International Journal of Prognostics and Health Management

Vibration-based Data-driven Fault Diagnosis of Rotating Machines Operating Under Varying Working Conditions

David Latil — 2025-07-01

The intelligent fault diagnosis of rotating machines has been significantly advanced by learning-based techniques in recent years. However, the performance of these techniques can drastically decrease under varying working conditions (VWC). This paper investigates the root causes of these decreased capabilities by analyzing the impact of VWC on each of the key steps in intelligent fault diagnosis for rotating machines. In addition, techniques proposed in the literature to mitigate these effects are reviewed and assessed for their relevance in industrial applications. A bibliometric study is also conducted to understand the evolution of research in this field over the past two decades. Beyond providing a synthesis of the existing literature, this review is intended for researchers, engineers, and industry professionals seeking to implement robust fault diagnosis systems under varying operational conditions. It offers insights on when and how these techniques can be effectively applied, depending on specific industrial scenarios and assumptions.

LSTM and Transformers based methods for Remaining Useful Life Prediction considering Censored Data

Jean-Pierre Noot — 2025-07-04

Predictive maintenance deals with the timely replacement of industrial components relatively to their failure. It allows to prevent shutdowns as in reactive maintenance and reduces the costs compared to preventive maintenance. As a consequence, Remaining Useful Life (RUL) prediction of industrial components has become a key challenge for condition-based monitoring. In many applications, in particular those for which preventive maintenance is the general rule, the prediction problem is made harder by the rarity of failing instances. Indeed, the interruption of data acquisition before the occurrence of the event of interest leads to right censored data. Recent deep-learning architectures, that show the best results of the literature for complete-life data, most often do not consider censoring, even though its rate in the industrial environment may be high.
The present article introduces a method which considers censored data for the Dual Aspect Self-Attention based on a Transformer proposed by (Z. Zhang, Song, & Li, 2022), and puts it into competition a modified version of the ordinal regression-based LSTMof (Vishnu, Malhotra, Vig, & Shroff, 2019). The evaluation of the resulting method on the CMAPSS and N-CMAPSS benchmark dataset shows that it is competitive compared to the state-of-the-art RUL prediction methods for a low censoring rate and more efficient for a high rate of censoring in large enough data sets. Finally, conformal prediction is used to estimate confidence intervals for the predictions.

Analytical Health Indices: Towards Reliability-Informed Deep Learning for PHM

Dersin Pierre — 2025-07-26

Deep learning has demonstrated significant potential for prognostics in complex systems (Fink et al., 2020). Recent advances in physics-informed machine learning have integrated physics-of-failure principles within data-driven models (AriasChao, Kulkarni, Goebel, & Fink, 2022). Beyond physical laws, fleet-level time-to-failure (TTF) distributions provide valuable prior knowledge for individual asset life predictions.In this paper we derive a probabilistic analytical health index(HI) model based on power-law degradation, enabling a probabilistic description that reconciles individual variability with fleet-wide trends. We show that, under Weibull, Gamma, and Pareto-distributed TTFs, the HI evolution follows an analytical form, allowing explicit characterization of time to reach intermediate degradation levels. Therefore, this work provides a theoretical foundation for integrating reliability principles with deep learning, advancing towards Reliability-Informed Deep Learning (RIDL). The approach is validated on synthetic turbofan engine data and real-world battery degradation datasets. This work establishes a rigorous basis for embedding reliability engineering principles into deep learning, improving predictive maintenance and remaining useful life (RUL) estimation.

Innovative Distributed Maintenance Concept

Maria Di Mascolo — 2025-07-26

This study proposes an integrated heuristic framework for the strategic optimization of distributed maintenance operations in geo-distributed production systems (GDPS). It introduces a dual-entity maintenance structure comprising a Centralized Maintenance Workshop (CMW) and a Mobile Maintenance Workshop (MMW), aimed at minimizing total long-term maintenance costs. The cost function incorporates transport, operations, and downtime penalties, optimized via a two-stage algorithmic approach: a Maintenance Planning Algorithm (MPA) based on predictive maintenance scheduling, and a Long-term Heuristic Scheduling Algorithm (LHSA) addressing a capacitated vehicle routing problem with time windows (CVRPTW). A novel contribution includes a heuristic for CMW location determination using the weighted barycentre of site failure probabilities and a discrete selection of MMW capacities. Mixed Integer Linear Programming (MILP) and a divide-and-conquer heuristic are utilized to handle the NP-hard nature of the problem. Experimental validation using Weibull-distributed failure data and various cost scenarios demonstrates that the proposed Optimised Maintenance and Capacitated Routing (OMCR) framework can reduce lifecycle maintenance costs by up to 50%, with increased scalability for systems exceeding 30 GDPS. The framework is applicable to sectors requiring high availability and centralized servicing, including aerospace, railway, and energy industries.

Constraint-Guided Learning of Data-driven Health Indicator Models

Yonas Tefera — 2025-08-14

This paper presents a constraint-guided deep learning (DL) framework to develop physically consistent health indicators (HIs) in bearing prognostics and health management. Conventional data-driven approaches often lack physical plausibility, while physics-based models are limited by incomplete knowledge of complex systems. To address this, we integrate domain knowledge into DL models via constraints, ensuring monotonicity, bounding output ranges between 1 and 0 (representing healthy to failed states, respectively), and maintaining consistency between signal energy trends and HI estimates. Using constraints eliminates the need for complex loss term balancing to incorporate domain knowledge. The constraint-guided gradient descent algorithm (CGGD) is used to train a DL model that satisfies specific constraints. Using time-frequency representations of accelerometer signals from the pronostia and XJTU-SY bearing datasets, the model learned using constraints generates more accurate and reliable representations of bearing health compared to conventional methods. It produces smoother degradation profiles that align with the expected physical behavior. Model performance is assessed using three metrics: trendability, robustness, and consistency. When compared to a conventional baseline model, the model learned using constraints shows a significant improvement in all three metrics. Another baseline incorporated the monotonicity behavior directly into the loss function using a soft-ranking approach. While this approach outperforms the model learned using constraints in trendability, due to its explicit monotonicity enforcement, the model learned with constraints performed better in robustness and consistency, providing stable and interpretable HI estimates over time. The ablation study confirms the importance of each constraint: the monotonicity constraint improves trendability, the boundary constraint ensures consistency, and the energy–HI consistency constraint enhances robustness. These findings demonstrate the effectiveness of CGGD in producing reliable and physically meaningful HIs for bearing prognostics and health management, offering a promising direction for future prognostic applications.

A Comparative Analysis of Anomaly Detection Techniques for Battery Telemetry Data in Low Earth Orbit Remote Sensing Satellites

Ahmed Adam — 2025-09-04

The article presents a comprehensive assessment of telemetry data of batteries used in low-earth orbit satellites. The study further performs an analysis of the performance of using different anomaly detection techniques, including Statistical (Z-Score), Machine Learning (One class support vector machine OCSVM, Isolation Forest), Deep Learning (Autoencoder), and Hybrid Approaches (Autoencoder and neural network and Autoencoder and Z-score). This study introduces and evaluates a hybrid anomaly detection framework combining deep learning-based feature compression (Autoencoder) with various downstream classifiers. The models are validated on real satellite telemetry data and benchmarked using medical electrocardiogram ECG datasets for generalizability. In addition, the study continues to analyze the system by detecting the faulty sensor that was responsible for the detected anomalies, which can help the operators to get a more accurate analysis of the system.

Remaining Useful Life Prediction Using Attention-LSTM Neural Network of Aircraft Engines

Marouane Dida — 2025-07-04

Accurate prediction of the Remaining Useful Life (RUL) is essential for the effective implementation of Prognostics and Health Management (PHM) in aerospace, particularly in enhancing aero-engine reliability and forecasting potential failures to reduce maintenance costs and human-related risks.

The NASA Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset, utilized in the 2021 PHM Data Challenge, serves as a widely recognized open-source benchmark, providing simulated turbofan engine data collected under realistic flight conditions. Previous deep learning approaches have leveraged this dataset to predict the remaining useful life of engine units.

However, data-driven methods for RUL prediction in aerospace often encounter challenges such as high model complexity, limited prediction accuracy, and reduced interpretability. To address these issues, this paper presents a novel hybrid framework that incorporates an attention mechanism to enhance aircraft engine RUL prognostics. Specifically, we employ a self-attention mechanism to effectively capture relationships and interactions among different features, enabling the transformation of high-dimensional feature spaces into lower-dimensional representations.

The proposed model, which integrates an LSTM network, demonstrates superior performance in predicting turbofan engine RUL. Experimental results validate its effectiveness, achieving RMSE values of 12.33 and 11.76, along with score values of 200 and 212 on the FD001 and FD003 sub-datasets, respectively. These results surpass those of other state-of-the-art methods on the C-MAPSS dataset.

Towards Reliable RUL Prediction

Oluwasegun Oluwole Gbore — 2025-09-11

This study investigates feature selection techniques for predicting the Remaining Useful Life (RUL) of aircraft engines, addressing the persistent challenge of inaccurate predictions due to suboptimal feature selection. In this context, a robust methodology was developed to select optimal features for enhancing the model’s predictive power. Using inferential statistical methods for analysing operation data from aircraft engines, the study involved data pre-processing to test its feasibility, feature engineering to minimise data variability, backward elimination for linear regression, random forest and gradient boosting for effective feature selection. The models’ performance was evaluated for predictive accuracy and reliability using various performance metrics. Findings show that the random forest model with an R-squared value of 0.86 surpassed linear regression (0.76) and gradient boosting (0.73). It further highlighted that the integration of advanced feature selection techniques in non-linear modelling substantially improved the prediction accuracy of RUL while also capturing the essential degradation patterns typical in aircraft engines, as depicted in the Partial Dependence Plots (PDPs). All the three models highlighted the critical importance of the 'time' (current age) feature in predicting RUL, accounting for more than half of the model's predictive power. The findings of this work not only supported some initial hypotheses regarding sensor relationships and operational settings' effects but also unveiled complex interactions previously unrecognized. By identifying and eliminating redundant sensors though a systematic approach of feature selection, this study significantly contributes to the field of predictive maintenance for aircraft engines in enhancing the robustness of predictive models.

Correlation-Enhanced Multi-Scale Residual Network for Bearing Fault Diagnosis in Noisy and Cross-Working Conditions

Panfeng Bao — 2025-08-06

Bearing fault diagnosis under noisy and cross-working conditions remains a challenging task due to complex signal variations and interference. To address this challenge, this paper proposes a Correlation-Enhanced Multi-Scale Residual Network (CE-MSRN), which effectively captures multi-scale fault features while enhancing correlation across different bearing faults. Our model integrates a residual learning framework with a multi-scale feature fusion mechanism, improving robustness against noise and generalization across diverse working conditions. Experimental evaluations on benchmark datasets demonstrate that CE-MSRN achieves superior diagnostic accuracy compared to mainstream methods, exhibiting strong adaptability to unseen fault patterns. These results confirm the potential of our approach for real-time and reliable bearing fault diagnosis in aero-engines and transmission systems.

Fault-Type-Aware Remaining Useful Life Prediction of Aircraft Engines Using an Integrated Deep Learning Framework

Junwon Seo — 2025-09-16

Accurate Remaining Useful Life (RUL) prediction is essential for reducing maintenance costs and improving operational efficiency in high-value, complex systems such as aircraft engines. Data-driven approaches have emerged as a primary methodology in RUL estimation research, demonstrating significant improvements in performance. However, discrepancies in degradation trajectories across multiple failure modes can adversely affect the prediction accuracy. To address this challenge, this study proposes an integrated framework based on TS K-Means–BiLSTM to perform RUL prediction considering different failure modes. Specifically, Time Series K-Means Clustering (TS K-Means) is used to cluster time series data into latent failure-mode groups, and a Bidirectional Long Short-Term Memory (BiLSTM) network is subsequently employed to predict the RUL for each group. The proposed framework is validated using the Commercial Modular Aero-Propulsion System Simulation dataset provided by NASA. Experimental results show that the proposed model outperforms existing methods. In addition, it achieves better results than the comparison Bi-LSTM model trained under the same conditions but without fault-type separation. This improvement likely results from minimizing interference between degradation patterns, allowing the model to better distinguish the unique behaviors associated with each fault type. Consequently, the proposed approach demonstrates strong potential for practical RUL prediction tasks.

Exploring Proactive Maintenance through Fault Detection Techniques for Rotating Machinery

Ahiamadu Okirie — 2025-09-22

Rotating machinery plays a crucial role in industrial production, where reliability and efficiency are essential for minimizing downtime and operational losses. This systematic review explores proactive maintenance through advanced failure detection techniques, assessing their effectiveness in optimizing maintenance strategies. The study identifies key fault detection methods, including vibration analysis, electrical signature analysis, thermal imaging, acoustic emission monitoring, oil analysis, and IoT-enabled real-time monitoring. While vibration analysis remains the most widely researched and applied method, emerging AI-driven predictive maintenance models and IoT-based real-time diagnostics are increasingly transforming industrial maintenance practices. Findings indicate that proactive maintenance enhances equipment reliability, reduces downtime, and improves safety. However, significant challenges hinder widespread adoption, including high implementation costs, data management complexities, and the need for specialized expertise. Additionally, research gaps persist in comparative evaluations of fault detection methods, cost-benefit analyses, and the standardization of key performance metrics for proactive maintenance effectiveness. The study emphasizes the need for integrating multiple fault detection techniques to improve accuracy and reliability.