Permutation-entropy methods for bearing fault diagnosis in wind turbines are typically validated on laboratory test rigs without enforcing specimen-level separation in cross-validation or quantifying axis-specific diagnostic contributions. This study addresses these limitations using the Fraunhofer LBF operational wind turbine bearing dataset, applying multiscale permutation entropy (MPE) to triaxial front-bearing accelerometer signals across 44 bearing specimens (18 healthy, 10 inner race, 4 outer race, 12 roller element), with two corrupted files excluded following data quality screening. GroupKFold cross-validation with unique specimen-level group identifiers prevents the data leakage that arises when temporally correlated analysis windows are split without regard to bearing identity - a limitation present in all ten studies identified in a systematic literature search. MPE achieves 97.67% +/- 2.10% window-level and 97.73% specimen-level accuracy using 12 features across three accelerometer axes, outperforming weighted permutation entropy (WPE, 91.83% window-level, 95.45% specimen-level) and matching a physically-augmented hybrid (MPE+Physical, 21 features) that contributes no additional specimen-level accuracy despite 33.0% feature importance within the combined set. The X-axis accelerometer (brng.f.x) accounts for 40.7% of classification importance, consistent with the primary radial load direction. Permutation entropy reveals a monotonic complexity hierarchy across fault types - Healthy (mu=0.702) < Roller Element (mu=0.889) < Inner Race (mu=0.981) < Outer Race (mu=0.992) - with Cohen's d > 0.9 for all pairwise comparisons. Roller element faults exhibit bimodal permutation entropy distributions, explained by load-zone-dependent impulsive generation. These results demonstrate that multiscale ordinal pattern analysis captures physically meaningful fault signatures in operational wind turbine data when evaluated under methodologically sound cross-validation protocols.
This study develops a robust acoustic fault diagnosis framework for mining conveyor idlers, addressing the challenge of detecting early-stage mechanical degradation in noisy and imbalanced industrial environments. A dual-channel temporal--spectral representation is constructed by combining multi-scale log-Mel spectrograms and raw waveforms to capture complementary spectral patterns and fine-grained temporal dynamics. A Dual-Stream Cross-Attention Convolutional Recurrent Neural Network (DSCA-CRNN) is proposed to model cross-stream dependencies and enhance feature fusion. Noise-perturbation augmentation and a triplet-based contrastive objective are employed to enrich minority fault samples and improve embedding discriminability. Experiments on a self-collected conveyor auscultation dataset with 2,495 segments across three health states show that DSCA-CRNN achieves an overall accuracy of 0.95 and a macro-F1 score of 0.90, outperforming representative machine learning and deep learning baselines. Severe-fault recognition reaches an F1-score of 0.80. Ablation studies and PCA visualization confirm the effectiveness of recurrent temporal modeling, cross-attention fusion, noise-perturbation learning, and contrastive representation shaping. The proposed auscultation-based framework provides a practical and deployable solution for safety-oriented conveyor condition monitoring.
This study presents an applied comparative evaluation of automated rolling bearing fault identification using Artificial Neural Networks (ANNs) optimized through Bayesian Optimization (BO), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA). Vibration signals were collected from bearings operating under five health conditions (healthy, outer ring fault, inner ring fault, ball fault, and combined faults), at three rotational speeds, and along three measurement directions. The acquired signals were preprocessed using filtering, normalization, and segmentation. Time-domain and Fast Fourier Transform (FFT)-based frequency-domain features were extracted and used to train ANN models. The ANN architectures, including hidden layers, neurons, and activation functions, were optimized using BO, PSO, and GA, resulting in six configurations. Since the problem is a multi-class classification task, performance was assessed using F1-score, Accuracy, precision, and recall. The optimized ANN models were also benchmarked against Support Vector Machine (SVM), K-Nearest Neighbors (kNN), and Random Forest (RF) classifiers using the same feature sets. Results show that FFT -based features consistently outperformed time-domain features, and ANN-PSO with FFT-based features achieved the best performance, with F1-score = 0.982, Accuracy = 0.998, precision = 0.982, and recall = 0.982. This work contributes a systematic applied comparison of ANN optimization strategies rather than a fundamentally new machine-learning architecture, highlighting optimized ANNs as competitive and computationally efficient solutions for bearing fault diagnosis.
The modern industry faces the challenge of prolonging the lifespan of high-performance systems while maintaining their operational efficiency and cost-effectiveness. Through well-planned maintenance strategies, industries can achieve significant reductions in life cycle costs while minimising operational downtime and improving system reliability and its availability. Prognostics and Health Management (PHM) is an integrative, system-level engineering framework that combines diagnostics, prognostics, and decision-support processes to enable informed asset health management throughout the system lifecycle. While diagnostics and prognostics are core components of condition-based maintenance (CBM) and predictive maintenance (PdM), PHM emphasises their systematic integration within a closed-loop process that links operational activities with organizational and life cycle considerations. Thus, PHM does not replace existing maintenance policies; rather, it provides a unified framework that aligns health information, prognostic outputs, and maintenance decisions with resource constraints, risk, and performance objectives. Despite rapid advances in algorithmic research, a significant gap exists in the foundational conceptual understanding required by practitioners and researchers who are new to the field. During a literature review of machine and deep learning applications in the field of Predictive Maintenance (PdM) and estimation of the Remaining Useful Life (RUL), the authors observed that while extensive attention is given to algorithmic developments and application-specific studies, a consolidated perspective on PHM's fundamental role, historical evolution, and strategic implementation of PHM remains notably absent. This perspective article addresses this gap by providing a clear conceptual framework and practical roadmap of maintenance models, ranging from corrective and preventive to condition-based and predictive approaches, rather than introducing new models. This study articulates PHM’s core functional elements of PHM and positions it as an integrative, system-level paradigm that contextualises established maintenance strategies, such as CBM and PdM. In addition, it discusses the technological and industrial drivers that led to the emergence of PHM, providing an accessible entry point for researchers and practitioners.
This paper proposes a deep transfer learning approach for detecting brake fluid leakage in gondola wagons using acoustic signals. Gondola wagons, also known as railroad gondolas, gondola cars, and open wagons are typically used for transporting dry cargo and rely on pneumatic brake systems that depend on compressed air components for effective braking. The study investigates three transfer-learning strategies—From-Scratch, Partial Fine-Tuning, and Full Fine-Tuning—to identify compressed air leakage based on sound emissions, mirroring the process performed by human wagon inspection professionals. Training data consists of waveform audio signals captured during real railcar inspections. In the proposed model, each audio file is processed in the time-frequency domain to obtain a melspectrogram, which is then used as input to pre-trained deep convolutional neural networks. Results demonstrate strong performance, achieving accuracy above 94%. The main scientific contribution lies in applying established deep learning techniques to a specific and underexplored industrial railway context, together with the development and validation of a real-world dataset collected under authentic operating conditions. These findings demonstrate the feasibility and practical effectiveness of the proposed approach for pneumatic brake leakage detection in operational environments.
The anticipation of automotive failures in general and the prediction of engine performance losses remain challenging for vehicle owners and automotive industry professionals. In this article, we start by analyzing the causes of engine performance loss to identify the significant parameters of this failure mode. These parameters are then identified as inputs for the implementation of Adaptive Neuro-Fuzzy Inference System (ANFIS) neurofuzzy models optimized by a Particle Swarm Optimization (PSO) algorithm that takes into account the four previous instants to predict the next instant. The model was used to predict the performance loss characteristic failures of engine overheating, air leakage, engine power loss, air-to-air heat exchanger fouling, and filter clogging. The proposed model is an Explainable solution that better compromises performance and complexity. The performance of the ANFIS-PSO algorithm was evaluated by comparing test data with actual data. Satisfactory results were obtained, with R2 of the order of 0.99 for the test and training data, Root Mean Square Error (RMSE) of the order of 10-14, Standard Deviation of prediction Errors (Error St. D) of the order of 10-15 and Mean Absolute Error of the order of 10-15 for a prediction horizon of 1800s. This is with a Central Processing Unit (CPU) time of 0.002s. It is clear that the ANFIS-PSO model, which considers the four previous time instants, is sufficiently performant to predict the phenomena associated with the loss of engine performance.
Diesel generator engines (DGEs) are critical safety and mission assets for naval platforms, providing continuous electrical power for navigation, communications, habitability, and training operations. In practice, maintenance of auxiliary generation on training ships still relies mainly on time-based tasks and reactive corrective actions, despite the increasing availability of onboard operational data. This paper presents a residual anomaly detection system with persistence logic for the diesel generator engines of a Colombian Navy training ship. The proposed approach targets early detection of abnormal thermal behavior under scarce fault labels through a traceable workflow that combines: (i) data quality gates and preprocessing, including plausibility filtering and multivariate inconsistency treatment; (ii) target and feature definition for nominal regression modeling; (iii) residual monitoring with EWMA smoothing and time varying control limits; and (iv) persistence rules for sustained event declaration. The methodology is organized as an end-to-end workflow aligned with CRISP-DM and a PHM detection-first strategy. Since historical fault labels are limited and not reliably aligned in time, offline evaluation combines predictive assessment on held-out healthy data with Monte Carlo validation under simulated fault scenarios. Results from historical generator monitoring data show that the nominal models provide stable residual baselines for key thermal variables, while the detector can identify simulated abnormal scenarios across different severity levels. This paper provides a traceable workflow for anomaly detection in sparse and irregular shipboard monitoring data, discusses the limitations imposed by manual logs and scarce labels, and outlines future work toward health in- dexing, operational feedback, and more robust diagnostic support.
TP53, PIK3CA, and MUC16 are somatic mutations that are useful in breast cancer progression and prognosis, but direct mutation profiling based on sequencing is not always practicable in practice. The data about gene expression can contain indirect transcriptomic patterns linked with mutational underlying states. This paper proposes an expression-based machine learning model to predict the status of mutations using METABRIC breast cancer cohort. Instead of directly estimating genetic changes, the suggested method estimates statistical relationships between transcriptomic phenotypes and binary somatic mutation states. A multi-stage gene features selection pipeline using variance filtering, mutual information ranking, and correlation pruning was used to reduce the number of genes (19,000). A hybrid predictive architecture was trained using these features that combined ElasticNet logistic regression and XGBoost that allowed balancing between linear regularization and nonlinear interaction modeling. The hybrid model with a combination of five-fold stratified cross-validation yielded mean ROC-AUC of 0.94 (TP53), 0.92 (PIK3CA), and 0.90 (MUC16) with the stability of the calibration and equal error rates. Coefficient analysis and SHAP-based explanations were used to investigate the interpretability of the models to describe the expression patterns on mutation status. The suggested framework is a hypothesis-generating, complementary method of transcriptomic analysis, which must be reevaluated by external validation to determine the wider generalizability.
Marine diesel propulsion engines are essential to naval platforms, enabling maneuvering, navigation readiness, and training operations. However, maintenance of propulsion consumables-particularly fuel filtration elements-often remains time-based and corrective despite the growing availability of onboard operational records. This paper presents the development and validation of a Remaining Useful Life (RUL) prediction model for the propulsion engine filtration system of the Colombian Navy (ARC) training ship, aiming to estimate time to replacement for cartridge-based filters. The proposed approach handles imperfect manual operational data and scarce, non-uniform maintenance labels through a Prognostics and Health Management (PHM) workflow guided by the Cross-Industry Standard Process for Data Mining (CRISP DM). It combines physics-informed data quality control using plausibility bounds, outlier mitigation, and time-series reconstruction; expert validation of representative operating cycles using a Delphi protocol; and event logging to align filter-replacement actions with gap-aware approximations. It was trained supervised regression models using an automated machine learning (AutoML) strategy implemented in PyCaret and refined through hyperparameter optimization in Optuna. A Random Forest Regressor model achieved the best performance, reaching a test root mean squared error (RMSE) of 52.92 hours with a coefficient of determination (R2) of 0.921.
Remaining useful life (RUL) prediction has become a critical task, as accurate prediction enables effective maintenance planning and minimizes unplanned downtime. Complex systems frequently operate under diverse operating conditions, with distributional differences and limited RUL labels. To address these issues, domain adaptation (DA) methods have attracted increasing attention. Mainstream RUL prediction DA methods mainly adopt an unsupervised setting (UDA), relying solely on unlabeled target-domain data. However, prediction performances of UDA methods are far inferior to Target-Only (TO) models trained on full labeled target data, and it fails to effectively transfer degradation-related knowledge for RUL prediction when source-target distribution differences are large. Additionally, existing research mainly focuses on in-dataset adaptation across operating conditions, lacking systematic exploration of cross-dataset RUL prediction. Therefore, this work proposes a semi-supervised domain adaptation (SSDA) workflow for RUL prediction between different datasets. We design an alignment pipeline to address feature discrepancies between datasets and sampling frequency differences. During SSDA training, the original RUL labels of the source domain are treated as noisy labels with respect to the target domain and are progressively modified throughout the adaptation process. We conduct two case studies, one on turbofan engines and the other on rolling bearings. The experimental results demonstrate that the proposed approach can effectively realize cross-dataset RUL prediction.
Accurate remaining useful life (RUL) prognostics for turbofan engines are critical for advancing aviation safety and optimizing condition-based maintenance. While conventional methods rely on static service intervals or purely data-driven models, this study introduces a physics-aware hybrid framework that synergizes low-cycle fatigue (LCF) dynamics with a multi-architecture neural network. The framework addresses two fundamental limitations in existing approaches: (1) the omission of material fatigue mechanisms in purely data-driven models, and (2) the limited generalizability of simulated training data to real-world operational variability. The proposed model achieves superior degradation tracking by integrating fatigue cycle analytics, derived from engine operational stress profiles, with a hybrid neural architecture comprising Long Short-Term Memory (LSTM) networks for temporal dependencies, Temporal Convolutional Networks (TCN) for local feature extraction, and Multi-Head Self-Attention (MHSA) layers for dynamic weighting of critical sensor inputs. Comprehensive evaluation on both a 12-year real-world fleet dataset (53 F100-PW-229 engines) and the NASA C-MAPSS benchmark demonstrates significant improvements in predictive performance, with RMSE reductions of 29-48% compared to state-of-the-art hybrid models. The results underscore the potential of physics-aware deep learning to revolutionize predictive maintenance practices in aviation.
Ensuring the safe and stable operation of large-scale chemical processes requires accurate fault detection and diagnosis under nonlinear dynamics and strong variable interactions. This study investigates deep learning-based fault diagnosis for the Tennessee Eastman Process (TEP), focusing on whether performance improvements beyond near-saturation Long Short-Term Memory (LSTM) baselines can be achieved. A standardised TEP dataset with 52 measured and manipulated variables is used, excluding the non-detectable fault cases (IDV 3, 9, and 15), resulting in a 19-class classification problem. A hybrid Convolutional Neural Network-Transformer (CNN-Transformer) architecture is proposed in which onedimensional convolutional layers capture local cross-variable correlations, while a Transformer encoder models long-range temporal dependencies through self-attention. To ensure fair comparison, both the proposed model and a strong LSTM baseline are trained and evaluated under identical preprocessing, optimization, and evaluation protocols. The CNN-Transformer achieves an overall classification accuracy of 99.92%, marginally outperforming the LSTM baseline (99.86%). Although the numerical improvement is slight, the proposed model consistently yields higher macro-averaged F1-scores and reduced fault-wise misclassification, indicating enhanced robustness in challenging fault scenarios. The key contribution of this work is demonstrating that combining convolutional feature extraction with attention-based temporal modelling provides consistent class-level robustness beyond near-saturated recurrent architectures, while maintaining a compact structure suitable for practical deployment.
TP53, PIK3CA, and MUC16 are somatic mutations that are useful in breast cancer progression and prognosis, but direct mutation profiling based on sequencing is not always practicable in practice. The data about gene expression can contain indirect transcriptomic patterns linked with mutational underlying states. This paper proposes an expression-based machine learning model to predict the status of mutations using METABRIC breast cancer cohort. Instead of directly estimating genetic changes, the suggested method estimates statistical relationships between transcriptomic phenotypes and binary somatic mutation states. A multi-stage gene features selection pipeline using variance filtering, mutual information ranking, and correlation pruning was used to reduce the number of genes (19,000). A hybrid predictive architecture was trained using these features that combined ElasticNet logistic regression and XGBoost that allowed balancing between linear regularization and nonlinear interaction modeling. The hybrid model with a combination of five-fold stratified cross-validation yielded mean ROC-AUC of 0.94 (TP53), 0.92 (PIK3CA), and 0.90 (MUC16) with the stability of the calibration and equal error rates. Coefficient analysis and SHAP-based explanations were used to investigate the interpretability of the models to describe the expression patterns on mutation status. The suggested framework is a hypothesis-generating, complementary method of transcriptomic analysis, which must be reevaluated by external validation to determine the wider generalizability.
The development and implementation of diagnostic and prognostic algorithms for smart maintenance purposes is hindered by the fundamental lack of relevant, complete and properly labeled data from fielded systems. This issue is partly tackled by the generation of well-defined datasets using numerical simulations or experimental set-ups in a laboratory environment. However, the widely varying formats of (the description of) these datasets make that data scientists need to invest heavily in interpreting the data and transforming it to a format that fit the model requirements. To reduce that effort and ensure a robust and consistent processing, this work proposes a standardized way of documenting such datasets. This ISA-PHM standard is based on the existing ISA metadata standard originating from life and biomedical sciences, that has been translated to the prognostics and health management (PHM) context. This is achieved by firstly structuring and generalizing the information required to document both diagnostic and prognostic (numerical or physical) experiments. This information is then carefully mapped to the ISA ontology, ensuring a complete and unambiguous documentation of any PHM-related test. The concept is demonstrated by application of ISA-PHM to three well-known public datasets (NLN-EMP, NASA milling data, CMAPSS) and a real failure. Finally, for the implementation, some practical software tools are presented (available on linked website) as well as the planned future extension towards collective data generation through distributed testing.
Predictive maintenance (PdM) under severe class imbalance challenges model evaluation and deployment, especially when probabilities inform maintenance decisions. Using the AI4I 2020 dataset, this study establishes a reproducible baseline for failure detection with emphasis on rigorous validation and probability calibration. Models such as Random Forest (RF), Multilayer Perceptron (MLP), and classical baselines were evaluated via nested cross-validation with strict leakage control. Metrics included Average Precision, recall, precision, Brier score, and Expected Calibration Error (ECE), while the impact of SMOTE on class imbalance was analyzed. RF achieved the most robust balance between discrimination and calibration reliability, whereas MLP with SMOTE improved sensitivity but incurred calibration and false-positive trade-offs. Torque and tool wear emerged as dominant predictors, aligning with physical degradation mechanisms. By explicitly linking predictive performance to probability calibration and operational cost considerations, this work provides an actionable, cost-aware reference baseline for PdM within Total Productive Maintenance frameworks.
This paper proposes a deep transfer learning approach for detecting brake fluid leakage in gondola wagons using acoustic signals. Gondola wagons, also known as railroad gondolas, gondola cars, and open wagons are typically used for transporting dry cargo and rely on pneumatic brake systems that depend on compressed air components for effective braking. The study investigates three transfer-learning strategies-From-Scratch, Partial Fine-Tuning, and Full Fine-Tuning-to identify compressed air leakage based on sound emissions, mirroring the process performed by human wagon inspection professionals. Training data consists of waveform audio signals captured during real railcar inspections. In the proposed model, each audio file is processed in the time-frequency domain to obtain a melspectrogram, which is then used as input to pre-trained deep convolutional neural networks. Results demonstrate strong performance, achieving accuracy above 94%. The main scientific contribution lies in applying established deep learning techniques to a specific and underexplored industrial railway context, together with the development and validation of a real-world dataset collected under authentic operating conditions. These findings demonstrate the feasibility and practical effectiveness of the proposed approach for pneumatic brake leakage detection in operational environments.
In aerospace maintenance, remaining useful life (RUL) prediction is critical for flight safety, system availability, and long-term sustainment. While data-driven and machine learning (ML) approaches have improved RUL accuracy, most methods provide only point estimates and either omit uncertainty quantification (UQ) or rely on fixed, fleet-wide safety margins. Without reliable uncertainty estimates, even accurate point predictions offer limited value for safety-critical maintenance decisions. This paper presents a comprehensive framework for RUL prediction that jointly addresses point estimation, uncertainty quantification, and aerospace risk preferences. The framework combines a gradient boosting regressor (GBR) for point predictions with asymmetric conformalized quantile regression (CQR) to produce prediction intervals that communicate uncertainty. The asymmetric formulation of CQR allocates miscoverage unequally between interval bounds to reduce the likelihood of overly optimistic predictions, thereby aligning interval construction with the preference to avoid late maintenance intervention. The framework is evaluated on NASA's Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) benchmark dataset. Across all four benchmark subsets, the framework achieves test RMSE values of 13.26-16.85 with empirical coverage of 88-92% at 90% nominal coverage. These results demonstrate accurate point predictions and well-calibrated uncertainty intervals aligned with the requirements of safety-critical maintenance planning.
The modern industry faces the challenge of prolonging the lifespan of high-performance systems while maintaining their operational efficiency and cost-effectiveness. Through well-planned maintenance strategies, industries can achieve significant reductions in life cycle costs while minimising operational downtime and improving system reliability and its availability. Prognostics and Health Management (PHM) is an integrative, system-level engineering framework that combines diagnostics, prognostics, and decision-support processes to enable informed asset health management throughout the system lifecycle. While diagnostics and prognostics are core components of condition-based maintenance (CBM) and predictive maintenance (PdM), PHM emphasises their systematic integration within a closed-loop process that links operational activities with organizational and life cycle considerations. Thus, PHM does not replace existing maintenance policies; rather, it provides a unified framework that aligns health information, prognostic outputs, and maintenance decisions with resource constraints, risk, and performance objectives. Despite rapid advances in algorithmic research, a significant gap exists in the foundational conceptual understanding required by practitioners and researchers who are new to the field. During a literature review of machine and deep learning applications in the field of Predictive Maintenance (PdM) and estimation of the Remaining Useful Life (RUL), the authors observed that while extensive attention is given to algorithmic developments and application-specific studies, a consolidated perspective on PHM's fundamental role, historical evolution, and strategic implementation of PHM remains notably absent. This perspective article addresses this gap by providing a clear conceptual framework and practical roadmap of maintenance models, ranging from corrective and preventive to condition-based and predictive approaches, rather than introducing new models. This study articulates PHM's core functional elements of PHM and positions it as an integrative, system-level paradigm that contextualises established maintenance strategies, such as CBM and PdM. In addition, it discusses the technological and industrial drivers that led to the emergence of PHM, providing an accessible entry point for researchers and practitioners.
Anomaly detection in rotating machinery is essential for reliable industrial operations, yet building accurate detectors remains difficult when fault labels in a new domain are scarce. Although transfer anomaly detection has been increasingly studied, most methods do not explicitly exploit the characteristic fault-frequency structure-i.e., the fact that only specific orders/frequency components are strongly diagnostic of emerging faults. Here, we extend our prior key-order transfer framework. In this framework, a key order is a spectral feature-weight vector that upweights diagnostically informative orders and down-weights less relevant components when computing the anomaly score, and we adapt it to the realistic regime in which a small (but growing) number of labeled target anomalies becomes available over time. We estimate key orders in both source and target domains and fuse them using uncertainty-aware Bayesian combination as well as robust heuristic rules. Experiments on automotive transmission vibration data from two manufacturing sites show that adaptive fusion consistently outperforms source-only or target-only weighting in label-scarce settings. Overall, these results highlight the value of uncertainty-aware transfer for practical industrial anomaly detection under domain shift.
Prognostics and health management for aeroengines is crucial, especially for reducing catastrophic loss of life and minimizing maintenance costs. Accurate Remaining Useful Life (RUL) estimation optimizes schedules, lowers costs, and enhances safety. Physics-based modeling for RUL assessment is challenging due to its inability to fully account for system-and environment-related dynamics. Therefore, machine and deep learning approaches are highly recommended. This paper proposes a new method for RUL prediction of turbofan engines using the well-known (CMAPSS) dataset. Generally, data from acquired engines may be corrupted by anomalies due to sensor failures or environmental disturbances, which can affect the accuracy of prediction models. Therefore, this paper combines advanced techniques for reducing irrelevant and redundant sensor signals by applying Shapley Additive Explanations (SHAP) to refine feature selection by quantifying each sensor's contribution to RUL prediction, ensuring both interpretability and efficiency. Then an anomaly detection and removal followed by RUL prediction with deep learning are used for such complex tasks. Specifically, Kmeans clustering, autoencoders, Temporal Convolutional Networks (TCN) and Long Short-Term Memory (LSTM) networks. This approach enables effective data preprocessing by detecting and removing anomalies, thus enhancing the quality of the training data. Moreover, uncertainty analysis is performed to assess the reliability of the predictions, including confidence intervals for the results. The experimental results show substantial improvements in predictive accuracy, confirmed by better evaluation metrics, numerical evaluations, and the coefficient of determination, compared to traditional approaches. The TCN-LSTM model achieved an average performance improvement of 24.05% on FD001, 12.44% on FD003, and 4.63% on FD002 (using Monte-Carlo uncertainty estimation). Furthermore, for the FD004, a performance decrease of-4.68
