Evaluation of predictive maintenance efficiency with the comparison of machine learning models in machining production process in brake industry.

Background: The utilization of technologies such as artificial intelligence (AI) and machine learning (ML) in industrial sectors has become a crucial requirement to enhance the efficiency and stability of production processes. Regular maintenance of machines and early detection of faults play a critical role in ensuring uninterrupted production and business continuity. Predictive maintenance practices, combined with sensors and data analysis methods, enable the collection, analysis, and transformation of machine-related data into meaningful insights. As a result, the anticipation of potential machine failures, the execution of planned maintenance activities, and the prevention of unexpected downtime become possible. These methods not only improve productivity in production processes but also contribute to reducing maintenance costs. Methods: This study aims to predict machine faults using data analysis methods and enhance the accuracy performance of these predictions for an industrial company that produces braking components. Comprehensive examination and analysis of data were conducted to understand the symptoms and relationships of machine failures. ML classification methods were employed in the relevant study. Results: Challenges such as the imbalance of class distributions in the dataset, the presence of missing and outlier values, and the high costs of necessary equipment and training pose significant barriers to implementation. Addressing these issues is critical for achieving effective predictive maintenance solutions. In order to achieve more accurate results, data splitting and k-fold cross-validation methods were applied during the learning and testing phases to overcome the imbalance problem in the dataset, undersampling techniques were applied, and outlier detection and normalization processes were used to improve data quality. The model performances, evaluated through accuracy, precision, recall, and F1-score, area under the curve (AUC), Cohen's Matthew's correlation coefficient (MCC) were compared. Hyperparameter optimization was also performed, resulting in significant improvements in model performance. This study contributes to the literature in terms of predictive maintenance application, classification, and data partitioning techniques. The findings highlight the importance of data preprocessing and advanced modeling techniques in predictive maintenance and emphasize how addressing data challenges can enhance the overall performance and reliability of ML models.