Using multilayer perceptron to avoid false-positive quench signals in Toroidal Field quench detection voltage system

In the framework of Europe-Japan collaboration for international development on fusion devices, the tokamak JT-60SA represents a significant milestone in the advancement of fusion energy research, as it is designed to operate under conditions similar to those of ITER and future fusion reactors. In this context, accurate measurement and control of plasma behavior are essential for optimizing fusion reactions. An important issue identified on JT-60SA is the presence of spurious signals in the Toroidal Field Quench Detection Voltage (TFQDV) system, a key component for maintaining plasma stability. These spurious signals, arising due to the interaction between Poloidal Field (PF) coils and the TFQDV system, can become false-positive quench detections, causing unnecessary TF system discharges and interfering with the experimental campaign. We propose a novel approach that utilizes a multilayer perceptron artificial neural network (ANN) to model the dynamics of the interlock system, bypassing the need for detailed parameter extractions. The objective is to provide a robust, generalized solution that improves performance across a variety of PF current patterns. We demonstrate that the ANN-based model can perform significant corrections in different scenarios with varying PF coil behaviors. The performance of the proposed model is validated using data from the Integrated Commissioning database, which includes a diverse set of plasma scenarios with different PF coil energization strategies.