Search

Electron beam welding (EBW) is a complex process used in manufacturing high-value components in the aerospace and nuclear industries. The Fourth Industrial Revolution is a fusion of advances in artificial intelligence, sensing techniques, data science, and other technologies to improve productivity and competitiveness in fast-growing markets. Although the EBW process can be monitored by characterisation of the electron beams before welding or using backscattered electron signals (BSE), the noise and lack of understanding of these signal patterns is a major obstacle to the development of a reliable, rapid and cost-effective process analysis and control methodology. In this thesis a controlled experiment was designed to be relevant to those industries and improve understanding of the relationship between beam and weld quality. The welding quality control starts before welding, continue throughout the welding process, and is completed with examination after welding. The same workflow was followed in this thesis, focusing on aforementioned QC stages, starting with beam probing experiments, followed by monitoring weld pool stability using high dynamic range camera and BSE signals, and ending with metallographic inspection on sections. The rapid development of computer vision methods brought an idea of classifying beam probing data before welding, which is first QC stage. Dataset of 3015 BeamAssure measurements was used in combination with deep learning, and various encoding methods such as Recurrence Plots (RP), Gramian Angular Fields (GAF), and Markov Transition Fields (MTF). The segmentation and classification results achieved a remarkable rate of 97.6% of accuracy in the classification task. This part of the work showed that use of time-series images enabled identification of the beam focus location before welding and providing recommended focus adjustment value. To replicate in-process QC step, titanium alloy (Ti-6Al-4V) plates were welded with a gap opened in a step

In-process penetration monitoring of the pulsed laser welding process remains a great challenge for achieving uniform and reproducible products due to the highly complex nature of the keyhole dynamics within the intense laser-metal interactions. The main purpose of this study is to investigate the feasibility of acoustic emission (AE) measurement for penetration monitoring based on acoustic wave characteristics and deep learning. Firstly, a series of laser welding experiments on aluminum alloys were conducted using high-speed photography and AE techniques. This allowed us to in-situ visualize the complete keyhole dynamics and elucidate the generation mechanism of acoustic waves originating from pressure fluctuations at the keyhole wall. Then, an adaptive time-frequency technique namely VMD (Variational Mode Decomposition) was proposed to characterize the acoustic energy distribution among the nine subsignals with low-frequency and high-frequency components under different welding penetrations. Lastly, a novel hybrid model combing CNN (Convolutional Neural Network) and LSTM (Long Short Term Memory) was designed to deeply mine the spatial and temporal acoustic features from the extracted frequency components. Extensive experiments demonstrate that our proposed approach yields a remarkable classification performance with a test accuracy of 99.8% and a standard deviation of 0.21, which obtains a high recognition rate. This work is a new paradigm in the digitization and intelligence of the laser welding process and contributes to an alternative way of developing an efficient end-to-end penetration monitoring system.

Terahertz time-domain spectroscopy has experienced significant progress in imaging, spectroscopy, and quality inspection, e.g., for semiconductor packaging or the automotive industry. Additive manufacturing alloys (also known as alloys for use in 3D printing) have risen in popularity in aerospace and biomedical industries due to the ability to fabricate intricate designs and shapes with high precision using materials with customized mechanical properties. However, these 3D-printed elements need to be polished thereafter, where the surface roughness is inspected using techniques such as the laser scanning microscope. In this study, we demonstrate the use of terahertz time-domain spectroscopy to assess the average roughness profile and height leveling of stainless steel for comparisons against the same parameters acquired using laser scanning microscopy. Our results highlight the potential of the proposed technique to rapidly inspect 3D-printed alloys over large areas, thus providing an attractive modality for assessing surface profiles of AM-manufactured terahertz components in the future.

Over the past two decades, laser additive manufacturing technology has evolved rapidly and has been applied in many industrial sectors [...]

The viscous layer generated on the anode surface during electrochem-ical polishing is essential for obtaining a smooth, mirror-like surface. However, the growth of the viscous layer for polishing Laser-Powder Bed Fusion (L-PBF) components with rough surface features of several hundred microns in height and wavelength differs from the micron/nanoscale surface finish. This study employed a Spatial Frequency Method to model rough L-PBF surfaces. The effect of height and frequency distribution parameters on the viscous layer thickness, geometry and uniformity was numerically investigated. NaCl-Ethylene Glycol-Ethanol and commercial A2 electrolytes were utilised to polish L-PBF 316L stainless steel for verification purposes. The results show that the viscous layer of the high-frequency surfaces is more homogeneous than the low-frequency surfaces, and the roughness reduction can reach 97.6% compared to 64% for the low-frequency surfaces.

The necessity for precise prediction of penetration depth in the context of electron beam welding (EBW) cannot be overstated. Traditional statistical methodologies, including regression analysis and neural networks, often necessitate a considerable investment of both time and financial resources to produce results that meet acceptable standards. To address these challenges, this study introduces a novel approach for predicting EBW penetration depth that synergistically combines computational fluid dynamics (CFD) modelling with artificial neural networks (ANN). The CFD modelling technique was proven to be highly effective, yielding predictions with an average absolute percentage deviation of around 8%. This level of accuracy is consistent across a linear electron beam (EB) power range spanning from 86 J/mm to 324 J/mm. One of the most compelling advantages of this integrated approach is its efficiency. By leveraging the capabilities of CFD and ANN, the need for extensive and costly preliminary testing is effectively eliminated, thereby reducing both the time and financial outlay typically associated with such predictive modelling. Furthermore, the versatility of this approach is demonstrated by its adaptability to other types of EB machines, made possible through the application of the beam characterisation method outlined in the research. With the implementation of the models introduced in this study, practitioners can exert effective control over the quality of EBW welds. This is achieved by fine-tuning key variables, including but not limited to the beam power, beam radius, and the speed of travel during the welding process.

Metallic powder bed fusion additive manufactured components have been extensively applied in the fields of aerospace, mobility, construction, etc. However, poor surface quality owing to residual powder, stair-step structure, un-melted track, etc. hinder its application in the market. Electrochemical polishing (EP) is a promising method for smoothing metal surfaces without inducing extra mechanical damage to the product while some disadvantages appeared when being applied on the Laser-Powder Bed Fusion components including low polishing efficiency, poor geometry control, and high cost of the experiment investigation. To solve the problems, this thesis investigated the EP effect of L-PBF 316L stainless steel (316L SS) and TC4 (Ti-6Al-4V) utilising the methods of numerical simulation, experiment, and machine learning. Firstly, a novel 2-dimensional EP model based on the Finite Element Method utilising the Spatial Frequency Method was proposed to simulate the viscous layer formation process with the consideration of the high surface roughness. In addition, the effect of parameters including diffusion coefficient, inlet velocity, inter-electrode distance, and more importantly, the surface textures on the thickness and uniformity of the viscous layer formation were investigated. Based on the simulation parameters, the EP effect of the current density ranging between 250 - 2000 mA/cm2 on the surface roughness, morphology, weight loss, and geometry changes was investigated, and a two-step EP process was proposed for optimisation. The experiment results were adopted to machine learning with six algorisms including the Adaptive Boosting algorithm, Random Forest, Multilayer Perceptron Regression, Ridge Regression, Support Vector Regression, and Classification and Regression Trees. Simulation results showed that the diffusion coefficient should be smaller than 1.010-7 m2/s to generate the viscous layer. The conditions of 0 mm/s inlet velocity, at least 3 mm inter-electrode dis