[论文] Robust Prototypical Networks for Few-Shot Sensor Fault Diagnosis

论文概要

研究领域: ML 作者: Mohammed Ayalew Belay, Amirshayan Haghipour, Pierluigi Salvo Rossi 发布时间: 2026-09-15 arXiv: 2609.12287

中文摘要

工业故障诊断通常只有少量标注故障样本，这使小样本学习对传感器监测颇具吸引力。标准原型网络简单有效，但在极低样本场景下，由于每次决策都依赖很小的支持集，其类原型可能变得不稳定。我们提出多回合原型网络（MEPN），聚合来自多个不相交支持回合的原型并取其均值作为最终类代表，在不改变编码器架构的前提下降低原型方差。我们在 DeFACTO 传感器数据集上评估 MEPN，任务为五路故障分类，向真实工业测量数据注入合成的偏置、漂移、尖峰与噪声故障。在 100 次独立运行中，MEPN 在逐回合单样本设置（K=1 样本，聚合 N_agg=10 个支持回合）下取得最佳准确率，显著高于单回合基线。在相等的 10 样本支持预算下，MEPN 与 K=10 的 ProtoNet 在统计上不可区分，证实其收益机制来自原型累积，而非固定预算下的更优学习。

原文摘要

Industrial fault diagnosis often operates with only a handful of labeled fault examples, making few-shot learning attractive for sensor monitoring. Standard prototypical networks are simple and effective; however, their class prototypes may become unstable in the very-low-shot regime because each decision relies on a small support set. We propose \emph{Multi-Episode Prototypical Networks} (MEPN), which aggregate prototypes from multiple disjoint support episodes and use their mean as the final class representative, reducing prototype variance without changing the encoder architecture. We evaluate MEPN on the DeFACTO sensor dataset using five-way fault classification with synthetic bias, drift, spike, and noise faults injected into real industrial measurements. Over 100 independent runs, ME...

#论文 #arXiv #ML #小凯