Speaker
Description
The J-PARC RCS uses four horizontal and two vertical paint-bump magnets to generate a high-intensity beam through painting injection. Their IGBT-chopper power supplies can reproduce current waveforms with an accuracy of better than 1%. Combining automatic generation of input voltage (IV) patterns with manual fine-tuning keeps the current deviation of the painting pattern (PP) within ±0.2%.
There are 90 waveform patterns in total, including two types: trapezoidal patterns for low-beam-loss studies and painting patterns for high-intensity beam production. As these patterns have different current demands and impose different loads on the power supplies, the tuning process becomes increasingly complex. Adjusting one PP takes approximately one hour and optimising all 90 patterns takes several days; therefore, reducing the adjustment time is essential.
To address this issue, a neural-network (NN) approach was applied to generate optimized IV patterns. Training the NN with existing IV data yielded highly accurate voltage patterns, improving PP reproducibility. This presentation reports on NN-based waveform optimisation and its application to beam operation.
| In which format do you inted to submit your paper? | Word |
|---|