An official website of the United States government
Early fault detection in rolling element bearings is pivotal for the effective predictive maintenance of rotating machinery. Deep Learning (DL) methods have been widely studied for vibration-based bearing fault diagnostics largely because of their capability to automatically extract fault-related features from raw or processed vibration data. Although most DL models in the current literature can provide fairly accurate classification outputs, the typical diagnostic procedure is performed in an offline environment utilizing powerful computers. This centralized approach can lead to unacceptable delays in safety-critical applications and can prohibit cost-sensitive wireless data collection. Meanwhile, very few studies have reported on deploying DL models on microprocessor-based Industrial Internet of Things (IIoT) devices, where edge computing can give users a real-time evaluation of bearing health without requiring expensive computational infrastructure. This paper demonstrates an IIoT deployment of a physics-informed DL model inside a commercially available wireless vibration sensor for online health classification. The diagnostic model here is developed and trained offline, and the trained model is then deployed inside the embedded system for online prediction. We demonstrate the model’s online diagnostic performance by imitating bearing vibration signals on a vibration shaker and by performing edge computing on the embedded system mounted on the shaker.
more »
« less
This content will become publicly available on August 1, 2026
Simultaneous fault prediction in evolving industrial environments with ensembles of Hoeffding adaptive trees
Abstract Predictive Maintenance (PdM) emerges as a critical task of Industry 4.0, driving operational efficiency, minimizing downtime, and reducing maintenance costs. However, real-world industrial environments present unsolved challenges, especially in predicting simultaneous and correlated faults under evolving conditions. Traditional batch-based and deep learning approaches for simultaneous fault prediction often fall short due to their assumptions of static data distributions and high computational demands, making them unsuitable for dynamic, resource-constrained systems. In response, we propose OEMLHAT (Online Ensemble of Multi-Label Hoeffding Adaptive Trees), a novel model tailored for real-time, multi-label fault prediction in non-stationary industrial settings. OEMLHAT introduces a scalable online ensemble architecture that integrates online bagging, dynamic feature subspacing, and adaptive output weighting. This design allows it to efficiently handle concept drift, high-dimensional input spaces, and label sparsity, key bottlenecks in existing PdM solutions. Experimental results on three public multi-label PdM case studies demonstrate substantial improvements in predictive performance of OEMLHAT over previous batch-based and online proposals for multi-label classification, particularly with an average improvement in micro-averaged F1-score of 18.49% over the second most-accurate batch-based proposal and of 8.56% in the case of the second best online model. By addressing a critical gap in online multi-label learning for PdM, this work provides a robust and interpretable solution for next-generation industrial monitoring systems for fault detection, particularly for rare and concurrent failures.
more »
« less
- Award ID(s):
- 2316003
- PAR ID:
- 10645385
- Publisher / Repository:
- Springer
- Date Published:
- Journal Name:
- Applied Intelligence
- Volume:
- 55
- Issue:
- 13
- ISSN:
- 0924-669X
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Accurate prediction of repair durations is a challenge in product maintenance due to its implications for resource allocation, customer satisfaction, and operational performance. This study aims to develop a deep learning framework to help fleet repair shops accurately categorize repair time given product historical data. The study uses an automobile repair and maintenance dataset and creates an end-to-end predictive framework by employing a multi-head attention network designed for tabular data. The developed framework combines categorical information, transformed through embeddings and attention mechanisms, with numerical historical data to facilitate integration and learning from diverse data features. A weighted loss function is introduced to overcome class imbalance issues in large datasets. Moreover, an online learning strategy is used for continuous incremental model updates to maintain predictive accuracy in evolving operational environments. Our empirical findings demonstrate that the multi-head attention mechanism extracts meaningful interactions between vehicle identifiers and repair types compared to a feed-forward neural network. Also, combining historical maintenance data with an online learning strategy facilitates real-time adjustments to changing patterns and increases the model’s predictive performance on new data. The model is tested on real-world repair data spanning 2013 to 2020 and achieves an accuracy of 78%, with attention weight analyses illustrating feature interactions.more » « less
-
Abstract This paper proposes a novel adaptive maintenance policy for degrading systems subject to hard failure. Compared with traditional condition‐based maintenance policies, the proposed predictive maintenance policy makes maintenance decisions adaptively based on model prognostic results. The prognostic model is continuously updated based on newly inspected data. The inspection times and preventive maintenance activities are scheduled online in a sequential manner based on the most current prediction of system reliability. A computationally efficient optimization scheme is proposed for obtaining optimal maintenance parameters. The proposed policy is demonstrated and its performance is evaluated through extensive simulations.more » « less
-
We present a dynamic risk-based process design and multi-parametric model predictive control optimization approach for real-time process safety management in chemical process systems. A dynamic risk indicator is used to monitor process safety performance considering fault probability and severity, as an explicit function of safety–critical process variables deviation from nominal operating conditions. Process design-aware risk-based multi-parametric model predictive control strategies are then derived which offer the advantages to: (i) integrate safety–critical variable bounds as path constraints, (ii) control risk based on multivariate process dynamics under disturbances, and (iii) provide model-based risk propagation trend forecast. A dynamic optimization problem is then formulated, the solution of which can yield optimal risk control actions, process design values, and/or real-time operating set points. The potential and effectiveness of the proposed approach to systematically account for interactions and trade-offs of multiple decision layers toward improving process safety and efficiency are showcased in a real-world example, the safety–critical control of a continuous stirred tank reactor at T2 Laboratories.more » « less
-
Predictive maintenance for underground high-pressure fluid-filled (HPFF) power cables remains a critical challenge due to the weak and intermittent nature of fault-induced signals and the limited accessibility of buried infrastructure. This paper proposes a physics-informed Seq2Seq-attentionautoencoder acoustic monitoring (Echo-AE) model for predictive maintenance in underground HPFF cable systems. The Echo-AE model is developed based on a physics-informed loss function that incorporates both physics-based constraints and prediction errors. A controlled experimental setup of underground HPFF cable systems was used to capture continuous acoustic monitoring data, where three fault severity levels were generated, resulting in 4 million acoustic samples spanning normal operations and 15 fault events, and producing an imbalanced dataset with a 117:1 normal-tofault ratio to simulate real-world scenarios in which early-stage faults are rare. Results demonstrated Echo-AE’s superior early-stage fault detection capability compared with traditional models, with an F1-score of 0.8313, precision of 0.7864, recall of 0.8816, and an accuracy of 0.9936. The model exhibits fast convergence (20 epochs) and an area under the receiver operating characteristic curve of 0.998. Threshold sensitivity analysis revealed an optimal operation point that balances false positives and false negatives.more » « less
