Engineering Applications of Artificial Intelligence最新文献

{"title":"Real-time inner wall surface defect detection based on multi-morphological feature fusion network","authors":"Zhenshen Qu ,&nbsp;Xinxu Cai ,&nbsp;Tianyi Zhang ,&nbsp;Jiazheng Xu ,&nbsp;Xintong Jiang ,&nbsp;Chuan Lin","doi":"10.1016/j.engappai.2025.111331","DOIUrl":"10.1016/j.engappai.2025.111331","url":null,"abstract":"<div><div>In industrial manufacturing, defects on the inner wall surface are crucial for quality and safety assessment. However, existing detection methods are limited by low resolution and glare interference. This study presents a Multi-morphological Feature Fusion Network for Object Detection (MFFN-OD) for 360°detection of inner wall image defects. First, it cleverly integrates panoramic imaging with conventional features through a dual-branch backbone and annular features, ensuring rotation invariance and holistic feature preservation. Second, we develop an Adaptive Multi-morphological Feature Alignment Module (AMFAM) that combats centrally polarized defects by automatically adjusting feature alignment, reducing noise, and increasing accuracy, as well as a feature interaction module with a focus on strengthening multiscale feature fusion. Third, we introduce an Asymptotic Feature Pyramid Network with Auxiliary Features (AFPN-AF) to further refine fusion, close semantic gaps, and improve performance. Experimental results show that MFFN-OD achieves 96.1% mean Average Precision (mAP) and 94.3% Average Precision (AP) for demanding faults with fast detection of 17 milliseconds per frame, meeting industrial requirements for accuracy and real-time performance.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111331"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144522961","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Addressing anomalies in smart grids: Methods for detecting and localizing high-current loads","authors":"Onder Civelek ,&nbsp;Sedat Gormus ,&nbsp;H.İbrahim Okumus ,&nbsp;Hasan Yilmaz","doi":"10.1016/j.engappai.2025.111223","DOIUrl":"10.1016/j.engappai.2025.111223","url":null,"abstract":"<div><div>The concept of a smart grid encompasses a wide range of advanced technologies that surpass the capabilities of traditional power grids, enabling enhanced monitoring, control and efficiency. In this context, the term “anomaly” denotes unusual or unexpected occurrences, such as abnormal consumption patterns, infrastructure failures, power outages, cyber attacks, or energy theft. Anomaly detection is a critical aspect of improving the reliability and operational efficiency of Smart grid systems. In this study, we focus on detecting high current loads as anomalies in low voltage networks and present a system designed for this purpose. The system architecture includes distribution transformers and customer meters equipped with current sensors, radio frequency (RF) modems for wireless data transmission, and a central management server for data analysis and storage. The system employs machine learning (ML) algorithms for real-time anomaly detection and localization. Therefore, in the feature extraction phase, three distinct methods are utilized: Pattern Analysis, Discrete Wavelet Transform (DWT), and Fast Walsh–Hadamard Transform (FHWT). The Pattern Analysis method achieved the best performance using the Matern Gaussian Process Regression (MGPR) approach, whereas the DWT and FHWT methods yielded the most accurate results with the Optimizable Boosting Method. The results of the Pattern Analysis method indicate a test root mean square error (RMSE) of 69.36, a coefficient of determination (R²) of 0.97, and a mean absolute error (MAE) of 45.40. The DWT method, employing a five-level wavelet transform with Daubechies 18 as the main wavelet, achieved a test RMSE of 53.44, an R² of 0.98, and an MAE of 37.25. The FHWT method resulted in a test RMSE of 83.99, an R² of 0.95, and an MAE of 53.71. Among these methods, the DWT method demonstrated the highest accuracy, with an average error rate of 3.02%, while the Pattern Analysis method and the FHWT method exhibited error rates of 4.09% and 4.86%, respectively. This innovative anomaly detection and localization approach provides a robust foundation for future research and development in power distribution networks, with the potential to reduce energy losses and improve grid stability. This system holds significant promise for reducing energy losses, preventing equipment damage, and enhancing grid stability by enabling rapid identification of unauthorized consumption and fault conditions. Its real-time monitoring capability empowers utilities to mitigate risks associated with energy theft and transient overloads, directly contributing to operational cost savings and improved service reliability.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111223"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Stability analysis of a homogeneous rock slope under steady-state seepage using artificial neural networks","authors":"M.A. Millán ,&nbsp;R. Galindo","doi":"10.1016/j.engappai.2025.111556","DOIUrl":"10.1016/j.engappai.2025.111556","url":null,"abstract":"<div><div>An accurate and effective procedure to determine the stability of rock slopes is paramount during the design and, even more so, during the management and maintenance of transportation infrastructures, when the rock slopes should be continuously monitored and checked against changing environmental conditions. The stability of rock slopes may be assessed using stability charts or limit equilibrium methods that assume an equivalent linear Mohr-Coulomb failure criterion for the rock or numerical models that can deal with complex configurations and rock behavior but demand more computational resources, expertise, and advanced software. In this study, an artificial neural network (ANN) was used to predict the safety factor of a two-dimensional, homogeneous rock slope incorporating several critical factors simultaneously, such as the rock's non-linear failure behavior, rock dilatancy, and different levels of the groundwater table associated with steady-state seepage flow through the rock. This greatly extends previous contributions in the field. A two-hidden layers ANN was trained using thousands of numerical simulations applying the Discontinuity Layout Optimization model, considering the rock mass toughness coefficient, non-dimensional height of the slope, artificial slope angle, dilatancy, and non-dimensional water level position. The ANN predictions closely matched the numerical results, demonstrating its potential as an easier and more accessible method for accurately evaluating the safety factor of rock slopes and as a reliable alternative to traditional numerical methods within specified input ranges. Its implementation is straightforward, using simple equations provided in the article, and it is easily scalable to perform intensive tasks and help complex engineering decision-making.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111556"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523215","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Adaptive prediction of Rate of Penetration while oil-well drilling: A Hoeffding tree based approach","authors":"Djamil Rezki ,&nbsp;Leïla-Hayet Mouss ,&nbsp;Abdelkader Baaziz ,&nbsp;Toufik Bentrcia","doi":"10.1016/j.engappai.2025.111465","DOIUrl":"10.1016/j.engappai.2025.111465","url":null,"abstract":"<div><div>Oil well drilling is an expensive process that needs a particular focus. For this reason, Rate Of Penetration (ROP) has been widely approved as a measure of drilling efficiency and adequate configuration parameters. Our aim in this work consists in the elaboration of a smart system using Hoeffding trees for predicting the Rate of Penetration (ROP) in oilfield drilling. The choice of Hoeffding trees to build our model is motivated by their adaptive learning capability and drift detection. They offer continuous, fast, and efficient learning both online on data streams and offline on batch data. To validate our approach, we used real drilling data from the “Hassi-Terfa” oilfield located in Southeast Algeria. The obtained results show in comparison to the eXtreme Gradient Boosting (XGBoost) algorithm that Hoeffding trees maintain their learning capacity and produce more accurate predictions even in the presence of drifts. This is thanks to the combination of the Adaptive Windowing (ADWIN) algorithm to manage drifts and least mean squares (LMS) filters to reduce noise. This observation highlights the effectiveness of our approach to predict the ROP while oil-well drilling. The proposed smart system offers more efficient solution to predict the ROP, whether in real-time or offline. By leveraging its adaptability to changes in data distribution, our approach ensures more accurate and adaptive predictions, facilitating drilling operations optimization and boosting the overall efficiency of the process.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111465"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144534444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"ESRPCB: An edge guided super-Resolution model and ensemble learning for tiny Printed Circuit Board defect detection","authors":"Xiem HoangVan ,&nbsp;Dang Bui Dinh ,&nbsp;Thanh Nguyen Canh ,&nbsp;Van-Truong Nguyen","doi":"10.1016/j.engappai.2025.111547","DOIUrl":"10.1016/j.engappai.2025.111547","url":null,"abstract":"<div><div>Printed Circuit Boards (PCBs) are critical components in modern electronics, which require stringent quality control to ensure proper functionality. However, the detection of defects in small-scale PCBs images poses significant challenges as a result of the low resolution of the captured images, leading to potential confusion between defects and noise. To overcome these challenges, this paper proposes a novel framework, named ESRPCB (edge-guided super-resolution for PCBs defect detection), which combines edge-guided super-resolution with ensemble learning to enhance PCBs defect detection. The framework leverages the edge information to guide the EDSR (Enhanced Deep Super-Resolution) model with a novel ResCat (Residual Concatenation) structure, enabling it to reconstruct high-resolution images from small PCBs inputs. By incorporating edge features, the super-resolution process preserves critical structural details, ensuring that tiny defects remain distinguishable in the enhanced image. Following this, a multi-modal defect detection model employs ensemble learning to analyze the super-resolved image, improving the accuracy of defect identification. Experimental results demonstrate that ESRPCB achieves superior performance compared to State-of-the-Art (SOTA) methods, achieving an average Peak Signal to Noise Ratio (PSNR) of 30.54 <span><math><mrow><mi>d</mi><mi>B</mi><mrow><mo>(</mo><mi>d</mi><mi>e</mi><mi>c</mi><mi>i</mi><mi>b</mi><mi>e</mi><mi>l</mi><mo>)</mo></mrow></mrow></math></span>, surpassing EDSR by <span><math><mrow><mn>0</mn><mo>.</mo><mn>42</mn><mi>d</mi><mi>B</mi></mrow></math></span>. In defect detection, ESRPCB achieves a mAP50(mean average precision at an Intersection over Union threshold of 0.50) of 0.965, surpassing EDSR (0.905) and traditional super-resolution models by over 5%. Furthermore, the ensemble-based detection approach further enhances performance, achieving a mAP50 of 0.977. These results highlight the effectiveness of ESRPCB in enhancing both image quality and defect detection accuracy, particularly in challenging low-resolution scenarios.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111547"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523122","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Bending analysis of cantilever microbeams with three porosity distributions using physics-informed neural network and modified couple stress theory","authors":"Aiman Tariq,&nbsp;Büşra Uzun,&nbsp;Babür Deliktaş,&nbsp;Mustafa Özgür Yaylı","doi":"10.1016/j.engappai.2025.111589","DOIUrl":"10.1016/j.engappai.2025.111589","url":null,"abstract":"<div><div>This study explores the use of a Physics-Informed Neural Network (PINN) framework to investigate the bending behavior of a cantilever microbeam made of porous material. PINN is a powerful approach that combines machine learning with physics principles to address the challenges of limited training data and enforce domain knowledge into the learning process, making them effective surrogate solvers for Partial Differential Equations (PDEs). In this work, a cantilever microbeam subjected to a uniformly distributed transverse load is examined, considering three different pore distributions including homogeneous, symmetric, and non-symmetric. The bending analysis incorporates the size effect by integrating the modified couple stress theory with the Euler-Bernoulli beam theory. First, the bending equation based on the modified couple stress theory is extended to include porous material properties. The governing equation is then solved using the Laplace transform. The PINN model is trained to approximate the solution by minimizing a loss function that accounts for residual errors at collocation points, as well as initial and boundary conditions. To enhance computational efficiency, the optimal hyperparameters of the PINN model are determined using a combination of Taguchi design of experiments and the Grey Relational Method. Taguchi-Grey approach effectively captures the trade-off between these objectives by normalizing and aggregating them into a single value to reflect the overall performance. The results are validated against analytical solutions based on Laplace transform, and the influence of key parameters such as microbeam length, length scale parameter, and porosity is systematically investigated.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111589"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523167","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Investigation of the dependence of fill factor and efficiency on gallium-doped silicon wafer characteristics using unsupervised learning","authors":"Denish Hirpara ,&nbsp;Paramsinh Zala ,&nbsp;Meenakshi Bhaisare ,&nbsp;Chandra Mauli Kumar ,&nbsp;Mayank Gupta ,&nbsp;Manoj Kumar ,&nbsp;Brijesh Tripathi","doi":"10.1016/j.engappai.2025.111598","DOIUrl":"10.1016/j.engappai.2025.111598","url":null,"abstract":"<div><div>This research analyses the performance parameters of bifacial silicon solar cells using the production line data of gallium-doped silicon wafers through unsupervised machine learning models and reports suitable input parameters. Gallium doping offers enhanced electronic properties, mitigating the light-induced degradation commonly seen in boron-doped silicon, thus making it a promising material for high-efficiency solar cells. The impact of silicon wafer thickness, resistivity, and carrier lifetime on fill factor, open-circuit voltage, and efficiency has been investigated. Employing unsupervised learning models, extensive production line data has been analysed to elucidate the complex dependencies between input parameters of silicon wafers (thickness, resistivity, and carrier lifetime) and performance parameters of fabricated solar cells using these silicon wafers (fill factor, open-circuit voltage, and efficiency). The findings indicate significant dependency of performance parameters on input parameters. Random forest classifier model demonstrated robust predictive capabilities, providing valuable insights for optimizing manufacturing processes. Based on analysis with unsupervised learning of data, following conclusions are drawn for input parameters of silicon wafers: (i) thickness range of 160–164 micro-meter gives the highest fill factor, (ii) resistivity range of 0.5–0.8 Ohm-centimetre is superior for higher fill factor, (iii) carrier lifetime of 0.9–1.05 micro-second results in better fill factor. A combination of all the above three parameters works in favour of high fill factor, it may not be a good idea to achieve one and deviate from the others. The results contribute to guiding future material and process optimizations, pushing the boundaries of solar cell efficiency.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111598"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523211","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Detection and classification of anomalies in oil well production using Open-World Learning","authors":"Lucas Gouveia Omena Lopes ,&nbsp;Thales Miranda de Almeida Vieira ,&nbsp;Pedro Esteves Aranha ,&nbsp;Eduardo Toledo de Lima Junior ,&nbsp;William Wagner Matos Lira","doi":"10.1016/j.engappai.2025.111514","DOIUrl":"10.1016/j.engappai.2025.111514","url":null,"abstract":"<div><div>Accurate anomaly detection and classification are critical for operational safety and efficiency in oil well production. While machine learning methods can identify known anomalies, detecting and classifying previously unseen anomalies remains a challenge. This paper presents the first Open-World Learning strategy applied to anomalies in oil well production data. The strategy detects anomalous behavior, classifies whether it belongs to a known labeled anomaly, and, if not, clusters it into newly proposed anomaly classes and learns to classify them. The approach integrates autoencoder reconstruction error for anomaly detection, autoencoder-based dimensionality reduction to extract latent features, binary classifiers to classify known anomalies, and clustering methods to group similar unseen anomalies. If an anomaly is detected via reconstruction error, the binary classifiers determine whether it belongs to a known class. If it does not, the clustering method groups similar events into new classes, which are validated by human experts. This validation enables the training of specific binary classifiers for the new classes and updates existing ones. Experiments on real anomalous oil well production data demonstrate that the discovered clusters align well with ground-truth labels. The clustering methodology achieves 81% accuracy overall, exceeding 95% for certain anomalies, while updated binary classifiers reach up to 99% accuracy. These findings demonstrate the proposed method’s effectiveness in dynamically adapting to novel anomalies, improving classification accuracy, and enhancing oil well monitoring.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111514"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523216","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A multi-agent deep reinforcement learning method for fully noisy observations","authors":"Kaiyu Wang ,&nbsp;Danni Wang ,&nbsp;Bohao Qu ,&nbsp;Menglin Zhang ,&nbsp;Xianchang Wang ,&nbsp;Ximing Li","doi":"10.1016/j.engappai.2025.111553","DOIUrl":"10.1016/j.engappai.2025.111553","url":null,"abstract":"<div><div>Multi-agent reinforcement learning (MARL) algorithms have achieved great breakthroughs in many aspects. The MARL algorithms can learn effective policies in ideal simulation environments. But different from the ideal simulation environments, noise is unavoidable in the real world. MARL algorithms need to learn effective policies in unavoidable fully noisy environments. In this paper, we consider a challenging multi-agent reinforcement learning problem: All agents cannot observe any noiseless observations from environments during the whole training process and MARL algorithms cannot learn effective policies in these fully noisy observation environments. To solve this problem, we propose a method called Robust <span><math><mi>P</mi></math></span>olicy <span><math><mi>L</mi></math></span>earning under Fully Noisy Observation vi<span><math><mi>A</mi></math></span> De<span><math><mi>N</mi></math></span>oising R<span><math><mi>E</mi></math></span>presentation Ne<span><math><mi>T</mi></math></span>work (PLANET), which enables MARL algorithms learning effective policies in fully noisy observation environments. The PLANET method learns the effective policy through two steps. (1) Extracting the noise characteristics and motion laws to obtain clean observations information from fully noisy observation histories. (2) Making MARL algorithms extract information from the noise characteristics and motion laws information, and learn effective policies. The results of a series of exhaustive experiments show that our method can mitigate the effects of noise and learn effective policies in fully noisy observation environments. Our Artificial Intelligence contribution lies in introducing the denoising representation network that learns noise characteristics and motion dynamics to recover clean observations from fully noisy observations. The proposed PLANET framework could be applied to real-world multi-agent robotic and sensor network systems, potentially improving policy robustness under fully noisy observation.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111553"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523212","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A comprehensive multimodal benchmark of neuromorphic training frameworks for spiking neural networks","authors":"Ying-Chao Cheng ,&nbsp;Wang-Xin Hu ,&nbsp;Yu-Lin He ,&nbsp;Joshua Zhexue Huang","doi":"10.1016/j.engappai.2025.111543","DOIUrl":"10.1016/j.engappai.2025.111543","url":null,"abstract":"<div><div>Spiking neural networks (SNNs) represent a promising paradigm for energy-efficient, event-driven artificial intelligence, owing to their biological plausibility and unique temporal processing capabilities. Despite the rapid growth of neuromorphic training frameworks, the lack of standardized benchmarks hinders both the effective comparison of these tools and the broader advancement of SNN-based solutions for real-world applications. In this work, we address this critical gap by conducting a comprehensive, multimodal benchmark of five leading SNN frameworks—SpikingJelly, BrainCog, Sinabs, SNNGrow, and Lava. Our evaluation system integrates quantitative performance metrics – including accuracy, latency, energy consumption, and noise immunity – across diverse datasets (image, text, and neuromorphic event data), along with qualitative assessments of framework adaptability, model complexity, neuromorphic features, and community engagement. Our results indicate that SpikingJelly excels in overall performance, particularly in energy efficiency, while BrainCog demonstrates robust performance on complex tasks. Sinabs and SNNGrow offer balanced performance in latency and stability, though SNNGrow shows limitations in advanced training support and neuromorphic features, and Lava appears less adaptable to large-scale datasets. Additionally, we investigate the effects of varying time steps, training methods, and data encoding strategies on performance. This benchmark not only provides actionable guidance for selecting and optimizing SNN solutions but also lays the foundation for future research on advanced architectures and training techniques, ultimately accelerating the adoption of energy-efficient, brain-inspired computing in practical artificial intelligence engineering.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"159 ","pages":"Article 111543"},"PeriodicalIF":7.5,"publicationDate":"2025-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523218","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}