Optimizing Maintenance Scheduling for Manufacturing Equipment Using Machine Learning Techniques

Thesis Abstract

Effective maintenance of manufacturing equipment is critical for optimizing production efficiency, minimizing downtime, and reducing operational costs. Traditional maintenance scheduling methods, often based on fixed intervals or reactive approaches, frequently result in suboptimal resource utilization and unanticipated equipment failures. This study aims to develop a data-driven, predictive maintenance framework that leverages machine learning techniques to optimize maintenance scheduling for manufacturing equipment. The specific objectives include identifying key predictive indicators of equipment failure, designing and validating machine learning models for failure prediction, and formulating an adaptive maintenance scheduling system based on model outputs to improve operational performance. The research adopts a mixed-methods approach, combining quantitative modeling with qualitative assessments. The study population comprises maintenance records, sensor data, and operational logs from a medium-sized manufacturing plant specializing in automotive parts production. A stratified random sampling technique was employed to select 500 equipment units over a three-year period, ensuring representation across different machine types and operational conditions. Data collection involved extracting historical maintenance logs, real-time sensor data capturing operational parameters such as vibration, temperature, and load, and conducting structured interviews with maintenance personnel to contextualize the data. For predictive modeling, machine learning algorithms including random forests, support vector machines (SVM), and artificial neural networks (ANN) were developed and compared using cross-validation techniques. Feature selection and engineering were performed through recursive feature elimination to identify the most significant predictors of equipment failure. The models' performance was evaluated based on metrics such as accuracy, precision, recall, F1-score, and Area Under the Receiver Operating Characteristic curve (AUC-ROC). Subsequently, the best-performing model was integrated into a simulation environment to develop an adaptive maintenance scheduling system that dynamically predicted failure probabilities and recommended timely interventions. The anticipated findings demonstrate that machine learning models can accurately forecast equipment failures with predictive accuracies exceeding 85%, significantly outperforming traditional maintenance approaches. The study is expected to reveal critical operational parameters that influence equipment health, providing actionable insights for maintenance planning. The simulation of the scheduled maintenance interventions suggests potential reductions in unplanned downtimes by up to 30%, lowering maintenance costs and enhancing production throughput. This research contributes to the body of knowledge in industrial engineering by empirically validating the application of advanced machine learning algorithms for predictive maintenance in manufacturing contexts. It bridges the gap between theoretical models and practical operational strategies, offering a scalable framework adaptable to various industrial settings. The study refines existing failure prediction theories by integrating sensor-based condition monitoring with empirical data analysis, aligning with the systems engineering and reliability-centered maintenance paradigms. In conclusion, the findings affirm that machine learning-driven maintenance scheduling optimizes resource allocation and extends equipment lifespan, thereby bolstering operational efficiency. Recommendations include the adoption of real-time sensor data analytics, continuous model updating with evolving data streams, and organizational restructuring to incorporate predictive maintenance practices. Future research should explore the integration of Internet of Things (IoT) technologies and explore their implications on maintenance decision-making processes, extending the framework's applicability and robustness.