IN
DEGREE PROJECT MECHANICAL ENGINEERING, SECOND CYCLE, 30 CREDITS
STOCKHOLM SWEDEN 2017,
Study of in-situ monitoring
methods to create a robust SLM process
Preventing collisions between recoater mechanism and part in a SLM machine
MICHAEL MATHIEU GERARDUS CAELERS
KTH ROYAL INSTITUTE OF TECHNOLOGY
Study of in-situ monitoring methods to create a robust SLM process
Preventing collisions between recoater mechanism and part in a SLM machine
Name author: Michael Caelers
Commissioning company:
Siemens Industrial Turbomachinery AB, Finspång Thesis supervisor:
P. Avdovic, Additive Manufacturing Innovation Manager Thesis initiator:
A. Graichen, Group Manager Additive Manufacturing Centre of Competence
University:
KTH Royal Institute of Technology, Stockholm Thesis supervisor:
Q. Fu, Post doctorate researcher Thesis examiner:
A. Rashid, Head of Machine and Process Technology department
Finspång, 14 June 2017
Summary
Within this thesis, a study of in-situ monitoring methods for a Selective Laser Melting process has been carried out. Focus of this study is the prevention of collisions between the recoating mechanism and the part.
The study of in-situ monitoring methods is based on a theoretical framework that consists of a state of the art literature review on in-situ monitoring methods for additive manufacturing processes. An analysis of the Selective Laser Melting process was performed to identify process signatures and possible defects resulting from the process. The sensors that have been described in the current state of the art are matched with the different process signatures and possible defects. A microphone has been chosen for a feasibility study, because of its potential to measure vibrations in the recoating mechanism, its novelty of application within a Selective Laser Melting process and its potential to monitor laser beam- powder interaction. An attempt has been made to predict collisions between the recoating mechanism and the part by using existing data on recoating time and speed. It can be concluded that the current sensor data on RecoatingDuration does not provide any information to predict a collision between the recoating mechanism and the part.
Sammanfattning
Inom denna rapport har en studie av in-situ mätteknikmetoder av modellen Selective Laser Melting-process genomförts. Fokus i denna studie är att förebygga kollisioner mellan recoating-mekanismen och den bearbetade delen. Studien av in-situ mätteknikmetoder bygger på en teoretisk ram som består av en state of the art litteraturöversikt om in-situ mätteknikmetoder för additiv tillverkning. En analys av Selective Laser Melting-processen utfördes för att identifiera processens utmärkande egenskaper samt identifiera de brister som härrör från att driva processen.
Hur de olika sensorerna kan kopplas till processens och hur de hör ihop med de utmärkande egenskaperna och eventuella defekter beskrivs i litteraturöversikten. En genomförbarhetsstudie utfördes genom att tillföra en mikrofon på recoating-mekanismen för att möjliggöra mätningar av mekanismens vibrationsförändringar. Att mäta vibrationsförändringar inom Selective Laser Melting-processen och dess potential att övervaka laserstråle-pulverinteraktion är helt nytt för detta ändamål. I ett försök att förutspå hur kollisionerna tillkommer mellan recoating-mekanismen och den bearbetade delen gjordes från mätningar av befintliga data såsom recoating-tid och -hastighet. Man kan dra slutsatsen att aktuella sensordata på RecoatingDuration inte ger någon vidare information för att förutse en kollision mellan recoating-mekanismen och delen.
Acknowledgements
This master thesis is the result of the degree project in the program of Production Engineering and Management, within the department of production engineering at KTH Royal Institute of Technology, Stockholm, in the period of February 2017 to June 2017.
Hereby, I express my gratefulness to all involved in this degree project.
Special thanks go out to Pajazit Avdovic as thesis supervisor and Andreas Graichen as thesis initiator at Siemens Industrial Turbomachinery AB for guiding me throughout this degree project. Furthermore, I am especially thankful to Qilin Fu as thesis supervisor at KTH Royal Institute of Technology who helped me to wrap up the degree project in this thesis report.
Finspång, 14 June 2017
Table of Content
SUMMARY ... I SAMMANFATTNING ... II ACKNOWLEDGEMENTS ... III
1. INTRODUCTION ... 1
1.1 COMPANY DESCRIPTION ... 1
1.2 SLM TECHNOLOGY DESCRIPTION ... 4
1.3 PROBLEM SITUATION ... 5
1.4 RESEARCH QUESTION ... 7
1.5 ASSIGNMENT AND SCOPING ... 7
1.6 RESEARCH GOAL AND METHODOLOGY ... 9
1.7 LIST OF TERMS ... 10
2. LITERATURE REVIEW ... 11
2.1 OPTICAL SENSORS WITHIN THE VISIBLE AND THERMAL RANGE ... 13
2.2 INTERFEROMETRY ... 31
2.3 TOMOGRAPHY ... 36
2.4 MICROSCOPY ... 38
2.5 ACOUSTIC ... 39
3. SLM PROCESS REPRESENTATION ... 45
3.1 REPRESENTATION MODEL SLM FORMING SYSTEM ... 49
3.2 REPRESENTATION MODEL OF RECOATING SUBSYSTEM ... 51
3.3 UNDERSTANDING POSSIBLE DEFECTS AND CAUSES ... 54
4. FEASIBILITY OF MICROPHONE SENSOR SYSTEM IN SLM ... 64
4.1 DETECT FRICTION BETWEEN RECOATER MECHANISM AND PART ... 65
4.2 DETECT SOUNDS OF AMBIENT AIR DISPLACEMENT ... 67
4.3 SENSOR FUSION OF ACOUSTIC AND OPTICAL SENSORS ... 72
5. ATTEMPT TO PREVENT COLLISIONS USING RECOATING DATA ... 73
5.1 CONCLUSION RECOATING SENSOR DATA ... 78
6. CONCLUSIONS ... 79
7. FUTURE WORK ... 80
7.1 DATA PROCESSING ... 80
7.2 ADAPTIVE CONTROL ... 80
LITERATURE ... 81
1. Introduction
This chapter will provide background information regarding Siemens Industrial Turbomachinery AB, the problem situation and this thesis research. The assignment and scoping of the assignment will be explained in detail. Finally, the research method will be discussed with which the quality of this thesis research is controlled.
1.1 Company description
Siemens Industrial Turbomachinery AB (SIT AB) holds office in Finspång in Sweden. The production of industrial turbomachinery has been taken place in Finspång since 1913, and became part of Siemens in 2003 [1]. SIT AB produces both steam and gas turbines that are used to generate electricity, steam and heat. Furthermore, the turbines are sold as power sources for compressors and pumps in the oil and gas industry (http://www.sit-ab.se/01_om_oss.html).
SIT AB adopted additive manufacturing in 2009. They industrialised Selective Laser Melting (SLM) technology in February 2016 by opening a dedicated production facility, called Siemens Additive Manufacturing Centre of Competence (Siemens, 2016). Nowadays, SIT AB uses the SLM process for, see Figure 1:
Rapid prototyping of for example a turbine blade of their SGT gas turbines;
Rapid repair of the burner tip of their SGT gas turbines;
Rapid manufacturing of the burner swirler of their SGT gas turbines.
Figure 1 Three pillars of AM applications at Siemens. Source:
http://www.automationregion.com/filer/docs/presentationer/160621_Hallberg.pdf.
SIT AB directs its additive manufacturing resources currently to the development of the combustion system and turbine area of the gas turbine, as shown in Figure 2.
Figure 2 Gas turbine focal areas for Additive Manufacturing. Source:
http://www.automationregion.com/filer/docs/presentationer/160621_Hallberg.pdf.
Siemens AG identified several market requirements that can be fulfilled by the use of additive manufacturing (AM). The market requirements they identified are “increased energy and resource efficiency, highly complex structures and designs, individualized mass production and shorter innovation cycles” [2]. Moreover, AM has some benefits compared to conventional machining, see Figure 3. The lead time of a burner tip can be reduced with 88.5% if produced by AM instead of conventional machining according to the example given in Figure 3.
Figure 3 Comparison of conventionally manufactured component vs. AM. Source:
http://www.automationregion.com/filer/docs/presentationer/160621_Hallberg.pdf.
1.2 SLM technology description
Selective Laser Melting (SLM) is an additive manufacturing technique.
Additive manufacturing technologies produce products in a layer-wise fashion. These technologies can broadly be grouped into seven different classes based on the mechanism in which each layer is formed [3]:
Photo-polymerisation;
Extrusion;
Sheet lamination;
Beam deposition;
Direct write and printing;
Powder bed binder jet printing;
Powder bed fusion.
SLM belongs to the powder bed fusion class. Layer by layer, this process fuses powder particles together with a laser beam. The recoater mechanism distributes a new layer of powder particles after the scanner that guides the laser beam is done. The layer thickness lies between the 20 m and 40 m. The atmosphere in the processing chamber usually consists of argon or nitrogen, to prevent oxidation of the metal during the fusion process. Oxygen levels are typically 0.1%.
A schematic visualisation of the SLM process at SIT AB, specifically showing the SGT burner tip repair, can be seen in Figure 4. The EOSINT M 280 machines at SIT AB are customised for this purpose. EOS GmbH created a new kind of building platform, in which the burner can be clamped and fixated. The new burner tip will be welded directly on top of the existing burner. Sequentially, a layer of metal powder will be distributed over the building platform by the dispenser and recoater mechanism. Thereafter, the laser will melt the powder according to the path that is defined by several input parameters. Examples of these input parameters are hatching strategy, laser power and the speed with which the laser is moved over the powder bed. These parameters are sent to the machine together with the CAD model before starting the process. After laser exposure, the building platform will drop and a new layer of powder will be distributed over the building platform by the dispenser and recoater.
Figure 4 Rapid burner tip repair at SIT AB. Source:
http://www.automationregion.com/filer/docs/presentationer/160621_Hallberg.pdf.
Several researchers suggest that the lack of process robustness, stability and repeatability are the main barriers for the industrial breakthrough of metal AM systems [3-5]. This validates the need for in-situ process monitoring and measurement in combination with adaptive closed loop control systems, and the research within this area.
1.3 Problem situation
SIT AB is dealing with collisions between the recoater mechanism in the SLM machine and elevated regions of the part that is being produced (super-elevations). These collisions damage both the part and recoater blade. This same problem has also been observed by Zur Jacobsmühlen [6]. Besides damage and cost of scrap or rework, the collision will also trigger the emergency stop function of the machine and the production will be stopped. This increases lead time.
The exact cause of these elevated regions in SIT AB’s SLM process is yet to be determined. In general it can be concluded that the SLM process is an open loop process that is not robust. According to Zur Jacobsmühlen et al.
[6] metal powder has a lower heat conductivity compared to solid metal by three orders of magnitude. It is likely that these elevated regions are caused by heat fluctuations after melting the powder material. Hot cracking is a familiar mechanism in welding processes. Hot cracking could lead to elevated regions in the part that collide with the recoating mechanism.
Another possible cause for super-elevated areas are curling effects that occur due to very high cooling rates during melting and solidification [7].
The residual stresses induce curling effects of the created layers.
A third possible cause for super-elevations is called ‘balling up’. Everton [4]
describe balling up as spheres formed by the powdered material that exceed the layer thickness, due to a presence of oxygen (>0.1%) in the build chamber leading to oxidation. Everton [4] state: “Subsequent layers amplify this discontinuity due to the resulting powder layers being non- uniform. This behaviour has been seen in stainless steel, iron and nickel based powders”. Balling up seems to be observed in a high laser power environment. Similar balling can however also be seen in processes with stainless steel powders and a low power environment. In that case, the laser power is insufficient to melt the powder fully. In Figure 5 an example of a SLM forming defect induced by balling is shown [8]. In this figure is referred to the recoating mechanism by ‘paving roller’.
Figure 5 Example of balling induced SLM forming defect. Source: Li, Liu, Shi, Wang & Jiang, 2012.
In order to prevent super-elevations in the first place, it is important to be able to identify what the exact cause for the super-elevated areas is.
Achieving that will require cross-functional knowledge on material science and welding. This thesis only focuses on the detection of super-elevated areas to create a robust process.
1.4 Research question
Following the before described problem situation the research question was formulated as:
“Which measures can be taken to prevent collisions between the recoater mechanism and the product being produced in the SLM machines?”
1.5 Assignment and scoping
The starting point of this thesis project is the SLM machine and sensor system, as in use to date at SIT AB. Historical data on recoater crashes can be retrieved from the machine server and internal documentation. This concerns information indicating how often recoater crashes occur, how severe the recoater crashes are and what the consequences are for the product being produced. This data forms the baseline measurement to which possible future measurements can be compared.
The literature research on current state of the art regarding in-situ monitoring and quality control in SLM forms the theoretical framework for this thesis research. The current state of the art gives direction to the thesis, to go into less-explored research areas regarding this topic.
Figure 6 depicts the scope of this thesis. Preventing collisions between the recoater and elevated regions of the part is the focus of this research. An important step towards realising this goal is the state of the art research that describes which sensors can be adapted in the SLM process. To prevent collisions between recoater and part, good understanding of the SLM process is required. Therefore, the SLM process, its characteristics and process signatures are thoroughly analysed and described by a closed loop representation model, based on control theory.
The literature review and the analysis of the SLM process result in the choice of one or more sensors. The choice is checked on feasibility and experiments are conducted. The experiments are however limited by the available time for this thesis research. Machine warranty, insurance, CE- marking and machine availability are other restrictions on the experiments that are conducted.
It is important to consider the related areas that are not part of the scope of the thesis. Even though these related areas are not part of the scope, a practical solution for the problem cannot be found without considering the effects of the chosen solution on these areas. However, this thesis does not provide solutions regarding data processing, data storage and possible feedback loops to machine control systems for example. This is part of possible future work.
Figure 6 Thesis scoping.
1.6 Research goal and methodology
The goal of this research is to investigate what sensors can be used to create a robust SLM process that specifically prevents collisions between the recoater and the part. The aim has been to find a preventive solution, rather than a reactive solution.
The foundation of this thesis research consists of a theoretical framework.
This theoretical framework is achieved by carrying out a literature review on the current state of the art in in-situ process monitoring and metrology in additive manufacturing. The started by studying research work that tries to give a comprehensive summary of currently available systems Thereafter, literature by Craeghs [9], Craeghs [10], focusing on powder bed monitoring was reviewed. Eventually, the literature review lead to research that focused on detecting elevated part regions within the powder bed [6, 11, 12].
A typical representation model for closed loop control systems is used to describe the SLM forming system. The closed loop representation model is based on closed loop control theory and often used to describe conventional machining systems, for example by Archenti [13] and Daghini [14]. Lastly, data on recoating time has been retrieved from the machine server in an attempt to predict collisions between the recoating mechanism and the part.
The structure of this thesis is as follows. This first chapter explains the background of this thesis. It will create context by explaining SIT AB’s activities and the SLM process. Thereafter, the problem situation, research question, assignment and scoping will be defined. This chapter will end with an introduction into the used research methods. The literature review will be discussed in the second chapter. The order in which the literature will be discussed is according to the different categories the sensors belong to. The third chapter will elaborate on the SLM process, its characteristics and process signatures and the representation model of the SLM forming system. In the fourth chapter the feasibility of the use of a microphone is investigated. Chapter five will discuss attempt to predict collisions by using recoating data. The sixth chapter will discuss the conclusions that can be drawn from this thesis research. Lastly, future work is discussed in chapter seven.
1.7 List of terms
AM Additive Manufacturing CCD Charge-Coupled Device
CMOS Complementary metal-oxide-semiconductor CNN Convolutional Neural Network
CoC Center of Competence
Exposure Activation of the laser and exposing of the powder to the active laser
GmbH Gesellschaft mit beschränkter Haftung HAZ Heat Affected Zone
IR Infra-red
LPBF Laser Powder Bed Fusion
PID Proportional Integral Derivative controller Powder
bed
Powder that has been newly distributed or solidified on the building platform and subsequent layers
Recoater Mechanism in the SLM machine that redistributes powder over the building platform
sCMOS Scientific CMOS
SIT AB Siemens Industrial Turbo Machinery Aktiebolag SLM Selective Laser Melting
UV Ultra Violet
2. Literature review
The lack of in-situ process control keeps the SLM process from industrialising [3-5]. This chapter gives a review of the research that has been done within the area of in-situ process monitoring of SLM processes.
An overview of different types of sensors that have been tested can be found in Figure 7. The different types of sensors are divided in either the category optical or acoustic. Within the optical category there are several subcategories, namely: cameras that capture visible & thermal wavelength, interferometry, tomography and microscopy. All different sensors are discussed in this chapter.
Topics like the mechanical instrumentation that is being used, the prerequisites that are part of the research, resolution of the sensors, how the acquired data is outputted and optional adaptive control systems are part of the research review. Sometimes multiple sensors are used in one and the same setup.
Figure 7 State of the art concerning in-situ process monitoring in SLM processes.
This research only targets non-contact monitoring and measuring methods. Touching the powder bed will distort the fine spreading of the metal powder over the building platform. Therefore, mechanical tactile probing is not considered.
Adaptive control systems and data processing are out of the scope of this thesis as discussed before. However, in some cases researchers integrated data processing and adaptive control in the sensor systems of their research. In those cases, the data processing and adaptive control are discussed briefly to create a foundation for future work.
2.1 Optical sensors within the visible and thermal range
This section examines the different optical sensors within the visible and thermal range. Most research has been done on sensors in this subcategory and some of them are already in commercial application.
2.1.1 High-speed camera
Mechanical instrumentation
Craeghs et al. [9, 10] use a high-speed CMOS camera and planar photodiode for the purpose of melt pool monitoring. The photodiode and camera are installed co-axially, meaning, the light they capture travels the same path as the laser, see Figure 8. The high-speed CMOS camera can only output an 8-bit grey value image. “Nevertheless, the grey values given by the camera can be related to temperature and a correlation between the melt temperature of the used material and a so-called melt grey value can be derived”. The camera and photodiode are sensitive for wavelengths in the range of 400 to 900 nm. It is important to choose the right wavelength bandwidth that needs to be observed, since there are multiple light sources that can be observed by this sensor system. The laser beam of the Yb-YAG fibre laser in this research’s setup has a wavelength of 1064 nm. Therefore, Craeghs et al. chose the upper boundary of the wavelength to be 950 nm. “The lower boundary needs to be higher than 700 nm because visible light (from e.g. illumination in the process chamber) is not of interest either”. The photodiode captures a larger area around the melt than the camera does.
Figure 8 Schematic overview of the experimental setup of the monitoring system. Source:
Craeghs et al., 2010.
Data processing
“The melt pool images are thresholded directly on the processing hardware of the camera into binary images: pixels having a higher grey value have a value of 1, lower grey values become 0. Based on the threshold image melt pool properties as melt pool length, width and area can be derived using particle analysis algorithm”.
Adaptive control system
The SLM machine that is being used in this research is developed in-house at K.U. Leuven, Department of Mechanical Engineering in Belgium.
Therefore, researchers from K.U. Leuven have easy access to the machine control system. This is not always the case. Since the researchers can access the control system, they integrated a PI-controller (Proportional- Integral controller). The photodiode is sensitive to spatter or sparks escaping from the melt pool. The spatter and sparks yield a high frequency noise in the photodiode signal. Thus, the bandwidth of the controller should not be too high.
2.1.2 Digital camera with CCD or CMOS image sensor
Mechanical instrumentation & image processing techniques
Digital cameras are widely used for in-situ process monitoring in SLM processes. They are usually used to inspect the powder bed, both before and after recoating. Digital cameras can however also be used of inspection of the melt pool. The most commonly used image sensors in digital cameras are CMOS (complementary metal-oxide-semiconductor) and CCD (charge-coupled device). The main difference between CCD and CMOS image sensors is, that within the CCD the photosites (pixels) are passive and within CMOS they are active [15]. The photosites in a CMOS sensor each have their own amplifier and therefore do local processing.
The photosites in a CCD sensor capture information and then this information is moved to an amplifier in the form of arrays. Processing of the information in CCD sensors is done when all the information is stored in the photosites, whereas the CMOS processes the information one photosite at a time. Photosites is where the information is captured and converted into light intensity and colours. Due to the properties of CMOS sensors, bending of the captured scene, called rolling shutter artefact, might occur. This is a property that should be taken into consideration when deciding which image sensor to use.
Craeghs et al. [10] utilise a digital camera to detect wear and local damage of the recoater blade. They do not specify which image sensor (CCD or CMOS) the digital camera has. They however, do specify the use of active illumination. They use active illumination from three different incidents in order to create shading around possible discontinuities in the powder bed, see Figure 9.
Figure 9 Principle of visual inspection system for monitoring of the powder layer top surface. Source: Craeghs et al., 2011.
With this method, the researchers try to identify two different defects based on different discontinuities in the powder bed. Namely, wear of the recoater blade and local damage of the recoater blade. Due to the horizontal movement of the recoater blade, it is possible to identify recoater wear or local damage by spotting horizontal lines in the powder bed that were created by shading from the different light incidents. This is shown in Figure 10 and Figure 11, where the upper images show a powder bed.
Figure 10 Ideally deposited powder bed.
Source: Craeghs et al., 2011.
Figure 11 Powder bed deposition with worn recoater blade. Source: Craeghs et al., 2011.
The graphs underneath the powder bed images in Figure 10 and Figure 11 show the grey values on its y-axis. The x-axis resembles the distance (in pixels) from top of the powder bed image to the bottom of the powder bed image. The white vertical lines that are shown in the upper images are the line profiles along which the mean grey value was determined. The upper control limits in the lower graphs are determined based on different factors like camera settings (e.g. exposure time), illumination (e.g. type of light source or luminosity) and the material the powder bed consists of.
These control limits can be based on empirical experience for example.
The five horizontal lines in the upper right powder bed image are represented by the five peaks that cross the control limits in the lower right graph.
This detection system could be used to identify recoater wear or local damage. An uneven distribution of the powder bed leads to an instable process, because there might be too much or too little powder in some areas. The input parameters like laser power and laser speed are based on a certain uniform layer thickness. In the areas where there is too little powder, this might cause for burning of the metal powder. In the areas where there is too much powder, this might cause for not complete melting of the metal powder.
Kleszczynski et al. [7, 11] and Zur Jacobsmühlen et al. [6, 12] utilise a digital camera and active illumination to detect discontinuities in the powder bed as well, see Figure 12.
Figure 12 Active illumination for surface inspection and detection of elevated areas. Source:
Kleszczynski et al., 2012.
The direction of the illumination is almost the same in both researches. A feature that can be found in Kleszczynski’s et al. [7] work and not in Craeghs’ et al. [10] work is the use of matt reflectors. According to Kleszczynski et al. [7], “diffuse lighting with a light source placed close to the working surface and opposite to the camera produces the best quality for surface images”. This is because the mirror-like metallic weld bead structures can cause specular reflections that could saturate the camera’s CCD sensor. The matt reflectors help to distribute homogenous lighting over the powder bed. “Active illumination can be arranged by spectrally narrow lamps or by laser [16]. According to Purtonen, Kalliosaari and Salminen [16] this can be arrange “by UV light emitting diodes during laser metal deposition of tool steel H13, by green laser or by fibre-coupled infrared laser diode during CO2 laser welding”.
For example, Liu et al. use a green laser as an active illumination source to monitor a laser hot-wire cladding process [17].
Whereas Craeghs et al. [10] use the camera and active illumination to detect recoater wear or local damages, Kleszczynski et al. [6, 7, 11, 12] use the camera and active illumination to detect super-elevations. Super- elevated areas are likely to collide with the recoater mechanism. The researchers use an externally mounted 29 megapixel monochrome CCD camera. “The resolution ranges from 20 to 30 m/pixel, depending on the field of view, which is approximately 150 mm by 110 mm”. The researchers assume that most pixels inside the image represented powder. In that case, it’s smart to crop an image in case it also shows part for the machine for example, so that only the powder bed is shown. The powder bed image is denoted by: Pz ℤ𝑀×𝑁. The image intensity was identified as Iz(x,y) with (x,y) Pz. The first step the researchers make was to compute the intensity mean of the powder:
𝐼̅ =𝑧 1
𝑀𝑁 ∑ 𝐼𝑧(𝑥, 𝑦)
𝑥,𝑦∈𝑃𝑧
The second step consists of identifying a threshold to detect outliers. The researchers define the threshold as:
𝑇𝑧 = 3 ∙ 𝜎
In this formula is the standard deviation of all pixel intensities. The factor of 3 was determined experimentally.
“Tz is computed for every layer and the detection result is obtained as a binary mask image” with:
𝐷𝑧(𝑥, 𝑦) = {0 𝐼𝑧(𝑥, 𝑦) ≤ 𝑇𝑧 1 𝑇𝑧 < 𝐼𝑧(𝑥, 𝑦)
The purpose of the next step is to extract the elevated area Az(x) at every x position inside every layer image from the detection mask Dz as:
𝐴𝑧(𝑥) = 𝐴𝑝𝑖𝑥𝑒𝑙∙ ∑ 𝐷𝑧(𝑥, 𝑦)
0≤𝑦<𝑀
In this formula Apixel is the area of one pixel in mm2.
Figure 13 Example of elevated region segmentation and analysis. Source: Kleszczynski et al., 2012.
Figure 13, shows the result of the steps described before. The researchers define a threshold for the critical elevated area per x position based on accelerometer measurements. Those measurements will be discussed in more detail later in the acoustics sensors section. The threshold for critical elevated areas per x position is Acritical = 0.1 mm2.
The researchers use a connected component analysis to analyse the different parts in a single build job individually. This allows for a comparison of process stability of identical parts built in a grid with varied process parameters. The downside is that the connected component analysis cannot differentiate between part and support structure.
Visualisation of the super elevated regions in an x/z-plot shows that the area of elevated regions is larger at the place where there is an overhanging structure, see Figure 14. The x/z-plot was created by integrating all elevated regions at specific xi positions over y:
𝐴(𝑥, 𝑦) = ∑ 𝐷𝑧(𝑥, 𝑦)
𝑦
In the left image in Figure 14 you can see the part geometry with overhanging angle. The right image shows the x/z-plot with the area of elevated regions in square millimetres.
Figure 14 Analysis of overhanging geometry in x/z-plot of elevated area. Source: Zur Jacobsmühlen, 2015.
Other image processing techniques
In the research presented before, active illumination is being used to create a clear profile of the powder bed surface. This technique is also called ‘shape from shading’ [18]. First, the gradients in both x- and y- direction are calculated from a captured greyscale image. Second, a height map is derived by integrating the gradients. This technique is not very complex and only requires basic image processing knowledge. The resolution that can be achieved with this technique is not clear, but it is not expected to have a high resolution. The resolution is dependent on the gradient distribution; hence the spatial resolution of the camera and the active lighting that is being used.
Another technique that is described by Weckenmann et al. [18] is called
‘depth from focus’, where the distance between the objective and the workpiece is varied, see Figure 15. Unfortunately, not much research about this technique was found.
Figure 15 Depth map based on focus series. Source: Weckenmann et al., 2009.
Industrial photogrammetry is a known metrology technique [19]. “Shape, size and position of objects are determined from cross-domain fusion of measurements made on two-dimensional images. It can also be classified as a passive triangulation principle, where three-dimensional coordinates of points of interest are calculated via optical triangulation from two or more images taken from different locations.” [18]. As seen in Figure 16, a photogrammetry setup can consist of multiple cameras. A distinction can be made between the so-called camera interior geometrical parameters and the exterior orientation parameters. The camera interior geometrical
parameters refer to the principal focal length, centre of image plane, lens distortion and the position and orientation parameters. These parameters must be known a priori. The exterior orientation parameters are the XYZ coordinates of the perspective centre and the three rotation angles between the image- and object-space coordinate axes. These are also known as the image position and orientation parameters.
The resolution of an industrial photogrammetry system greatly depends on the spatial resolution of the chosen camera. Off-the-shelf cameras offer high resolution and performance at an affordable price. Some industrial photogrammetry systems use the Nikon D3x with a 24.5-megapixel colour sensor and an option to transfer images directly wireless to a computer, according to Bösemann [19]. However, there are also special photogrammetry cameras available, like the MoveInspect HF camera from AICON. This camera has “a build-in processor for fast data processing, protective housing for use in rough industrial environments and adaptive illumination”.
Figure 16 Multi-station photogrammetry setup. Source: Weckenmann et al., 2009.
Digital fringe projection is a technique that is used to measure surface topography. The fringe projection system developed by Zhang et al. [20] is capable of measuring several powder bed signatures. These powder bed signatures are:
Powder layer flatness;
Surface texture;
Average height drop of the fused regions;
Characteristic length scales on the surface;
Splatter drop location and dimension.
A fringe projection system consists of a camera and a projector. The projector produces structured light patterns like a sequence of sinusoidally varying intensity patterns on the object’s surface. The camera captures these patterns from a different viewing angle. The fringe projection systems in this research of Zhang et al. [20] has a camera resolution of 4160 by 2091 pixels, a field of view of about 28 by 15 mm, hence a lateral resolution of 6.8 m/pixel. The researchers work with a customized projection lens that is mounted outside of the building chamber. The projection lens creates a projected image of about 45 by 28 mm. This is relatively small considering the fact that the build surface of the EOSINT M 290 is 250 by 250 mm. However, the small image is necessary to generate dense fringes of 0.35 mm/cycle and achieve the desired height resolution.
Some of the results of tests that have been carried out can be seen in Figure 17. The average height of the fused surface is lower than the unfused surface due to solidification. The average height difference between the fused and unfused surface can be calculated. The results of this technique seem promising. Measurements can be carried out with a very high resolution. However, the field of view is very limited and the fringe projection method is complicated.
Figure 17 Height maps of powder bed before and after laser fusion measured on every other layer. Source: Zhang et al., 2016.
The different image processing techniques discussed before require fusion of image data to achieve dimensional information. As mentioned before, the image processing of ‘shape from shading’ is quite rudimental compared to the more complex fringe projection technique. Complexity, computational requirements, accuracy, resolution and speed are important factors that differ much between these techniques.
2.1.3 Photodiode
The photodiode commonly used to monitor the melt pool during the SLM process [3-5, 9, 16, 21]. Two different types of implementations of the photodiode sensors can be distinguished; respectively the Lagrangian reference frame and the Eulerian reference frame [3]. Nevertheless, other sensor modalities such as cameras and pyrometers can also be implemented with these kind of reference frames.
The Lagrangian reference frame is a moving reference frame along the same optical path as the laser. This kind of reference frame is therefore called an on-axis or co-axial reference frame. By following the same optical path as the laser, the monitoring is incident with the focal point of the laser and therefore the melt pool and heat affected zone (HAZ). By following the same optical path as the laser, the Lagrangian reference frame does not monitor temperature history, such as the cooling rate.
As soon as a part of the powder is melted, the galvanometer-scanners will continue to another part of the powder bed to melt the powder there.
The Eulerian reference frame is an off-axis reference frame because it does not follow the optical path of the laser. In this case, the photodiode is monitoring a fixed point or area on the build surface. Since the photodiode is monitoring a fixed point or area, it can record the thermal history of that point or area. Both melting and cooling will be monitored.
Photodiodes convert light into current or voltage and can be used in different wavelength ranges [16]. The photodiodes are usually used to measure size, shape and light intensity of the melt pool. These parameters give away information about the amount of energy that is absorbed by the powder bed. “The length of the melt pool has been shown to be a proxy for material cooling rates which ultimately may indicate the amount of residual stress in the part” [3]. “Typically, silicon photodiodes are used for UV and visible wavelengths and indium-gallium-arsenide (InGaAs) photodiodes for visible and IR wavelengths” [16].
2.1.4 Pyrometer
“A pyrometer can be classified by the number of spectral bands of the detection system. They can be divided into single-, dual- or multiband systems”. According to Purtonen et al. [16], there is no consistent proof that a multi-wavelength pyrometer is more accurate than a monochromatic pyrometer. A pyrometer takes non-contact temperature measurements. Grasso et al. discuss a pyrometer setup in their state of the art research. The pyrometer they discuss includes “two InGaAs photodiodes with a temperature range of 1200 – 2900 K, a transmission spectrum at central wavelength 1260 nm with a 100 nm bandwidth, and a measured area with a diameter of 560 m” [5]. As discussed before, a pyrometer can both be implemented on-axis (Lagrangian) and off-axis (Eulerian).
A pyrometer can be used to measure the temperature of the melt pool area. Furthermore, can the pyrometer be used to “monitor the temperature evolution during the SLM of each slice and to observe the spatters ejected by the beam-material interaction” [5].
2.1.5 Infrared camera
The photodiode used in the research of Craeghs et al. [22] is sensitive to wavelengths in the range of 400 to 900 nm. Figure 18 shows the electromagnetic spectrum and as you can see, the photodiode only detects visible light. According to Craeghs et al., Planck’s law shows that the radiation energy at the melting point of metals, which lies roughly around 1500K, is highest in the near infrared region around 1000 nm.
Therefore, for detection of mid- and long wave infrared radiation an infrared camera is needed.
Figure 18 The electromagnetic spectrum. Source: J. Gustafsson, personal communication, September 14, 2016.
Infrared cameras are alternative non-contact thermal measurement sensors for pyrometers. Infrared cameras have a greater capture rate and are more accurate than pyrometers according to Everton et al. [4].
Purtonen et al. divide infrared cameras into two categories, namely infrared cameras with cooled or uncooled detectors [16]. Amongst uncooled detectors they distinguish two types, respectively ferroelectric detectors and microbolometers. “The most common type is a vanadium oxide (VOx) microbolometer, which allows a relatively high spatial resolution and higher sensitivity compared to ferroelectric detectors”.
2.1.6 Optical spectrometer
Optical spectroscopy is usually used to monitor the behaviour of the plasma plume that is formed by ionization and excitation of the metal vapour [17].Spectroscopy can provide information about the chemical composition, electron temperature and electron density of the plasma plume.
The research by Liu et al. is carried out on a laser hot-wire cladding process and not on a SLM process. It is expected that a spectrometer could also be used to monitor the plasma plume characteristics of a SLM process to control the weld quality. Optical spectrometers have been used in conventional laser welding to monitor the plasma plume [23, 24].
Although, the biggest obstacle to overcome would be the distance from the spectrometer to the actual laser-powder interaction, and thus the plasma plume. Figure 19 shows a schematic image of the experimental setup Liu et al. used to monitor their laser hot-wire cladding process.
Figure 19 Schematic image of process control for laser hot-wire cladding. Source: Liu et al., 2014.
Lui et al. demonstrate that the plasma electron temperature is a reliable indicator for the dynamic change in the cladding process and the variation of the clad results. The Boltzmann plot is used to calculate the electron temperature Te [17]:
𝑙𝑛 (𝑙𝑚𝑛𝜆𝑚𝑛
𝐴𝑚𝑛𝑔𝑚) = 𝑙𝑛 (𝑁ℎ𝑐 𝑍 ) −𝐸𝑚
𝑘𝑇𝑒
In this equation lmn is the emission line relative intensity, mn is the wavelength, Amn is the transition probability, gm is statistical weight, N is the total population density of the element, h is Planck’s constant, c is the light velocity, Z is the partition function, Em is the upper level energy and k is the Boltzmann constant.
2.2 Interferometry
Both laser interferometry and white light interferometry are optical probing techniques. These metrology tools are categorised as optical profilers and are commonly used in surface characterisation. Microscopes are also categorised as optical profiles, but they will be discussed in another section of this literature review. Optical profilers are often used to study surface damage, erosion and wear of engineering surfaces [25].
According to Miyoshi, an optical profiler can “profile heights ranging from
<1 nm to 5000 m at speeds up to 10 m/s with a 0.1 height resolution and profile areas as large as 100 mm by 100 mm”.
Figure 20 Representation of a typical interferometer. Source: J. Gustafsson, October 2, 2016.
Figure 20, represents the working of a typical interferometer. The light source projects a beam towards a beam splitter that directs one beam towards a reference mirror (upper mirror) and another beam towards the object of interest (the movable mirror in Figure 20). When both beams are reflected to the beam splitter they interfere with each other and create an interference pattern, as seen in the image on the right of Figure 20. The heights of the surface are derived from deviations in this interference pattern.
There are several factors that should be taken into consideration deciding the suitability of interferometry as a metrology tool. According to Miyoshi [25], if the surface roughness is greater than 1.5 mm, “the interference fringes can be scattered to the extent that topography cannot be determined”. Moreover, the environmental conditions determine the accuracy of the interferometer, see Figure 21.
Figure 21 Environmental factors affecting interferometer accuracy. Source:
https://www.renishaw.com/en/how-do-interferometric-systems-work--38612 on March 8, 2017.
The operational wavelength of the laser beam depends on the refractive index of the air through which it passes. The refractive index of the air depends on the air temperature, relative humidity and air pressure as shown in Figure 21. In dimensional metrology, there is an international standard called ISO 1 Geometrical Product Specifications, which states that all dimensional measurements should be reported at a standard reference temperature of 20 C. If a measurement is performed at another temperature than the standard reference temperature, corrections need to be made afterwards. This is a relatively easy task if the temperature is stable. This also applies for relative humidity and air pressure. However, this is not (always) the case in a SLM process.
The temperature at the level of the building platform typically gets preheated up to 80 C, whereas the temperature at the top of the processing chamber is approximately 40 C. Furthermore, there is an argon supply of 100 liters per minute within the processing chamber. Therefore, the environment in the processing chamber of a SLM process is assumed to be quite turbulent. Wyant has presented a technique to reduce the effects of air turbulence [26]. This technique is called dynamic single-shot interferometry. It is said to be insensitive to vibrations and it reduces the effects of air turbulence by averaging many measurements.
In this research, the application is described as, being used for measuring (aspheric) optics / optical surfaces. Therefore, it is not clear if the dynamic single-shot interferometry would also work in a SLM process.
Regardless of the obstacles described before, Neef et al. integrated a low coherence interferometer (LCI) in an EOS M 250 SLM machine, see Figure 22 [27].
Figure 22 Co-axial low coherence interferometer in SLM machine. Source: Neef et al., 2014.
Neef et al. use the optical path of the laser to sample the surface topography. This is done sequentially, since the area captured is only 3 mm by 3 mm. The intended use of this LCI sensor is to inspect the powder bed, core regions and counter regions. Some results of their experiments are shown in Figure 23.
Figure 23 Surface areas (a) powder bed and (b) SLM structure. Resulting profile scans (c) powder bed and (d) SLM structure. Source: Neef et al., 2014.
Figure 23a shows a defect in the powder bed. This defect is approximately 50 m in depth, as can be seen in Figure 23c. The scan profile also shows single powder particles of 20 m to 40 m. Figure 23d shows the weld tracks with height differences of 50 m. The resolution of the LCI sensor seems to be good enough to measure small discontinuities in the powder bed or part. However, the researchers do not discuss the accuracy of their LCI sensor and the influences the environmental conditions have on it.
They do mention that the LCI sensor cannot be used at the same time as the laser. This means that in-situ quality monitoring with this sensor will automatically increase lead time significantly.
The LCI sensor of Neef et al. is based on the Precitec IDM (In-Process Depth Meter) sensor [28-30]. The IDM sensor is based low-coherence interferometry and is used in conventional laser welding. The purpose of the IDM sensor is to measure the depth of the keyhole during welding.
Thus, the IDM sensor must be aligned co-axially with the laser optical system and measure the keyhole depth at the same time as the laser creates the keyhole. It is unclear why the IDM sensor can be used at the same time as the laser and the LCI sensor cannot be used at the same time as the laser. Nevertheless, it is thinkable that the LCI sensor has the potential to measure keyhole depths in a SLM process after further development.
2.3 Tomography
The optical tomography technique was developed in Germany by MTU Aero Engines and is now owned by EOS GmbH [31]. EOS is currently in the process of commercialising the optical tomography technique under the name EOSTATE Exposure [32]. The technique utilises a 5-megapixel thermal stabilised sCMOS-camera to capture the thermal radiation resulting from the melt pool. The spectrum of the thermal radiation lies within the visible wavelength between 380 to 780 nm. An infrared filter is used to eliminate the reflected laser light at 1064 nm, see Figure 24. The 5- megapixel camera has a field of view of 250 by 250 mm, which results into a geometric resolution of approximately 0.1 mm per pixel [33].
Figure 24 Quantum efficiency of the sCMOS camera and spectral filter. Source: Zenzinger et al., 2015.
Whereas the melt pool intensity and shape is typically captured by photodiodes during exposure of the laser, the OT-camera takes one long time exposure image during exposure [31]. That way, the melting and solidification process can be captured and hot or cold spots can be detected for example, see Figure 25. The layer images as shown in Figure 26, can be stacked on top of one other to create a 3-dimensional visualisation of the part and the location of possible cold or hot areas.
Figure 25 OT-captured image of one layer showing hot and cold areas. Source: Zenzinger et al., 2015.
Figure 26 OT-images stacked 3D-visualisation. Source:
Bamberg et al., 2016.
2.4 Microscopy
As mentioned before, microscopy is considered an optical profiler. It retrieves information by optically probing a surface. There are many different types of microscopes, amongst others there are confocal microscopes, scanning tunnelling microscopes, atomic force microscopes and scanning electron microscope. However, the microscopes operate within close range to the inspection surface [20]. This is not possible in the SLM machine due to the recoater movement over the powder bed.
Therefore, no further analysis for microscopic sensors has been performed.
2.5 Acoustic
Acoustic sensors are commonly used in conventional welding processes.
These sensors aim to capture air-borne or structure-borne process emissions, see Figure 27. Air-borne emissions can consist of process signatures resulting from the displacement of ambient air by vaporised metal that is escaping from the keyhole [24]. These acoustic emissions usually lie within the human audible range between 20 Hz to 20 kHz. The structure-borne emissions can consist of hot- or cold-cracking within the welded part or emissions by the scanning mirrors or recoater for example.
Figure 27 Typical experimental setup for capturing acoustic emissions. Source: Shao & Yan, 2005.
Several sensors that can capture both the air-borne and structure-borne acoustic emissions are discussed in this section.
2.5.1 Accelerometer
Accelerometers are used in the research of Kleszczynski et al. [6, 7, 11, 12]
to detect recoating vibrations. The information that was gathered on the vibrations of the recoating mechanism, due to contact with the part, is being correlated with the information that is retrieved from processing images of the powder bed after exposure.
The researchers set thresholds for the vibration intensity and area of super-elevations in the powder bed images. The super-elevated area is considered problematic when the vibrations and measured super-elevated area are too large and exceed their thresholds. This has been described before in section 2.1.2. EOS GmbH owns the patent on the acceleration measurement system used to measure the recoating mechanism vibrations.
Figure 28 shows the output of the a) accelerometer measurement and b) elevated area measurement from the c) powder bed image.
Figure 28 a) Measured acceleration plotted over x-axis at layer z=1.120 mm, b) accumulated elevated area based on image processing, c) image of the layer z=1.120 mm.
Source: Kleszczynski et al., 2014.
In Figure 28c, four parts are shown, respectively labelled from number one to four from left to right. Part number four is the reference part that was produced with the optimal prescribed processing parameters for the specific layer thickness, powder material and part geometry. For part three the laser power was increased by 20%, for part two the laser power was increased 40% and for part one the laser power was increased with 60% compared to the reference. In both a) and b) in Figure 28 the signals increase.
Balling occurs due to a too high power density, hence more/larger elevated areas are observed. In a) the acceleration is largest in the first part due to increased contact between the recoating mechanism and the part, which can be seen in b). There an increase area of elevations is displayed.
By combining the elevation detection in the images with the vibration information acquired by the accelerometer, the researchers have found a method that accurately detects contact between the recoating mechanism and the part. This information can be used to detect trends of increasing contact and increasing area of elevations. If these trends are detected before collision between the recoating mechanism and the part occurs, an operator could be warned and measure could be taken in order to prevent the collision from happening.
2.5.2 Microphone
Microphones have been used in conventional laser welding processes for the purpose of process quality monitoring. However, no research was found that integrates a microphone into a SLM machine. It is expected that a microphone can be used to monitoring the quality of a SLM process.
A challenge to overcome is the environment in the processing chamber.
There is a certain temperature gradient between the bottom and the top of the processing chamber. Furthermore, to prevent oxidation during the welding process, an argon flow of 100 liters / minute is created, which is expected to be turbulent and not laminar. These conditions will be discussed in more detail in the feasibility study section.
According to Gu and Duley, optical and acoustic resonance can be excited by modulating the laser beam intensity at a frequency that coincides with the natural (Eigen) frequency of the keyhole vibrational mode [34]. It is assumed that resonance is an unwanted property since it is expected to increase process instability. The analogy is made with forced vibrations in traditional machining. In that case, if the tooth passing frequency coincides with the natural (Eigen) frequency of the workpiece, resonance occurs and the process becomes unstable.
Farson, Sang and Ali explain that emitted optical power and sound pressure emissions correspond to the laser power absorbed in the powder bed and the formation of a keyhole [35]. According to Farson et al.
deviations in the absorption of the laser power influence the plume temperature which can be observed in the optical signal. Deviations in the absorbed laser power will also influence the evaporation rate of molten metal and displacement of ambient air, which can be observed in the acoustic signal.
2.5.3 Ultrasonic transducer
Rieder, Dillhöfer, Spies, Bamberg and Hess describe the use of ultrasonic transducers in monitoring dynamics of “the layer build-ups, the interface coupling, the local material properties as well as the formation of porosity and distortions due to residual stresses” [36]. The experiment is performed in an EOSINT M 280 SLM machine. Some of the characteristics of the experimental setup are:
A bandwidth range of 400 kHz – 30 MHz;
Sampling rate of 250 Mega-samples per second;
14-bit resolution;
Ultrasonic transducer was mounted underneath the building platform.
The sampling is triggered by the process and can acquire data during build- jobs of up to eight hours duration [36]. According to Rieder et al.
“recording up to 1000 data A-scans per second results in several Giga- bytes of data to be stored, further processed and evaluated.
The researchers conclude that ultrasonic online monitoring of the SLM process through the building platform is feasible for parts with non- complex geometries.
2.5.4 Acoustic spectrometer
Smith et al. describe a method that acquires spatial acoustic information.
To be more precise, the researchers introduce the inspection method of spatially resolved acoustic spectroscopy (SRAS). “This method uses surface acoustic waves (SAWs) to probe the material to a depth of a few 10’s or 100’s of m” [37]. Figure 30 displays a schematic representation of the SRAS instrument. Smith et al. explain that “a pulsed laser passes through a chrome grating and is imaged to the sample surface, where the absorbed pulses thermoelastically generate acoustic waves”. The spacing of the lines on the grating determines the wavelength of the surface acoustic wave.
The characteristic frequency of the waves 𝑓 is defined by 𝑓 = 𝑣/𝜆, where 𝑣 is the SAW velocity of the material under the generation patch (𝜙𝑝) and 𝜆 is the acoustic wavelength [37].
Figure 29 Schematic representation of the SRAS instrument, incl. pulsed generation laser, project mask, continuous wave detection laser and detector. Source: Smith et al., 2016.
According to Smith et al. it is possible to determine the orientation of the grains by taking multiple velocity images with different acoustic wave propagation. The depth sensitivity of the SRAS instrument can be adjusted by changing the acoustic wavelength through modifying the line spacing of the projection grating. According to Smith et al. this allows the inspection of a single or several build layers at once.
The current SRAS instrument uses a 2 KHz repetition rate for the laser, which results in approximately 1000 points per second after data acquisition dead time and scanning overheads are included.
Currently, the SRAS instrument is being used off-line, that is to say, after production of the parts. The surfaces of the test parts have been prepared by grinding and polishing to create a good specular reflection. The SRAS instrument is currently not able to inspect optically rough surfaces, as it requires a specular reflection from the sampling surface. According to Smith et al. the SLM samples have a roughness of approximately 10 m Ra, which cannot be detected by the current sensor. The waves must propagate to the detection spot and this is not the case for very rough surfaces. Therefore, it is not likely that this inspection method can currently be used for in-situ metrology.
3. SLM process representation
The state of the art review in the chapter before showed that extensive research work has been carried out on sensors that can monitor the process quality in a SLM machine. These different sensors can monitor, and sometimes measure, a wide range of process parameters, see Figure 30. Monitoring and measuring all these process parameters will enable the creation of a robust SLM process.
Figure 30 SLM process parameters. Source: Spears et al., 2016.
Some of the parameters in Figure 30 are predefined, for example laser frequency and packing density of the powder particles. Other parameters in Figure 30 can be controlled, for example scan velocity of the laser and layer thickness of the powder bed.
However, it is often unsure what the mechanical properties and dimensions of the part will be when it comes out of the process.
Therefore, the SLM machine is presented as a black box with an unknown output in Figure 31.
Figure 31 Black box representation of a SLM machine.
Seven categories are defined that cover all the process signatures, as can be seen in Figure 31. These categories are respectively [5]:
Melt pool;
Slice (powder bed after exposure & before recoating);
Powder bed (before exposure & after recoating);
Recoating system;
Part & building platform;
Weld emissions;
Keyhole.
Within these categories there are several parameters and process signatures that can be measured, for example size and shape of the melt pool. Then there are several different sensors that can monitor or measure these parameters and process signatures.
The consistency of the size and shape of the melt pool provides information about the consistency of the SLM process. An increase in size of the melt pool might suggest too little powder in that area.
In that case, the power density will be too high, which can lead to more molten powder, hence a bigger melt pool. By detecting deviations in size and shape of the melt pool it is possible to retrieve information about the process consistency, but not about one specific failure mode, rather a range of possible failure modes. Combining this shape and size information with the intensity information of the melt pool, it is possible to narrow down the possible failure modes. If the shape of the melt pool changes, but the intensity value of the melt pool stays the same, it is possible that there is a contamination in the powder bed.
The slice can provide information about the geometrical and dimensional accuracy of the product that is being printed. With image processing techniques, like edge detection, it is possible to measure the dimensions of the part in the layer that has just been exposed.
The powder after recoating should be homogeneous and equally spread over the previous layer. A camera and accelerometer are being used in the research of Kleszczynski et al. to detect super-elevated regions that protrude through the powder bed [6, 7, 11, 12]. As discussed before, the accelerometers measure acceleration due to vibration of the recoating mechanism. Craeghs et al. use images to detect disturbances in the powder bed by recoater wear for example [9, 10, 22].
The building platform has a prescribed surface flatness. Imagine if the building platform would not be completely flat, but angled for example.
That would immediately influence the quality of the powder spreading and thus the quality of the SLM process and final part. The low coherence interferometry (LCI) sensor described by Neef et al. can assure the flatness of the building platform and confirm that the building platform is not mounted tilted [27].
Emissions from the welding process are secondary process signatures of the laser-powder (tool-material) interface. These emissions provide information that can be directly correlated with the actual interaction between the laser beam and the powder. Several researchers observed that the optical emissions correspond to the absorbed laser power [23, 35].
