Description
Understanding how precursor phenomena combine to trigger disruptions remains critical for reliable avoidance and mitigation in future devices. TCV offers a unique disruption testbed, with a broad diversity of naturally occurring precursor types and a large database of disruptive and non-disruptive discharges. In this work, we present results from an event-chain analysis performed on a large database ($\sim 4000$ shots) aimed at identifying typical pathways to the first minor disruption and their relations to subsequent final disruption in the TCV tokamak.
We first adapt and deploy a ``Melt Event'' detector based on the heat-flux-factor (HFF) [1], an engineering proxy for surface temperature during heat-flux pulses: $\mathrm{HFF}\propto \dot W_{\mathrm{th}}\sqrt{\tau},$ where $\dot W_{\mathrm{th}}$, the rate of change of kinetic energy, is inferred from the diamagnetic-loop measurements during a heat-flux pulse time $\tau$ using the toroidal flux relation $\Phi = \frac{2\kappa}{1+\kappa^2} \frac{(\mu_0I_p)^2}{8\pi B_{\phi0}}(1-\beta_\theta)$ [2]. Although TCV's stored energy is below melting-relevant levels, the detector provides an automated and physics-motivated timestamp for rapid stored-energy drops and can be scaled to SPARC [3] level heat-fluxes to do projections of potential melting event. The detector correlates well with thermal-quench like energy losses and allows to highlight the most significant minor disruption events.
Using TCV's disruption event database, assembled within the DEFUSE framework [4], we then examine the events occurring prior to the first minor disruption. A first direct result is the events frequency ranking indicating that MHD activity is the most common precursor, followed by loss of vertical control, edge cooling, MARFEs, and radiation-collapses. Building on these trends, we compute event-chain statistics: transition probabilities in form of directed graphs, inter-event timing and event-based warning times prior to disruptions. We compare event-chains across outcomes, separating sequences that precede benign events from melt events. We further classify the analysis by discharge phase (current ramp-up, flat-top, and ramp-down) and confinement regimes to test whether precursor ordering and warning times depend on the operational stage.