An Active Learning Strategy for Online Batch Fault Detection and Diagnosis

Online batch fault detection and diagnosis in modern manufacturing increasingly relies on machine learning detectors trained on large volumes of labeled data. However, labeling all samples collected by high-speed, high-dimensional sensors in real time is often impractical due to limited labeling resources. Besides, most collected samples correspond to a normal production environment, resulting in a severely imbalanced class distribution. Furthermore, evolving operating conditions give rise to novel, previously unseen fault types. These constraints pose challenges for applying conventional machine learning approaches for online FDD, which typically depend on complete and balanced data sets to effectively train the fault detector. To address these constraints, a bilevel active learning framework is proposed, where a batch of samples is strategically selected at each time step to update the fault detector in real time for online FDD with limited labeling resources. At the first level, a joint exploitation-exploration sampler is developed to select a batch of samples that are highly informative for the current fault detector and likely to reveal rare or unseen fault types while explicitly minimizing the intrabatch redundancy. At the second level, a minority-aware replay sampler is designed to enhance the robustness of the query set from the first level by replacing less informative samples with more informative ones from the historical data set while maintaining the size of the query set. Finally, the resulting data set from the bilevel active learning framework is used to update the fault detector incrementally by a continual learning method. The framework is tested on synthetic data sets and real-world cases, demonstrating its superior performance compared with state-of-the-art benchmarks. History: Bianca Maria Colosimo served as the senior editor for this article. Supplemental Material: The online appendix is available at https://doi.org/10.1287/ijds.2026.0172 .

Authors

Institutions

Publication Details

Journal
INFORMS Journal on Data Science
Published
2026-11-17
DOI
https://doi.org/10.1287/ijds.2026.0172
Primary Topic
Fault Detection and Control Systems
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An Active Learning Strategy for Online Batch Fault Detection and Diagnosis

Yue Jiang, Xiaochen Xian, Ana María Estrada Gómez
3 links
INFORMS Journal on Data Science
Fault Detection and Control Systems
article

An Active Learning Strategy for Online Batch Fault Detection and Diagnosis

Yue Jiang, Xiaochen Xian, Ana María Estrada Gómez
article en

Abstract

Online batch fault detection and diagnosis in modern manufacturing increasingly relies on machine learning detectors trained on large volumes of labeled data. However, labeling all samples collected by high-speed, high-dimensional sensors in real time is often impractical due to limited labeling resources. Besides, most collected samples correspond to a normal production environment, resulting in a severely imbalanced class distribution. Furthermore, evolving operating conditions give rise to novel, previously unseen fault types. These constraints pose challenges for applying conventional machine learning approaches for online FDD, which typically depend on complete and balanced data sets to effectively train the fault detector. To address these constraints, a bilevel active learning framework is proposed, where a batch of samples is strategically selected at each time step to update the fault detector in real time for online FDD with limited labeling resources. At the first level, a joint exploitation-exploration sampler is developed to select a batch of samples that are highly informative for the current fault detector and likely to reveal rare or unseen fault types while explicitly minimizing the intrabatch redundancy. At the second level, a minority-aware replay sampler is designed to enhance the robustness of the query set from the first level by replacing less informative samples with more informative ones from the historical data set while maintaining the size of the query set. Finally, the resulting data set from the bilevel active learning framework is used to update the fault detector incrementally by a continual learning method. The framework is tested on synthetic data sets and real-world cases, demonstrating its superior performance compared with state-of-the-art benchmarks. History: Bianca Maria Colosimo served as the senior editor for this article. Supplemental Material: The online appendix is available at https://doi.org/10.1287/ijds.2026.0172 .

INFORMS Journal on Data Science
Georgia Institute of Technology (US), Purdue University West Lafayette (US)
Fault Detection and Control Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon