Disruption Precursors Research Articles

Adaptive anomaly detection disruption prediction starting from first discharge on tokamak

Abstract Plasma disruption presents a significant challenge in tokamak fusion, especially in large-size devices like ITER, where it causes severe damage. While current data-driven machine learning methods perform well in disruption prediction, they require extensive discharge data for model training. However, future tokamaks will begin operations without any prior data, making it difficult to train data-driven disruption predictors and select appropriate hyperparameters during the early operation period. In this period disruption prediction also aims to support safe exploration of operation range and accumulate necessary data to develop advanced prediction models. Thus, predictors must adapt to evolving plasma states during this exploration phase. To address these challenges, this study further develops the Enhanced Convolutional Autoencoder Anomaly Detection (E-CAAD) predictor and proposes a cross-tokamak adaptive transfer method based on E-CAAD. By training the E-CAAD model on data from existing devices, the predictor can effectively distinguish between disruption precursor and non-disruption samples in new device, enabling disruption prediction from the first shot on the new device. Additionally, adaptive learning from scratch and alarm threshold adaptive adjustment strategies are proposed to enable model automatically adapt to changes in the discharge scenario. The adaptive learning strategy enables the predictor to fully use scarce data during the early operation of the new device while rapidly adapting to changes in the discharge scenario. The threshold adaptive adjustment strategy addresses the challenge of selecting alarm thresholds on new devices where the validation set is lacking, ensuring that the alarm thresholds adapt to changes in the discharge scenario. Finally, the experiment transferring the model from J-TEXT to EAST exhibit that this method enables disruption prediction from the first shot on EAST, allowing the predictor to adapt to changes in the discharge scenario and maintain high prediction performance.

A self-organised partition of the high dimensional plasma parameter space for plasma disruption prediction

Abstract This paper introduces a disruption predictor constructed through a fully unsupervised two-dimensional mapping of the high-dimensional JET operational space. The primary strength of this disruption predictor lies in its inherent self-organization capability. Diverging from both supervised disruption predictors and earlier approaches suggested by the same authors, which were based on unsupervised models such as Self-Organizing or Generative Topographic Maps, this predictor eliminates the need for labeling data of disruption terminated pulses during training. In prior methods, labels were indeed required post-mapping to inform the model about the presence or absence of disruption precursors at each time instant during the disrupted discharges. In contrast, our approach in this study involves no labeling of data from disruption-terminated experiments. The Self-Organizing Map, operating without any a priori information, adeptly identifies the regions characterizing the pre-disruptive phase. Moreover, SOM discovers non-trivial relationships and captures the complicated interplay of device diagnostics on the internal plasma states from the experimental data. The provided model is highly interpretable; it allows the visualization of high-dimensional data and facilitates easy interrogation of the model to understand the reasons behind its correlations. Hence, utilizing SOMs across various devices can prove invaluable in extracting rules and identifying common patterns, thereby facilitating extrapolation to ITER of the knowledge acquired from existing tokamaks.

Cross-tokamak deployment study of plasma disruption predictors based on convolutional autoencoder

In the initial stages of operation for future tokamak, facing limited data availability, deploying data-driven disruption predictors requires optimal performance with minimal use of new device data. This paper studies the issue of data utilization in data-driven disruption predictor during cross tokamak deployment. Current predictors primarily employ supervised learning methods and require a large number of disruption and non-disruption shots for training. However, the scarcity and high cost of obtaining disruption shots for future tokamaks result in imbalanced training datasets, reducing the performance of supervised learning predictors. To solve this problem, we propose the Enhanced Convolutional Autoencoder Anomaly Detection (E-CAAD) predictor. E-CAAD can be trained only by non-disruption samples and can also be trained by disruption precursor samples when disruption shots occur. This model not only overcomes the sample imbalance in supervised learning predictors, but also overcomes the inefficient dataset utilization faced by traditional anomaly detection predictors that cannot use disruption precursor samples for training, making it more suitable for the unpredictable datasets of future tokamaks. Compared to traditional anomaly detection predictors, the E-CAAD predictor performs better in disruption prediction and is deployed faster on new devices. Additionally, we explore strategies to accelerate the deployment of the E-CAAD predictor on the new device by using data from existing devices. Two deployment strategies are presented: mixing data from existing devices and fine-tuning the predictor trained on existing devices. Our comparisons indicate that the data from existing device can accelerate the deployment of predictor on new device. Notably, the fine-tuning strategy yields the fastest deployment on new device among the designed strategies.

Tokamak plasma disruption precursor onset time study based on semi-supervised anomaly detection

Plasma disruption in tokamak experiments is a challenging issue that causes damage to the device. Reliable prediction methods are needed, but the lack of full understanding of plasma disruption limits the effectiveness of physics-driven methods. Data-driven methods based on supervised learning are commonly used, and they rely on labelled training data. However, manual labelling of disruption precursors is a time-consuming and challenging task, as some precursors are difficult to accurately identify. The mainstream labelling methods assume that the precursor onset occurs at a fixed time before disruption, which leads to mislabeled samples and suboptimal prediction performance. In this paper, we present disruption prediction methods based on anomaly detection to address these issues, demonstrating good prediction performance on J-TEXT and EAST. By evaluating precursor onset times using different anomaly detection algorithms, it is found that labelling methods can be improved since the onset times of different shots are not necessarily the same. The study optimizes precursor labelling using the onset times inferred by the anomaly detection predictor and test the optimized labels on supervised learning disruption predictors. The results on J-TEXT and EAST show that the models trained on the optimized labels outperform those trained on fixed onset time labels.

Disruption prediction based on fusion feature extractor on J-TEXT

Predicting disruptions across different tokamaks is necessary for next generation device. Future large-scale tokamaks can hardly tolerate disruptions at high performance discharge, which makes it difficult for current data-driven methods to obtain an acceptable result. A machine learning method capable of transferring a disruption prediction model trained on one tokamak to another is required to solve the problem. The key is a feature extractor which is able to extract common disruption precursor traces in tokamak diagnostic data, and can be easily transferred to other tokamaks. Based on the concerns above, this paper presents a deep feature extractor, namely, the fusion feature extractor (FFE), which is designed specifically for extracting disruption precursor features from common diagnostics on tokamaks. Furthermore, an FFE-based disruption predictor on J-TEXT is demonstrated. The feature extractor is aimed to extracting disruption-related precursors and is designed according to the precursors of disruption and their representations in common tokamak diagnostics. Strong inductive bias on tokamak diagnostics data is introduced. The paper presents the evolution of the neural network feature extractor and its comparison against general deep neural networks, as well as a physics-based feature extraction with a traditional machine learning method. Results demonstrate that the FFE may reach a similar effect with physics-guided manual feature extraction, and obtain a better result compared with other deep learning methods.

IDP-PGFE: an interpretable disruption predictor based on physics-guided feature extraction

Disruption prediction has made rapid progress in recent years, especially in machine learning (ML)-based methods. If a disruption prediction model can be interpreted, it can tell why certain samples are classified as disruption precursors. This allows us to tell the types of incoming disruption for disruption avoidance and gives us insight into the mechanism of disruption. This paper presents a disruption predictor called interpretable disruption predictor based on physics-guided feature extraction (IDP-PGFE) and its results on J-TEXT experiment data. The prediction performance of IDP-PGFE with physics-guided features is effectively improved (true positive rate = 97.27%, false positive rate = 5.45%, area under the ROC curve = 0.98) compared to the models with raw signal input. The validity of the interpretation results is ensured by the high performance of the model. The interpretability study using an attribution technique provides an understanding of J-TEXT disruption and conforms to our prior comprehension of disruption. Furthermore, IDP-PGFE gives a possible mean on inferring the underlying cause of the disruption and how interventions affect the disruption process in J-TEXT. The interpretation results and the experimental phenomenon have a high degree of conformity. The interpretation results also gives a possible experimental analysis direction that the resonant magnetic perturbations delays the density limit disruption by affecting both the MHD instabilities and the radiation profile. PGFE could also reduce the data requirement of IDP-PGFE to 10% of the training data required to train a model on raw signals. This made it possible to be transferred to the next-generation tokamaks, which cannot provide large amounts of data. Therefore, IDP-PGFE is an effective approach to exploring disruption mechanisms and transferring disruption prediction models to future tokamaks.

Integrated deep learning framework for unstable event identification and disruption prediction of tokamak plasmas

The ability to identify underlying disruption precursors is key to disruption avoidance. In this paper, we present an integrated deep learning (DL) based model that combines disruption prediction with the identification of several disruption precursors like rotating modes, locked modes, H-to-L back transitions and radiative collapses. The first part of our study demonstrates that the DL-based unstable event identifier trained on 160 manually labeled DIII-D shots can achieve, on average, 84% event identification rate of various frequent unstable events (like H-L back transition, locked mode, radiative collapse, rotating MHD mode, large sawtooth crash), and the trained identifier can be adapted to label unseen discharges, thus expanding the original manually labeled database. Based on these results, the integrated DL-based framework is developed using a combined database of manually labeled and automatically labeled DIII-D data, and it shows state-of-the-art (AUC = 0.940) disruption prediction and event identification abilities on DIII-D. Through cross-machine numerical disruption prediction studies using this new integrated model and leveraging the C-Mod, DIII-D, and EAST disruption warning databases, we demonstrate the improved cross-machine disruption prediction ability and extended warning time of the new model compared with a baseline predictor. In addition, the trained integrated model shows qualitatively good cross-machine event identification ability. Given a labeled dataset, the strategy presented in this paper, i.e. one that combines a disruption predictor with an event identifier module, can be applied to upgrade any neural network based disruption predictor. The results presented here inform possible development strategies of machine learning based disruption avoidance algorithms for future tokamaks and highlight the importance of building comprehensive databases with unstable event information on current machines.

An Application of Machine Learning for Plasma Current Quench Studies via Synthetic Data Generation

Electromagnetic forces, thermal loads, and radiation loads experienced by the in-vessel components or vacuum vessels at the time of the tokamak plasma current quench (CQ) significantly affect the overall plasma device’s health. Thus the mitigation of plasma CQ is of paramount importance, which requires a proper identification of the disruption precursors. Using new Machine Learning (ML) and Artificial Intelligence (AI) approaches, it is possible to identify disruption precursors; however, such approaches require training the ML models. This training of models requires a massive amount of experimental data, which sometimes may not be available for different tokamaks. This necessitates the need for accurate synthetic disruption data generation presenting different types of the CQ profiles observed experimentally. A novel approach for synthetic CQ data generation, considering the experimental aspect of the CQ profile shape for a wide range of tokamak plasma discharges, is designed to train ML/AI models. The trained model results are also elaborated here, which includes identifying current before disruption and classification of CQ profile types in time-space.

Deep Learning for the Analysis of Disruption Precursors Based on Plasma Tomography

The JET baseline scenario is being developed to achieve high fusion performance and sustained fusion power. However, with higher plasma current and higher input power, an increase in pulse disruptivity is being observed. Although there is a wide range of possible disruption causes, the present disruptions seem to be closely related to radiative phenomena such as impurity accumulation, core radiation, and radiative collapse. In this work, we focus on bolometer tomography to reconstruct the plasma radiation profile, and on top of it, we apply anomaly detection to identify the radiation patterns that precede major disruptions. The approach makes extensive use of machine learning. First, we train a surrogate model for plasma tomography based on matrix multiplication, which provides a fast method to compute the plasma radiation profiles across the full extent of any given pulse. Then, we train a variational autoencoder to reproduce the radiation profiles by encoding them into a latent distribution and subsequently decoding them. As an anomaly detector, the variational autoencoder struggles to reproduce unusual behaviors that include not only the actual disruptions but their precursors as well. These precursors are identified based on an analysis of the anomaly score across all baseline pulses in two recent campaigns at JET.

Progress Toward Interpretable Machine Learning–Based Disruption Predictors Across Tokamaks

In this paper we lay the groundwork for a robust cross-device comparison of data-driven disruption prediction algorithms on DIII-D and JET tokamaks. In order to consistently carry on a comparative analysis, we define physics-based indicators of disruption precursors based on temperature, density, and radiation profiles that are currently not used in many other machine learning predictors for DIII-D data. These profile-based indicators are shown to well-describe impurity accumulation events in both DIII-D and JET discharges that eventually disrupt. The univariate analysis of the features used as input signals in the data-driven algorithms applied on the data of both tokamaks statistically highlights the differences in the dominant disruption precursors. JET with its ITER-like wall is more prone to impurity accumulation events, while DIII-D is more subject to edge-cooling mechanisms that destabilize dangerous magnetohydrodynamic modes. Even though the analyzed data sets are characterized by such intrinsic differences, we show through a few examples that the inclusion of physics-based disruption markers in data-driven algorithms is a promising path toward the realization of a uniform framework to predict and interpret disruptive scenarios across different tokamaks. As long as the destabilizing precursors are diagnosed in a device-independent way, the knowledge that data-driven algorithms learn on one device can be re-used to explain a disruptive behavior on another device.

Multi-device study of temporal characteristics of magnetohydrodynamic modes initiating disruptions

Disruptions in tokamaks are often preceded by magnetohydrodynamic (MHD) instabilities that can rotate or become locked to the wall. Measurements from magnetic diagnostics in the presence of MHD mode precursors to disruptions can yield potentially valuable input to the plasma control system, with a view to disruption avoidance, prediction and mitigation. This paper presents an exploratory analysis of the growth of MHD modes and corresponding time scales on the basis of magnetic measurements in multiple tokamaks. To this end, a database was compiled using disruptive discharges from COMPASS, ASDEX Upgrade, DIII-D and JET, manually classified according to disruption root cause, and characterized by a great diversity of operational conditions and mode dynamics. The typical time during which a mode can be detected using saddle coils and the duration of the locked mode phase in the database both extend over several orders of magnitude, but generally the time scales increase with plasma size. Several additional factors are discussed that can influence these durations, including the disruption root cause. A scaling law for the locked phase duration was estimated, yielding predictions toward ITER of the order of hundreds of milliseconds or even seconds. In addition, a scaling law for the mode amplitude at the disruption onset, proposed earlier by de Vries et al. (2016), is applied to the database, and its predictive capabilities are assessed. Despite significant uncertainty on the predictions from both scaling laws, encouraging trends are observed of the fraction of disruptions that may be detected with sufficient warning time to allow mitigation or even avoidance, based solely on observations of MHD mode dynamics. When combined with similar analysis of measurements from diagnostics that are sensitive to other disruption precursors, our analysis methods and results may contribute to the reliability, robustness and generalization of disruption warning schemes for ITER.

Outline Design of ITER Radial X-Ray Camera Diagnostic

The radial X-ray camera (RXC) is designed to measure the poloidal profile of plasma X-ray emission with high spatial and temporal resolution. Its primary diagnostic role includes measuring low (m, n) magnetohydrodynamic modes, sawteeth and disruption precursors, H-mode, edge-localized modes, and L-H transition. The RXC comprises two subsystems, i.e., in-port and ex-port cameras that view the outer and core regions, respectively, through vertical slots in the diagnostics shield module of an equatorial port plug. Detailed camera design is in progress including design of the camera structure, electronics, data acquisition and control, calibration, and pretest on the EAST tokamak. The sight path and neutron shielding have been optimized. The secondary vacuum, heat insulation, cooling, positioning, and calibration have been designed. The structure analysis results for the external camera indicate that even under five times gravity acceleration, the maximum stress was still below the allowable stress. The heat analysis results indicate that the maximum temperature on the detector box was ~56°C, which is within the detector operation temperature limit. The neutronics analysis results indicate that the detectors can be operated during the whole deuterium-deuterium phase without detector replacement. The electronics group and instrumentation and control group have also made good progress.

Research on the effect of resonant magnetic perturbations on disruption limit in J-TEXT tokamak

The effect of resonant magnetic perturbations (RMPs) on high density limit and low-q limit discharges is studied in J-TEXT tokamak. It is found that a moderate amplitude of applied RMPs increases the density limit and delays the disruption, and the disruption precursor is suppressed with a slight reduction in toroidal plasma rotation. A too large amplitude of RMPs, however, leads to a lower density limit and an earlier disruption. For the optimal RMPs amplitude, the disruption is delayed by about 150 ms. For low-q discharges, the applied static RMP can either lower the limit of the edge safety factor qa from 2.15 to nearly 2.0 or sustain stable plasma discharge with qa = 2.1, and the disruptive precursor is also suppressed.

Modelling and control for plasma disruption avoidance and mitigation

This paper proposes an approach aimed at avoiding or mitigating the consequences of a plasma disruption in a tokamak. The control strategy is based on a Disruption Predictive Control (DPC) where the use of a set of disruption predictors could allow actively modification of some plasma parameters when a disruption warning is issued. The DPC approach requires the continuous monitoring of various plasma performance indicators, which play the role of disruption precursors. The active control is based on the availability of magnetic and nonmagnetic plasma parameter response models as well as multivariable control algorithms and architectures.

Avoidance of disruptions at high βN in ASDEX Upgrade with off-axis ECRH

Experiments on disruption avoidance have been carried out in H-mode ASDEX Upgrade plasmas: the localized perpendicular injection of ECRH (1.5 MW ∼ 0.2Ptot) onto the q = 2 resonant surface has led to the delay and/or complete avoidance of disruptions in a high βN scenario (Ip = 1 MA, Bt = 2.1 T, q95 ∼ 3.6, with NBI ∼7.5 MW). In these discharges (at low q95 and low density) neoclassical tearing modes (NTMs) are excited: the growth and locking of the m/n = 2/1 mode leads to the disruption. The scheme of the experiment is successfully applied in the same way as in previous disruption avoidance experiments in FTU and ASDEX Upgrade. As soon as the disruption precursor signal (the locked mode detector and/or the loop voltage) reaches the preset threshold, the ECRH power is triggered by real-time control. A poloidal scan in deposition location (ρdep) has been carried out by setting the poloidal launching mirrors at different angles in each discharge. The results depend on ρdep: complete disruption avoidance can be achieved when the power is injected close to or onto the 2/1 island. When ECRH is injected outside the island (either at radii inside or outside the q = 2 surface), the discharge is disrupted as in the reference case.

Disruption control on FTU and ASDEX upgrade with ECRH

The use of ECRH has been investigated as a promising technique to avoid or postpone disruptions in dedicated experiments in FTU and ASDEX Upgrade. Disruptions have been produced by injecting Mo through laser blow-off (FTU) or by puffing deuterium gas above the Greenwald limit (FTU and ASDEX Upgrade). The toroidal magnetic field is kept fixed and the ECRH launching mirrors have been steered before every discharge in order to change the deposition radius. The loop voltage signal is used as disruption precursor to trigger the ECRH power before the plasma current quench. In the FTU experiments (Ip = 0.35–0.5 MA, Bt = 5.3 T, PECRH = 0.4–1.2 MW) it is found that the application of ECRH modifies the current quench starting time depending on the power deposition location. A scan in deposition location has shown that the direct heating of one of the magnetic islands produced by magnetohydrodynamic (MHD) resistive instabilities (either m/n = 3/2, 2/1 or 3/1) prevents its further growth and also produces the stabilization of the other coupled modes and the delay of the current quench or its full avoidance. Disruption avoidance and complete discharge recovery are obtained when the ECRH power is applied on rational surfaces. The modes involved in the disruption are found to be tearing modes stabilized by a strong local ECRH heating. The Rutherford equation has been used to reproduce the evolution of the MHD modes. In the ASDEX Upgrade experiments L-mode plasmas (Ip = 0.6 MA, Bt = 2.5 T, PECRH = 0.6 MW ∼ POHM) the injection of ECRH close to q = 2 significantly delays the 2/1 onset and prolongs the duration of the discharge: during this phase the density continues to increase. No delay in the onset of the 2/1 mode is observed when the injected power is reduced to 0.35 MW.

A New Criterion for Disruption Prediction on HL-2A

A new criterion has been proposed to predict the major disruptionscaused by tearing mode instabilities. According to the HL-2Aexperimental results, the statistical analyses are employed toinvestigate the relationships between MHD activities and the plasmadisruptions. Two kinds of the tearing mode activities can finally causethe disruption on HL-2A operations. By introducing a new parameter,i.e. an integral of poloidal magnetic field over time, asthe criterion of disruption precursor, almost all of the disruptionscan be predicted.

Heterodyne ECE diagnostic in the mode detection and disruption avoidance at TEXTOR

The paper describes the detection of the m/n = 2/1 disruption precursor at TEXTOR. This frequently observed precursor is detected by realtime cross-correlation of two radially separated channels of the heterodyne electron cyclotron emission diagnostic. An electronic module developed to detect the precursor and its functionality is briefly described. A disruption statistic for TEXTOR is presented, from which the probability of the occurrence of the m/n = 2/1 mode is seen. A method to generate reproducible disruption precursors at TEXTOR is also presented. Experiments to avoid disruptions by inducing toroidal velocity shear by tangential neutral beam injection and mitigate disruptions by He-puffing are presented. In the experiments performed, up to 50% of the disruptions could be avoided. Furthermore, calculations of the precursor frequency and its evolution are described and compared with measurements.

MHD stability of optimized shear discharges in JET

The main limitation to the performance of JET optimized shear (OS) discharges is due to MHD instabilities, mostly in the form of a disruptive limit. The structure of the MHD mode observed as a precursor to the disruption, as measured from SXR and ECE diagnostics, shows a global ideal MHD mode. The measured mode structure is in good agreement with the calculated mode structure of the pressure driven kink mode. The disruptions occur at relatively low normalized beta (1 < βN < 2), in good agreement with calculated ideal MHD stability limits for the n = 1 pressure driven kink mode. These low limits are mainly due to the extreme peaking factor of the pressure profiles. Other MHD instabilities observed in JET OS discharges include, usually benign, chirping modes. These modes, which occur in bursts during which the frequency changes, have the same mode structure as the disruption precursor but are driven unstable by fast particles.

Growth of ideal magnetohydrodynamic modes driven slowly through their instability threshold: Application to disruption precursors

The growth of an ideal magnetohydrodynamic (MHD) instability in a high-temperature plasma is calculated in the case where the plasma β is driven slowly through its instability threshold. The MHD perturbation grows faster than exponentially, approximately as exp[(t/τ)3/2]. Its characteristic growth time τ∼(32)2/3γ̂MHD−2/3γh−1/3 is a hybrid of the ideal MHD incremental growth rate γ̂MHD and the heating rate γh. This simple model agrees well with the observed growth of disruption precursors in high β DIII-D [J. L. Luxon and L. G. Davis, Fusion Technol. 8, 441 (1985)] discharges having strongly peaked pressure profiles, where the observed growth times of ⩾10−4 s are significantly slower than the typical ideal MHD time scale of ⩽10−5 s.