Automated Homogeneity Inspection of U3O8-Al Dispersion Fuel Plates Using X-ray Radiography, Deep Learning and Chimp Optimization Algorithms

Zeinab F. Elsharkawy¹, Refaat M. Fikry¹, Nadia M. Nawwar¹, A. G. Ahmed², Mohammed E. El-Bedawy²

¹ Engineering Department, Nuclear Research Center, Egyptian Atomic Energy Authority (EAEA), Egypt
² Nuclear Metallurgy Department, Nuclear Research Center, Egyptian Atomic Energy Authority (EAEA), Egypt

Abstract

The uniform distribution of U3O8 particles within an aluminum matrix is crucial for the optimal performance and safety of plate-type nuclear fuel produced by the dispersion technique. X-ray radiography is a primary non-destructive inspection method for evaluating U3O8-Al dispersion fuel plate homogeneity, wherein an X-ray beam passes through the fuel plates and creates images based on differential material density absorption, revealing the internal particle distribution and enabling identification of inhomogeneities.

This study presents an automated inspection system that combines deep learning with metaheuristic optimization algorithms to classify fuel plates as "homogeneous" or "non-homogeneous" (including observed and defected categories). Multiple optimization techniques were evaluated, with the Chimp Optimization Algorithm (CHOA) selected for its superior performance. Leveraging a pre-trained EfficientNet for feature extraction and CHOA for feature selection, the system identifies optimal discriminative features, while a support vector machine (SVM) classifier—optimized via CHOA—achieves classification accuracies of 99.77% for binary classification and 99.72% for ternary classification. This approach significantly improves inspection reliability and efficiency, ensuring robust nuclear fuel quality control.

Keywords: X-ray radiography, U₃O₈-Al fuel plates, deep learning, Chimp Optimization Algorithm (CHOA), automated inspection, nuclear quality control

1. Introduction

Since 1998, the Fuel Manufactured Pilot Plant (FMPP) has been producing plate-type nuclear fuel for the Egyptian Second Research Reactor (ETRR-2). This fuel is a U3O8-Al dispersion type (19.75% U235), where U3O8 serves as the chemical compound containing fissile material dispersed within an aluminum matrix, forming a composite material encased in 6061 nuclear-grade aluminum alloy cladding using the "picture frame" technique \cite{1}.

The distribution of U3O8 within the aluminum powder matrix is critical for fuel performance during reactor operation. Uranium homogeneity—the uniform distribution of uranium within a nuclear fuel plate—is essential for consistent fuel behavior, optimal reactor performance, and safety. The distribution of U3O8 particles directly impacts heat generation patterns; inhomogeneities lead to localized hotspots, increasing thermal stresses and risking fuel clad failure. Non-uniform distribution also affects reactivity and neutron moderation, reducing fuel efficiency and creating operational challenges \cite{2,3}.

X-ray radiography provides a valuable non-destructive tool for examining U3O8-Al dispersion fuel plate homogeneity \cite{4,5}. In this technique, an X-ray source emits high-energy electromagnetic radiation that passes through the fuel plates and is absorbed differentially based on material density and thickness. Areas with higher density (uranium oxide particles) absorb more X-rays, reducing detector intensity, while lower-density regions (aluminum matrix) allow greater transmission. The resulting image reveals internal structure variations, enabling inspectors to identify particle distribution patterns. Uniform distributions appear as consistent patterns, while accumulations or inhomogeneities manifest as deviations. Transmission densitometry, which measures transmitted light intensities through a small aperture using a photoelectric sensor, is commonly employed for quantitative evaluation \cite{6}.

Examples of fuel meats with good and acceptable homogeneity are shown in Fig. 1 [FIGURE:1], representing reference radiographs based on field experience. Variations in uranium density within specific fuel meat areas are reflected in corresponding densitometer readings, with statistical analysis performed on collected data to determine homogeneity levels \cite{7}. While X-ray radiography captures detailed internal images without physical alteration (Fig. 2 [FIGURE:2]), traditional manual assessment is time-consuming, subjective, and suffers from declining accuracy in high-volume production. These limitations have driven adoption of machine learning (ML) and deep learning (DL) systems, which demonstrate superior detection accuracy, processing speed, and consistency.

Recent advancements in computer vision and artificial intelligence show significant potential for automating defect detection and material characterization. Deep convolutional neural networks (CNNs) \cite{8,9} can extract high-level features from radiography images for accurate defect classification, though their effectiveness depends heavily on optimal feature selection and classifier optimization. Traditional dimensionality reduction techniques like Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) \cite{24} often struggle with high-dimensional data's nonlinear relationships. Metaheuristic optimization algorithms \cite{10-16}, including the Chimp Optimization Algorithm (CHOA) \cite{17-20}, have shown remarkable success in enhancing ML models by efficiently navigating complex search spaces to identify optimal features and parameters.

This study presents an automated inspection framework combining X-ray radiography, deep learning, and metaheuristic optimization for high-accuracy classification of U₃O₈-Al fuel plates. The system employs pre-trained EfficientNet for robust feature extraction, CHOA-based feature selection to identify discriminative attributes, and CHOA-optimized SVM to categorize plates into binary ("Homogeneous" vs. "Non-Homogeneous") and ternary ("Regular," "Observed," or "Defected") classes. Rigorous evaluation against alternative optimization techniques demonstrates CHOA's superior performance and highest classification accuracies.

Key Contributions:

1. Automated Feature Extraction: An effective deep learning-based method (EfficientNet) that improves detection reliability while eliminating manual inspection subjectivity and inefficiencies.

2. Robust Feature Selection: Integration of CHOA to enhance feature discriminability, reducing computational complexity without sacrificing accuracy.

3. Optimized SVM Parameters: CHOA is employed to select optimal SVM parameters that maximize classification accuracy.

4. High-Performance Classification: A comparative study of optimization algorithms confirming CHOA's effectiveness in nuclear fuel plate inspection.

5. Scalable Quality Control: A robust framework applicable to industrial-scale nuclear fuel manufacturing, representing a significant advancement in nuclear fuel quality assurance.

2. Related Work

Automated inspection of nuclear fuel plates has evolved significantly over two decades, leveraging advancements in non-destructive testing (NDT), image processing, and machine learning. This section reviews key methodologies, focusing on X-ray radiography, texture analysis, deep learning, and optimization algorithms.

X-ray radiography has long served as the primary NDT method for evaluating U₃O₈-Al dispersion fuel plate homogeneity. Early approaches relied on manual expert inspection using transmission densitometers to measure radiographic film density variations correlated with uranium particle distribution \cite{6,7}. While effective, this method was labor-intensive and subjective, causing inconsistencies in large-scale production.

To overcome manual inspection limitations, researchers increasingly adopted ML techniques. Early automated systems employed texture analysis, with Keyvan et al. \cite{21} applying supervised neural networks (backpropagation and fuzzy ARTMAP) and unsupervised ART2-A for pellet inspection, finding supervised learning outperformed unsupervised methods. Berbar et al. \cite{22} proposed a system using Haralick's Gray Level Co-occurrence Matrix (GLCM) to extract textural features (angular second moment, entropy, contrast) from U₃O₈-Al fuel plate X-ray images, achieving 96% accuracy with a Backpropagation Neural Network (BPNN). However, this approach was limited by handcrafted features that may not generalize to complex or subtle defects.

Recent studies have shifted toward deep learning for higher-accuracy defect detection. Guo et al. \cite{23} applied Faster R-CNN with ResNet-101 to detect scratches on fuel assemblies, achieving 98% true positive rate (TPR) at 10% false positive rate (FPR), highlighting CNNs' superiority in capturing spatial features. These studies underscored DL's potential while identifying challenges such as the need for large annotated datasets and computational resources.

ML-based inspection system performance critically depends on optimal feature selection and model parameter tuning. While traditional techniques like PCA and LDA \cite{24} demonstrate utility in simple applications, their linear assumptions often fail to capture high-dimensional data's complex nonlinear relationships. Metaheuristic optimization algorithms have emerged as powerful alternatives, including Whale Optimization Algorithm (WOA), Dragonfly Algorithm (DA) \cite{11,15}, Salp Swarm Algorithm (SSA) \cite{10,15}, Sine-Cosine Algorithm (SCA) \cite{11,15}, and Grey Wolf Optimizer (GWO) \cite{13-15}, all showing promise in feature selection and classifier optimization. Among these, CHOA demonstrates particular efficacy due to superior convergence speed and balanced exploration-exploitation dynamics, making it especially suitable for high-dimensional optimization challenges in defect classification \cite{17-20}.

Current nuclear fuel inspection systems face three fundamental limitations: (1) manual feature dependency in traditional GLCM-based approaches restricting adaptability to novel defect types, (2) computational inefficiency of deep learning models hindering real-time deployment, and (3) oversimplified binary classification frameworks lacking granularity for multi-defect analysis. This study proposes an integrated framework combining EfficientNet-based deep feature extraction for robust pattern recognition, CHOA-optimized feature selection to enhance discriminative capability while maintaining computational efficiency, and CHOA-tuned SVM classification achieving superior accuracy in both binary (defective/non-defective) and ternary (regular/observed/defected) tasks, thereby addressing all three limitations simultaneously.

3. Methodology

The proposed system leverages EfficientNet-B0 as a deep feature extractor, ensuring robust pattern recognition through its highly optimized architecture. CHOA then refines extracted features to enhance discriminative capability while minimizing computational overhead, retaining only the most relevant features to improve classification efficiency. Subsequently, CHOA fine-tunes SVM hyperparameters, optimizing key parameters through a dual-stage framework that significantly boosts performance, achieving high accuracy in both binary (Hom/Non-Hom) and ternary (regular/observed/defected) classification tasks.

3.1 EfficientNet-B0

EfficientNet-B0 serves as an optimal feature extractor for nuclear fuel plate inspection, combining computational efficiency with robust pattern recognition \cite{25-27}. As the EfficientNet family baseline, it employs compound scaling to balance depth, width, and resolution. The architecture consists of an initial 3×3 convolution followed by 16 MBConv layers (mixing 3×3 and 5×5 kernels) and global max pooling, outputting a 1280-D feature vector from raw images (Fig. 3 [FIGURE:3]). With only 5.3M parameters, its pre-trained weights enable effective transfer learning, extracting hierarchical features directly from U₃O₈-Al radiographs without preprocessing or large datasets. The network's ability to capture multi-scale spatial patterns makes it ideal for nuclear fuel inspection tasks where defect manifestations vary significantly in size and morphology. By leveraging EfficientNet's deep feature representations, the system reduces manual feature engineering reliance while improving defect detection robustness.

3.2 Hybrid CHOA-SVM Optimization Framework

The Chimp Optimization Algorithm (CHOA) is a population-based metaheuristic inspired by chimpanzees' intelligent social and hunting behaviors \cite{17-20}. It demonstrates strong global optimization capabilities particularly suited for high-dimensional search spaces, making it effective for feature selection and parameter tuning in machine learning tasks. CHOA operates in a two-stage hierarchical optimization process to enhance SVM performance for nuclear fuel plate defect classification.

In Stage 1 (Feature Selection), CHOA operates on a binary-encoded population where each chimp represents a candidate feature subset. The fitness of each subset is evaluated by training an SVM with default parameters and measuring cross-validation accuracy. CHOA's position update rules—modeling group hunting dynamics with chaotic exploration—iteratively refine feature subsets until convergence, effectively eliminating redundant or noisy features. Stage 2 (Parameter Optimization) then optimizes SVM hyperparameters (e.g., regularization strength C, kernel coefficient γ) using continuous-valued chimps, where fitness is assessed via cross-validation on the optimal feature subset identified in Stage 1. The algorithm's adaptive chaos factor balances exploration of the parameter space with exploitation of promising regions. Finally, the selected features and tuned parameters train a high-performance SVM classifier, achieving superior generalization by simultaneously addressing dimensionality reduction and model calibration. This hybrid approach leverages CHOA's strengths in both discrete (feature selection) and continuous (parameter optimization) search spaces while maintaining computational efficiency through parallelizable fitness evaluations. The pseudo-code for the proposed Hybrid CHOA-SVM is shown in Fig. 4 [FIGURE:4].

4. Experimental Results and Discussion

4.1 Data Description

Experiments were conducted using a dataset of fuel plate images manufactured for the ETRR-2 reactor. The classification task was performed under two scenarios: binary (Homogeneous vs. Non-Homogeneous) and ternary (Regular, Observed, or Defected). A meticulously curated dataset was utilized, with data augmentation applied to improve model generalization through vertical/horizontal flipping, blurring, cropping, and random rotations (20° to 280°). The binary classification task employed 4,330 images (2,570 homogeneous and 1,760 non-homogeneous), while the ternary task used 3,570 images (1,810 regular, 1,496 observed, and 264 defected).

4.2 Evaluation Metrics

Classification accuracy (CA), confusion matrix (CM), area under the receiver operating characteristics curve (AUC), precision (P), specificity (Sp), recall (R), and F1-score (F1) were used to evaluate model performance. These metrics are calculated as follows:

• Precision: ( P = \frac{TP}{TP + FP} )
• Recall: ( R = \frac{TP}{TP + FN} )
• Specificity: ( Sp = \frac{TN}{TN + FP} )
• F1-score: ( F1 = 2 \times \frac{P \times R}{P + R} )
• Classification Accuracy: ( CA = \frac{TP + TN}{TP + TN + FP + FN} )

where TP, FP, TN, and FN represent true positives, false positives, true negatives, and false negatives, respectively.

4.3 Results and Discussion

To achieve optimal performance, the optimized SVM was fed with robust, minimal feature sets acquired by CHOA using various optimization procedures. All experiments were conducted on an Intel® Core™ i7-10870H 8-core processor with 16GB RAM and NVIDIA GeForce RTX 3060 GPU, running Windows 11 with MATLAB 2024a.

The CHOA implementation used a population size of 50 chimps and 100 iterations. The dataset was partitioned into 90% for training (with 10-fold cross-validation) and 10% for testing. Deep features were extracted using EfficientNet-B0, followed by feature selection. Six metaheuristic algorithms—SSA, WOA, GWO, DA, SCA, and CHOA—were employed to optimize feature selection and SVM hyperparameters, with performance systematically compared.

Binary Classification Performance: As shown in Table 1 [TABLE:1], CHOA achieved the most efficient feature selection, reducing the initial 1,028 features to only 191 (an 81.4% reduction) while yielding the highest classification accuracy for both feature selection and SVM optimization. Table 2 [TABLE:2] compares binary classification performance across algorithms. Although SSA exhibited the highest training CA, CHOA demonstrated superior testing performance and overall robustness. The confusion matrix and ROC curves (Figs. 5 [FIGURE:5] and 6 [FIGURE:6]) validate model effectiveness: only one homogeneous plate image was misclassified while all non-homogeneous images were correctly identified, with AUC values of 1.0 for both classes confirming near-perfect performance.

Ternary Classification Performance: For the three-class scenario (Regular, Observed, Defected), CHOA again outperformed competing algorithms, selecting only 315 features while achieving the highest CA in both feature selection and SVM optimization (Table 3 [TABLE:3]). When these features trained an optimized SVM, CHOA-based classification surpassed all other methods (Table 4 [TABLE:4]), reducing the feature space by 69.4% (315/1,028 features) and enhancing computational efficiency without sacrificing accuracy.

The proposed model's three-class performance is detailed in Table 5 [TABLE:5], demonstrating 100% accuracy for Regular plates with high performance maintained for other classes. The confusion matrix (Fig. 7 [FIGURE:7]) shows correct classification of all 1,810 Regular and 1,496 Observed plate images, with only 1 of 26 Defected images misclassified. ROC curves (Fig. 8 [FIGURE:8]) confirm robustness with AUC values of 1.0 (Regular), 0.9967 (Observed), and 0.9985 (Defected), indicating strong discriminative power across all classes.

Comparative Analysis: Table 6 [TABLE:6] summarizes key studies in nuclear fuel plate inspection, emphasizing methodologies, datasets, and performance metrics. The proposed model advances existing approaches by integrating optimized feature selection with multi-class SVM classification, achieving superior accuracy and generalizability compared to prior work.

5. Conclusion

This study presented an advanced automated inspection system for evaluating U₃O₈-Al dispersion fuel plate homogeneity by integrating X-ray radiography, deep learning, and metaheuristic optimization. The framework leverages EfficientNet-B0 for robust deep feature extraction, followed by CHOA for optimal feature selection and SVM hyperparameter tuning. The system achieved 99.77% binary accuracy (Hom/NonHom) and 99.72% ternary accuracy (Regular/Observed/Defected), outperforming competing algorithms (SSA, WOA, GWO, DA, SCA). CHOA reduced features by 81.4% (binary) and 69.4% (ternary), enhancing computational efficiency without sacrificing accuracy. The model demonstrated near-perfect discriminative capability with 100% accuracy for Regular plates and 99.72% for Defected plates, validated by AUC values of 1.0, 0.9967, and 0.9985 for respective classes. This approach addresses critical limitations—manual feature dependency, computational inefficiency, and oversimplified binary classification—offering a fast, reliable, and automated solution for nuclear fuel quality assurance.

References

[1] W. Z. a. I. Elseaidy, "General Description and Production Lines Of The Egyptian Fuel Manufacturing Pilot Plant," in IGORR 7: 7th meeting of the International Group On Research Reactors, Germany, 1999: IAEA.

[2] I. A. E. AGENCY, Standardization of Specifications and Inspection Procedures for Leu Plate-type Research Reactors (Report of A Consultants meeting, Geesthacht, 16-18 April 1986). Vienna: IAEA-TECDOC-467, IAEA, 1988.

[3] H. B. Peacock, "Properties of U sub 3 O sub 8 -aluminum cermet fuel," 01 October 1989. [Online]. Available: https://doi.org/10.2172/7262763

[4] K. L. Ali et al., "Development of low enriched uranium target plates by thermo-mechanical processing of UAl2–Al matrix for production of 99Mo in Pakistan," Nuclear Engineering and Design, vol. 255, pp. 77-85, 2013/02/01/ 2013, doi: https://doi.org/10.1016/j.nucengdes.2012.10.014.

[5] M. Durazzo, E. Vieira, E. F. Urano de Carvalho, and H. G. Riella, "Evolution of fuel plate parameters during deformation in rolling," Journal of Nuclear Materials, vol. 197-210, https://doi.org/10.1016/j.jnucmat.2017.04.018, 2017/07/01/ 2017.

[6] R. A. Salinas, U. Raff, and H. Coronado, "Uranium Density Measurements and Homogeneity Assessment in Quality Control of Low Enriched 235U Fuel Plates Using Machine Vision," Measurement and Control, vol. 36, no. 10, pp. 305-308, 2003/12/01 2003, doi: 10.1177/002029400303601002.

[7] N. A. Mansour, M. E. El-Bedawy, and M. M. Ghoneim, "Effect of delay time between mixing and compacting for producing nuclear fuel plates," Journal of Radiation Research and Applied Sciences, vol. 14, no. 1, pp. 429-439, 2021/12/01/ 2021, doi: https://doi.org/10.1080/16878507.2021.1987127.

[8] A. S. Elkorany and Z. F. Elsharkawy, "Efficient breast cancer mammograms diagnosis using three deep neural networks and term variance," Scientific Reports, no. 0123456789, pp. 1-12, 2023, doi: 10.1038/s41598-023-29875-4.

[9] A. S. Elkorany and Z. F. Elsharkawy, "Optik COVIDetection-Net: A tailored COVID-19 detection from chest radiography images using deep learning," Optik, 166405-166405, February, 2021, doi: 10.1016/j.ijleo.2021.166405.

[10] A. S. Elkorany and Z. F. Elsharkawy, "Automated optimized classification techniques for magnetic resonance brain images," Multimedia Tools and Applications, vol. 79, no. 37-38, pp. 27791-27814, 2020, doi: 10.1007/s11042-020-09493-2.

[11] A. S. Elkorany, M. Marey, K. M. Almustafa, and Z. F. Elsharkawy, "Breast Cancer Diagnosis Using Support Vector Machines Optimized by Whale Optimization and Dragonfly Algorithms," IEEE Access, vol. 10, no. June, pp. 1-1, 2022, doi: 10.1109/access.2022.3186021.

[12] Z. F. Elsharkawy, A. S. Elkorany, and S. M. Elhalafawy, "Deep Learning-Based Classification of Magnetic Resonance Brain Images," vol. 34, no. 2, 2025.

[13] H. M. Ahmed, B. A. B. Youssef, A. S. Elkorany, Z. F. Elsharkawy, A. A. Saleeb, and F. A. El-Samie, "Hybridized classification approach for magnetic resonance brain images using gray wolf optimizer and support vector machine," Multimedia Tools and Applications, vol. 78, no. 19, pp. 27983-28002, 2019, doi: 10.1007/s11042-019-07876-8.

[14] H. M. Ahmed, B. A. B. Youssef, A. S. Elkorany, A. A. Saleeb, and F. Abd El-Samie, "Hybrid gray wolf optimizer–artificial neural network classification approach for magnetic resonance brain images," Appl. Opt., vol. 57, no. 7, pp. B25-B31, 2018/03/01 2018, doi: 10.1364/AO.57.000B25.

[15] A. B. Gabis, Y. Meraihi, S. Mirjalili, and A. Ramdane-Cherif, "A comprehensive survey of sine cosine algorithm: variants and applications," Artificial Intelligence Review, vol. 54, no. 8, pp. 5469-5540, 2021/10/01 2021, doi: 10.1007/s10462-021-10026-y.

[16] A. S. Elkorany and Z. F. Elsharkawy, "Deep Learning Based Classification of Magnetic Resonance Brain Images using YOLOv5 and Lion Swarm Optimization," Menoufia Journal of Electronic Engineering Research (MJEER), pp. 12-19, 2025, doi: 10.21608/mjeer.2025.325982.1096.

[17] M. Khishe and M. R. Mosavi, "Chimp optimization algorithm," Expert Systems with Applications, vol. 113338-113338, 2020, doi: 10.1016/j.eswa.2020.113338.

[18] W. Lei, K. Jia, X. Zhang, and Y. Lei, "Research on Chaotic Chimp Optimization Algorithm Based on Adaptive Tuning and Its Optimization for Engineering Application," Journal of Sensors, vol. 2023, 2023, doi: 10.1155/2023/5567629.

[19] S. Khosravi and A. Chalechale, "Chimp Optimization Algorithm to Optimize a Convolutional Neural Network for Recognizing Persian/Arabic Handwritten Words," Mathematical Problems in Engineering, vol. 2022, no. 1, p. 4894922, 2022/01/01 2022, doi: https://doi.org/10.1155/2022/4894922.

[20] L. Deng and S. Liu, "Exposing the chimp optimization algorithm: A misleading metaheuristic technique with structural bias," Applied Soft Computing, vol. 158, 2024/06/01/ 2024, doi: https://doi.org/10.1016/j.asoc.2024.111574.

[21] S. Keyvan, X. Song, and M. Kelly, "Nuclear fuel pellet inspection using artificial neural networks," Journal of Nuclear Materials, vol. 264, no. 1, pp. 141-154, 1999/01/01/ 1999, doi: https://doi.org/10.1016/S0022-3115(98)00464-4.

[22] K. H. Mohamed, B. M. Abdou, E. S. Mohamed, and H. A. Hassan, "Automated Image Analysis System for Homogeneity Evaluation of Nuclear Fuel Plates," in 2005 International Conference on Neural Networks and Brain, 13-15 Oct. 2005 2005, vol. 2, pp. 1178-1183, doi: 10.1109/ICNNB.2005.1614825.

[23] Z. Guo et al., "Defect detection of nuclear fuel assembly based on deep neural network," Annals of Nuclear Energy, vol. 137, p. 107078, 2020/03/01/ 2020, doi: https://doi.org/10.1016/j.anucene.2019.107078.

[24] H. F. Aamir Khan, "Principal Component Analysis-Linear Discriminant Analysis Feature Extractor for Pattern Recognition," International Journal of Computer Science Issues, vol. 8, no. 6, pp. 267-270, 2011.

[25] P. Bansal, K. Gehlot, A. Singhal, and A. Gupta, "Automatic detection of osteosarcoma based on integrated features and feature selection using binary arithmetic optimization algorithm," Multimedia Tools and Applications, vol. 81, no. 6, pp. 8807-8834, 2022/03/01 2022, doi: 10.1007/s11042-022-11949-6.

[26] M. M. U. S. Shakin et al., "Squeeze and Excitation Attention Meets Modified EfficientNetB0 Architecture: Multi-Class Brain Tumor Classification Using Explainable Artificial Intelligence," in 2023 26th International Conference on Computer and Information Technology (ICCIT), 13-15 Dec. 2023 2023, pp. 1-6, doi: 10.1109/ICCIT60459.2023.10441407.

[27] C. Su and W. Wang, "Concrete Cracks Detection Using Convolutional Neural Network Based on Transfer Learning," Mathematical Problems in Engineering, vol. 2020, no. 1, p. 7240129, 2020/01/01 2020, doi: https://doi.org/10.1155/2020/7240129.