International Journal of Prognostics and Health Management

Efficiency Monitoring of a Cooling Water Pump based on Machine Learning Techniques

Marta Casero — 2025-01-22

This paper presents a method for efficiency monitoring of two circulating water pumps working in a combined cycle power plant for cooling the steam coming from a water-steam turbine. The method is based on monitoring the performance of the pumps over time using machine learning techniques that try to discover patterns in the data observed from the pumps. This permits the maintenance staff to assess the possible degradation of the pumps and evaluate the effect of the corrective and preventive maintenance implemented. Some examples of real cases will be presented in the paper to illustrate the method proposed.

Breast Cancer Detection Analysis Using Different Machine Learning Techniques: South Iraq Case Study

Salma Abdulbaki Mahmood — 2025-02-26

Contemporary oncology has seen a growing interest in digital technologies, whose integration with extensive healthcare and clinical data has raised new aspirations in managing patient profiles and organizing treatment plans. Among the commonly used digital technologies are Machine Learning (ML) methods that can perform many tasks, such as prediction, classification, and description, based on previously stored big data with high precision and speed. This study aims to develop a predictive ML model for early prediction of breast cancer based on a set of medically categorized risk factors. The locally collected database contained 415 instances from Al-Sadr Teaching Hospital in Basrah, Iraq, 219 (53%) of which were breast cancer patients, whereas 196 (47%) of them were control, respectively non-patients. It trained seven machine learning methods, namely Decision Tree (DT), Random Forest (RF), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Logical Regression (LR), Multinominal Naïve Bayes (NB), and Gaussian NB. The dataset was cleaned and balanced before being used. The results proved the superiority of the Decision Tree model with 96% accuracy, 96% sensitivity, and 96% specificity, the Random Forest model with 94% accuracy, 100% sensitivity, and 87% specificity, and SVM model with 92% accuracy, 96% sensitivity, and 87% specificity, respectively. Other models gave diverging results. The current study concluded that modern technologies should be employed to raise awareness and control diseases. The need to adopt Electronic Health Records (EHR) to ensure the integration of clinical data of different types recorded over time for patients contributes to building accurate and reliable prediction models.

Uncertainty Assessment Framework for IGBT Lifetime Models. A Case Study of Solder-Free Modules

Ander Zubizarreta — 2024-12-30

Insulated gate bipolar transistors (IGBTs) are ubiquitous semiconductor devices used in diverse electronic power applications. The reliability and lifetime assessment of IGBTs is intricate and influenced by different ageing processes. One of the main ageing mechanisms is the bond wire lift-off failure mode. The model used to describe this failure mode and estimate the IGBT lifetime is influenced by different variables and factors, which are stochastic, and tend to be specifically adjusted for different IGBT modules and applications. However, unless these variables are not assessed with respect to potential sources of uncertainty, the IGBT lifetime estimate leads to a single-value deterministic estimate, which, frequently, results inaccurate. In this context, assessing the influence of the variability of these variables on the lifetime model is a crucial activity for an uncertainty-aware IGBT lifetime estimate and adoption of appropriate sensing technology. Accordingly, this paper presents a methodology to evaluate the impact of the uncertainty of IGBT lifetime parameters on the lifetime estimate. The approach is first validated on three different experimental IGBT operation profiles, demonstrating the impact of variations of certain variables on the damage estimation. The approach has been tested here for a single lifetime model, but it is generally applicable to other IGBT lifetime models.

Adaptive Res-LSTM Attention-based Remaining Useful Lifetime Prognosis of Rolling Bearings

Boubker Najdi — 2024-12-30

Predicting the Remaining Useful Lifetime (RUL) of bearings is crucial for the maintenance and reliability of rotating machinery. This paper presents a novel approach utilizing PRONOSTIA and XJTU-SY datasets for RUL prediction. The proposed methodology leverages Synchrosqueezing Wavelet Transform (SSWT) and Random Projection (RP) to extract significant features from vibration signals. These features are then fed into a Residual Network (ResNet) combined with a temporal attention layer, followed by a Long Short-Term Memory (LSTM) model, referred to as the Adaptive Residual Attention LSTM (ARAL), to assess the Health Indicator (HI) of the bearings. Notably, an exponential data labeling technique is employed instead of traditional linear labeling, enhancing the robustness of the HI assessment. Following the HI estimation, the three-sigma method is applied to identify the degradation starting point. Subsequently, Gaussian Process Regression (GPR) is utilized to predict the RUL from this point forward. The proposed method demonstrates superior performance compared to existing techniques, providing more accurate and reliable RUL predictions. Experimental results show that this integrated approach effectively captures the complex degradation patterns of bearings, making it a valuable tool for prognostics and health management in industrial applications.

Robust Kalman Filter with Recursive Measurement Noise Covariance Estimation Against Measurement Faults

Chingiz Hajiyev — 2025-01-25

A new innovation-based recursive measurement noise covariance estimation method is proposed. The presented algorithm is used for Kalman filter tuning, as a result, the robust Kalman filter (RKF) against measurement malfunctions is derived. The proposed innovation-based RKF with recursive estimation of measurement noise covariance is applied for the model of Unmanned Aerial Vehicle (UAV) dynamics. Algorithms are examined for two types of measurement fault scenarios; constant bias at measurements (additive sensor faults) and measurement noise increments (multiplicative sensor faults). The simulation results show that the proposed RKF can accurately estimate UAV dynamics in real time in the presence of various types of sensor faults. Estimation accuracies of the proposed RKF and conventional KF are investigated and compared. In all investigated sensor fault sceneries, the Root Mean Square (RMS) errors of the proposed RKF estimates are lower. The conventional KF gives inaccurate estimation results in the presence of sensor faults.

Pump Health Monitoring Test Environment for Diagnosing the Erosive Effects from Cavitation

Tedja Verhulst — 2025-05-19

Cavitation occurs frequently in pumps. The subsequent erosion that is caused by cavitation can significantly reduce the operational efficiency and Remaining Useful Life (RUL) of the pump. This study describes a new hybrid Health Monitoring (HM) test environment, used to diagnose permanent damage caused by cavitation erosion in a two-stage centrifugal pump. Flowrate, pressure and motor current measurements are made and compared to Computational Fluid Dynamic (CFD) results. The hybrid-based HM carried out using the three methods provide the facility to develop diagnostics for cavitation erosion damage. The suggested methods will not only aid in HM development, but also select the best operating conditions to carry it out. The Gray Level Method (GLM) is implemented using CFD to predict the erosion areas in the centrifugal pump. A Simscape model is devised to enable development of health monitoring algorithms. Few works have attempted to detect for erosion caused by cavitation. It was found that a high-level agreement was achieved between the Simscape, CFD and test-rig results, with an average error of 0.8%, 2.5%, and 2.0% for current, pressure and flow measurements respectively. The results from this research show the feasibility of developing HM algorithms to detect cavitation erosion in aircraft fuel pumps by fusing model and data-based methods. This is an enabler for a move from time-based to condition-based maintenance, thus reducing aircraft operating costs.

Proficiency of Physics Informed Machine Learning in Multi-component Fault Recognition of Rotational Machines under Different Speed Conditions

S Sowmya — 2025-05-16

Understanding the limitations of incorporating conventional machine learning synergy led to the inclusion of physics knowledge. This study presents the potency of physics-informed feature engineering for machine learning to enhance fault detection in gears, shafts, and bearings at three constant-speed running conditions. AI models such as Decision Tree (DT), Random Forest (RF), and Support Vector Machine-Radial Basis Function (SVM-RBF) are constructed to verify traditional statistical performance metric and physics-based signal descriptors. Additionally, time-frequency domain representation as spectrogram images is fed into the CNN-oriented ResNet-152 architecture to demonstrate the skillfulness of the model’s ability. Based on the results obtained, RF is observed to be supreme with 98.42% upon applying physics-centric parameters when compared with statistical variables. To make an inference, further comparison of the best classification model’s accuracy using physics expertise when accounted with ResNet image-based categorization, physics-grounded RF models have premier achievements. Thus, it is concluded that physical laws are expedient in offering exceptional outcomes for identifying various defects in complex industrial rotary machines in different operating modes.

Diagnostics and Prognostics of Boilers in Power Plant Based on Data-Driven and Machine Learning

Achmad Widodo — 2025-01-31

This paper reports diagnostics and prognostics study of boiler in power plant using actual boiler operating data. This study aims to early detect anomalies that occur in the boiler and to predict the remaining useful life (RUL) after anomalies are detected. The proposed method utilizes machine learning techniques through support vector machine (SVM) and random forest algorithm (RFA) for anomaly detection and similarity-based method of dynamic time warping (DTW) for RUL prediction. The developed method is validated by testing the prediction models using real operating data acquired from three boilers in power plant. The results show that some anomalies are successfully detected by prediction model even though there are anomalies that give low accuracies in predictions. RUL prediction also provides fair results given the limitations of the real data used in building prediction models. Overall, the results of this study have potential to be applied in real system as an auxiliary tool in the boiler condition monitoring to support boiler maintenance programs.

Comparative Analysis of LSTM Variants for Fault Detection and Classification in Aircraft Control Surfaces

Muhammad Fajar — 2025-05-09

Aircraft control surfaces play a critical role in ensuring safe and efficient flight. Faults in these surfaces could lead to catastrophic consequences. This paper investigates the application of Long Short-Term Memory (LSTM) networks for fault detection and classification in aircraft control surfaces. Four deep learning models: LSTM, Stacked-LSTM, Bi-LSTM, and Attention-based LSTM (ALSTM), were trained, validated, and tested to classify faults based on residual features. The methodology involved data generation, preprocessing, normalization, and training the models over 200 epochs. Evaluation metrics, including confusion matrices, precision, recall, and F1-scores, were used to assess model performance. Results show that Bi-LSTM achieved the highest accuracy (98.93%) and lowest loss (0.0264), significantly outperforming other models in fault detection, particularly for challenging fault types such as hard-over and lock-in-place. ALSTM followed closely, with notable performance improvements over standard and stacked LSTM models.

Dynamic Relationship Between Oil Temperature and BGCI in Bell 407 Helicopter

lotfi Saidi — 2025-05-10

The primary objective of this study was to investigate the dynamic relationship between oil temperature and the Bearing Gearbox Condition Indicator (BGCI) values of the Bell 407 helicopter. The study aims to simplify the fault diagnosis process by proposing a method that utilizes only one vibration sensor and one temperature sensor per bearing. To achieve this goal, we employ robust econometric tools, such as unit root tests, cointegration tests, and Autoregressive Distributed Lag (ARDL) models, for both long-run and short-run estimates. Our findings indicate that the variable temperature tends to converge to its long-run equilibrium path in response to changes in other variables. The results of the ARDL analysis confirmed that spectral kurtosis, inner race, cage, and ball energy significantly contributed to the increase in temperature. Furthermore, we utilized the Impulse Response Function (IRF) to trace the dynamic response paths of the shocks to the system. The identification of a cointegrating relationship between oil temperature and BGCI values suggests a practical and significant connection that can potentially be used to predict hazardous changes in oil temperature using BGCI values, which is an important implication for enhancing the safety and reliability of helicopter operations.

The study presents a promising direction for condition monitoring (CM) in rotating machinery, emphasizing the potential of integrating temperature data to simplify the diagnostic process while still achieving reliable results.

Adaptive Wavelet-Based Physics-Informed CNN for Bearing Fault Diagnosis

Reza Hassannejad — 2025-04-05

With the increasing expansion of data science into various fields, the application of deep neural networks in the fault diagnosis of rotating machines has attracted significant attention from researchers. However, in the methods available in the literature, the physical characteristics of the problem are not incorporated into the structure of deep networks. In most existing methods, fault diagnosis is performed solely based on features extracted by convolutional layers, with no additional layers utilized to enhance or refine these features. This work introduces a novel physics-based neural network for bearing fault diagnosis, in which specific layers are designed based on signal processing methods to extract the physical features of faults. These layers, referred to as physics-based layers, are constructed using adaptive analytical wavelet filterbanks. The features extracted by these layers are then classified using convolutional layers, enabling the diagnosis of bearing faults. A key advantage of this physics-based network is that it does not rely on a fixed architecture for feature extraction and classification. Instead, the characteristics of the network layers adapt to the fault characteristics present in the bearing vibration signals. The classification accuracy of the proposed method has been evaluated using experimental data from two studied cases. The results demonstrate that the newly introduced network achieves higher accuracy in classifying bearing signals with different faults compared to similar methods.

Integrating Machine Learning-Based Remaining Useful Life Predictions with Cost-Optimal Block Replacement for Industrial Maintenance

YOUNGSUK CHOO — 2025-04-21

This study presents a preventive maintenance methodology to predict the remaining useful life (RUL) of mechanical systems and determine cost-effective replacement schedules. The approach integrates machine learning for RUL prediction, Weibull distribution for reliability analysis, and a block replacement model with minimal repair to optimize preventive maintenance. Many existing studies rarely incorporate RUL prediction results into determining optimal maintenance actions due to the high uncertainty in RUL prediction. To address this, the proposed methodology emphasizes not stopping at the prediction stage but integrating RUL predictions into actionable maintenance strategies to reduce uncertainty and improve applicability in industrial contexts. A case study using the open CMAPSS dataset demonstrates the feasibility of the approach. The value of this study lies in proposing a methodology that not only utilizes prediction-based proactive outcomes instead of predefined replacement intervals but also integrates them with subsequent maintenance strategies, providing practical and cost-effective solutions for industrial applications.

A Comparative Study of Deep Learning Model Based Equipment Fault Diagnosis and Prognosis

Xianpeng Qiao — 2025-05-18

Bearing fault diagnosis and prognosis are crucial for the effective management of industrial equipment. Due to the automatic feature extraction of Deep Learning (DL) models, many recent studies have focused on using DL for these tasks. However, most studies address only one of these tasks. This study aims to present DL models and their powerful ML tools capable of both fault diagnosis and prognosis on industrial equipment. To identify the best DL model for both tasks, a comparative study is conducted on various DL models and ML tools, including Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), CNN in parallel with LSTM (CNN-LSTM), Bidirectional LSTM (Bi-LSTM), and transformer models. The ML tools investigated include Recurrent Dropout, Residual Network (ResNet), and Monte Carlo Dropout (MC Dropout). These models are validated using online datasets from Case Western Reserve University (CWRU) and Xi’an Jiao Tong University (XJTU-SY) for the task of fault diagnosis. For fault prognosis, datasets from XJTU-SY and IEEE PHM are used. The results demonstrate the superiority of the ResCNN-LSTM model in both fault diagnosis and prognosis tasks. It achieves prediction accuracy of 99.87% and 96.39% and F1-scores of 0.998 and 0.964 for fault diagnosis on the CWRU and XJTU-SY datasets, respectively. Additionally, it shows a Root Mean Square Error (RMSE) of 8.56 and Mean Absolute Error (MAE) of 12.16 for fault prognosis on the XJTU dataset, and an RMSE of 12.18 using the IEEE PHM bearing dataset. These high performance metrics indicate the model's effectiveness in accurately diagnosing faults and predicting failures.

Liner Wear Prediction Using Bayesian Regression Models and Clustering

Jacob Van Den Broek — 2025-03-31

Chutes, bins, and hoppers are critical assets in bulk commodity handling. Sacrificial wear liners are employed to protect these assets from abrasive wear. An essential maintenance challenge is optimising the timing of liner replacements. Traditionally, episodic human inspections have been in place, but now, real-time wireless IoT sensing systems that measure liner thickness are being used. We propose a novel approach to estimate the remaining useful chute liner life. Instead of linear extrapolation based on individual sensor wear rates (commonly used in industry), we leverage a Clustered Bayesian Hierarchical Modeling (BHM). Two models are developed: Model 1 (Cluster Exemplar) uses parameters from the closest cluster exemplar, while Model 2 (Spatial and Temporal BHM) incorporates data from the active sensor, with prior distribution informed by Model 1. Data are drawn from a single hopper with 88 sensors, 20 of which reached their end-of-life threshold. Both Model 1 and Model 2 outperform the industry regression approach, significantly reducing over-prediction. Notably, Model 2 predicts remaining useful life within 95% credible intervals and identifies anomalous sensor performance. This innovative use of historical and adjacent sensor data enhances wear degradation prediction, contributing valuable insights to the literature.