3D Vision and Digital Twin for In-process Geometric Deviation Mapping and Defect Monitoring in Near-net Shape Metal Additive Manufacturing

Cold spray additive manufacturing (CSAM) is a solid-state manufacturing technique that offers many possibilities for manufacturing complex free-form metallic parts with a high deposition rate. As a melt-less technology, it minimises oxidation and deposits solid-state coatings while maintaining the original properties of the feed material. CSAM is increasingly employed in aerospace, automotive, healthcare, and defence industries, where structural integrity and functionality are critical. CSAM technology, despite its advantages, is not fully matured. It faces significant challenges related to the geometric accuracy of components and issues with quality and consistency in production. Moreover, the complex feeding mechanism makes the process unstoppable once it is started. In the current process, system parameters are optimised by several trial-and-error experiments to estimate process parameters. However, they are material-specific, requiring repeated experiments for each material. In addition, the variability in processing parameters leads to defects such as overbuild and underbuild. More importantly, they are amplified in each layer, leading to part failure, causing significant waste of expensive raw materials and increasing overall processing costs. Addressing these challenges requires an uninterrupted in-process monitoring system providing real-time metrology and defect detection capabilities. To address these challenges, this thesis aims to develop and calibrate a three-dimensional (3D) vision system for real-time metrology, generate a digital twin of the printed part and detect geometric defects during the additive manufacturing (AM) process to ensure the dimensional accuracy of the component. A multi-sensor 3D vision system was developed as real-time metrology equipment for the CSAM process. This setup provides various benefits over using a single scanner. By strategically positioning the scanners, the blind spot in the sensor's Field Of View (FOV) is minimised, allowing for continuous capture of parts as they are constructed. To accurately combine data from the scanners and create a comprehensive 3D point cloud of the developing object, it is crucial to determine the relative transformation between the scanner frame and the robot's end-effector frame (where the plate is mounted). A novel automated hand-eye calibration technique was developed to establish this transformation relationship between the robot and each sensor. The method is efficient, requires no specialised knowledge, and can be conducted by non-expert operators. The developed 3D scanning system was employed to collect real-time profile data during the CSAM process and mapped into the global coordinate system using the hand-eye calibration result. However, the raw data collected during deposition contains dense, redundant, incremental points representing each single-track deposition. A novel geometric digital twin framework, geometric digital twin in additive manufacturing (gDT-AM), was developed for high deposition rate robotic additive manufacturing (HDRRAM). It continuously captures and maps the part's geometry in real-time during printing. It includes two experimentally validated alternative techniques for quickly and accurately reconstructing surfaces from sparse spatio-temporal two-dimensional (2D) line profiler scans. The gDT-AM framework was further enhanced by integrating a geometric deviation detection system. The deviation detection system leverages a reference model representing near-net manufactured components without sharp edges. A 3D deviation map was generated by comparing the reference model with the part digital twin model. The system also tracks the shape and size of the segmented local deviations layer by layer. A streamlined pipeline for real-time 3D reconstruction of the growing part and direct quantitative comparison with reference near-net model enables the identification of potential defects during production, facilitating timely interventions that reduce waste and ensure quality. Experiments were conducted on various complex CSAM builds to evaluate the efficacy of the proposed defect detection system. The defect data and toolpath information were correlated to identify patterns and conditions where defects occurred, closing the gap in the literature. The findings underscored the potential of a laser scanning system for defect detection and its importance in CSAM process control for the geometric accuracy of parts.<p></p>