valve blockage/fault inference

This study addresses valve stiction—a critical cause of industrial process instability, equipment wear, and increased maintenance costs—by proposing a machine learning framework that enables real-time detection and up to four-hour-ahead prediction of stiction faults using only standard controller output (OP) and process variable (PV) signals, without requiring additional sensors. A novel data-driven labeling approach based on slope ratio analysis is introduced to train convolutional neural network (CNN), CNN–support vector machine (SVM), and long short-term memory (LSTM) models. Among these, the LSTM model demonstrates superior performance, achieving high prediction accuracy on real-world industrial data. This work represents the first successful implementation of early stiction prediction solely from routine process data, offering a practical foundation for predictive maintenance and significantly reducing unnecessary hardware replacements.

To address the scarcity of early-failure samples and insufficient detection robustness in industrial pump systems, this paper proposes an intelligent fault diagnosis framework integrating domain knowledge with data augmentation. We innovatively design a dual-threshold labeling strategy and inject physical constraints to generate high-fidelity synthetic fault signals, effectively mitigating the shortage of rare and emerging fault samples in real-world scenarios. Leveraging multivariate time-series sensor data—including vibration, temperature, flow rate, pressure, and current—we ensemble tree-based models (e.g., Random Forest and XGBoost), evaluated via confusion matrices and temporal visualization. Experimental results demonstrate significant accuracy improvements for minority fault classes, alongside high overall detection accuracy and strong generalization capability. The framework exhibits practical efficacy, robustness against data scarcity, and scalability to diverse industrial settings.

Low algorithmic trustworthiness and insufficient operator confidence hinder intelligent diagnosis of sudden anomalies (e.g., leaks, contamination) in water distribution networks. Method: This paper proposes an explainable event diagnosis framework centered on the novel concept of “counterfactual event fingerprints”—a first-of-its-kind construct that quantifies and visualizes the divergence between the current diagnostic output and its nearest alternative explanation, thereby revealing model decision rationale. The framework integrates counterfactual reasoning, fault diagnosis algorithms, and graphical explanation techniques. Contribution/Results: Evaluated on the L-Town benchmark and a real-scale water network, the framework demonstrates robust performance. Experiments show significant improvements in operators’ comprehension of algorithmic logic and their trust in automated outputs, enabling more accurate and reliable human–AI collaborative decision-making for anomaly response.

Industrial fault detection often lacks rigorous risk control and uncertainty quantification. Method: This paper proposes a novel framework integrating statistical significance testing with conformal prediction: model residuals are treated as nonconformity measures, transforming fault detection into a hypothesis test with formal risk guarantees. Significance testing is, for the first time, embedded within the conformal prediction framework, enabling users to specify a risk level α and guaranteeing nominal 1−α coverage while supporting controllable trade-offs between risk and efficiency. The method computes p-values and constructs prediction sets solely from calibration data—no model retraining is required. Results: Experiments demonstrate that the approach maintains stable coverage under distributional shift and exhibits strong robustness even when point prediction accuracy degrades.

Addressing data imbalance and label scarcity in intelligent condition monitoring of industrial equipment, this paper presents a systematic review of AI-based fault detection and diagnosis methods. Using the Tennessee Eastman Process (TEP) as a unified benchmark, it conducts the first comprehensive comparative evaluation—across accuracy, robustness, and generalizability—of representative ML/DL models including SVM, random forests, LSTM, autoencoders, GANs, and graph neural networks. We propose an integrated framework combining resampling, semi-supervised learning, and uncertainty quantification to mitigate data bias and annotation deficiency. Our contributions are threefold: (1) establishing a holistic review framework covering algorithm selection, performance evaluation, and uncertainty management; (2) providing reproducible, TEP-based comparative results across key metrics (e.g., detection rate, false alarm rate); and (3) delivering theoretical insights and practical guidelines for industrial intelligent maintenance research and deployment.

This study addresses the challenge of root-cause localization in automotive software testing, where high-dimensional sensor data generated during hardware-in-the-loop (HIL) simulations render traditional threshold-based methods ineffective. Existing data-driven approaches often require extensive labeled data and lack interpretability, failing to meet ISO 26262 traceability requirements. To overcome these limitations, the authors propose a two-stage diagnostic framework: first, safety requirements are automatically verified on a dSPACE real-time platform to filter anomalous test records; then, sliding windows of sensor signals are abstracted into statistical, relational, and contextual descriptors, which are fed as fixed prompts to an open-source large language model fine-tuned with 4-bit low-rank adaptation (LoRA). Evaluated on six fault types injected into a gasoline engine, the approach achieves 81.6% accuracy with a minimal 2B-parameter model—comparable to larger models—while operating entirely on a single consumer-grade GPU. This work pioneers the use of instruction-tuned large language models for sensor-level automotive fault diagnosis, demonstrating that diagnostic performance hinges more on task-specific adaptation convergence than on model scale, thereby achieving high accuracy, data efficiency, and explainable decision-making.

To address the challenge of real-time early warning for equipment failures and process disruptions in steel hot-rolling—leading to high unplanned downtime costs—this paper proposes a lightweight vision-sensor fusion analytical framework. The method deploys industrial cameras to continuously capture video streams of mill status and strip motion; an edge-based anomaly detection model reduces PLC computational load, while a central video server fuses visual features with multi-source process data (e.g., temperature, current, vibration) to enable early fault identification and generate interpretable maintenance recommendations. Its key innovation lies in embedding anomaly detection directly into the production control loop, ensuring cross-line scalability and operational adaptability. Field validation at a major steel plant demonstrates a 32.7% reduction in unplanned downtime and a 24.5% decrease in maintenance costs, significantly enhancing production line reliability and operational efficiency.

This study addresses the lack of open datasets in marine engine predictive maintenance that combine controlled fault experiments, well-defined operating conditions, and system-level measurements. The authors conducted bench tests on a turbocharged, intercooled three-cylinder marine diesel engine, covering normal operation across 30–90% load and five physical fault types: cooling water pump cavitation, air filter clogging, intercooler fouling, fuel injector clogging, and increased turbocharger exhaust backpressure. High-dimensional, multi-source time-series data were synchronously collected under these conditions. This dataset is the first to integrate controlled fault injection, multi-load operation, and comprehensive system-level sensing on a real marine platform. It demonstrates physical consistency between data and expected fault mechanisms, with distinct signatures across varying fault severities, thereby providing a high-quality, structured, and reusable open benchmark for anomaly detection and fault diagnosis research.