PhD.dissertartion:ARASH EBRAHIMI ARAGHIZAD
MACHINE LEARNING-BASED MODELING AND MONITORING OF MACHINING PROCESSES AND TOOL WEAR
ARASH EBRAHIMI ARAGHIZAD
Manufacturing Engineering Ph.D. Dissertation, July 2024
Thesis Jury
Prof. Erhan Budak (Thesis Advisor),
Prof. Mustafa Bakkal
Assoc.Prof. Lütfi Taner Tunç
Assoc.Prof. Bekir Bediz
Assoc.Prof. Umut Karagüzel
Assoc. Prof. Kemal Kılıç (Thesis Co-Advisor)
Date & Time: 23rd JULY 2024 – 15:30
Place: FASS G049
Keywords : Intelligent Monitoring – Machine Learning – Hybrid Milling Modeling– Tool Wear
Abstract
Milling processes are a basis of manufacturing across a variety of industries, including aerospace, automotive, and heavy machinery. Effective monitoring of these processes is essential to ensuring high-quality production, minimizing downtime, and extending tool life. Inefficient monitoring can lead to defects in machined parts, excessive tool wear, and ultimately, significant financial and material waste. Traditionally, the monitoring of milling operations has relied heavily on direct human supervision, periodic testing, and sensor-based monitoring systems, which are based on time-consuming teach cycles. These existing monitoring systems, mainly targeted at tool condition monitoring, cannot accurately identify the source of variation or fault, making them both error-prone and resource-intensive. This thesis seeks to revolutionize this traditional approach by implementing advanced machine learning (ML) techniques integrated with physics-based simulations, significantly reducing the reliance on extensive physical testing and manual oversight.
The enhanced monitoring capabilities offered by these advanced technologies enable precise control over the milling process, ensuring optimal tool engagement and machining parameters. This leads to improved consistency in product quality and a substantial reduction in waste, which is crucial for maintaining competitiveness in fast-paced markets. Moreover, intelligent monitoring systems can predict tool wear and potential failures before they occur, allowing for preemptive maintenance and scheduling. This not only extends the lifespan of milling equipment but also ensures continuous production without unexpected interruptions, thereby enhancing overall manufacturing efficiency.
The first section of the thesis presents a series of innovative hybrid models, known as physics-informed machine learning (PIML), which excel in predicting milling forces, tool wear, and tool-related faults. By combining limited experimental data with detailed simulation outputs, these models achieve predictive accuracies up to 98%. Demonstrated across various materials and tool configurations, the models’ adaptability and scalability underscore their potential for widespread industrial application.
Subsequent sections elaborate on the development of an advanced fault detection system, designed specifically for real-time applications in unmanned manufacturing environments. This system, employing refined force models and sophisticated machine learning algorithms, not only detects deviations with over 96% accuracy but also pinpoints the source of these faults. By accurately identifying not just the occurrence of anomalies but also their origins, the system enables targeted interventions, thereby optimizing manufacturing processes and significantly reducing operational costs. This dual capability of detection and precise source identification enhances the system’s effectiveness in maintaining continuous production flow and minimizing downtime.
Additionally, the thesis explores tool wear prediction using hybrid modeling approaches that integrate mechanistic insights with diverse ML algorithms. This approach significantly reduces the need for extensive wear testing, facilitating a more streamlined and economically viable monitoring process.
In conclusion, the research presented in this thesis not only advances the field of intelligent manufacturing monitoring by providing robust predictive tools but also establishes a solid foundation for future enhancements. By transforming traditional monitoring methods into more intelligent, efficient, and adaptive systems, this work pioneers a new era of manufacturing that aligns with the demands of modern industry.