Investigation of the data augmentation technique in an industrial problem

Cultured stone is a versatile building material that offers an economical and practical alternative to natural stone by imitating its aesthetic properties. While processing and transporting natural stone is both difficult and costly, cultured stone eliminates these challenges. It is widely used in both interior and exterior applications, including facade cladding, interior decoration, landscape design, and the restoration of historical buildings. The aim of this study is to examine the cultured stone production process and determine the optimal production parameters. A major quality issue during production is the cracking of the stones, which leads to material waste and the need for reprocessing. This negatively affects production efficiency. Therefore, identifying and preventing the causes of these cracks is crucial for improving the overall manufacturing process. In the initial phase of the research, the main process parameters affecting product quality were identified. For this purpose, the Plackett-Burman design, a method under the Design of Experiments (DOE) framework, was used. This approach allows for the rapid analysis of variable effects using a minimal number of experiments. Six different process parameters were tested at two levels, and for each combination, the percentage of cracked stones was measured and evaluated. Experimental data were analyzed using Linear Regression (LR) and Random Forest (RF) algorithms. Model performance was assessed using statistical metrics such as the coefficient of determination (R²), Mean Squared Error (MSE), and Mean Absolute Error (MAE). While both models provided satisfactory results under certain conditions, they struggled with outliers and the limited diversity of the dataset. This showed that their generalization capability was limited. To address the lack of diversity and imbalance in the data, data augmentation techniques were applied. Gaussian noise was added to the process parameters to introduce small, random variations, enabling the model to learn from a wider range of scenarios. Additionally, Bootstrap sampling was used to generate synthetic samples by repeatedly selecting subsets from the original data. This significantly increased the volume of the dataset. The results demonstrated that these data augmentation methods considerably improved model performance. Gaussian noise made the model more sensitive to slight variations in the data, while the Bootstrap technique helped reduce overfitting and increased robustness against extreme values. Furthermore, the physical and mechanical properties of cultured stones were evaluated using standard test methods, including compressive strength, splitting tensile strength, density, water absorption rate, and freeze-thaw resistance. The experimental results revealed how different production parameters affect the durability and quality of the final product. This comprehensive approach allowed producers to better understand and improve their processes using both experimental and data-driven methods. Analyzing data collected from the production line helped build predictive models to detect potential issues in advance. This proactive approach reduced production losses and offered significant advantages in terms of quality assurance, customer satisfaction, and cost control. In conclusion, this study demonstrated that machine learning and data augmentation techniques can be effective tools for process control and quality assurance in cultured stone production. However, one limitation is that augmented data still remains close to the original training data, which may restrict model generalization. In the future, integrating these methods with real-time monitoring systems and autonomous control mechanisms will contribute to the development of smart manufacturing systems in line with Industry 4.0 goals. These data-driven and automated methods not only improve cost efficiency but also support environmentally responsible industrial practices. Enhancing production quality and reducing reprocessing rates in cultured stone manufacturing can be achieved by carefully monitoring each stage of the process. Specifically, the demolding and drying phases are critical, as temperature, humidity, and duration must be precisely controlled. Otherwise, the internal structure of the concrete may not cure properly, resulting in decreased strength. Dimensional accuracy and surface quality are also evaluated through visual inspections at the end of the production line. Defects such as bubbles, cracks, and contamination are detected, and defective products are removed. This is essential for customer satisfaction, especially in exterior cladding applications where visual consistency is key. Therefore, keeping the molds clean is critical. Contaminated or deformed molds can negatively affect both the appearance and strength of the products. The integration of data science and artificial intelligence into industrial processes has gained increasing importance in recent years. The machine learning models and data augmentation methods used in this study can be applied not only to cultured stone production but also to various manufacturing fields. These technologies are also useful in factory automation, predictive maintenance, and energy efficiency monitoring. Especially for small-scale manufacturers, such technologies offer opportunities to enhance productivity and accelerate growth. The contribution of data augmentation to production quality is not limited to improving the accuracy of statistical models. It also helps systems with limited data to learn faster, handle outliers, and reduce the effects of data imbalance. Since it is not always possible to observe all scenarios in real-life production environments, synthetic data generation allows for simulating potential conditions in advance. This enables production systems to be tested and prepared for unexpected situations. The application section of this study also provided suggestions on how augmented models can be implemented in the production line. It was proposed that data collected from integrated sensors can be transmitted to the model in real time, allowing operators to be alerted or enabling the system to adjust itself automatically. This would help minimize human error, maintain quality standards, and maximize production efficiency. Finally, the methods discussed in this study support the vision of sustainable manufacturing. By predicting breakages in advance, reprocessing rates can be reduced, thus minimizing material waste and energy consumption. This results in reduced carbon emissions, lower labor costs, and significant environmental and economic benefits. These improvements offer not only short-term operational gains but also long-term advantages for corporate sustainability and environmental responsibility. This holistic approach to cultured stone production combines data science, materials science, and industrial engineering, offering an interdisciplinary solution model. Especially in developing industrial regions, such innovative approaches are essential for local manufacturers to gain competitive advantages. As similar studies are applied to other sectors, digitalization and automation in production processes will accelerate, supporting Turkey's goals for industrial transformation.