Papers by Keyword: Prediction

Abstract: Previous research shows that predicting width deviation inherits a central importance in hot rolling processes, so that the pass planning in the hot strip mill (HSM) can be optimized. These predictions can be enabled using machine learning, complementing analytical formulations of width spread. For reliable production, it is important for the plant operator to be able to control the geometry with high accuracy across the entire plant. Therefore, the width must be accurately known throughout the entire HSM. This paper aims on the prediction of width deviation in early product stages during the roughing mill processes, where the major deformation takes place, and thus also has the most significant influence on the width spread. Therefore, this paper takes industrial data into account which is also used for the roll passing planning. To achieve a prediction during rough rolling for the width after exiting the mill (future state of the strip width), various machine learning algorithms were implemented and tested. The prediction results are evaluated against an inline width measurement, where the XGB model performs best with a Root Mean Squared Error (RMSE) of 1.11 mm. Subsequently, feature importance analyses are used to examine which features are relevant for the prediction result and to elaborate which significance process-and geometry data has on the same strip.

Abstract: Natural fiber-reinforced composites offer lightweight and sustainable alternatives for automotive and aerospace applications. However, similar to synthetic composite options, forming-induced defects such as wrinkles can reduce targeted performance. Predicting these defects is particularly challenging for natural fiber reinforcements due to the inherent variability in the fiber geometry and ensuing fabric properties. This study applies a machine learning approach to predict formability in flax woven fabrics (2×2 twill and biaxial non-crimp) using a glass fabric as a synthetic fabric reference. The fabrics’ shear, bending, tensile, and friction behaviors were experimentally characterized to capture forming-relevant mechanical properties. The fabrics were subsequently formed in single-, dual-, and triple-layer configurations over a square tool, followed by 3D scanning to quantify wrinkle distributions. Forming-induced surface deformations were transformed into grayscale maps, from which Haralick texture features were extracted. Combined with the fabric design parameters such as weave, orientation, number of layers, grammage, and thickness, the mechanical properties features were used to train linear regression models, reliably predicting select Haralick features, cross-validated using Monte Carlo simulations. Results showed that flax twill reinforcements exhibited the lowest formability, while the glass fabric formed smoothly, and biaxial non-woven fabric showed primarily localized folds. Increasing the fabric orientation from 0° to 45° improved forming performance for most woven reinforcements; but not the biaxial non-woven alternative. Linear regression models accurately predicted the defect severity via homogeneity (R² = 0.81) and dissimilarity (R² = 0.73) texture features, demonstrating that integrating texture-based image analysis with fabric parameters and mechanical properties provides a promising machine-learning-based framework for predicting surface quality upon forming of fabric-reinforced composites.

Abstract: The carbonized materials structure levels from molecular to macrostructure is analyzed. A study was conducted to find correlations between the granular carbonized materials electrical resistance and other substance physicochemical properties. It is proven theoretically and experimentally that determining the electrical resistance for a granular material, rather than a finely ground sample, is a more informative indicator for reflecting the microstructural features of the material, its reactivity, strength, and clarifying the carbonization conditions. A method is developed for determining the granular materials electrical resistance in the rotating drum interelectrode space, with determination of the indicator value in the cold and hot state for the substance under investigation. An indicator of the granular matter electrical resistance based on the heating time to 670 °C at a drum power supply constant voltage is introduced. Calculated dependencies is obtained for predicting, based on experiment results, some quality indicators for such a carbonized material as metallurgical coke: structural strength, apparent density, reactivity, gasification degree, and electrical resistivity. Better correlations is achieved with these indicators than with the standard electrical resistance on the "micropress" device, which indicates a better reflection of the carbonized materials substance supramolecular structure.

Abstract: This study is undertaken to provide a viable alternative to destructive test assessment which is prohibitively costly and makes it difficult to fully capture the bending strength distribution of timber. The bending strength of glued laminated Nigerian wood specie which from previous research have clearly not been sufficiently studied were carefully evaluated under controlled conditions. Five wood species—Afara (Terminalia superba), Akomu (Pycnanthus angolensis), Melina (Gmelina arborea), Iroko (Milicia excelsa), and Omo cedar (Stereospermum accuminatissmum)—were tested for physical and mechanical properties at a mean moisture content level of 13%. Using standard procedures, density was determined for the wood species and found to be within low and medium densities. The glulam beams were produced using Phenol resorcinol formaldehyde, Urea formaldehyde, and Polyurethane adhesives. Furthermore, the bending strength of the beams was assessed parallel and perpendicular to the glue line using central point loading. Based on the strength assessment, four variables were evaluated as determinants for bending strength. These were wood density, wood species, adhesive type, and load direction. Analysis of variance was conducted to evaluate the significance of these variables as bending strength determinants in the beams. Furthermore, a stepwise regression method was used to develop the models, from which three models emerged with density, wood species, and load direction as predictors. Finally, these models were validated using an in-sample technique by splitting the data into a validation group and a training group in a ratio of 20% to 80%, respectively. The Pearson correlation between the predicted and experimental data was 0.847, 0.917, and 0.916 in the validation group and 0.824, 0.875, and 0.877 in the training group for models 1, 2, and 3, respectively. Higher correlation with the experimental data was found in the validation group than in the training group, thereby validating the models. These models are recommended for use because of the simplicity and efficiency of prediction.

Abstract: A music chart is one way to measure the success and popularity of a song. One of the companies that presents music charts is Billboard Publication which serves as a critical reference point. However, many popular songs struggle to maintain longevity on the Billboard chart. This study focuses on predicting song longevity on the music charts, specifically the Billboard chart. The model incorporates characteristic data from previously charted popular songs on the Top 100 Billboard Chart and additional attributes from Spotify to ensure accurate predictions. The findings of this research will offer valuable insights to upcoming artists and producers by identifying the attributes they must focus on improving to enhance the popularity’s longevity of their music. Four machine learning models were utilized: Random Forest, Logistic Regression, Neural Network, and XGBoost. The tuned Random Forest model achieved an overall metric average of approximately 91.3%, followed by XGBoost with around 89.9%. These results demonstrate the effectiveness of decision tree models for this prediction task. Furthermore, artist-popularity, loudness, song-duration-ms, instrumentalness, and speechiness proved significant in this context.

Abstract: Diabetes mellitus is a chronic disease that has become a serious global health problem. The high prevalence of diabetes mellitus and the lack of public awareness of the risk of diabetes are serious problems. Early detection and prevention of this disease are important. However, early detection is often not optimized. The development of information and communication technology causes the use of technologies such as machine learning and web-based applications to be a potential solution to increase public awareness of the risks and early detection of diabetes mellitus. Therefore, this research develops a web-based diabetes mellitus prediction application using machine learning technology with history and advice features that can be used as a health assistant for the general public regarding their diabetes risk.

Abstract: This study aims to predict the hardness of electrodeposited Ni-W alloy coatings by combining machine learning methods based on a small dataset, with the goal of streamlining the trial-and-error process and reducing experimental costs. In this study, 11 features comprised of electrolyte compositions and process parameters (including current density, pH value, bath temperature, and agitation) were utilized as input parameters, with coating hardness serving as the target value. Two machine learning models (KNN and Elastic-Net) were employed to predict coating hardness, and hyperparameters were tuned using Randomized-Search CV (CV=5). The results demonstrate that the KNN model exhibits the highest predictive accuracy, with R²=0.942 and RMSE=0.0658. The SHAP method was used to analyze the importance of features and their impact on hardness. It is found that bath temperature, current density, and ammonium chloride concentration have the most significant influence on coating hardness. This study demonstrates how machine learning can enhance electrodeposition to predict coating hardness, offering insights for improving Ni-W alloy coatings in mechanical applications.

Abstract: This study presents the application of tree-based algorithms to predict springback in the V-bending process of sheet metals, particularly for SUS304 material. V-bending, a critical process in metal forming, often faces challenges due to springback, which affects dimensional accuracy and product quality. Using virtual experiments conducted via ANSYS software, the study evaluates the influence of variables such as die angle, die radius, material thickness, and punch displacement on springback. Four tree-based algorithms—Decision Trees, Random Forest, Gradient Boosting Machines (GBM), and Extra Trees—were used to predict springback, with the Decision Tree model showing the highest accuracy (Mean Absolute Error MAE=0.35; Mean Square Error=0.20). The findings indicate that incorporating predictive models into die and punch displacement selection can enhance operational efficiency and ensure consistent product quality. This methodology offers a cost-effective and efficient approach to improving the precision of sheet metal V-bending processes, with potential applications across various sheet metal materials.

Abstract: Water quality is important for both environmental sustainability and public health. This research introduces an innovative method for forecasting water quality using Random Forest Regression, optimized through Genetic Algorithm (GA) techniques. The goal is to enhance prediction accuracy and offer meaningful insights for better water resource management. The study employed the “Water Quality Data” dataset, encompassing 11 essential water quality parameters from different locations. After thorough data preprocessing, the Random Forest model, refined with GA optimization, achieved a Mean Squared Error (MSE) of 0.3476 and an accuracy rate of 91.77%, surpassing conventional methods. This approach highlights the effectiveness of merging machine learning algorithms with evolutionary optimization techniques to achieve superior predictive outcomes. Although the dataset was of moderate size, the results show considerable improvements in model accuracy. This work advances the field of water quality prediction by leveraging sophisticated algorithms and emphasizes the significance of hyperparameter tuning. Future research should focus on using larger datasets and examining the specific regions from which the data is collected.

Abstract: Stroke is the second factor of mortality in the world. According to the World Health Organization (WHO), stroke is an acute brain dysfunction. The effects of stroke are disability and mortality. Therefore, this is a concern for world health. In early 2019, the Pandemic Covid-19 attacked the world and caused many mortalities. Especially, people who have complications with diseases such as heart attack, stroke, and asthma. The purpose of this research is to predict stroke diseases with input parameters (age, glucose level, heart rate, and BMI) and to test the accuracy of the system. Moreover, analysis of the management of stroke patients’ strategy. ANFIS is a combination of ANN and FIS. It can construct a network realization of IF/THEN rules. This method was used by many researchers to predict and test the accuracy of the system. According to the result, the error of this system is 0.04 and the accuracy is 94%. Thus, it was good for predicting stroke diseases. According to the severity of the stroke, there are stroke management strategies that can be conducted by the patients; self-management and medical management. For self-management, problem-solving, goal setting, decision-making, and coping skills can help recovery. On the other way, there are five categories for medical management; stroke acute care, reperfusion, rehabilitation, cognitive decline, and neuroprotection and repair.Your manuscript will be reduced by approximately 20% by the publisher. Please keep this in mind when designing your figures and tables, etc.