发表机构
Phuong Hai JSC; International College of Management Sydney; RMIT; FPT（芳海股份公司; 悉尼国际管理学院; RMIT大学; FPT集团）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究评估C-MAPSS上三种RUL模型归因方法的忠实性与稳定性，发现积分梯度与遮挡更可靠，Transformer注意力仅宜作可视化，为工程决策提供解释方法选择依据。
AI 中文摘要
在 NASA C-MAPSS 上，深度剩余使用寿命（RUL）模型现已常规使用，为传感器和时间步着色的热力图同样如此。一张看起来机械化的热力图并不等同于工程师可据此采取行动的解释。我们在官方 FD001 和 FD003 划分上训练了三种标准架构——一维 CNN、LSTM 和小型 Transformer 编码器，采用分段 RUL 上限 125 个周期和官方 PH08 非对称评分。随后，我们附加了三种归因图（积分梯度、遮挡、最后一层注意力），并用 XAI-for-PdM 文献仍鲜有报道的检查来评估它们：删除/插入忠实性、传感器尺度噪声下的 Spearman 稳定性、跨训练种子的一致性，以及 RUL 区间内的余弦一致性。预测误差是前提，而非主张。核心在于：当移除解释方法的顶部单元时，哪种方法会移动 RUL 输出；哪种图能在 5% 输入扰动下存活。积分梯度和遮挡在 LSTM 上同样忠实；Transformer 注意力廉价且时间平滑，但忠实性较弱。所有三种图在 5% 输入噪声下几乎不变，然而 IG/遮挡在两个 LSTM 种子间仅中度一致——对传感器抖动的稳定性不同于对重训练的稳定性。对 AI4I 2020 故障数据集的次要表格检查显示，树重要性存在相同的删除模式。我们建议在供维护工程师阅读的任何 C-MAPSS 风格报告中采用遮挡或积分梯度，并将原始注意力权重仅视为可视化工具。
英文摘要
Deep remaining-useful-life (RUL) models on NASA C-MAPSS are now routine, and so are heatmaps that colour sensors and timesteps. A heatmap that looks mechanical is not the same as an explanation an engineer can act on. We train three standard architectures - a 1D CNN, an LSTM, and a small Transformer encoder - on the official FD001 and FD003 splits with the piecewise RUL cap of 125 cycles and the official PHM08 asymmetric score. We then attach three attribution maps (Integrated Gradients, occlusion, last-layer attention) and evaluate them with the checks the XAI-for-PdM literature still under-reports: deletion/insertion faithfulness, Spearman stability under sensor-scale noise, agreement across training seeds, and cosine consistency inside RUL bins. Prediction error is a prerequisite, not the claim. The headline is which explanation method moves the RUL output when its top cells are removed, and which map survives a 5% input perturbation. Integrated Gradients and occlusion are similarly faithful on the LSTM; Transformer attention is cheap and temporally smooth but weakly faithful. All three maps are almost unchanged under 5% input noise, yet IG/occlusion agree only moderately across two LSTM seeds - stability to sensor jitter is not the same as stability to retraining. A secondary tabular check on the AI4I 2020 failure dataset shows the same deletion pattern for tree importances. We recommend occlusion or IG for any C-MAPSS-style report that will be read by a maintenance engineer, and we treat raw attention weights as a visualisation only.
Comments6 pages, 2 figures, 3 tables
