Real-time digital twin and machine learning solutions for hole quality and tool condition monitoring in robotic drilling of composite materials

Robotic drilling is increasingly used in manufacturing, especially in the aerospace industry, due to its flexibility, reach, and efficiency. Unlike traditional CNC machines, robotic systems can handle complex geometries and large-scale drilling tasks needed for aircraft components such as wings. The use of carbon fibre-reinforced polymer (CFRP) in aviation has grown due to its high strength-to-weight ratio, but drilling CFRP is challenging because of its anisotropic and heterogeneous nature, leading to defects such as delamination and fibre pull-out.

To address these challenges, precise monitoring and control of the drilling process are essential, making digital twin technology an ideal solution. A digital twin is a digital representation of a physical system that interacts in real-time through data exchange, enabling continuous monitoring, predictive maintenance, and process optimisation. This technology can enhance the efficiency and quality of robotic drilling by detecting and preventing defects before they occur.

This thesis introduces a digital twin framework for robotic drilling. Initially, a generic reference model outlines the critical components of robotic drilling operations. A detailed digital twin architecture is then established according to ISO 23247 standards. To demonstrate the framework’s capabilities, a real-time visualisation system for monitoring drilling parameters is implemented, serving as a foundational step toward a fully functional digital twin system.

To evaluate the quality of drilled holes in-situ, a hybrid classification model was developed as part of a unified product digital twin. The classifier, trained and tested with a random selection of drilled holes, achieved approximately 90% overall prediction accuracy on unseen holes. This machine learning approach, using a convolutional neural network and support vector machine classifier, can improve inspection reliability while reducing production time for drilled composite components.

Subsequently, a process digital twin combining a machine learning model and real-time sensor data of a robotic drill was developed to estimate hole quality and tool condition. An ensemble neural network model, combining an artificial neural network with a genetic algorithm, was used to assess drilled hole quality. The model was tested on the machined holes to relate process input parameters and drilling torques to hole quality. Model predictions were validated with six unseen datasets, with five predicted accurately. A full factorial study using analysis of variance (ANOVA) showed that tool condition is the largest contributor to drilling torque. This real-time monitoring method can improve manufacturing productivity and ensure high-quality drilled components.

This thesis on developing both product and process digital twins represents a crucial step towards a fully unified digital twin system. The developed methods, integrating both the product and manufacturing processes, have shown potential in enhancing the precision, efficiency, and quality of robotic drilling operations. By combining real-time data with advanced machine learning models, the approach not only improves in-situ inspection and predictive maintenance but also streamlines manufacturing workflows. As the research progresses, this digital twin framework paves the way for more sophisticated, interconnected systems that can revolutionise production practices in aerospace and other industries.