A hybrid predictive maintenance framework that combines Temporal Convolutional Networks for temporal feature extraction with Genetic Algorithms to select optimal features and Variational Autoencoders for detecting anomalies and Transformer-based models to classify failures and predict RUL is proposed.
Abstract
Predictive maintenance has become an important strategy in the aviation industry for improving aircraft reliability, safety, and operational efficiency. With the increasing availability of sensor data from aircraft engines, Machine Learning (ML) techniques can be used to detect degradation patterns and predict failures before they occur. However, many previous studies focus on isolated tasks such as fault detection or Remaining Useful Life (RUL) prediction and often rely on complex Deep Learning (DL) models without efficient feature selection, leading to high computational cost and limited interpretability. The study proposes a hybrid predictive maintenance framework that combines Temporal Convolutional Networks (TCN) for temporal feature extraction with Genetic Algorithms (GA) to select optimal features and Variational Autoencoders (VAE) for detecting anomalies and Transformer-based models to classify failures and predict RUL. The model is evaluated using the NASA CMAPSS turbofan engine dataset. Experimental results indicate that the proposed method is able to achieve a high classification accuracy of 98.71% and reduce costs associated with maintenance by 46.25%, compared to traditional methods of maintaining an aircraft. The results of this study indicate that the proposed hybrid method has the potential to improve maintenance-related decision-making, improve aircraft reliability, and reduce operational cost. Additionally, these results show that there are a significant number of opportunities to deploy intelligent predictive maintenance systems to aviation operations across the globe.
To improve the operational readiness of combat aircraft engines and reduce unplanned maintenance costs, accurately estimating the remaining useful life (RUL) is critical. Traditional maintenance often proves insufficient under dynamic mission profiles. In this study, a deep learning-based predictive maintenance model capable of autonomously extracting features from multivariate sensor data was developed. Using the NASA C-MAPSS FD001 and FD004 datasets, data were converted into sequential blocks via 50- and 30-step sliding windows, respectively. The model's architectural superiority in autonomously extracting temporal degradation features was validated against RF, CNN-LSTM, and BiLSTM baselines. On FD001, it achieved an R-squared (R2) of 0.8901, a 13.28 RMSE, and a 320.34 NASA risk score, demonstrating generalizability on the multi-regime FD004 dataset with a 15.71 RMSE. The proposed maintenance protocol achieved a 0.9973 AUC at the critical 30-cycle threshold, ensuring high reliability. Additionally, a decision-support simulator has been developed to validate this protocol under aggressive combat flight profiles.
A structured methodology for ML-based PdM frameworks, covering data-driven, physics-based, and hybrid approaches, including supervised, unsupervised, and deep learning models is proposed, offering valuable insights for developing efficient and scalable PdM solutions.
Sithik Shah· International Journal of App...· 0 citations
Predictive maintenance has become a critical component of Industry 4.0 by enabling intelligent mechanical systems to minimize unexpected equipment failures, reduce maintenance costs, and improve operational reliability. This study proposes a machine learning-based predictive maintenance framework for machine failure prediction and fault diagnosis using the AI4I 2020 Predictive Maintenance Dataset. The dataset comprises 10,000 machine observations with operational parameters, including air temperature, process temperature, rotational speed, torque, tool wear, and machine type. A comprehensive comparative analysis was conducted using Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, and Extreme Gradient Boosting (XGBoost). To address the highly imbalanced failure distribution, multiple evaluation metrics, including Accuracy, Precision, Recall, F1-score, Receiver Operating Characteristic–Area Under the Curve (ROC–AUC), and Matthews Correlation Coefficient (MCC), were employed. The experimental results demonstrated that XGBoost achieved the best predictive performance, obtaining an accuracy of 98.95%, an F1-score of 82.93%, a ROC–AUC of 98.00%, and an MCC of 0.8289, outperforming the remaining machine learning models. Furthermore, SHapley Additive exPlanations (SHAP) identified torque, torque–speed ratio, and tool wear as the most influential predictors of machine failure, providing transparent model interpretation and supporting informed maintenance decision-making. The proposed framework offers an accurate and interpretable solution for predictive maintenance, contributing to enhanced equipment reliability, reduced downtime, and intelligent maintenance strategies in modern manufacturing environments.
Abhishek Sharma, Sujesh Kumar, Ramkrishna Mohan Kambli et al.· Journal of Intelligent Decis...· 0 citations
Predictive maintenance (PdM) has emerged as one of the key strategies in the contemporary industrial set-ups, with the ambition of improving the reliability of the machineries, minimizing downtimes, and optimizing the cost operational patterns. The conventional methods of maintenance such as preventive and corrective methods are usually reactive and ineffective leading to unnecessary costs, and breakdown of the system without any prior issues. This study reports on the use of highly-developed neural network (NN) models to reach the desired high-precision predictive maintenance by accurately predicting equipment failures and anomalies. We explore various types of neural networks, such as feedforward neural network (FNNs), recurrent neural networks (RNNs), and convolutional neural network (CNNs), to capture more complicated industrial data. It is proposed that the sensor data processing, feature extraction, and model training will be linked in a framework enabling the prediction of the failure events with a high degree of accuracy. Large-scale evaluations are performed on publicly available predictive maintenance data and the performance is measured in terms of accuracy, precision, recall, F1-score and mean absolute error (MAE). It has been found that deep learning models, especially LSTM-based RNNs, are more effective at revealing temporal correlations in time-series sensor data, thereby providing data with a better predictive aspect. This research gives a complete methodology of implementing neural network-based predictive maintenance systems such as the architecture, preprocessing of data techniques, hyperparameter optimization, and the evaluation of the model. The results support the fact that neural network models have a tremendous potential to revolutionize the predictive maintenance practice, which can be applied in practice by industrial practitioners and scholars. The suggested solution will help to reduce downtimes, make equipment operational longer, and decrease operations costs in industries significantly.
Vijaya Ragavan, Neela Rohit, S. Mohammed· International Journal of Mod...· 0 citations
Timely asset maintenance remains a critical challenge in Industry 4.0 environments. Predictive Maintenance aims to anticipate failures and estimate Remaining Useful Life (RUL), enabling cost reduction and minimizing production downtime. However, real-world industrial scenarios are often characterized by noisy telemetry data, incomplete information about operating conditions, and weak degradation signals, which limit the effectiveness of conventional data-driven approaches. This paper proposes a methodology for RUL prediction under such challenging conditions, leveraging raw sensor telemetry without requiring detailed knowledge of machine operating characteristics. The approach introduces a novel Degradation Index, combined with a Health Index, to better represent degradation patterns. Additionally, signal preprocessing techniques, including Savitzky–Golay and Kalman filters, are applied to mitigate noise and improve data quality. The methodology integrates statistical analysis, similarity-based pattern extraction, and machine learning techniques, including Convolutional Neural Networks and Long Short-Term Memory models, for feature selection and prediction. Experiments conducted on real-world industrial datasets demonstrate that the proposed approach significantly improves prediction performance, achieving high accuracy and enabling failure anticipation up to five days in advance. The results highlight the importance of feature engineering and signal processing in PdM applications, showing that combining degradation modeling with deep learning yields robust, generalizable RUL predictions, even in noisy, partially observed environments.
Tiago Zonta, C. D. da Costa, F. Zeiser et al.· Scientific Reports· 0 citations
Electrified transportation is becoming a crucial component of the global energy transition. Therefore, ensuring the reliability and resilience of its subsystems is essential for its sustainable operations. The most important subsystem is the battery system whose failures can affect operational efficiency. While the machine learning predictive maintenance method has shown promise for early fault detection, most existing approaches rely on offline and statically trained models that are unable to adapt to sensor degradation and concept drift. In this paper, we present the evaluation of an adaptive machine learning method for predictive maintenance in EV battery systems.
We proposed a framework using the NASA Prognostics Center of Excellence (PCoE) lithium-ion battery dataset that uses an online adaptive learning and concept drift detection method to maintain model accuracy. A random forest regression model has been developed to predict battery capacity degradation based on extracted statistical and degradation-related features. The drift detection mechanism that is based on rolling mean absolute error (MAE) was used to identify performance degradation and trigger model retraining using a sliding window approach. The model result showed high predictive performance by achieving a high level of agreement between predicted and actual battery capacity degradation.
The statistical result shows a MAE of 0.061, a root mean square error (RMSE) of 0.11, and a coefficient of determination (R2) of 0.911. This shows its ability to capture nonlinear battery degradation trends accurately. The integration of a decision layer also enables maintenance to be prioritized by balancing system reliability and energy efficiency. This adaptive method addressed concept drift by capturing patterns and maintaining predictive accuracy through continuous model updates in EV battery systems.
S. A. Ajala, S. A. Animashaun, A. Olaleye· SPE Nigeria Annual Internati...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
