文章背景与核心概要
工业传感器诊断通常依赖于复杂的预处理、表征和分类流水线,这使得自动化流水线搜索(AutoML/AutoDL)在减少人工设计成本方面发挥了重要作用。然而,现有的自动化机器学习/深度学习报告通常只记录成功拟合的试验和获胜模型,而忽略了无效、被剪枝、被跳过、被缓存或未拟合的候选方案,导致审计信号约束、预算利用率和未评估替代方案变得十分困难。
为了解决这一局限性,本文作者引入了候选状态记账法(candidate-fate accounting),这是一个专为诊断搜索轨迹设计的候选级审计框架。该框架将每个观测到的候选方案记录为可审计的证据,具体包括:利用哈希值合并重复观测、通过合法性检查标记无效候选、通过分配理由解释预算决策,以及通过封闭状态分类账为每个候选分配单一的终极状态。在三个轴承诊断数据集上的实验表明,该框架能够成功检测无效候选,找回传统仅记录拟合试验的报告所遗漏的 30–41 个候选,并在保持具有竞争力的诊断性能的同时,验证了完整的候选审计过程。
候选状态记账法:实现透明的传感器诊断流水线搜索
arXiv ID: arXiv:2608.18665 [cs.AI]
提交时间: 2026年8月19日
作者: Haotao Xie, Yutian Chen, Yangqi Liu, Xiaoyu Jiang
代码仓库: GitHub - candidate-fate-accounting
📌 摘要
Automated machine and deep learning (AutoML/AutoDL) for industrial sensor diagnostics typically records only successful, fitted trials and winning models, omitting invalid, pruned, skipped, cached, or unfitted candidates. This lack of transparency makes it difficult to audit signal constraints, budget utilization, and unevaluated alternatives.
To resolve this limitation, the authors introduce candidate-fate accounting, a candidate-level audit framework designed for diagnostic search traces. This framework records every observed candidate as auditable evidence using: * Hashes to merge repeated observations, * Legality checks to flag invalid candidates, * Allocation rationales to explain budget decisions, and * A closed fate ledger that assigns a single terminal fate to each candidate.
Experiments across three bearing-diagnostic datasets demonstrate that the framework successfully detects invalid candidates, uncovers 30–41 candidates previously omitted by traditional fitted-trial-only reports, and verifies complete candidate accounting while preserving competitive diagnostic performance.
工业传感器诊断依赖于预处理、表征和分类流水线,这使得自动化流水线搜索能够有效降低人工设计成本。然而,现有的自动化机器学习/深度学习(AutoML/AutoDL)报告通常仅保留拟合试验、评分和获胜模型,而忽略了无效、被剪枝、被跳过、被缓存或未拟合的生成候选。这种缺失限制了评审人员检查信号约束、预算使用情况以及未评估的合法替代方案的能力。为了解决这一问题,我们提出了候选状态记账法,这是一个针对诊断搜索轨迹的候选级审计框架。它将每个观测到的候选记录为可审计的证据:通过哈希值合并重复观测,通过合法性检查标记无效候选,通过分配理由解释预算决策,并通过封闭状态分类账为每个候选分配一个终极状态。在三个轴承诊断数据集上的实验表明,该框架能够检测无效候选,并识别出传统仅记录拟合试验的报告所遗漏的 30–41 个候选,封闭的状态记录验证了完整的候选审计,同时维持了具有竞争力的诊断性能。
📑 摘要原文
Industrial sensor diagnostics relies on preprocessing, representation, and classification pipelines, making automated pipeline search useful for reducing manual design cost. However, existing automated machine/deep learning (AutoML/AutoDL) reports typically retain only fitted trials, scores, and winners, omitting generated candidates that are invalid, pruned, skipped, cached, or unfitted. This omission limits reviewers' ability to check signal constraints, budget use, and unevaluated legal alternatives. To address this, we propose candidate-fate accounting, a candidate-level audit framework for diagnostic search traces. It records each observed candidate as auditable evidence: hashes merge repeated observations, legality checks flag invalid candidates, allocation rationales explain budget decisions, and a closed fate ledger assigns one terminal fate to each candidate. Experiments on three bearing-diagnostic datasets show that the framework detects invalid candidates and identifies 30--41 candidates omitted by fitted-trial-only reports, with closed fate records verifying complete candidate accounting while maintaining competitive diagnostic performance.
工业传感器诊断依赖于预处理、表征和分类流水线,这使得自动化流水线搜索有用於降低人工设计成本。然而,现有的自动化机器学习/深度学习(AutoML/AutoDL)报告通常仅保留拟合试验、评分和获胜模型,而忽略了无效、被剪枝、被跳过、被缓存或未拟合的生成候选。这种遗漏限制了评审人员检查信号约束、预算使用情况以及未评估的合法替代方案的能力。为了解决这一问题,我们提出了候选状态记账法,这是一个针对诊断搜索轨迹的候选级审计框架。它将每个观测到的候选记录为可审计的证据:哈希值合并重复的观测,合法性检查标记无效候选,分配理由解释预算决策,封闭的状态分类账为每个候选分配一个终极状态。在三个轴承诊断数据集上的实验表明,该框架检测到无效候选,并识别出被仅记录拟合试验的报告所遗漏的 30--41 个候选,封闭状态记录验证了完整的候选记账,同时维持了具有竞争力的诊断性能。
🔗 资源链接
- 全文格式: 查看 PDF | HTML (实验性) | TeX 源码
- 数字对象唯一标识符 (DOI): 10.48550/arXiv.2608.18665
- 引用与指标: Google Scholar | Semantic Scholar | NASA ADS