Here in this paper, the vulnerabilities of deep neural networks (DNNs) to adversarial attacks within the context of time series classification (TSC) are explored. Expanding upon previous theoretical work in this field, a broader spectrum of models and augmentation techniques is applied to assess the performance and robustness of self-ensemble methods. Findings indicate that increasing the number of self-ensemble models significantly enhances the robustness of TSC systems. Augmentations such as Jitter and Smooth are particularly effective in reducing the Attack Success Rate (ASR), thus improving the security of DNNs against adversarial threats. Moreover, this study refines and modularizes previous methodologies, increasing the flexibility for future research to integrate and test various model architectures with different augmentation methods.