A Comprehensive Survey of Convolutions in Deep Learning: Applications, Challenges, and Future Trends

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 41180 - 41218

Published: Jan. 1, 2024

In today's digital age, Convolutional Neural Networks (CNNs), a subset of Deep Learning (DL), are widely used for various computer vision tasks such as image classification, object detection, and segmentation. There numerous types CNNs designed to meet specific needs requirements, including 1D, 2D, 3D CNNs, well dilated, grouped, attention, depthwise convolutions, NAS, among others. Each type CNN has its unique structure characteristics, making it suitable tasks. It's crucial gain thorough understanding perform comparative analysis these different understand their strengths weaknesses. Furthermore, studying the performance, limitations, practical applications each can aid in development new improved architectures future. We also dive into platforms frameworks that researchers utilize research or from perspectives. Additionally, we explore main fields like 6D vision, generative models, meta-learning. This survey paper provides comprehensive examination comparison architectures, highlighting architectural differences emphasizing respective advantages, disadvantages, applications, challenges, future trends.

Language: Английский

---

Deep learning based approaches for intelligent industrial machinery health management and fault diagnosis in resource-constrained environments

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

Published: Jan. 7, 2025

Industry 4.0 represents the fourth industrial revolution, which is characterized by incorporation of digital technologies, Internet Things (IoT), artificial intelligence, big data, and other advanced technologies into processes. Industrial Machinery Health Management (IMHM) a crucial element, based on (IIoT), focuses monitoring health condition machinery. The academic community has focused various aspects IMHM, such as prognostic maintenance, monitoring, estimation remaining useful life (RUL), intelligent fault diagnosis (IFD), architectures edge computing. Each these categories holds its own significance in context In this survey, we specifically examine research RUL prediction, edge-based architectures, diagnosis, with primary focus domain diagnosis. importance IFD methods ensuring smooth execution processes become increasingly evident. However, most are formulated under assumption complete, balanced, abundant often does not align real-world engineering scenarios. difficulties linked to classifications IMHM have received noteworthy attention from community, leading substantial number published papers topic. While there existing comprehensive reviews that address major challenges limitations field, still gap thoroughly investigating perspectives across complete To fill gap, undertake survey discusses achievements domain, focusing IFD. Initially, classify three distinct perspectives: method processing aims optimize inputs for model mitigate training sample set; constructing model, involves designing structure features enhance resilience challenges; optimizing training, refining process models emphasizes ideal data process. Subsequently, covers techniques related prediction edge-cloud resource-constrained environments. Finally, consolidates outlook relevant issues explores potential solutions, offers practical recommendations further consideration.

Language: Английский

---

Transfer learning-based deep learning models for proton exchange membrane fuel remaining useful life prediction

Fuel, Journal Year: 2024, Volume and Issue: 367, P. 131461 - 131461

Published: March 20, 2024

Language: Английский

---

A Deep Transfer Learning-Based Visual Inspection System for Assembly Defects in Similar Types of Manual Tool Products

Sensors, Journal Year: 2025, Volume and Issue: 25(6), P. 1645 - 1645

Published: March 7, 2025

This study introduces an advanced inspection system for manual tool assembly, focusing on defect detection and classification in flex-head ratchet wrenches as a modern alternative to traditional methods. Using deep learning R-CNN approach with transfer learning, specifically utilizing the AlexNet architecture, accurately identifies classifies assembly defects across similar tools. demonstrates how pre-trained model older models can be efficiently adapted new only moderate amounts of samples fine-tuning. Experimental evaluations at three stations show that achieves accuracy 98.67% station highest variety, outperforming randomly initialized weights. Even 40% reduction sample size products, maintains 98.66%. Additionally, compared R-CNN, it improves average effectiveness by 9% efficiency 26% all stations. A sensitivity analysis further reveals proposed method reduces training 50% similarity while enhancing 13.06% 5.31%.

Language: Английский

---

Cutting-Edge AI Approaches with MAS for PdM in Industry 4.0: Challenges and Future Directions

Journal of Applied Data Sciences, Journal Year: 2024, Volume and Issue: 5(2), P. 455 - 473

Published: May 15, 2024

Integrating Artificial Intelligence (AI) within Industry 4.0 has propelled the evolution of fault diagnosis and predictive maintenance (PdM) strategies, marking a significant shift towards smarter paradigms in mechatronics sector. With advent 4.0, mechatronic systems have become increasingly sophisticated, highlighting critical need for advanced methodologies that are both efficient effective. This paper delves into confluence cutting-edge AI techniques, including machine learning (ML) deep (DL), with multi-agent (MAS) to enhance precision facilitate PdM context 4.0. Specifically, we explore use various ML models, Support Vector Machines (SVMs) Random Forests (RFs), DL architectures like Convolutional Neural Networks (CNNs) Recurrent (RNNs), which been effectively oriented analyses complex industrial data. Initially, study examines progress algorithms accelerate identification by leveraging data from system operations, sensors, historical trends. AI-enabled rapidly detects irregularities discerns fundamental causes, thereby minimizing downtime enhancing reliability efficiency. Furthermore, this underscores adoption AI-driven approaches, emphasizing prognostics predict Remaining Useful Life (RUL) machinery. capability allows strategic scheduling activities, optimizing resource use, prolonging lifespan expensive assets, refining management spare parts inventory. The tangible advantages employing showcased through case authentic implementations. highlights successful implementations, documenting real-world challenges such as integration issues interoperability, elaborates on strategies deployed navigate these obstacles. results demonstrate improved operational cost savings shed light pragmatic considerations solutions MAS applications. also navigates prospective research avenues applying domain setting stage ongoing innovation exploration transformative domain.

Language: Английский

---

Self-adaptive fault diagnosis for unseen working conditions based on digital twins and domain generalization

Reliability Engineering & System Safety, Journal Year: 2024, Volume and Issue: unknown, P. 110560 - 110560

Published: Oct. 1, 2024

Language: Английский

---

A new dual-channel convolutional neural network and its application in rolling bearing fault diagnosis

Measurement Science and Technology, Journal Year: 2024, Volume and Issue: 35(9), P. 096130 - 096130

Published: June 14, 2024

Abstract Recently, deep learning has received widespread attention in the field of bearing fault diagnosis due to its powerful feature capability. However, when actual working conditions are complex and variable, information a single domain is limited, making it difficult achieve high accuracy. To overcome these challenges, this paper proposes method based on Markov transition field, continuous wavelet transform (CWT), dual-channel convolutional neural network (CNN). The combines descriptive ability model for state transfer, time-frequency analysis CWT signal, excellent performance CNN with mechanism extraction classification. Specifically, we first propose multi-channel method, which combined obtain two different representations two-dimensional (2D) images. comprehensively mine information, further an mechanism. design structure aims extract multi-level features from types 2D At same time, designed embedded enable focus more extracting effective features, thereby improving accuracy network. verify effectiveness proposed three datasets were used empirical research. results show that exhibits superior higher compared traditional methods.

Language: Английский

---

Hybrid Predictive Maintenance for Building Systems: Integrating Rule-Based and Machine Learning Models for Fault Detection Using a High-Resolution Danish Dataset

Buildings, Journal Year: 2025, Volume and Issue: 15(4), P. 630 - 630

Published: Feb. 18, 2025

This study evaluates the effectiveness of six machine learning models, Artificial Neural Networks (ANN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression (LR), for predictive maintenance in building systems. Utilizing a high-resolution dataset collected every five minutes from office rooms at Aalborg University Denmark over ten-month period (27 February 2023 to 31 December 2023), we defined rule-based conditions label historical faults HVAC, lighting, occupancy systems, resulting 100,000 fault instances. XGBoost outperformed other achieving an accuracy 95%, precision 93%, recall 94%, F1-score 0.93, with computation time 60 s. The model effectively predicted critical such as “Light_On_No_Occupancy” (1149 occurrences) “Damper_Open_No_Occupancy” (8818 occurrences), demonstrating its potential real-time detection energy optimization management Our findings suggest that implementing frameworks can significantly enhance accuracy, reduce waste, improve operational efficiency.

Language: Английский

---

Digital Cloning as a Self-Adaptive Multicriteria Optimization Process

Procedia Computer Science, Journal Year: 2025, Volume and Issue: 253, P. 37 - 48

Published: Jan. 1, 2025

Language: Английский

---

Comparison and Optimization of Generalized Stamping Machine Fault Diagnosis Models Using Various Transfer Learning Methodologies

Sensors, Journal Year: 2025, Volume and Issue: 25(6), P. 1779 - 1779

Published: March 13, 2025

The integration of artificial intelligence (AI) with stamping technology has become increasingly critical in smart manufacturing, driven by advancements both fields. Total clearance, a crucial determinant process and product quality operations, significantly impacts cutting precision, material deformation, the longevity equipment. Consequently, real-time monitoring prediction total clearance are essential for effective control fault diagnosis. However, heterogeneity machine designs necessitates development numerous machine-specific models, posing significant challenge practical implementation. This research addresses this developing generalized diagnosis model applicable across multiple types. Specifically, is designed to monitor four distinct models: OCP-110, G2-110, G2-160, ST1-110. Vibration data, acquired using accelerometers strategically placed at two sensor locations on each machine, serve as primary input model. Four prominent deep learning architectures—a 10-layer convolutional neural network (CNN), CNN residual connections (CNN-Res), VGG16, ResNet50—were rigorously evaluated conjunction fine-tuning strategies determine optimal architecture. resulting achieved an average accuracy, recall rate, F1 score exceeding 99%, demonstrating its efficacy reliability real-world applications. proposed approach offers potential scalability additional types operational conditions, thereby streamlining deployment predictive maintenance systems equipment manufacturers.

Language: Английский

---