A sound-vibration physical-information fusion constraint-guided deep learning method for rolling bearing fault diagnosis

Reliability Engineering & System Safety, Journal Year: 2024, Volume and Issue: unknown, P. 110556 - 110556

Published: Oct. 1, 2024

Language: Английский

---

Enhanced recurrent capsule network with hyrbid optimization model for shrimp disease detection

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 26, 2025

Disease detection plays an important role in shrimp aquaculture to ensure the health and sustainability of farming operations. Specifically, detecting viral infections at early stages can prevent significant losses. Image processing applications have been developed detect different types diseases shrimp. However, theaccuracy models needs improvement various through a single model. Therefore, this research presents novel disease model using Enhanced Recurrent Capsule Network (ERCN) with hybrid optimization for enhanced performance. The proposed ERCN utilizes dynamic routing capsules extract spatial hierarchies patterns images, while recurrent layer extracts temporal dependencies. Performance is further improved by incorporating channel attention select optimal regions features images fusion process. dual-level feature procedure combines local global features, providing final fused data classify diseases. Additionally, work incorporates that Harris Hawks Optimization (HHO) Marine Predator Algorithm (MPA) fine-tune classifier parameters. Experiments evaluate performance metrics such as accuracy, precision, recall, specificity, Matthews correlation coefficient, F1-score. resutls confirms superior precision 94.9%, recall 93.5%, F1-score 94.6% accuracy 95.2% over conventional Neural (RNN), Convolutional (CNN), Gated Unit (GRU), Long Short Term Memory (LSTM) Networks.

Language: Английский

---

A novel rolling bearing fault diagnosis method based on time-series fusion transformer with interpretability analysis

Nondestructive Testing And Evaluation, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 27

Published: Nov. 17, 2024

Rolling bearing fault diagnosis enhances equipment reliability, reduces maintenance costs, and enables effective non-destructive testing (NDT). However, current research often emphasizes model design performance optimization, overlooking the long-term dependencies of signals need for interpretability. This study proposes a rolling utilizing time-series fusion transformer with interpretability analysis. The introduces multi-scale feature adaptive to automatically capture integrate features across different scales, enhancing global pattern detection in data. A dynamic patch auto-encoder module transforms embeddings into low-dimensional space better retain local information. model's design, particularly decoding layer Transformer, is optimized multi-head self-attentive mechanism, multi-dimensional attention weights visualization methods are employed clarify extraction process. Quantitative visualizations throughout training improve insight learning dynamics. Experimental results indicate that this surpasses state-of-the-art approaches on benchmark datasets, proving its generalizability robustness diverse scenarios.

Language: Английский

---

Rolling bearing remaining useful life prediction using deep learning based on high-quality representation

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 10, 2025

In the realm of intelligent manufacturing, accurately predicting remaining useful life (RUL) rolling bearings is essential for maintaining high reliability and optimized performance rotating machinery. To address challenges associated with efficiently representing degradation states capturing temporal dependencies in RUL prediction, this paper proposes a deep learning-based approach. The proposed method integrates one-dimensional convolutional autoencoder (1D-DCAE) high-quality feature extraction multilevel bidirectional long short-term memory (Bi-LSTM) network pattern attention (TPA) mechanism to capture dependencies. 1D-DCAE extracts health indicators (HIs) from vibration signals, which serve as representations state. These HIs, along self-labelled data, are fed inputs into Bi-LSTM + TPA model, enhancing quality data used prediction network. Experimental results on PHM2012 bearing dataset demonstrate that effectively signal features outperforms traditional labelling methods, achieving higher accuracy robustness. Furthermore, model exhibits strong generalizability transferability across diverse operating conditions, underscoring its potential real-world applications.

Language: Английский

---

Research on interpretability of writer identification based on multi-label classification of writing style

Engineering Research Express, Journal Year: 2025, Volume and Issue: 7(1), P. 015250 - 015250

Published: Jan. 8, 2025

Abstract Writer identification based on deep learning has shown great potential in fields such as forensic analysis and financial security due to its high efficiency accuracy. However, the specificity of neural networks limits acceptance adoption their results these fields.This is ‘opacity’ networks. To address this issues, paper proposes an interpretable framework for writer multi-label classification writing styles, implemented using residual attention mechanisms. Firstly, study selects five style types commonly used experience manual identification.Based Chinese handwriting dataset HWDB2.0, annotation was carried out construct HWDB-STYLE. Next, a convolutional network combined with channel-spatial module backbone network. Finally, number structure classifiers are improved multi-task model obtained which performs both styles. This can provide identity different types, interpret output through type. Experiments HWDB-STYLE demonstrate that not only maintains accuracy but also accurately classifies each sample. The consistent human observations, providing level interpretability results.

Language: Английский

---

An intelligent fault diagnosis model for bearings with adaptive hyperparameter tuning in multi-condition and limited sample scenarios

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 24, 2025

Abstract Bearing fault diagnosis under multiple operating conditions is challenging due to the complexity of changing environments and limited availability training data. To address these issues, this paper presents an advanced method using a hybrid Grey Wolf Algorithm (HGWA)-optimized convolutional neural network (CNN) Bidirectional long short-term memory (BiLSTM) architecture. The proposed model leverages CNN for extracting spatial features BiLSTM capturing temporal dependencies. Through HGWA, hyperparameters are efficiently optimized, achieving 100% diagnostic accuracy across four with CWRU dataset. Additionally, optimized CNN–BiLSTM demonstrated high when applied as pre-trained in new environments, even minimal not only improves performance but also enhances optimization efficiency, faster results within same time frame. This approach mitigates challenges manually tuning effectively addresses bearing constrained sample conditions, representing meaningful contribution field rolling diagnostics.

Language: Английский

---

A Bearing Fault Diagnosis Method Combining Multi-Source Information and Multi-Domain Information Fusion

Machines, Journal Year: 2025, Volume and Issue: 13(4), P. 289 - 289

Published: March 31, 2025

In modern industries, bearings are often subjected to challenges from environmental noise and variations in operating conditions during their operation, which affects existing fault diagnosis methods that rely on signals single types of sensors. These fail provide comprehensive stable information, thereby affecting the diagnostic performance. To address this issue, paper introduces a multi-source multi-domain information fusion method for (M2IFD) bearings, integrating an attention mechanism enhance process. The proposed is structured into three main stages: initially, original signal undergoes transformation frequency time–frequency domains using envelope spectral transform (EST) Bessel (BT) extract richer features. second stage, features extracted independently each transformed domain combined with channel feature fusion, preserving unique source. Finally, further fused through improve classification accuracy. Extensive comparison experiments conducted Paderborn dataset illustrate M2IFD significantly enhances recognition accuracy across various conditions, showcasing its adaptability robustness. This approach presents new avenues bearing diagnosis, significant implications both theoretical practical applications.

Language: Английский

---

A graph representation learning-based method for fault diagnosis of rotating machinery under time-varying speed conditions

Nonlinear Dynamics, Journal Year: 2025, Volume and Issue: unknown

Published: April 11, 2025

Language: Английский

---

EADCN-BCSR: A novel framework for accurate and real-time waste detection and classification

Earth Science Informatics, Journal Year: 2025, Volume and Issue: 18(2)

Published: April 17, 2025

Language: Английский

---

A hybrid deep learning model for robust aeroengine remaining useful life prediction

Signal Image and Video Processing, Journal Year: 2025, Volume and Issue: 19(7)

Published: May 12, 2025

Language: Английский

---