Pilot capability evaluation of a feedback electronic imaging system prototype for in-process monitoring in electron beam additive manufacturing

Electron beam additive manufacturing (EBAM) is an additive manufacturing (AM) technique increasingly used by many industrial sectors, including medical and aerospace industries. The application of this technology is still, however, challenged by many technical barriers. One of the major issues is the lack of process monitoring and control system to monitor process repeatability and component quality reproducibility. Various techniques, mainly involving infrared (IR) and optical cameras, have been employed in previous attempts to study the quality of the EBAM process. However, all attempts lack the flexibility to zoom-in and focus on multiple regions of the processing area. In this paper, a digital electronic imaging system prototype and a piece of macroscopic process quality analysis software are presented. The prototype aims to provide flexibility in magnifications and the selection of fields of view (FOV). The software aims to monitor the EBAM process on a layer-by-layer basis. Digital electronic images were generated by detecting both secondary electrons (SE) and backscattered electrons (BSE) originating from interactions between the machine electron beam and the processing area using specially designed hardware. Prototype capability experiments, software verification and demonstration were conducted at room temperature on the top layer of an EBAM test build. Digital images of different magnifications and FOVs were generated. The upper range of the magnification achieved in the experiments was 95 and the demonstration verified the potential ability of the software to be applied in process monitoring. It is believed that the prototype and software have significant potential to be used for in-process EBAM monitoring in many manufacturing sectors. This study is thought to be the necessary precursor for future work which will establish whether the concept is suited to working under in-process EBAM operating conditions.


Introduction
Electron beam additive manufacturing (EBAM) is an additive manufacturing (AM) technique that makes use of an accelerated electron beam to melt metallic powder on a layer-by-layer basis, manufacturing components based on computer-aided design (CAD) models [1]. The ability of the EBAM process to form components from metallic powder arises from electron interactions with metallic materials. In the EBAM process, an electron beam is accelerated by an anode, focused onto a powder bed by an electromagnetic focusing coil and subsequently deflected to specific locations by an electromagnetic deflection coil, the electrons penetrate the powder grains, whereupon they decelerate, converting their kinetic energy into thermal energy. If the energy input is sufficient, the temperature of the powder particles rises above their melting point and the particles melt. When the beam is raster-scanned across the preheated powder bed in a tightly controlled, predefined pattern, melt tracks are solidified to form fully dense cross sections of the desired single or multi-components constituting the build. This process is repeated with the additional requirement that the underlying solid is also partially melted to ensure adequate bonding between the underlying and newly formed layers, ensuring that full density is achieved [1].
EBAM shows great promise in the manufacture of orthopaedic implants and aerospace components. In an evaluation study on powder-based EBAM technology, it has been concluded that the EBAM process would enable the manufacture of a wide range of complex and difficult-to-fabricate aerospace and biomedical components [2]. The increased design freedom of the EBAM process enables the economic manufacture of porous bone ingrowth surfaces for orthopaedic implants [3] whilst the reduced thermal residual stress and the high-vacuum process environment is beneficial for the production of aircraft components [4]. Nevertheless, both of these industries are highly regulated and their current standard manufacturing processes are well established [5,6]. Despite the perceived benefits of the EBAM process, the transition from current standard manufacturing techniques to a layered manufacturing approach will not be possible unless many current technology gaps, including having a rigorous in-process monitoring and validation system for realtime control of the EBAM process, are bridged [7].
Academic research groups have built various monitoring systems to assess the quality of the EBAM process. These systems are thermal/optical imaging-based, involving the use of either IR (wavelength between 700 nm and 1 mm), visible light (wavelength between 400 nm and 700 nm) or X-ray (wavelength between 0.01 nm and 10 nm). Table 1 summarises the monitoring systems developed.
Each of the systems reported in Table 1 has many useful features; however, significant limitations exist in all current systems. Table 2 summarises these limitations with particular emphasis on the requirements of EBAM.
We postulate that the use of a digital electronic imaging system, following the methodology used in electron beam welding (EBW) and digital scanning electron microscopes (SEMs), would tackle the major functional inadequacies in existing thermal/optical data collection systems described previously.
EBW is a technique which processes materials with a focused, accelerated electron beam. EBW is commonly used for joining metals (including refractory metals) due to the available concentrated thermal energy, up to 10 8 W/cm 2 [27], delivered by the electron beam. During EBW, process artefacts including secondary electrons (SE), backscattered electrons (BSE) and/or electrons in the plasma plume are generated above the welding zone [28]. By capturing some or all of these artefacts, electronic images [27,29] and electron signal-time series plots [28][29][30][31] can be generated post and during EBW. Electron beam focus quality [28] and weld-quality attributes including weld-seam quality [27] and keyhole depth [32] are commonly monitored in industry.
In thermal and optical imaging, the adjustments of FOVand magnification are either not possible or involve movements of physical lenses. On the contrary, when conducting electronic imaging (by the collection of SE and/or BSE) with a SEM, the change in FOV and magnification only requires scanning the primary electron beam over a different region of interest (ROI) [33]. Given this flexibility and the ability for electronic imaging to reveal the topography of a conductive area [34], the incorporation of an electronic imaging system in the EBAM process will bridge the technology gap and contribute to the effective live EBAM process monitoring.  [16][17][18][19][20] Electron beam profile and position X-ray detection X-ray Semiconductor detector [21]  Monitoring-irrelevant areas lead to an unnecessary increase in dataset size and data processing time [8,10,23] Inflexible FOV Only capable of either monitoring the whole processing area or one fixed region [24][25][26] Inflexible magnification Unable to switch from monitoring the full processing area to zoom-in to specific regions [24] 2 Electronic imaging system design and experiments 2.1 Electronic imaging system prototype A bespoke digital electronic imaging system prototype [35], shown in Fig. 1a, was used for experiments investigating image magnification and FOV. The in-house developed prototype consists of an electron sensor (Fig. 1b), a signal differential amplifier (Fig. 1c), data logger ( Fig. 1d), an image generation software and a standard computer. It was designed to generate digital images from the secondary electrons (SE) and backscattered electron (BSE), hereafter collectively referred to as feedback electrons, originating from the interactions between a primary electron beam and an imaging target. The electron sensor was made of electronically conductive materials; amplifier and data logger were put together with off-the-shelf, standard electronic components and an Fig. 1 Schematic of the a electronic imaging system prototype [35], b electron sensor, c signal differential amplifier and d data logger Fig. 2

Experimental arrangements
The prototype was interfaced with a commercial EBAM machine (Arcam A1-GE Additive, USA), and two sets of singlelayer electronic imaging experiments were conducted at room temperature. The first set aimed to generate images with a range of magnifications whilst the second concerned image FOV. The two sets of experiments were carried out on a standard electron imaging test build [35] shown in Fig. 2. The test build was designed to investigate various electronic imaging abilities including image contrast, magnification and spatial resolution. The tests build contained gaps to trap sintered powder and contained geometric features including squares and holes for image features extraction purposes. The test build was designed to have nine imaging locations, each with a dimension of 60 mm × 60 mm, representing nine virtual components packed into an EBAM processing area of 210 mm × 210 mm. In a typical industrial EBAM build, multiple components are packed into the processing area to maximise productivity; Fig. 3 shows an example of an EBAM hip prosthesis design and its EBAM build setup extracted from the literature [36]. This paper focuses on (1) image magnification and (2) image FOV. The geometric features of the test build will not be discussed in this study. Table 3 details the settings of the prototype whilst Table 4 gives the EBAM machine settings and the range of monitoring areas involved in the two sets of experiments. The electron beam current and scan speed presented in Table 4 were dictated by the signal amplifier gain, data logger sampling rate and input range    Table 3. The imaging areas were chosen based on the design of the test build, i.e. dimension and design of the imaging locations.
For the evaluation of image magnification, the machine primary electron beam raster-scanned across a range of imaging areas across imaging location 5 of the test build. When investigating the FOV, the beam raster-scanned across different ROIs with the same imaging area, covering the full machine processing area. Table 5 gives the magnification and FOV tests setup. With regard to trials on image FOV, only the 60 mm × 60 mm image area was chosen to be involved in this pilot FOV experiment as a demonstration of the prototype capability. Based on the test build depicted in Fig. 2a, a

Macroscopic process quality analysis
This section presents the development, verification and demonstration setup of an image quality analysis software.

Software development
In a typical EBAM build, multiple components are packed into the processing area to maximise productivity. Carrying out EBAM process monitoring with electronic imaging opens up the opportunity to monitor individual areas of components by zooming in to the corresponding ROI to achieve the required FOV. A piece of software was developed in an open-source programming language, Python, aiming to assess individual component quality on a layer-by-layer basis on a macroscopic scale. Only the bulk quality was designed to be assessed, i.e. detailed component geometry with edge detection was not involved. This macroscopic quality analysis targeted process anomalies and defects from over-melting, peeled-off metallisation [37] and lack of powder deposition in the local processing area. At the beginning of the analysis, the software generated a stack of two-dimensional (2D) reference images from a threedimensional (3D) stereolithography (STL) design according to layer height and user-defined ROI. It then compared the reference images with the workpiece images from the same layer height and ROI generated by the electronic imaging system prototype. In order to evaluate the macroscopic process quality, both the reference and workpiece images were first binarised to form black and white images. Comparison between these two images was made and the differences between the two were expressed as white pixels, in the resultant image. Bulk process quality was quantitatively assessed by evaluating the ratio of the number of white pixels to that of black pixels in the resultant binary image. Figure 4 describes the software process flow for an arbitrary layer.

Software verification
Once the software was developed, a benchtop verification experiment was carried out to prove the correct design and operation of the software. In the experiment, a verification "reference-virtual workpiece" image set was prepared. The imaging location 5 area from the Ti-6Al-4V test build design was chosen to be the ROI for verification. A reference image was generated from this location and a virtual workpiece image was generated from the same ROI with the 8 mm × 8 mm square on the top right-hand corner removed artificially. Figure 5 depicts how this virtual workpiece image was generated from the imaging location 5 test build design. Macroscopic process quality analysis was conducted on the "reference-virtual workpiece" image set. The virtual workpiece image was overlapped onto the reference image and the difference between the images was evaluated and shown as white pixels in a resultant image. If the

Software demonstration
A demonstration trial was carried out with the test build depicted in Fig. 2a. The demonstration aimed to showcase the software ability to monitor a local ROI, i.e. the individual imaging locations 1-9 (representing individual component in a real EBAM build). Upon completion, the test build was removed from the EBAM chamber with subsequent removal of excess sintered powder. The test build was replaced in the chamber and further electronic imaging was carried out with the prototype as shown in Fig. 1a-d. During imaging, the electron beam raster-scanned across the 60 mm × 60 mm ROI of location 5 and a workpiece image revealing the real topographical features of location 5. A reference image covering the same ROI was extracted from the top layer of the test build STL design by the software. The software then carried out macroscopic process quality analysis to evaluate the quality of the top layer by overlapping the workpiece image onto the reference image.

Results
This section presents the experimental results of electronic image generation and macroscopic process quality analysis.

Digital electronic image generation time
Digital electronic image generation utilised a piece of C/C++ software receiving data from the data logger, i.e. Arduino DUE microcontroller break-out board (depicted in Fig. 1d), and a piece of Python software to conduct image pixel intensity allocation. Figure 6 is a screenshot from the computer running the software and displays a typical operational environment during electronic imaging generation. Table 6 summarises a typical set of image generation times. It shows that the typical total image generation time is less than 2 s with the settings described in Table 3.

Digital image processing
Raw digital images were generated from the electronic imaging experiments. Image noise was removed by applying a median filter, and image contrast was enhanced by carrying out histogram equalisation. is the pixel value of the filtered image at (x,y) g(s,t) is the pixel value of the raw image at (s,t) S xy represents the set of coordinates within a user-defined area of an image is the cumulative probability distribution of the normalised histogram y k is an integer, the equalised number of pixel with a pixel value of k

Image magnification
A range of image magnifications was achieved in the first set of experiments. Table 7 gives details on the achievement and Fig. 7a-d shows a typical set of processed images.

Image FOV
The prototype also generated images from different FOVs in the second set of experiment. Nine images were generated from the nine imaging locations of the standard electronic imaging test build. Figure 8a-d presents a selected set of processed images.  Fig. 9a, as depicted in Fig. 5. Table 8 gives a quantitative summary of the macroscopic process quality analysis. The image size of the resultant image, Fig.  9c, is 1800 pixel × 1800 pixel, representing an image ROI of 60 mm × 60 mm whilst the difference in the number of white pixels found is 57,600. This difference in white pixel corresponds to an image area of 64 mm 2 across the image ROI. Figure 10a shows a raw electronic workpiece image of the top surface of the build at location 5. This raw image has gone through noise removal and histogram equalisation by the application of Eqs. 1 and 2. After thresholding the processed image, the binary workpiece image, shown in Fig. 10b, has been generated. Figure 10c shows a binary reference image of the top layer of imaging location 5, covering the same image ROI as that of the workpiece image. It is extracted from the test build STL design. Macroscopic process quality analysis was carried out with the software and Fig. 10d is the resultant image. The white pixels in this bitmap reflect the deviations between the binary workpiece image and the binary reference image. Table 9 gives a quantitative summary of the macroscopic process quality analysis. The size of the resultant image, Fig. 10d, is 1800 pixel × 1800 pixel, representing an image ROI of 60 mm × 60 mm whilst the difference in the number of white pixels found is 31,891. This difference in white pixel corresponds to an image area of 35.4 mm 2 across the image ROI.

Discussion
In this section, discussions on image generation time, FOV, magnification, and the macroscopic process quality analysis are presented. Table 6 shows that the typical image generation time is 1.37 s, and is additional to the EBAM process layer time. As EBAM build design varies, no single EBAM layer time could be used as a reference. In order to estimate the effect of the addition time attributing to image generation in a real-life situation, two different EBAM designs were brought in as a demonstrator design to carry out two sets of estimation. Table 10 summarises the EBAM build report obtained from the Arcam A1 EBAM machine and additional layer time. Figure 11a, b depicts the two designs. Results show that the additional time is 2.67 and 1.36% of the two selected demonstrator designs. Clearly, the additional time incurred by the use of electron imaging is a burden to the overall build time of the system; Fig. 9 Verification of the macroscopic process quality analysis software. a Reference image. b Virtual workpiece image. c Analysis resultant image with white pixels revealing the deviations between the virtual workpiece and reference image however, the benefit of image generation outweighs such additional process layer time particularly when one considers the possibility of in-process build correction or process control.

Image magnification
The prototype was capable of generating images (Fig. 7a-d) of the test build with different user-defined magnifications. As Table 7 suggests, the image magnification is 2.6 when a 180 mm × 180 mm area across the whole test build is imaged and the maximum magnification achieved is 95 when a 5 mm × 5 mm area is imaged. The ability to achieve userdefined image magnification indicates that the prototype has the potential to zoom-in to specific ROIs, revealing local component features or defects when fully integrated with an EBAM machine.
Despite the achievement in image magnification, two issues can be observed in the images. Firstly, horizontal black stripes are detected at the top or at both the top and bottom in each of the resultant images (Fig. 7a-d). It is suspected that the stripes are caused by a delay in the EBAM machine electron gun to deliver the required beam current during experiment. With no interactions between the machine electron beam and the test build, no feedback electrons were generated and thus the prototype data logger registered no responses resulting in black stripes in the images. Apart from the delay, the second issue observed is the limitation in spatial resolution. In theory, if the machine electron beam is infinitely small, the image spatial resolution would increase with higher image magnification. In reality, the image spatial resolution cannot keep up with the increase in magnification, and thus results in visually blurry images (Fig. 7c, d). The blurriness implies that the electron beam diameter is larger than the pixel size in Fig.   Fig. 10 Demonstration of the macroscopic process quality analysis. a Electronic workpiece image (raw) of imaging location 5. b Processed and binarised workpiece image of location 5. c Binary electronic reference image of imaging location 5, extracted from the STL design. d Analysis resultant image with white pixels revealing the deviations between the workpiece and the reference image

Image FOV
The prototype was capable of generating images (Fig. 8a-d) from the test build with different user-defined FOVs. The ability to achieve user-defined image FOV indicates that the prototype can allow specific ROI on the processing area to be imaged during future in-process EBAM monitoring. This ability is complementary to the imaging of the whole processing area. For the effective use of the machine processing area, multiple components of the same or different designs are always packed into the same EBAM build. Whilst an image of the whole processing area can provide information of the overall quality of an additive layer, an image of a specific ROI opens up opportunities to enable the effective analysis of individual components.

Macroscopic process quality analysis software verification
Verification was carried out on the macroscopic process quality analysis software. The virtual workpiece image (Fig. 9b) was overlapped on top of the reference image (Fig. 9a) and a resultant image (Fig. 9c) was produced. Table 8 shows that the difference between the reference and virtual workpiece image is 57,600 pixels, which corresponds to 64 mm 2 (as 60 mm corresponds to 1800 pixel). This area is exactly the same as that covered by the artificially removed 8 mm × 8 mm square on the top right-hand corner of the design depicted in Fig. 5. The result verifies the capability and correct operation of the software.

Macroscopic process quality analysis demonstration
Macroscopic process quality analysis demonstration was carried out on electronic images. The binary workpiece image (Fig. 10b) was overlapped on top of the binary reference image (Fig. 10c) and a resultant image (Fig. 10d) was produced. Table 9 shows that the difference between the reference and workpiece image is 31,891 pixels, which corresponds to 35.4 mm 2 (as 60 mm corresponds to 1800 pixels). This demonstration gives a quantitative measure, serving as an indicator of the macroscopic process quality. Due to the wide range of variation on design and tolerance of different EBAM components, it is thought that no universal benchmark can be set using this macroscopic indicator, for the definition of good process quality. There are two main root causes for the difference between the workpiece (Fig. 10b) and reference image (Fig. 10c). The first root cause is thought to be the misalignment between the test build location and the movements of the machine electron beam. In the demonstration, the test build was taken out of the chamber and put back manually for the removal of excess sintered powder. Due to the lack of fixation points on the EBAM machine processing area and the inaccuracy involved in manual handling, the location of the test build after powder removal was different from that when it was built, leading to a misalignment in the build location when electronic imaging experiments were carried out with the machine electron beam. The second root cause is thought to come from human error as well. Figure 10d shows that the left-hand side of the inner square containing the figure number 5 has the largest deviation. Referring to the binary workpiece image Fig. 10b, there is a horizontal bridge connecting the outer frame and the inner square. According to the binary reference image Fig. 10c, generated from the same ROI and layer height of the STL design as the workpiece image, the bridge should not be visible. The explanation is thought to be that the  Unlike the demonstration, when the software is operating in a real, in-process EBAM monitoring condition, the workpiece image will be generated once the processing of a layer is complete. As a result, without any manual disturbances to the build location or the top surface of the processed layer, human errors will not contribute to the deviations observed. Any deviations measured in the macroscopic process quality analysis in a real setting will be due to errors and imperfections of the EBAM process. It is thought that the software could be used as an "imaging go/no-go gauge". It has the potential to detect process errors including over-melting, peeled-off metallisation [37], the lack of powder deposition in local processing area and undesired electron beam movement due to stray magnetic field. The described errors are expected to lead to a detectable difference in the macroscopic process quality analysis output. This software is thought to have the potential to contribute to the decision making of the acceptance or rejection of a processed layer by the EBAM machine.

Conclusions
Applications of an electronic imaging system prototype have been presented in this paper. The prototype was designed to interface with a commercial EBAM machine. With the use of the prototype, single-layer electronic imaging experiments were carried out with the EBAM machine at room temperature. Digital electronic images were generated. Four different magnifications were achieved with the maximum value being 95. In addition, nine 60 mm × 60mm images with different FOVs were generated across the machine processing area. Moreover, macroscopic process quality analysis software was developed, verified and demonstrated. The demonstration showcased the capability of the software. It is thought that the software has the potential to be used for in-process EBAM macroscopic process quality analysis for individual components. With regard to in-process EBAM monitoring, it is thought that the electronic imaging system prototype has the potential to serve as an alternative to systems which employ thermal or optical imaging. There will be challenges moving to in-process monitoring in order to realise the systems potential. They are thought to include carrying out imaging on multi-layer for the whole additive manufacturing process, working at elevated temperature as the EBAM cycle includes preheating of the processing area and dealing with metallization generated from vaporisation of metal powder during the EBAM process.