Performance of eddy currents for the in-situ detection of defects during PBF-LB metal AM

Near-surface and sub-surface defects are one of the main factors currently hindering the growth of metal Additive Manufacturing applications. This is because they reduce the performance of the produced parts and lead to significant production, qualification, and certification costs. It is therefore desirable to accelerate the development of a technique enabling early detection of these defects – ideally during the process. 

Only a few physical principles are available to achieve this goal: eddy currents, ultrasound, thermography, and X-ray. These detection technologies are based on the physical changes in the material caused by the defects (difference in electrical conductivity, acoustic properties, heat-transfer, x-ray absorption, etc.). They can, however, only be used in-situ if they meet certain detection requirements (accuracy, defect types) within the constraints of the Additive Manufacturing process (speed, surface roughness, machine environment) and machine integrability [1, 2].

In this study by Carl Zeiss AG, AMiquam, and EOS, the performance of eddy currents was assessed in this context. In-situ measurements were taken using the AMiquam Eddy Current W1 to document potential industry use cases and assess the actual performance of the product. The eddy current measurements have been achieved by instrumenting the machine recoater with shielded absolute coils (5.8 mm outer diameter (OD) and 200 kHz interrogating frequency, resulting in a theoretical electromagnetic penetration depth of 0.95 mm in the Inconel 718 material from which the components are made).

About fifty cuboid samples were additively manufactured on an EOS M280 Laser Beam Powder Bed Fusion (PBF-LB) machine at the EOS Applications Centre. These contained process-induced defects (lack of fusion and keyhole) and seeded/designed voids; additional samples for post-processing measurements were also manufactured. These samples were additively manufactured such that the sensor’s sensitivity could be assessed in different use cases:

1. Process variation without seeded defects (see Fig. 2 and Fig. 4), where the volumetric energy density has been varied. The volumetric energy density E_v is defined as E_v (J/mm³) = P / (v · d · t) where P stands for output laser power (W), v stands for scan speed (mm/s), d stands for scan spacing (mm), and t stands for powder layer thickness (mm)
2. Seeded defects in regular array geometries without process variation (Fig. 3 and Fig. 7)
3. Single seeded defect without process variation (Fig. 5 and Fig. 6)

Seeded defects in regular array geometries were investigated to mimic an averaged defect density, which is calculated by the number of defects within the sensor’s probe volume. By varying the distance between these defects, the effect of different defect densities can be assessed.

By transforming the in-phase and out-of-phase electrical signals of the eddy current measurement, the lift off distance (the distance from the sensor to the top of the consolidated metallic part) and the electrical conductivity of the sample can be determined. The process variations produced homogenous porosity that linearly affected the electrical conductivity for small porosity. In this linear regime, the normalised electrical conductivity can be directly related to the density of the sample with a 0.1% resolution.

CT scans have been performed on one sample where three different parameter sets (a-b-c, see Fig. 2) have been applied one after the other and are presented below.

These data confirm previous results obtained by AMiquam showing that the detection resolution is 0.1% by volume, meaning a porosity increase of 0.1% can detected during the process using eddy currents. CT scans can be used to calibrate the eddy current response. Indeed, as different physical processes are involved in the two techniques, we expect some difference in the amplitude of the response of the two techniques. Using the CT scan result as the reference measurement, one concludes that a normalised electrical conductivity variation of 0.3% corresponds to a porosity of 0.12%.

In general, the detectability of defects depends on the defect density. If it is above a certain threshold, the SNR becomes larger than one, meaning that small defects of a certain size (<0.3 mm for example) can be detected – though not resolved individually – if there are enough defects present in the probe volume (i.e. if it meets a certain threshold).

Precisely determining the performances of the setup requires the separation and assessment of the various causes that affect and obscure the signal of the ideal pore: the probe vibration, material inhomogeneities, fused surface topology, and electric noise alongside the fact that the pore does not have the desired shape (Fig. 5c).

To reach this goal, a gantry scanner was calibrated to measure off-machine four 16 x 16 mm coupons, each manufactured with a a single pore of different diameters (0.7, 0.5, 0.3, 0.1 mm) located at their respective centres, 0.16 mm below the surface (the surface has been polished after the manufacturing, removing about 0.1 mm of material, 0.16±0.02 mm is therefore the thickness of the material on top of the ‘ideal’ defect). First, the scanner was equipped with the same experimental setup as the one used for the in-situ measurements (sensor ABS58) and then with a higher sensitivity sensor consisting of two 2.5 mm OD coils mounted in a send-receive configuration (sensor SR25).

The results are summarised in Table 1. The signal-to-noise ratio (SNR) is defined by the ratio between the maximal amplitude of the signal on the pore and the noise value. For the ABS58 sensor, the noise value has been evaluated to 20 mV with the signal variation caused by the vibrations of the scanner dominating the electronic noise (about 3 mV). For the SR25 sensor, the noise value is 5 mV, and the noise caused by the vibrations is mostly present in the other phase. We note that different sensor configurations may have different lift-off vs electrical conductivity responses. The resolution can be estimated using the spatial extension of the region with measurable pore signal, which is 5 mm for the ABS58 sensor and 3 mm for the SR25 one.

The C-scans on Figs. 5 and 6 show different responses to the same defects because of the different probe configurations. The ring response of the ABS58 probe is the consequence of the ferrite pot core surrounding the wire coil. This confines the magnetic field in a ring shape, which becomes visible when homogeneities smaller than the coil OD (a pore for instance) are measured. In addition, the usage of this ring response with a deconvolution can improve the detection capability of the system.

The impact of the surface roughness on the sensitivity of the reading has been investigated by comparing the data of the as-manufactured and the processed surfaces. As-manufactured surfaces do not exhibit a significantly worse SNR compared to the polished ones because the characteristic length of the surface roughness is much smaller than the sensor OD. Thus, the details of the surface topography do not have a significant impact on the EC signal. On the other hand, the EC probes may pick up some features of the laser trajectories, but we do not yet have systematic data about it.

Obviously, the ABS58 sensor operated with the standard parameters and along with the gantry scanner does not have the sensitivity to detect pores of size 0.1 mm, even with a polished surface. The filtering of the signal to remove the scanner vibrations (20 mV) should increase the SNR, potentially making the pore detectable. As expected, the SR25 sensor provides a better SNR, explained by the smaller coils and the send-receive configuration. Integration of arrays of these sensors in PBF-LB Additive Manufacturing machines is possible at the cost of the number of sensors that is about twice that of ABS58 arrays.

Several parameters can be optimised to improve the detection capabilities of the system. This includes the operating electrical parameters (frequency, gain, etc.) and the probe configuration (smaller coils, send-receive configuration). The reduction of the vibration of the scanner or recoater should also help to improve the SNR. However, the separation of the lift-off and electrical conductivity signals was successful in the in-situ experiments. Finally, to establish the ultimate sensitivity of a specific setup, the manufacture of pores with a high contrast compared to the base material (using, for instance, subtractive laser or spark erosion techniques) is needed. For PBF-LB, it is a challenging task to generate small pores in a controlled manner (i.e. with dedicated size, geometry, and position). On the other hand, the effort of testing and data analysis is much higher for stochastically generated, real process defects.

Conclusion

As stated in the ASTM 3166:20 standard, eddy currents are recommended to detect a variety of surface and near-surface defects, including cracking, porosity, inclusions, lack of fusion, residual stress, and surface defects. In this study, we documented the difference in the eddy current response between three types of discontinuities: lack of fusion; arrays of pores; and regions in the XY plane manufactured with the ‘skin’ parameters. The defects show different behaviours, especially when the two phases of the eddy current signals are considered. These results have been added to a larger set of data from which a robust classifier has been designed. Moreover, the penetration depth of the eddy currents allows us to track the presence of pores even when subsequent dense layers are consolidated over the pores, therefore enabling to determine whether the defects are ‘healed’ (i.e. the region around the pore is remelted during subsequent layer deposition so that the pore disappears) or remain in the part. The sensitivity to subsurface defects is the key advantage of this technology over all optical systems.

We emphasise that the sensitivity of the EC system presented in this study outperforms the state-of-the-art NDT technique by one order of magnitude. Indeed, NDT applications based on eddy currents do not usually enable submillimetre defects to be detected, especially with coils exceeding 5 mm OD as used in this study. An explanation comes from the regular motion of the scanner/recoater and the 2D nature of the deposition process.

Authors

Jonatan Wicht
Alain Berthoud
Bernard Revaz
AMiquam SA, CH 1196 Gland,
Switzerland

Harald Krauss
EOS GmbH Electro Optical Systems, D 82152 Krailling,
Germany

Frank Widulle
Carl Zeiss AG, Carl-Zeiss-Str. 22, 73447 Oberkochen,
Germany

Julian Schulz
Edson Costa Santos
Carl Zeiss Industrielle Messtechnik GmbH, Carl Zeiss-Str. 22, 73447 Oberkochen,
Germany

Contact

[email protected]
www.amiquam.ch

References

[1] Mission possible: The five-year plan to gain FAA and EASA acceptance of in-process monitoring, MAM Vol. 9 No. 4 p. 147ff
[2] ASTM Strategic Guide: AM In-Situ Monitoring TRL, ASTM E 3166: 2020, Standard Guide for Nondestructive Examination of Metal Additively Manufactured Aerospace Parts After Build