🤖 AI Summary
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.
📝 Abstract
This study presents a practical approach for early fault detection in industrial pump systems using real-world sensor data from a large-scale vertical centrifugal pump operating in a demanding marine environment. Five key operational parameters were monitored: vibration, temperature, flow rate, pressure, and electrical current. A dual-threshold labeling method was applied, combining fixed engineering limits with adaptive thresholds calculated as the 95th percentile of historical sensor values. To address the rarity of documented failures, synthetic fault signals were injected into the data using domain-specific rules, simulating critical alerts within plausible operating ranges. Three machine learning classifiers - Random Forest, Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) - were trained to distinguish between normal operation, early warnings, and critical alerts. Results showed that Random Forest and XGBoost models achieved high accuracy across all classes, including minority cases representing rare or emerging faults, while the SVM model exhibited lower sensitivity to anomalies. Visual analyses, including grouped confusion matrices and time-series plots, indicated that the proposed hybrid method provides robust detection capabilities. The framework is scalable, interpretable, and suitable for real-time industrial deployment, supporting proactive maintenance decisions before failures occur. Furthermore, it can be adapted to other machinery with similar sensor architectures, highlighting its potential as a scalable solution for predictive maintenance in complex systems.