Machine learning-guided data augmentation may support the development of technologies in the physical sciences, such as nuclear fusion tokamaks. Here we endeavor to study the problem of detecting disruptions - i.e., plasma instabilities that can cause significant damage, impairing the reliability and efficiency required for their real-world viability. Machine learning (ML) prediction models have shown promise in detecting disruptions for specific tokamaks, but they often struggle to generalize to the diverse characteristics and dynamics of different machines. This limits the effectiveness of ML models across different tokamak designs and operating conditions, which is a critical barrier to scaling fusion technology. Given the success of data augmentation in improving model robustness and generalizability in other fields, this study explores the use of a novel time series viewmaker network to generate diverse augmentations or “views” of training data. Our results show that incorporating views during training improves AUC and F2 scores on DisruptionBench tasks compared to standard or no augmentations. The effectiveness of viewmaker networks appears scenario-dependent: while they hindered zero-shot transfer between tokamaks, they showed significant promise for few-shot transfer and may help satisfy the increased data demands of larger, more complex prediction models. This approach represents a promising step towards developing more broadly applicable ML models for disruption avoidance, which is essential for advancing fusion technology and ultimately addressing climate change through reliable and sustainable energy production.