Surfaces generated by Additive Manufacturing or laser texturing can involve the solidification of droplets of liquid, which can give rise to overhanging features on the solidified surface. Overhanging features add a layer of complexity to the surface topography and are undetectable by standard surface roughness measurement techniques such as profilometry. Such features are important because they can have a considerable effect on surface properties such as wettability. New techniques and algorithms are therefore required to analyse and quantify convoluted surfaces with overhanging (re-entrant) features. Earlier work by the authors introduced the concept of using X-ray micro-computed tomography (Micro-CT) to identify the directions of vectors normal to the surface at any point and thus indicate the presence or absence of overhanging features. This paper divides overhanging features into two types; simple and compound, and introduces new, size independent, analysis techniques which measure what proportion of each type is on the surface. Another extension of the analysis is the comparison of surface profiles taken in different directions in order to identify any surface roughness anisotropies.

Three-dimensional particle positioning from inline holograms is performed using convolutional neural networks. The faster R-CNN architecture is implemented for multi-particle identification and lateral positioning, and a second network estimates the depth position. Supervised learning is used to train the network using simulated holograms.

Convoluted rough surfaces involving overhanging features can be a natural consequence of laser additive manufacturing and other spray techniques or can be generated deliberately by laser surface texturing, e.g., to aid osseointegration. Overhanging features add an extra level of complexity to the topography of a rough surface and can have a substantial effect on wettability, etc. However, features of this type are invisible to traditional surface roughness measurement techniques. This work presents a computer-based surface analysis method that gives useful information about the presence and nature of overhanging features on rough technical surfaces. The technique uses micro-computer tomography to generate a typical cross section of the surface under investigation. The angles of the vectors normal to the surface can then be analyzed to reveal the presence of overhanging features, which can also be indicated by the standard deviation of the normal vector distribution. Titanium surfaces generated by six different techniques were compared. The characteristics of these surfaces varied strongly, as did the shapes of the overhangs involved. These variations are reflected by different statistical distributions of the normal vectors.

This thesis can be conveniently divided into three sections as follows;

Part I. Monitoring of laser welding 

In laser materials processing there has always been a need for suitable methods to supervise and monitor the processes on-line, to ensure correct production quality or to trigger alarms when failures are detected. Numerous investigations have been made in this field, including experimental and theoretical work. It is common practice in this field to monitor surface temperature, plasma radiation and back-reflected laser light, coaxially with the laser beam.  Traditionally, the monitoring systems involved carry out no statistical analysis of the signals received – they merely involve thresholds.

The first two papers in this thesis look at the feedback collected during laser welding using a co-axial setup from a Digital Signal Processing point of view and also uses high speed video photography to correlate signal perturbations with process anomalies. Digital signal processing techniques such as Kalman filtering, Principal Component Analysis and Cluster Analysis have been applied to on-line measurement data and have generated new ways to describe laser welding behaviour using parameters such as reflected pulse shape. The limitations of commercially available welding supervision systems have been studied and design suggestions for the next generation of on-line weld monitoring equipment have been formulated.

Having progressed from thin section welding with continuous wave lasers to pulsed laser welding, the thesis then moves on to pulsed laser surface melting.

Part II Analysis of surface texturing of titanium.

The second part of the thesis concentrates on laser structuring of titanium surfaces for medical implants. The two papers in this section present an analysis of the laser-material interactions which create surfaces suitable for osseointegration (bone attachment). The work concentrates on a commercially available surface used for screw implants in dentistry; BioHelix™. This surface is generated by an intense bombardment of laser pulses and the surface is thus disrupted during solidification. The formation of various levels and types of roughness are analysed and it is noted that laser generated rough surfaces are fundamentally different from those with a mechanically produced roughness. One key point is that laser generated rough surfaces can include overhanging features. This finding lead to the research carried out in part III of the thesis.

Part III. Analysis and classification of laser generated surfaces.

The final section of the thesis presents research which uses statistical techniques to identify whether or not a roughened surface includes overhanging features. The presence or otherwise of such features is important because they can affect the wettability of surfaces and thus their suitability for implant surfaces, adhesive bonding and lubrication etc. Micro Computer Tomography was used to generate a typical cross section of the surface under investigation. At equally spaced positions the profile of this cross section is then allocated vectors which are normal to the profile at each point. The angles of the vectors can then be analyzed to reveal the presence or otherwise of overhanging features. The presence of overhangs on the material surface is indicated by the existence of normal vectors with angles that exceed 180°. The papers in this section also investigate possibilities and limitations of using statistical methods in conjunction with Micro Computer Tomography.

The papers have in common processing of data for laser materials processing, by advanced methods to identify and extract essential information from the processes and the resulting material properties.

Accelerated bone grow (osseointegration) can be achieved by modifying the surface of medical implants. For this purpose, pulsed lasers can be used to successfully texture such beneficial surfaces on titanium, e.g. a BioHelix™ structure. This surface typically includes ridges and droplets with a size range between 1 and 20 μm. This paper presents the results of an experimental program where a range of laser parameters was used to create different surface textures on titanium substrates, using pulsed laser spattering. The resultant surfaces are analysed by scanning electron microscope and X-ray Micro Computer Tomography. It is shown that optimisation of the laser parameters results in a robust process which produces a surface that has proven to be beneficial for osseointegration. The results are also deeper analysed, explaining how different types of surface are created by the laser-material interaction under different conditions. Further, droplet flight distances and the formation of the spongeous nano-scale surface that characterizes the surface structure depends on very fast cooling and is also evaluated.

Pulsed lasers can be used to modify the surface of medical implants in order to accelerate bone growth (osseointegration). A surface covered in attached droplets with diameters between 1 and 20 μm is a beneficial surface for rapid osseointegration. This paper presents the results of an experimental program in which a broad range of laser parameters and different atmospheres were used to create different surface textures on titanium substrates, including the desired "attached droplet" topology. The resulting surfaces were analyzed by scanning electron microscopy and micro-computer tomography. The paper explains how different types of surfaces are created by the laser-material interaction under different conditions and focus characteristics. It is shown that optimization of the laser parameters results in a robust process, which produces a surface that is fundamentally different from those created by nonlaser methods.

Several techniques can be used to modify implant surfaces in order to accelerate bone growth around titanium implants. One method is to generate a surface structure which stimulates bone growth and remodelling. This paper describes and explains a non-ablative method for producing osseointegrating (structural and functional bone bonding) surfaces on titanium implants using laser processing. The focus is especially on surface texturing of dental implant screws, where the ability of a Nd:YAG laser to generate ‘splashy’ surfaces covered in resolidified micro scale droplets coated with nano-scale surface oxides is assessed. The surfaces produced were analysed by scanning electron microscope (SEM) and energy dispersive x-ray spectroscopy (EDS). It is concluded that laser processing using Q-pulsed Nd:YAG lasers can generate surfaces which match the demands set by clinical experience. One important characteristic of the surfaces discussed here is that they involve overhanging features which are suitable for trapping red blood cells and which cannot be created by mechanical or chemical roughening techniques.

Laser-assisted cold-spray has been recognized for over a decade as a technique capable of depositing high quality coatings. By laser heating (and hence softening) the surface being coated, deposition can occur at particle velocities lower than those normally associated with the cold spray process. This can be used to increase deposition rate. However, it can also be used to facilitate the deposition of higher hardness material combinations, normally more out of the reach of the conventional cold spray process. Laser heating can also reduce the requirements of the process on gas usage and gas heating for a given combination, making it more cost-effective. In the work reported below, the capability of a novel co-axially laser-assisted system (COLA) to deposit higher hardness materials, relevant to a range of different industrial applications, has been evaluated. This system can be retrofitted to conventional cold spray equipment. 

This paper describes two new techniques for monitoring the quality of laser welds by statistical analysis of the reflected light signal from the weld surface. The first technique involves an algorithm that analyses the variance of the peak values of the reflected signal as a measure of the stability of the surface during pulsed Nd:YAG laser welding in the heat conduction mode. Kalman filtering is used to separate a useful signal from the background noise. A good correlation between weld disruption and signal fluctuation has been identified. This technique could be used in tandem with the present practice of simply using the peak values of reflected (or emitted) light as an indicator of weld quality. The second technique investigated involves an assessment of the temporal shape of the power distribution of individual reflected pulses in comparison with an average of the results from a high quality weld. Once again a high correlation between a poor signal match and inferior quality welding was discovered, which may pave the way towards a new generation of optical weld monitoring devices.

This paper considers the point that it is not possible to interpret individual weld perturbations from the raw electromagnetic feedback collected from laser weld zones. The presentation of electromagnetic data as a 3D cloud is presented as a new, useful tool in the analysis of this feedback. It is shown that there is a very low correlation between the plasma or thermal signals and the reflected light signal from the weld zone, and that a strong correlation exists between the plasma and thermal signals. It is also demonstrated that data points from a weld perturbation form a different 3D cluster to those from the stable welding process. A strategy for future real time data analysis is presented utilising a suitably shaped data cloud envelope. The rates of data fit to the various segments of such an envelope could be correlated with specific weld anomalies.