UAV Anomaly Detection Method Based on Convolutional Autoencoder and Support Vector Data Description with 0/1 Soft-Margin Loss DOI Creative Commons
Huakun Chen, Yongxi Lyu,

Jingping Shi

и другие.

Drones, Год журнала: 2024, Номер 8(10), С. 534 - 534

Опубликована: Сен. 29, 2024

Unmanned aerial vehicles (UAVs) are becoming more widely used in various industries, raising growing concerns about their safety and reliability. The flight data of UAVs can directly reflect health status; however, the rarity abnormal spatiotemporal characteristics these represent a significant challenge for constructing accurate reliable anomaly detectors. To address this, this study proposes an detection framework that fully considers temporal correlations distribution data. This first combines one-dimensional convolutional neural network (1DCNN) with autoencoder (AE) to establish feature extraction model. model leverages capabilities 1DCNN reconstruction AE thoroughly extract features from UAV Then, adaptive thresholds, research nonlinear support vector description (SVDD) utilizing 0/1 soft-margin loss, referred as L0/1-SVDD. replaces traditional hinge loss function SVDD function, goal enhancing accuracy robustness detection. Since is bounded, non-convex, non-continuous paper Bregman ADMM algorithm solve Finally, difference between reconstructed actual value employed train L0/1-SVDD, resulting hypersphere classifier capable detecting experimental results using real show that, compared methods such AE, LSTM, LSTM-AE, proposed method exhibits superior performance across five evaluation metrics.

Язык: Английский

Aircraft takeoff speed prediction with machine learning: parameter analysis and model development DOI
Nazire Nur KARABURUN, S. Arık Hatipoğlu, Mehmet Konar

и другие.

The Aeronautical Journal, Год журнала: 2025, Номер unknown, С. 1 - 16

Опубликована: Фев. 19, 2025

Abstract With developing technology, the usage areas of aircraft are constantly expanded. In designed for different missions, it is an important issue to evaluate many design possibilities and make optimum designs by taking into account parameters that not directly connected each other with equal importance. this context, issues such as safety performance come fore in designs. One critical situations affecting flight takeoff landing phases aircraft. The speed changes occur these stages issue. study, was predicted machine learning algorithms using data Boeing B-737-300 type Linear regression, support vector classification regression trees, random forest Extreme Gradient Boosting were selected from prediction. Base models created training data. Considering obtained results, feature engineered applied minimise error values proposed base models. developed applying engineered, reduced better observed Takeoff actual presented comparatively first time literature. simulation results emphasise can be used effective alternative method

Язык: Английский

Процитировано

1
Fault Diagnosis in Drones via Multiverse Augmented Extreme Recurrent Expansion of Acoustic Emissions with Uncertainty Bayesian Optimisation DOI Creative Commons
Tarek Berghout, Mohamed Benbouzid

Machines, Год журнала: 2024, Номер 12(8), С. 504 - 504

Опубликована: Июль 26, 2024

Drones are a promising technology performing various functions, ranging from aerial photography to emergency response, requiring swift fault diagnosis methods sustain operational continuity and minimise downtime. This optimises resources, reduces maintenance costs, boosts mission success rates. Among these methods, traditional approaches such as visual inspection or manual testing have long been utilised. However, in recent years, data representation deep learning systems, achieved significant success. These learn patterns relationships, enhancing diagnosis, but also face challenges with complexity, uncertainties, modelling complexities. paper tackles specific by introducing an efficient method denoted Multiverse Augmented Recurrent Expansion (MVA-REX), allowing for iterative understanding of both representations model behaviours gaining better dependencies. Additionally, this approach involves Uncertainty Bayesian Optimisation (UBO) under Extreme Learning Machine (ELM), lighter neural network training tool, tackle uncertainties reduce Three main realistic datasets recorded based on acoustic emissions involved tackling propeller motor failures drones conditions. The UBO-MVA REX (UBO-MVA-EREX) is evaluated many, error metrics, confusion matrix computational cost uncertainty quantification confidence prediction interval features. Application compared the well-known long-short term memory (LSTM), optimisation approximation error, demonstrates performances, certainty, efficiency proposed scheme. More specifically, accuracy obtained UBO-MVA-EREX, ~0.9960, exceeds LSTM, ~0.9158, ~8.75%. Besides, search time UBO-MVA-EREX ~0.0912 s, which ~98.15% faster than ~4.9287 making it highly applicable challenging tasks diagnosis-based emission signals drones.

Язык: Английский

Процитировано

5
Photodiode Signal Patterns: Unsupervised Learning for Laser Weld Defect Analysis DOI Open Access
Erkan Caner Ozkat

Processes, Год журнала: 2025, Номер 13(1), С. 121 - 121

Опубликована: Янв. 5, 2025

Laser welding, widely used in industries such as automotive and aerospace, requires precise monitoring to ensure defect-free welds, especially when joining dissimilar metallic thin foils. This study investigates the application of machine learning techniques for defect detection laser welding using photodiode signal patterns. Supervised models, including Support Vector Machine (SVM), k-Nearest Neighbors (kNN), Random Forest (RF), were employed classify weld defects into sound welds (SW), lack connection (LoC), over-penetration (OP). SVM achieved highest accuracy (95.2%) during training, while RF demonstrated superior generalization with 83% on validation data. The also proposed an unsupervised method a wavelet scattering one-dimensional convolutional autoencoder (1D-CAE) network anomaly detection. its effectiveness achieving accuracies 93.3% 87.5% training datasets, respectively. Furthermore, distinct patterns associated SW, OP, LoC identified, highlighting ability signals capture dynamics. These findings demonstrate combining supervised methods detection, paving way robust, real-time quality systems manufacturing. results indicated that could offer significant advantages identifying anomalies reducing manufacturing costs.

Язык: Английский

Процитировано

0
Pedestrian trajectory prediction via physical-guided position association learning DOI
XU Yue-yun, Hongmao Qin, Yougang Bian

и другие.

Engineering Science and Technology an International Journal, Год журнала: 2025, Номер 64, С. 102008 - 102008

Опубликована: Март 8, 2025

Язык: Английский

Процитировано

0
Deep Learning for Anomaly Detection in CNC Machine Vibration Data: A RoughLSTM-Based Approach DOI Creative Commons
Rasim Çekіk, Abdullah Turan

Applied Sciences, Год журнала: 2025, Номер 15(6), С. 3179 - 3179

Опубликована: Март 14, 2025

Ensuring the reliability and efficiency of computer numerical control (CNC) machines is crucial for industrial production. Traditional anomaly detection methods often struggle with uncertainty in vibration data, leading to misclassifications ineffective predictive maintenance. This study proposes rough long short-term memory (RoughLSTM), a novel hybrid model integrating set theory (RST) LSTM enhance CNC machine data. RoughLSTM classifies input data into lower, upper, boundary regions using an adaptive threshold derived from RST, improving handling. The proposed method evaluated on real-world milling machines, achieving classification accuracy 94.3%, false positive rate 3.7%, negative 2.0%, outperforming conventional models. Moreover, comparative performance analysis highlights RoughLSTM’s competitive or superior compared CNN–LSTM WaveletLSTMa across various operational scenarios. These findings highlight potential improve fault diagnosis maintenance, ultimately reducing downtime maintenance costs settings.

Язык: Английский

Процитировано

0
UAV Anomaly Detection Method Based on Convolutional Autoencoder and Support Vector Data Description with 0/1 Soft-Margin Loss DOI Creative Commons
Huakun Chen, Yongxi Lyu,

Jingping Shi

и другие.

Drones, Год журнала: 2024, Номер 8(10), С. 534 - 534

Опубликована: Сен. 29, 2024

Unmanned aerial vehicles (UAVs) are becoming more widely used in various industries, raising growing concerns about their safety and reliability. The flight data of UAVs can directly reflect health status; however, the rarity abnormal spatiotemporal characteristics these represent a significant challenge for constructing accurate reliable anomaly detectors. To address this, this study proposes an detection framework that fully considers temporal correlations distribution data. This first combines one-dimensional convolutional neural network (1DCNN) with autoencoder (AE) to establish feature extraction model. model leverages capabilities 1DCNN reconstruction AE thoroughly extract features from UAV Then, adaptive thresholds, research nonlinear support vector description (SVDD) utilizing 0/1 soft-margin loss, referred as L0/1-SVDD. replaces traditional hinge loss function SVDD function, goal enhancing accuracy robustness detection. Since is bounded, non-convex, non-continuous paper Bregman ADMM algorithm solve Finally, difference between reconstructed actual value employed train L0/1-SVDD, resulting hypersphere classifier capable detecting experimental results using real show that, compared methods such AE, LSTM, LSTM-AE, proposed method exhibits superior performance across five evaluation metrics.

Язык: Английский

Процитировано

1