Digital Twin-Based Cyber-Physical Quality System for Autonomous Manufacturing

Abstract

The paradigm of Manufacturing built on Cyber-Physical Systems (CPS) embodies a dynamic and transformative realm of knowledge, empowering the creation of intricately designed components through the precision of Computer Numerical Controlled (CNC) machines. Within this domain, the fusion of technology and production holds immense promise, boasting the capacity to analyze vast datasets. Yet, amid this promise lies a formidable challenge: ensuring the seamless maintenance of product quality and consistency throughout the CNC manufacturing process, a task rendered complex by the intricate dynamics inherent in such endeavours. In acknowledgement of this pivotal gap, the author of this thesis has discerned a critical imperative for industrial manufacturers: the adoption and strategic utilization of machine learning (ML) and deep learning (DL) technologies. Their integration stands poised to deliver real-time prognostications of manufacturing part quality with an astonishing accuracy rate of 96.58%. Prior frameworks, whether grounded in machine data, sensor data, or image data, have faltered in unifying the realms of manufacturing and ML into a cohesive entity capable of accurate quality prediction. The central objective of this thesis thus crystallizes into the development of a domain specific framework for Cyber-Physical Quality Surveillance (CPQS), nestled at the confluence of ML, DL, and manufacturing methodologies. This framework harmonizes disparate data streams from machines, sensors, and images, meticulously tailored to yield predictions of quality surpassing the 95% threshold for components forged from Advanced High-Strength Steel (AHSS) via CNC machining. To this end, three novel methodologies have been conceived and executed: Strategic Sensor Placement: Leveraging modal and finite element analyses to pinpoint optimal sensor locations, ensuring the acquisition of precise data for ML/DL analysis. This curated sensor data furnishes invaluable insights into critical parameters such as speed, feed, depth of cut, and vibration. • Integrated Feature Extraction: Effectively distilling features from machine, sensor, and image data to construct a cohesive framework capable of accurate quality predictions. Employing an XG Boost classifier, this methodology surpasses the classification performance of convolutional neural networks and ResNet models. • Tool Wear Analysis: Unveiling the nexus between tool wear and product quality by scrutinizing processed images of cutting tools via advanced image processing techniques. This approach excels in discerning nuanced information about tool wear from image data. These methodologies epitomize a synergy between sensor data and innovative data analytics, fostering interdisciplinary research and propelling technological frontiers. Integrated seamlessly into a CPQS framework deployed on Digital Twin (DT) technology, operational in real-time via the Google Colab platform, this model heralds a new era of flawlessly manufactured components, bereft of defects or imperfections.