach was successfully employed for both the simulated storage ring and in real machine operation.  Further details can be found in the internal report and in the IPAC-21 and ICALEPCS-21 conference papers.