Real-time AM monitoring opens up new process control opportunities

Additive manufacturing (AM) in metals opens up new product design possibilities, but process development and qualification can be challenging. Laser powder-bed fusion (LPBF) works on a small scale, using rapidly moving melt pools that selectively melt and solidify the powder feed-stock, transforming it into dense, intricate components. A concern for manufacturers, particularly in applications where components will be subject to cyclic loading, is that minor inconsistencies in this melting behaviour could produce local anomalies that impact on the operational performance and longevity of the part.

Eliminating unwanted process characteristics can involve long process development times as we use extensive post-build testing - destructive and non-destructive - to evaluate part performance. This then leads to an iterative cycle of process parameter optimisation, part re-design and further testing to certify that the output of our production process is now acceptable. This all takes time and money. Subsequent process control and post-processing costs may also be significant.

Minimising development lead times and process control costs is critical to more widespread adoption of metal AM as a mainstream production process. We need to find a robust way to detect, identify and eliminate the cause of defects without complete dependence on post-build testing. In-process detection and root cause analysis gives us the opportunity to systematically eliminate defects through better process design, and ultimately to repair defects that arise from inherent process variation. Such advances bring the possibility of right-first-time AM components closer to reality.

Image above - InfiniAM Spectral 3D visualisation of delivered laser power (left) and melt pool data (right)

This article focuses on new developments in real-time monitoring of the laser melting process that make this sort of rapid process development and robust process control possible. Before we look at these technological advances, we will review how anomalies can arise and consider their impact on part performance.

Where do LPBF anomalies come from?

Laser powder-bed fusion builds up parts layer-by-layer using a focused laser beam that generates a small (typically c. 150-200 micron wide) and rapidly moving melt pool. Once the laser spot moves on, this melt pool solidifies into a weld track that overlaps with previous scans and partially re-melts previous layers to form a dense, detailed component.

Ideally, we want our solidified metal to exhibit 100% density, with no pores or defects that reduce its strength and durability. This requires stability of the melting process, with consistent melting conditions and thermal history across all regions of the part. Of course, we try to design our build process to achieve this, but there are a number of failure modes that can lead to minor defects.

1. Scan vector errors

A build file consists of hatching patterns and border scans, whilst special parameters may be chosen for down-skin regions and support structures. When the component is sliced and the scan vectors and parameters are determined, problems can occur at the interfaces. Poor quality triangulation of the surface models can also make parts difficult to slice and scan correctly. These issues can sometimes lead to hot spots or dead zones where the energy delivery is inconsistent. Build preparation software can simulate the exposures and demanded energy input for each layer to highlight such anomalies, enabling alternative strategies to be used.

Image above - 'heat map' in QuantAM build processor, showing total energy delivered across a build layer, highlighting regions of overlapping scans with higher local energy input.

2. Local processing conditions

We choose our process parameters to deliver a specified amount of energy to achieve the desired melting effect. We want the melt pool to be of a controlled size and depth, such that it overlaps with the neighbouring weld track and securely welds to the layer below, whilst avoiding excessive penetration of the laser energy into the substrate. I explored these points in my article X marks the spot - find ideal parameters for your metal AM parts.

The temperature of the substrate affects how the material responds to the new energy that we are delivering. If heat that has been input in previous layers is unable to dissipate, then the substrate will be at a higher temperature when the next layer is processed. We now have a higher effective energy input than we planned, such that we move our operating point towards the keyhole formation zone:

We typically see temperature build-up in overhanging regions, where there is limited conduction of heat down through the component. This effect is particularly marked in materials with low thermal conductivity, such as Ti6Al4V. It can result in discoloration of down-skin regions, as well as internal porosity in the bulk of the overheating region due to keyhole formation. We may want to adjust our parameters in such regions to reduce overheating.

As I explained in Gone with the wind - how gas flow governs LPBF performance, variations in gas flow across the build plate can also lead to changes in laser energy absorption.

3. Scanning precision

Once we have a good scan strategy in place, the next challenge is to execute it precisely. The optical system is required to direct the laser spot along the programmed scan vectors, turning the laser on and off as it does so to create the individual melt tracks. Any small inaccuracies in positioning the spot, or inconsistencies in the response of the laser, will result in local variations in the amount of delivered laser energy. Too little energy can result in lack of fusion, whilst too much can result in formation of a deep 'keyhole', both of which can produce unwanted porosity.

These effects are most likely to be seen at the beginning or end of scan vectors, where the galvo mirrors have to re-position the laser spot as the laser is switched on and off. This image shows examples of keyhole porosity in an etched sample (ringed in red), where the relative timing of galvo movement and laser firing was not optimal.

Whilst we can optimise the settings on our optical system to minimise these sorts of systematic failure modes, some level of variation in positioning and firing is unavoidable.

4. Powder condition

Another inherent source of variation is the powder feed-stock. Powder comes in a range of particle sizes and shapes - this image shows an example of poor quality powder. The thickness of the layer and the amount of energy absorbed can therefore vary slightly across the powder bed. Some grains of powder can also contain trapped gas from the atomisation process, which will react to the laser in a different manner to a solid grain, and can result in pores in the solidified part.

5. Dosing

The powder re-coating mechanism and the machine Z-axis work together to create even layers of new powder as the build progresses. Wear or damage to the re-coater and poor flow of the powder, can lead to uneven dosing and hence inconsistent melting. Severe cases of short-dosing can result in de-lamination and build failure.

6. Spatter

The LPBF process involves some disturbance of the powder bed and various particles are emitted from the melt pool. Emissions include fine condensate, entrained powder particles, and welding spatter. In Gone with the wind - how gas flow governs LPBF performance, I discussed how some of the process emissions, especially heavier spatter particles, can land elsewhere in the powder bed. Where the laser encounters a large spatter particle on top of the powder bed, it will have more work to do to melt this extra material. This can lead to powder below the spatter particle being shielded from the laser energy, leading to lack-of-fusion porosity.

Image above - spatter shielding leading to lack of fusion porosity.

Spatter can also lead to defects as a result of irregular dosing. Oversize particles that are loose on the surface of the powder bed will generally be swept away by the next dosing layer. However, where spatter has been partially melted, either by the laser or by semi-sintering to the hot top surface of the component, it may stand proud of the surface of the previous layer, creating an obstacle for the re-coater which can lead to localised short dosing behind the spatter particle:

Short dosing can lead to porosity in the next layer, as there will be insufficient material to produce the weld track, and the laser energy will penetrate deeper into the layers below in the that location.

So, there can be a range of causes of porosity in our build. Some of these we can avoid through process refinement, whilst others are endemic and thus must be either tolerated or detected and corrected.

What are the consequences of process anomalies?

Fatigue is a progressive failure phenomena associated with the initiation and propagation of cracks to an unstable size. When the crack reaches a critical dimension, any further stress causes sudden failure. Fatigue is dangerous as it occurs over time and does so at stresses that often lower than both the ultimate tensile strength (UTS) and the yield strength. 

In LPBF parts, fatigue failures generally start from a surface or sub-surface defect which acts as a crack initiator. Spherical pores are relatively benign, whereas irregular pores and cracks with small internal radii are more concerning. Under cyclic stress, the defects slowly grow due to a high stress concentration around the crack tip each time the stress becomes tensile. If multiple defects are close to one another, these small cracks can join to form larger defects. Eventually, the crack becomes large enough to weaken the component to the point that further cyclic loading causes brittle failure of the remainder of the material. The rate of crack growth is dependent on the applied stress level, with higher stress leading to shorter life.

LPBF parts perform well in comparison with cast and wrought parts under fatigue loading. The graph below shows high-cycle axial fatigue test results for Ti6Al4V specimens built with 30 micron layers, generated by the EU-funded AMAZE project.

The challenge for manufacturers seeking to use LPBF to produce components with good fatigue performance is to assure themselves that their AM process is producing parts that are free from the types of defect that will lead to premature fatigue failure. This often involves destructive testing and inspection of parts and test artefacts, extensive use of CT scanning to spot the tiny defects that can act as crack initiators, and post-process surface treatments or hot isostatic pressing (HIP) to suppress any defects that remain. It is not unusual for aerospace companies to spend more on quality assurance and post-processing of AM parts than they do on building them in the first place.

Detecting defects as they occur

The following discussion relates to Renishaw's new InfiniAM Spectral AM process monitoring technology, available on RenAM systems from 2018.

If we can detect inconsistencies in our process as they arise, then we have a chance to reduce this dependence on post-build analysis and post-process modification. We can react to gross errors as they occur, stopping the process and avoiding wasted time and potential damage to equipment that can arise if we continue. A detailed view of process behaviour also supports investigation into more subtle defects, opening up opportunities for process refinement as well as, potentially, for adaptive process control.

Process monitoring challenges

So what information do we need to be confident that our parts are conforming? The following questions are important:

• Did the re-coater deliver an even dose of powder that fully covers the previous layer? Is there any evidence of re-coater damage?
• What laser energy has been delivered to each part of our build layer? Is this consistent with what we planned?
• How did the material react to the energy that we delivered? Did we achieve a stable, consistent melt pool? Were there any anomalies that might indicate over- or under-melting?
• How did this build compare to previous builds of the same part?
• Where we see anomalies in the melting process, can we use this information to direct our non-destructive testing? Can we quantify and sentence any potential defects that we highlight in this way?
• Can we take action to correct minor defects through controlled, localised re-work before we proceed to the next layer? Can we stop a minor defect in one layer from snowballing into a fatal flaw?

Layer camera

To address the first question, we need a layer camera to monitor the re-coating process, checking the condition of the powder bed at the beginning and end of each layer. Image analysis can reveal inconsistencies in the dosing process, such as damage to the re-coater leading to short-dosing and build failure, as shown below:

Image above - photographs of the build area at the start of the layer immediately after dosing (left) and at the end of the layer (right). In this case, localised part distortion has damaged the re-coater blade, creating dark stripes where insufficient powder has been spread. These in turn led to flaws in the built component.

Melt pool and laser sensing

When you look through the window of your LPBF machine, you will see a glowing, rapidly moving melt pool, with a bright plasma above it and sparks emerging from it. The melting process emits radiation across a broad spectrum and so we need a range of high-speed sensors to monitor its behaviour.

• A photo-diode tuned to the laser frequency, mounted within the optical module, to measure the delivered laser energy.
• A visible wavelength sensor that measures the intensity of the plasma.
• An infra-red frequency sensor that assesses the thermal emissions from the melt pool. This must have a narrow field of view that tracks the laser spot so that it is not affected by other hot objects on the build plate.
• High resolution position feedback from the galvanometer mirror encoders to record the actual spot position - where the laser went, rather than just the commanded position. This is spatially and temporally synchronised with the other sensor data so that all of the information can be accurately mapped and visualised in 4 dimensions: x, y, z and t (time).

The sampling rate for these sensors must be fast enough to keep pace with the moving melt pool whilst providing sufficient resolution to spot fluctuations that indicate potential defects. The rapid melting and cooling phenomena require a sample rate of 100 kHz, or one reading every 10 micro-seconds.

Image above - optical scheme for Renishaw's MeltVIEW and LaserVIEW process monitoring system, available on RenAM machines from 2018. The optical sensors are all passive and do not impinge on the optical delivery path, so processing parameters are unaffected.

Analysing process data

With readings coming in from multiple sensors at 100 kHz sample rate, we quickly generate large amounts of data. We require tools to help us to visualise and analyse this data to transform it into useful information and actionable insight.

3D visualisation

Visualisation of sensor data in three dimensions is invaluable for building an understanding of process behaviour in the context of a particular build. We can view the whole part or zoom in to inspect regions of interest, using thresholds to pick out anomalies for further investigation.

For instance, we can review delivered laser power to identify any areas of over- and under-melting that might arise from deficiencies in our scan strategy or the dynamic performance of our optical system. Here we are expecting consistent data with very few anomalies.

It is also possible to use this energy input signal as a benchmark to contrast the melt pool emissions signals against. Any marked deviations in the two signals would indicate a potentially anomalous region requiring further investigation.

We can also review the melt pool response to this energy input, highlighting hidden 'hot spots' inside the component that indicate heat build-up and keyhole formation, as well as areas of weaker melt pool emissions due to local variations in the powder bed, or where spatter is encountered.

Image above - InfiniAM Spectral 3D vidualisation of MeltVIEW sensor data.

2D layer analysis

When we spot an anomaly, we will want to investigate further. This is where 2D analysis is helpful, looking at the data from a single layer, or scrolling up and down through successive layers to understand defect propagation. In this image we are looking at delivered laser power across a layer at a resolution of 40 microns, highlighting stripe overlaps in our hatching scans.

Another key analytical tool is comparison with other data sets, such as those from the other machine sensors for temperature, pressure, oxygen content and other machine events. We can also use previous data sets by evaluating differences with previous known-good builds, which may have undergone further post process analysis such as non-destructive testing (NDT) or cut-ups. These data sets act as a 'golden reference' and allow us to understand process variation from build-to-build and machine-to-machine. With a sufficient body of data behind them, these may ultimately offer a reference point for us to confidently infer part quality from the process emissions sensor data alone. 

Image above - InfiniAM Spectral comparison of the same layer for two different builds, showing consistent melting performance

Data handling

With large volumes of sensor information to collate and analyse, an appropriate data handling architecture is needed. To facilitate real-time analysis, raw sensor data is exported from the AM machine at the end of each layer to a data collector PC that compiles it into a more compact volume file for immediate viewing by visualisation clients. Layer data can be viewed as it arrives for immediate diagnosis, whilst 3D visualisations are built up on demand as the build progresses.

Image above - InfiniAM Spectral data handling. The data collector PC features substantial data storage, whilst visualisation client computers are equipped with high-performance graphics cards.

Sending data layer-by-layer removes the need for very large file transfers, maintaining compatibility with existing network infrastructure - up to 10 lasers can be monitored simultaneously on a 1 GB network.

Process control opportunities

This high fidelity AM process information opens up future opportunities for process control, reducing post-processing costs and potentially eliminating defects at source.

Process traceability

The first obvious benefit of high fidelity process data is that we now have a detailed record of the actual delivered energy and the melting behaviour. This provides evidence of successful production process execution and is helpful when diagnosing faults during process development.

Directed CT inspection

Comparison of data with known-good builds enables us to spot inconsistencies that experience may tell us correspond to defects in the build. If we set a threshold for the difference in melt pool response between our known-good part and each production build, we can record the location of any such anomalies in the build. The locations of these anomalies could be used to direct micro-CT scanning to check for porosity in those regions only, reducing time-consuming 100% inspection.

Future versions of InfiniAM Spectral will enable overlaying of MeltVIEW and CT data to identify correlations, contributing towards machine learning to make better sense of these large data sets.

In-layer defect correction

As we have seen, some causes of melting process variation are inherent to the process and cannot be eliminated altogether through process or machine design. Defects as a result of encountering spatter on top of the powder bed, for instance, could be detected by changes in the melt pool signature, indicating that insufficient melting has taken place.

It is unlikely that changes to the melting process could be made sufficiently quickly to counteract these transient events as they occur - by the time the change in melt pool response is detected and reacted to, the laser will have moved on. However, if we can spot such anomalies in the melt pool response on the AM machine as the layer is produced, then we could perform corrective re-work in just those precise locations at the end of the layer. The idea is to prevent small defects from accumulating in subsequent layers.

Whilst such adaptive techniques are not the locked-down process that many manufacturers tend to prefer, it may be possible to demonstrate defect reduction and superior fatigue performance in the as-built component, which is surely an attractive proposition. Further work is needed to correlate spatter shielding with melt pool signatures, and to work out optimum re-work techniques, but this is a highly promising field of research.

Summary

LPBF builds up parts from millions of laser exposures which must be delivered with great precision if the part is to be fit for purpose. The melting process also exhibits some inherent sources of variation that occur over very short time periods.

New real-time spectral monitoring technologies can provide the necessary high-speed, high-resolution data from our melting process, enabling traceable production and rapid process optimisation.

Such data also opens up new possibilities for process control, detecting defects as they arise and dealing with them as the build progresses, driving us closer to the ideal of defect-free AM components.

More details

Web: InfiniAM Spectral AM process monitoring technology

Video: InfiniAM Spectral laser and melt pool monitoring

Brochure: H-5800-3916