Artificial Datasets Research Articles

Minimizing the Limitations in Improving Historical Aerial Photographs with Super-Resolution Technique

Compared to natural images in artificial datasets, it is more challenging to improve the spatial resolution of remote sensing optical image data using super-resolution techniques. Historical aerial images are primarily grayscale due to single-band acquisition, which further limits their recoverability. To avoid data limitations, it is advised to employ a data collection consisting of images with homogeneously distributed intensity values of land use/cover objects at various resolution values. Thus, two different datasets were created. In line with the proposed approach, images of bare land, farmland, residential areas, and forested regions were extracted from orthophotos of different years with different spatial resolutions. In addition, images with intensity values in a more limited range for the same categories were obtained from a single year’s orthophoto to highlight the contribution of the suggested approach. Training of two different datasets was performed independently using a deep learning-based super-resolution model, and the same test images were enhanced individually with the weights of both models. The results were assessed using a variety of quality metrics in addition to visual interpretation. The findings indicate that the suggested dataset structure and content can enable the recovery of more details and effectively remove the smoothing effect. In addition, the trend of the metric values matches the visual perception results.

A Multi-center Dental Panoramic Radiography Image Dataset for Impacted Teeth, Periodontitis, and Dental Caries: Benchmarking Segmentation and Classification Tasks.

Panoramic radiography imaging plays a crucial role in the diagnostic process of dental diseases. However, current artificial intelligence research datasets for panoramic radiography dental image processing are often limited to single-center and single-task scenarios, making it difficult to generalize their results. To address this, we present a multi-center, multi-task labeled dataset. In this study, our dataset comprises three datasets obtained from different hospitals. The first set has 4940 panoramic radiography images and corresponding labels from the Stemmatological Hospital of the General Hospital of Ningxia Medical University. The second set includes 716 panoramic radiography images and labels from the People's Hospital of Yinchuan City, Ningxia. The third dataset contains 880 panoramic radiography images and labels from a hospital in Shenzhen, Guangdong Province. This comprehensive dataset encompasses three types of dental diseases: impacted teeth, periodontitis, and dental caries. Specifically, it comprises 2555 images related to impacted teeth, 2735 images related to periodontitis, and 1246 images related to dental caries. In order to evaluate the performance of the dataset, we conducted benchmark tests for segmentation and classification tasks on our dataset. The results show that the presented dataset could be effectively used for benchmarking segmentation and classification tasks critical to the diagnosis of dental diseases. To request our multi-center dataset, please visit the address: https://github.com/qinxin99/qinxini .

An improved YOLOv5-based model for automatic PCB defect detection

The detection of PCB defects plays an important role in PCB production. To meet the requested quality standard, systematic research was invested in PCB defect detection. The existing PCB defect detection methods are mainly trained by using an artificial defect image dataset, which is much more ideal than the real PCB production process. To deal with the real PCB production defect detection, we collected 3239 image samples and labeled each of them by label-image tool. This paper proposed a new model improved from the basic YOLOv5 framework, adding a scale and self-attention mechanism. The results revealed that this model performed well on defect location and classification with its mAP0.5 reaching 63.4%.

Detecting fast neutrino flavor conversions with machine learning

Neutrinos in dense environments like core-collapse supernovae (CCSNe) and neutron star mergers (NSMs) can undergo fast flavor conversions (FFCs) once the angular distribution of neutrino lepton number crosses zero along a certain direction. Recent advancements have demonstrated the effectiveness of machine learning (ML) in detecting these crossings. In this study, we enhance prior research in two significant ways. First, we utilize realistic data from CCSN simulations, where neutrino transport is solved using the full Boltzmann equation. We evaluate the ML methods’ adaptability in a real-world context, enhancing their robustness. In particular, we demonstrate that when working with artificial data, simpler models outperform their more complex counterparts, a noteworthy illustration of the bias-variance tradeoff in the context of ML. We also explore methods to improve artificial datasets for ML training. In addition, we extend our ML techniques to detect the crossings in the heavy-leptonic channels, accommodating scenarios where νx and ν¯x may differ. Our research highlights the extensive versatility and effectiveness of ML techniques, presenting an unparalleled opportunity to evaluate the occurrence of FFCs in CCSN and NSM simulations. Published by the American Physical Society 2024

An ensemble learning method with GAN-based sampling and consistency check for anomaly detection of imbalanced data streams with concept drift.

A challenge to many real-world data streams is imbalance with concept drift, which is one of the most critical tasks in anomaly detection. Learning nonstationary data streams for anomaly detection has been well studied in recent years. However, most of the researches assume that the class of data streams is relatively balanced. Only a few approaches tackle the joint issue of imbalance and concept drift. To overcome this joint issue, we propose an ensemble learning method with generative adversarial network-based sampling and consistency check (EGSCC) in this paper. First, we design a comprehensive anomaly detection framework that includes an oversampling module by generative adversarial network, an ensemble classifier, and a consistency check module. Next, we introduce double encoders into GAN to better capture the distribution characteristics of imbalanced data for oversampling. Then, we apply the stacking ensemble learning to deal with concept drift. Four base classifiers of SVM, KNN, DT and RF are used in the first layer, and LR is used as meta classifier in second layer. Last but not least, we take consistency check of the incremental instance and check set to determine whether it is anormal by statistical learning, instead of threshold-based method. And the validation set is dynamic updated according to the consistency check result. Finally, three artificial data sets obtained from Massive Online Analysis platform and two real data sets are used to verify the performance of the proposed method from four aspects: detection performance, parameter sensitivity, algorithm cost and anti-noise ability. Experimental results show that the proposed method has significant advantages in anomaly detection of imbalanced data streams with concept drift.

A Comprehensive Literature Review on Artificial Dataset Generation for Repositioning Challenges in Shared Electric Automated and Connected Mobility

In the near future, the incorporation of shared electric automated and connected mobility (SEACM) technologies will significantly transform the landscape of transportation into a sustainable and efficient mobility ecosystem. However, these technological advances raise complex scientific challenges. Problems related to safety, energy efficiency, and route optimization in dynamic urban environments are major issues to be resolved. In addition, the unavailability of realistic and various data of such systems makes their deployment, design, and performance evaluation very challenging. As a result, to avoid the constraints of real data collection, using generated artificial datasets is crucial for simulation to test and validate algorithms and models under various scenarios. These artificial datasets are used for the training of ML (Machine Learning) models, allowing researchers and operators to evaluate performance and predict system behavior under various conditions. To generate artificial datasets, numerous elements such as user behavior, vehicle dynamics, charging infrastructure, and environmental conditions must be considered. In all these elements, symmetry is a core concern; in some cases, asymmetry is more realistic; however, in others, reaching/maintaining as much symmetry as possible is a core requirement. This review paper provides a comprehensive literature survey of the most relevant techniques generating synthetic datasets in the literature, with a particular focus on the shared electric automated and connected mobility context. Furthermore, this paper also investigates central issues of these complex and dynamic systems regarding how artificial datasets could be used in the training of ML models to address the repositioning problem. Hereby, symmetry is undoubtedly a crucial consideration for ML models. In the case of datasets, it is imperative that they accurately emulate the symmetry or asymmetry observed in real-world scenarios to be effectively represented by the generated datasets. Then, this paper investigates the current challenges and limitations of synthetic datasets, such as the reliability of simulations to the real world, and the validation of generative models. Additionally, it explores how ML-based algorithms can be used to optimize vehicle routing, charging infrastructure usage, demand forecasting, and other important operational elements. In conclusion, this paper outlines a series of interesting new research avenues concerning the generation of artificial data for SEACM systems.

A new method to design energy-conserving surrogate models for the coupled, nonlinear responses of intervertebral discs

The aim of this study was to design physics-preserving and precise surrogate models of the nonlinear elastic behaviour of an intervertebral disc (IVD). Based on artificial force-displacement data sets from detailed finite element (FE) disc models, we used greedy kernel and polynomial approximations of second, third and fourth order to train surrogate models for the scalar force-torque -potential. Doing so, the resulting models of the elastic IVD responses ensured the conservation of mechanical energy through their structure. At the same time, they were capable of predicting disc forces in a physiological range of motion and for the coupling of all six degrees of freedom of an intervertebral joint. The performance of all surrogate models for a subject-specific L4|\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\vert$$\\end{document}5 disc geometry was evaluated both on training and test data obtained from uncoupled (one-dimensional), weakly coupled (two-dimensional), and random movement trajectories in the entire six-dimensional (6d) physiological displacement range, as well as on synthetic kinematic data. We observed highest precisions for the kernel surrogate followed by the fourth-order polynomial model. Both clearly outperformed the second-order polynomial model which is equivalent to the commonly used stiffness matrix in neuro-musculoskeletal simulations. Hence, the proposed model architectures have the potential to improve the accuracy and, therewith, validity of load predictions in neuro-musculoskeletal spine models.

A hybrid sampling method for highly imbalanced and overlapped data classification with complex distribution

Class imbalance and overlap coupling is the primary cause of performance degradation for classifiers. Unfortunately, it always occurs. In this paper, we propose a hybrid sampling method derived from optimized generative adversarial network and natural neighbor search (HD-GNNS) for this scenario. The new approach considers both the global distribution and local distribution of the minority class to improve the data distribution fundamentally. First, natural neighbor search with Fisher’s discriminant ratio is conducted to screen overlapped sample subset and remove noise samples. It effectively overcomes the parameter sensitivity by adaptively determining the search radius. Then, an encoder with squeeze and excite block is introduced into generative adversarial network, and the structure of generative adversarial network is optimized with cross-layer and low-rank matrix. It better captures the distribution characteristics of minority samples in overlapped subset for oversampling. Afterwards, the local density of majority samples in overlapped subset is calculated by the aforementioned natural neighbor search method, and Thornton’s Separation Index is used to implement under-sampling adaptively. We evaluate the proposed approach on 1 artificial dataset, 14 UCI datasets and 8 real-word datasets. The experimental results show that the proposed HD-GNNS exhibits more impressive performance compared to other benchmark methods.

An adaptive evolutionary multi-objective clustering based on the data properties of the base partitions

Evolutionary multi-objective clustering (EMOC) is a modern clustering technique in which the general concepts of evolutionary multi-objective optimization are applied to the clustering problem. The design and definition of clustering are difficult problems, and the choice of the objective functions and parameter setting of the algorithms are still challenges. In this paper, we propose the adaptive evolutionary multi-objective clustering approach based on data properties (AEMOC). AEMOC considers a new metric to evaluate the base partitions (candidate solutions used in the initial population) generated by minimum spanning tree clustering. This metric is applied to: (1) determine properties of the data, such as separation and overlap; and (2) define an offline selection of objective functions and parameter settings of the multi-objective algorithm. The information regarding the data properties allows AEMOC to avoid unnecessary data processing in datasets in which the initialization provides optimal solutions and optimization is not required. AEMOC presented promising results considering a diverse set of artificial and real-life datasets, based on two aspects: (1) it succeeded in the determination of the data properties of the base partitions and verifying the potential clustering quality, presenting a correlation of 0.8 with a reference metric, and (2) it provided better clustering results than reference EMOC approaches, achieving a general gain in the clustering performance of 3% in real-life datasets and 7% in artificial datasets.

MOBFinder: a tool for mobilization typing of plasmid metagenomic fragments based on a language model.

Mobilization typing (MOB) is a classification scheme for plasmid genomes based on their relaxase gene. The host ranges of plasmids of different MOB categories are diverse, and MOB is crucial for investigating plasmid mobilization, especially the transmission of resistance genes and virulence factors. However, MOB typing of plasmid metagenomic data is challenging due to the highly fragmented characteristics of metagenomic contigs. We developed MOBFinder, an 11-class classifier, for categorizing plasmid fragments into 10 MOB types and a nonmobilizable category. We first performed MOB typing to classify complete plasmid genomes according to relaxase information and then constructed an artificial benchmark dataset of plasmid metagenomic fragments (PMFs) from those complete plasmid genomes whose MOB types are well annotated. Next, based on natural language models, we used word vectors to characterize the PMFs. Several random forest classification models were trained and integrated to predict fragments of different lengths. Evaluating the tool using the benchmark dataset, we found that MOBFinder outperforms previous tools such as MOBscan and MOB-suite, with an overall accuracy approximately 59% higher than that of MOB-suite. Moreover, the balanced accuracy, harmonic mean, and F1-score reached up to 99% for some MOB types. When applied to a cohort of patients with type 2 diabetes (T2D), MOBFinder offered insights suggesting that the MOBF type plasmid, which is widely present in Escherichia and Klebsiella, and the MOBQ type plasmid might accelerate antibiotic resistance transmission in patients with T2D. To the best of our knowledge, MOBFinder is the first tool for MOB typing of PMFs. The tool is freely available at https://github.com/FengTaoSMU/MOBFinder.

Biobjective gradient descent for feature selection on high dimension, low sample size data.

Even though deep learning shows impressive results in several applications, its use on problems with High Dimensions and Low Sample Size, such as diagnosing rare diseases, leads to overfitting. One solution often proposed is feature selection. In deep learning, along with feature selection, network sparsification is also used to improve the results when dealing with high dimensions low sample size data. However, most of the time, they are tackled as separate problems. This paper proposes a new approach that integrates feature selection, based on sparsification, into the training process of a deep neural network. This approach uses a constrained biobjective gradient descent method. It provides a set of Pareto optimal neural networks that make a trade-off between network sparsity and model accuracy. Results on both artificial and real datasets show that using a constrained biobjective gradient descent increases the network sparsity without degrading the classification performances. With the proposed approach, on an artificial dataset, the feature selection score reached 0.97 with a sparsity score of 0.92 with an accuracy of 0.9. For the same accuracy, none of the other methods reached a feature score above 0.20 and sparsity score of 0.35. Finally, statistical tests validate the results obtained on all datasets.

Isolation Forests for Symbolic Data as a Tool for Outlier Mining

Aim: Outlier detection is a key part of every data analysis. Although there are many definitions of outliers that can be found in the literature, all of them emphasise that outliers are objects that are in some way different from other objects in the dataset. There are many different approaches that have been proposed, compared, and analysed for the case of classical data. However, there are only few studies that deal with the problem of outlier detection in symbolic data analysis. The paper aimed to propose how to adapt isolation forest for symbolic data cases. Methodology: An isolation forest for symbolic data is used to detect outliers in four different artificial datasets with a known cluster structure and a known number of outliers Results: The results show that the isolation forest for symbolic data is a fast and efficient tool for outlier mining. Implications and recommendations: As the isolation forest for symbolic data appears to be an efficient tool for outlier detection for artificial data, further studies should focus on real data sets that contain outliers (i.e. credit card fraud dataset), and this approach should be compared with other outlier mining tools (i.e. DBCSAN). The authors recommend using the same initial settings for the isolation forest for symbolic data as the settings that are proposed for the isolation forest for classical data. Originality/value: This paper is the first of its kind, focusing not only on the problem of outlier detection in general, but also extending the well-known isolation forest model for symbolic data cases.

Improved limited view ultrasound tomography via machine learning.

Tomographic reconstruction is used extensively in medicine, non-destructive testing and geology. In an ideal situation where measurements are taken at all angles around an object, known as full view configuration, a full reconstruction of the object can be produced. One of the major issues faced in tomographic imaging is when measurements cannot be taken freely around the object under inspection. This may be caused by the size and geometry of the object or difficulty accessing from particular directions. The resulting limited view transducer configuration leads to a large deterioration in image quality, thus it is very beneficial to employ a compensation algorithm. At present, the most effective compensation algorithms require a large amount of computing power or a bespoke case-by case approach, often with numerous arbitrary constants which must be tuned for a specific application. This work proposes a machine learning based approach to perform the limited view compensation. The model is based around an autoencoder architecture. It is trained on an artificial dataset, taking advantage of the ability to generate arbitrary limited view images given a full view input. The approach is evaluated on ten laser-scanned corrosion maps and the results compared to positivity regularisation - a limited view compensation algorithm similar in the speed of execution and generalisation potential. The algorithms are compared for root mean squared error (RMSE) across the image, and maximum absolute error (MAE). Furthermore, they are visually compared for subjective quality. Compared to the conventional algorithm, the ML-based approach improves on the MAE in eight out of the ten cases. The conventional approach performs better on mean RMSE, which is explained by the ML outputting inaccurate background level, which is not a critical ability. Most importantly, the visual inspection of outputs shows the ML approach reconstructs the images better, especially in the case of irregular corrosion patches. Compared to limited view images, the ML method improves both the RMSE and MAE by 41%.

Worst-Case Discriminative Feature Learning via Max-Min Ratio Analysis.

We propose a novel discriminative feature learning method via Max-Min Ratio Analysis (MMRA) for exclusively dealing with the long-standing "worst-case class separation" problem. Existing technologies simply consider maximizing the minimal pairwise distance on all class pairs in the low-dimensional subspace, which is unable to separate overlapped classes entirely especially when the distribution of samples within same class is diverging. We propose a new criterion, i.e., Max-Min Ratio Analysis (MMRA) that focuses on maximizing the minimal ratio value of between-class and within-class scatter to extremely enlarge the separability on the overlapped pairwise classes. Furthermore, we develop two novel discriminative feature learning models for dimensionality reduction and metric learning based on our MMRA criterion. However, solving such a non-smooth non-convex max-min ratio problem is challenging. As an important theoretical contribution in this paper, we systematically derive an alternative iterative algorithm based on a general max-min ratio optimization framework to solve a general max-min ratio problem with rigorous proofs of convergence. More importantly, we also present another solver based on bisection search strategy to solve the SDP problem efficiently. To evaluate the effectiveness of proposed methods, we conduct extensive pattern classification and image retrieval experiments on several artificial datasets and real-world ScRNA-seq datasets, and experimental results demonstrate the effectiveness of proposed methods.

Workpiece Pose Classification Framework for Flexible Part Feeding Systems Using Machine Learning and Synthetic Datasets

In previous work, a concept for a novel flexible part feeding system based on aerodynamic feeding was presented. In contrast to conventional part feeding systems, such as vibratory bowl feeders, aerodynamic part feeding systems use controlled pressurized air jets to reorient the workpieces. The proposed concept aims to increase flexibility and eliminate retooling times to meet the challenges of modern automated production with regard to increasing uncertainties, shorter product life cycles and higher cost pressure. Regarding these objectives, the determination of the workpiece poses in the feeding system requires flexible image processing, which will be enabled using machine learning and synthetic datasets.This work presents a framework which enables the classification of workpiece poses using a standard industrial camera and convolutional neural networks (CNN). The training of CNNs usually requires large datasets. However, in order to eliminate the need to create and label datasets of real images, which would cause machine downtimes and increase the retooling effort, the CNNs will be trained with synthetic datasets. Based on the CAD-data of the workpieces, artificial images of each pose are rendered using the open source engine Blender. Several CNN architectures are selected, trained with the artificial datasets and evaluated using real images of the workpieces. The results show that high classification accuracies of over 99 % can be achieved. It is concluded that the presented approach enables the classification of workpiece poses with high accuracy, while eliminating the need for the cumbersome creation of real image datasets. The approach is not limited to the use in the aerodynamic part feeding system but can be transferred to other applications in which the orientation of a workpiece has to be classified.

Empirical analysis of monthly electricity consumption prediction in manufacturing industry using machine learning techniques

Abstract Electricity consumption prediction is an important part of power planning and the basis of power dispatch planning. The SVM and GBDT algorithms that were optimized by the PSO algorithm are used to build the machine learning-based electricity consumption prediction model in this paper. The decomposed and reconstructed high-frequency signals and low-frequency signals are optimized by the particle swarm algorithms SVM and GBDT for power consumption prediction, respectively. Improve the clustering performance of the traditional fuzzy C-mean algorithm for unbalanced data. The performance of the algorithms is analyzed in two different application scenarios, namely, artificial dataset and power users' real measurement dataset. Power consumption prediction in the manufacturing industry is performed. It is found that the ARI, FMI, and AMI index values of the improved algorithms proposed in this paper are 0.9543, 0.9347, and 0.9344, respectively, on the grid user-measured dataset, while the traditional DPC and K-means clustering algorithms are less effective. The machine learning algorithm optimized after wavelet decomposition increased R² by 8.98%, MAPE decreased by 19.78%, and RMSE decreased by 11.53% compared to PSO-GBDT, and the predictive evaluation indexes were all improved, and the two machine learning algorithms in this paper optimized by wavelet decomposition combined with PSO have good predictive effect R² increased from 0.801 to 0.842, and the two machine learning algorithms designed in this paper based on wavelet decomposition have good predictive effect R² increased from 0.801 to 0.842. The designed machine learning model for electricity consumption prediction based on wavelet decomposition and PSO has excellent performance, and the design expectation has been fulfilled. This paper makes a useful exploration and proposes an effective method for accurate prediction of electricity consumption in the manufacturing industry.

HBC: halo-based clustering using local comparative density

<p>Clustering is a fundamental technique for revealing structure within data. Density-based clustering has the advantage of detecting structure without prior knowledge, such as the distribution or the number of clusters. However, existing density-based clustering algorithms usually suffer from limitations in latent density patterns, which hinder their applicability to complex datasets. We have showed that this may be caused by the bias of standard methods for point density estimation, such as kernel density estimation (KDE), which tends to favor compact clusters. To address the limitations of existing methods, we introduced the concept of local comparative density. We then proposed a specialized technique well-suited for capturing the nature of local comparative density. This approach allowed us to balance the density variations across clusters. Building on this foundation, we presented a halo-based clustering method that can effectively discover clusters with arbitrary shapes, heterogeneous densities, peakless distributions, and unbalanced cluster sizes. Additionally, this method is capable of identifying high-quality noise within the data. The superiority of the proposed algorithm was demonstrated by comparing it with state-of-the-art density-based clustering algorithms on artificial and real-world datasets.</p>

Self-Supervised Random Forest on Transformed Distribution for Anomaly Detection.

Anomaly detection, the task of differentiating abnormal data points from normal ones, presents a significant challenge in the realm of machine learning. Numerous strategies have been proposed to tackle this task, with classification-based methods, specifically those utilizing a self-supervised approach via random affine transformations (RATs), demonstrating remarkable performance on both image and non-image data. However, these methods encounter a notable bottleneck, the overlap of constructed labeled datasets across categories, which hampers the subsequent classifiers' ability to detect anomalies. Consequently, the creation of an effective data distribution becomes the pivotal factor for success. In this article, we introduce a model called "self-supervised forest (sForest)", which leverages the random Fourier transform (RFT) and random orthogonal rotations to craft a controlled data distribution. Our model utilizes the RFT to map input data into a new feature space. With this transformed data, we create a self-labeled training dataset using random orthogonal rotations. We theoretically prove that the data distribution formulated by our methodology is more stable compared to one derived from RATs. We then use the self-labeled dataset in a random forest (RF) classifier to distinguish between normal and anomalous data points. Comprehensive experiments conducted on both real and artificial datasets illustrate that sForest outperforms other anomaly detection methods, including distance-based, kernel-based, forest-based, and network-based benchmarks.

Memetic segmentation based on variable lag aware for multivariate time series

The memetic segmentation algorithm has been proposed to obtain precise segmentation results through efficient local refinement. However, this algorithm overlooks the consideration of time lags among variables or spatial locations in multivariate time series data. Although some researchers have adopted dynamic programming to eliminate time lags and obtain segmentation results iteratively, they have failed to achieve simultaneous optimization for both lag awareness and segmentation. To address this problem, we propose a memetic segmentation algorithm based on variable lag-aware for multivariate time series (MSVLAMTS), which enhances the encoding operation of the memetic algorithm to optimize both time-lag awareness and segmentation positions. Furthermore, MSVLAMTS incorporates the self-distribution of multivariate time series and the correlation between variables to construct a maximum likelihood function for fitness evaluation, and further improves the optimization strategy of the memetic algorithm. Experiment results on artificial and real datasets demonstrate that MSVLAMTS exhibits superior performance in both time-lag awareness and segmentation. It effectively resolves the challenge of variable time-lag awareness in multivariate time series data and enhances the accuracy of collaborative segmentation.

Enhancing perovskite solar cell performance with a novel theoretical method and artificial neural networks

One of the most promising solutions for photovoltaics in space is perovskite solar cells (PSCs), which have intrinsic microscales in thickness. PSCs have power conversion efficiencies (PCEs) that are almost 26%, however, the stability issue prevents their commercialization. One of the PSC's primary operating difficulties is phase velocity performance, which should be enhanced stability in complex scenarios. This study investigates the size-dependent dynamic behavior of PSCs made of 3D-FG material at the metal layer site. The current relevant structure in the real-world situation that takes into consideration surface boundary domains is described mathematically using the 3D-elasticity theory while taking into account all strain–stress components. By building a multi-scale dynamics framework on the modified couple stress theory, the wave propagation analysis is broadened to encompass microscale composites. Considering the possibility of solving the size-dependent motion equations using the jointed analytical–numerical techniques. In addition to being modeled mathematically, artificial neural networks are used to train the findings in order to analyze solar cells in the future for the production of green energy. The artificial neural network's training and test datasets are used in batches to create the aforementioned artificial intelligence. The results show that, with an axial radius ratio of more than 1, the panel becomes more stable because from this point the change in sensitivity of the phase velocity to the 3D functionally graded material changes only slightly and gradually. Another important outcome is that with an increase in the y-direction curvature ratio, the sensitivity decreases, but this change is maximally 10%. The study's conclusions will help improve the impact-bearing capability and safety of the PSC energy-collecting devices by enhancing their astronautic design and space deployment.