Browsing by Author "Çiftçioglu, AÖ"

Now showing 1 - 8 of 8
  • No Thumbnail Available
    Item
    Fire resistance evaluation through synthetic fire tests and generative adversarial networks
    Çiftçioglu, AÖ; Naser, MZ
    This paper introduces a machine learning approach to address the challenge of limited data resulting from costly and time-consuming fire experiments by enlarging small fire test data sets and predicting the fire resistance of reinforced concrete columns. Our approach begins by creating deep learning models, namely generative adversarial networks and variational autoencoders, to learn the spatial distribution of real fire tests. We then use these models to generate synthetic tabular samples that closely resemble realistic fire resistance values for reinforced concrete columns. The generated data are employed to train state-of-the-art machine learning techniques, including Extreme Gradient Boost, Light Gradient Boosting Machine, Categorical Boosting Algorithm, Support Vector Regression, Random Forest, Decision Tree, Multiple Linear Regression, Polynomial Regression, Support Vector Machine, Kernel Support Vector Machine, Naive Bayes, and K-Nearest Neighbors, which can predict the fire resistance of the columns through regression and classification. Machine learning analyses achieved highly accurate predictions of fire resistance values, outperforming traditional models that relied solely on limited experimental data. Our study highlights the potential for using machine learning and deep learning analyses to revolutionize the field of structural engineering by improving the accuracy and efficiency of fire resistance evaluations while reducing the reliance on costly and time-consuming experiments.
  • No Thumbnail Available
    Item
    Estimation of Natural Frequencies of Pipe-Fluid-Mass System by Using Causal Discovery Algorithm
    Dagli, BY; Ergut, A; Çiftçioglu, AÖ
    This paper employs a novel approach to investigating the dynamic behavior of a pipe conveying fluid and the relationship between the variables that influence it, based on causal inference. The pipe is modeled as a beam with Rayleigh beam theory and Hamilton's variation principle is demonstrated to obtain the equation of motion. Concentrated mass at various locations is introduced using the Dirac delta function. The fluid in the pipe has no compression properties and no viscosity. The non-dimensional equations of motion of the pipe-fluid-mass system are achieved by using the approach of the fluid-structure interaction problem. The non-dimensional partial differential equations of motion are converted into matrix equations and the values of natural frequencies are obtained by using the Finite Differences Method. The relationship between the variables is investigated by causal discovery using the produced natural vibration frequencies dataset. Moreover, the Bayesian Network's probability distribution is fitted to the discretized data using the structural model created through causal discovery, resulting in trustworthy predictions without the need for sophisticated analysis. The findings highlighted that the proposed causal discovery can be an alternative practical way for real-time applications of pipe conveying fluid systems.
  • No Thumbnail Available
    Item
    Exploring failure mechanisms in reinforced concrete slab-column joints: Machine learning and causal analysis
    Çiftçioglu, AÖ
    Reinforced concrete slab-column construction comprises interconnected slabs and columns that constitute the structural system of a building. While providing architectural flexibility and ease of construction, these types of structures are prone to failure because the structural arrangement beneath the slabs is not always considered thoroughly. This research uses machine learning models to conduct an in-depth study and categorizes failure modes into three main types: flexure, punching, and combined flexure-punching modes. These failure modes are classified with appreciable accuracy by eight machine learning approaches: RAGN-L, Random Forest, Extra Trees, K-Nearest Neighbors, Adaptive Boosting, Support Vector Machine, Logistic Regression, and Gaussian Naive Bayes classifiers, optimized using hyperparameter tuning. The results indicate that the RAGN-L achieves the highest accuracy at 0.99, followed by the Random Forest model with an accuracy of 0.98. The study extends the machine learning analysis by investigating the deep causes that rule the complex interactions among key structural parameters. SHAP analysis revealed the influence of features like slab thickness, reinforcement ratio, and punching shear strength on failure modes. Counterfactual analyses further revealed how changes in these parameters can change failure modes and indicate their sensitivity and robustness. The results imply that reducing or optimizing certain parameter values will change the sample types and thus make them change between failure modes. By combining machine learning, SHAP analysis, causal analysis, and counterfactual methods, this study offers valuable insights into the failure mechanisms of slab-column joints and provides actionable recommendations to enhance structural safety and reliability.
  • No Thumbnail Available
    Item
    MACHINE LEARNING BASED PREDICTION OF COMPRESSIVE STRENGTH IN CONCRETE INCORPORATING SYNHTHETIC FIBERS
    Erdem, RT; Çiftçioglu, AÖ; Gücüyen, E; Kantar, E
    Different types of fibers are added to the concrete mixture to improve its behavior under different loading cases. This study intends to investigate the compressive strength of concrete cubic samples in which synthetic macro fibers are added in different amounts. For this purpose, a total of 72 cubic samples are produced in the experimental program. Axial pressure test is applied to cubic samples and 7 and 28 days compressive strength values are obtained in the end. However, a lot of effort has been spent to complete the time-consuming laboratory tests. To overcome this situation, four machine learning methods-Xgboost, Random Forest, Decision Tree, and Multiple Linear Regression-are adapted for efficient compressive strength forecasting. Moreover, four metrics are employed for a more meaningful evaluation of models: R2, RMSE, MAE, and MAPE. Remarkably, all models achieved R2 values exceeding 90%, with Xgboost notably reaching an impressive R2 value of 97%. This highlights the effectiveness of integrating machine learning in predicting compressive strength, offering a viable alternative to traditional laboratory tests. Incorporating the Shapley Additive exPlanation (SHAP) method, the study provides a detailed analysis of the models' interpretability. SHAP analysis revealed that Day and Fiber have been identified as crucial features influencing compressive strength predictions. Localized SHAP analyses for specific samples further enhanced the understanding of individual predictions, emphasizing the practicality and transparency of machine learning in structural engineering. The promising results of this study indicate the potential for further advancements in enhancing performance, utilizing machine learning insights.
  • No Thumbnail Available
    Item
    RAGN-L: A stacked ensemble learning technique for classification of Fire-Resistant columns
    Çiftçioglu, AÖ
    One of the main challenges in using reinforced concrete materials in structures is to comprehend their fire resistance. The assessment of fire resistance can be performed in a laboratory environment using fire. However, such tests are time-consuming and expensive, and they may not provide a complete assessment of all relevant properties of a particular tested specimen. To that end, the implementation of machine learning (ML) in the investigation of fire-resistant structural performance would be beneficial, as it would also contribute to the reduction of time and cost problems related to traditional techniques. Here, this research proposes a novel ensemble ML approach to classify columns according to their fire resistance characteristics, supporting the application of ML techniques by fire engineers and scientists. The proposed model, named RAGN-L, combines Random Forest, Adaptive Boosting, and Gradient Naive Bayes, and is stacked using the Logistic Regression approach. RAGN-L is evaluated on real-world databases of reinforced concrete columns and concrete-filled steel tube columns, as well as a synthetic database generated by the TVAE deep learning model. The performance of the proposed solution is compared with ten different ML classifiers based on common statistical metrics, accu-racy, precision, recall, and f1-score, and validated using the k-fold cross-validation approach. The developed algorithm outperforms ten different classifiers in all databases, with classification accuracies of 86.6%, and 99.6% for the real-world and synthetic databases of reinforced concrete columns, respectively, and 88.1% for the real-world database of concrete-filled steel tube columns.
  • No Thumbnail Available
    Item
    Gradient-based grey wolf optimizer with Gaussian walk: Application in modelling and prediction of the COVID-19 pandemic
    Khalilpourazari, S; Doulabi, HH; Çiftçioglu, AÖ; Weber, GW
    This research proposes a new type of Grey Wolf optimizer named Gradient-based Grey Wolf Optimizer (GGWO). Using gradient information, we accelerated the convergence of the algorithm that enables us to solve well-known complex benchmark functions optimally for the first time in this field. We also used the Gaussian walk and Le ' vy flight to improve the exploration and exploitation capabilities of the GGWO to avoid trapping in local optima. We apply the suggested method to several benchmark functions to show its efficiency. The outcomes reveal that our algorithm performs superior to most existing algorithms in the literature in most benchmarks. Moreover, we apply our algorithm for predicting the COVID-19 pandemic in the US. Since the prediction of the epidemic is a complicated task due to its stochastic nature, presenting efficient methods to solve the problem is vital. Since the healthcare system has a limited capacity, it is essential to predict the pandemic's future trend to avoid overload. Our results predict that the US will have almost 16 million cases by the end of November. The upcoming peak in the number of infected, ICU admitted cases would be mid-to-end November. In the end, we proposed several managerial insights that will help the policymakers have a clearer vision about the growth of COVID-19 and avoid equipment shortages in healthcare systems.
  • No Thumbnail Available
    Item
    Designing energy-efficient high-precision multi-pass turning processes via robust optimization and artificial intelligence
    Khalilpourazari, S; Khalilpourazary, S; Çiftçioglu, AÖ; Weber, GW
    This paper suggests a novel robust formulation designed for optimizing the parameters of the turning process in an uncertain environment for the first time. The aim is to achieve the lowest energy consumption and highest precision. With this aim, the current paper considers uncertain parameters, objective functions, and constraints in the offered mathematical model. We proposed several uncertain models and validated the results in real-world case studies. In addition, several artificial intelligence-based solution techniques are designed to solve the complex nonlinear problem. We determined the most efficient solution approach by solving various test problems. Then, simulated several scenarios to demonstrate the robustness of our results. The results showed that the solutions provided by the offered model significantly reduce energy consumption in different setups. To ensure the reliability of the results, we carried out worst-case sensitivity analyses and found the most critical parameters. The results of the worst-case analyses indicated that the offered robust model is efficient and saves a significant amount of energy comparing to traditional models. It is shown that the provided solution by the presented robust formulation is reliable in all situations and results in the lowest energy and the best machining precision.
  • No Thumbnail Available
    Item
    COVID-19 chest CT scan image segmentation based on chaotic gravitational search algorithm
    Rather, SA; Das, S; Çiftçioglu, AÖ
    Image segmentation is a pivotal phase in the image processing pipeline, offering detailed insights into various image features. Traditional segmentation methods grapple with challenges such as local minima and premature convergence when navigating intricate pixel search spaces. Additionally, these algorithms experience prolonged processing times as the number of threshold levels increases. To mitigate these issues, we implemented the Chaotic Gravitational Search Algorithm (CGSA), a robust optimizer, for the multi-level thresholding of COVID-19 chest CT scan images. CGSA amalgamates the Gravitational Search Algorithm (GSA) for exploration with chaotic maps for exploitation. Kapur's entropy method was employed to partition the sample images based on optimal pixel values. The segmentation performance of CGSA was rigorously assessed on various COVID-19 chest CT scan imaging datasets from Kaggle, utilizing metrics such as Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Feature Similarity Index Measure (FSIM). The qualitative analysis encompassed convergence curves, segmented graphs, colormap images, histogram curves, and boxplots. Statistical validation was conducted using the signed Wilcoxon rank sum test, and eight sophisticated heuristic algorithms were enlisted for comparative analysis. The comprehensive evaluation unequivocally demonstrated CGSA's superiority in terms of processing time efficiency and its ability to provide optimal values for image quality metrics, establishing it as a powerful tool for quickly assessing COVID-19 disease severity.