Machine learning-based system for vacuum leakage detection in EAST fusion device

Abstract

In order to achieve real-time monitoring of partial pressures of residual gases and timely detection of vacuum leakage in fusion devices, a machine learning-based system was successfully developed on the LabVIEW software platform for its first-time application in the EAST fusion device. This system acquires partial pressure data from the quadrupole mass spectrometer (QMS) in both histogram and P&T scanning modes to identify the nature of vacuum leakage. The backpropagation (BP) neural network algorithm is employed for leakage detection. To train the BP neural network model, historical data from normal and leakage scenarios on the EAST device are utilized as inputs. Subsequently, the trained model is integrated into the system. The BP neural network model exhibited an identification accuracy of 95.27 % on the test set, effectively detecting water and gas leakage on EAST. Consequently, this system enables rapid detection of vacuum leakage in fusion devices, facilitating prompt mitigation of their impact on plasma experiments and ensuring the device's safety and stability.