Optical Detection of Defects during Laser Metal Deposition: Simulations and Experiment

While laser metal deposition is a rapidly evolving method for 3D printing and the fabrication of high-quality parts by depositing materials on substrates, the quality of production depends on instrumental design and operational parameters that require constant quality control during the process. Igor Gornushkin and colleagues at BAM Federal Institute for Materials Research and Testing in Berlin, Germany studied the feasibility of using optical spectroscopy as a control method for laser metal deposition, and he recently spoke to us about this work. Gornushkin is the 2022 recipient of the Lester W. Strock Award from the New England Chapter of the Society for Applied Spectroscopy (SAS). This interview is part of an ongoing series of interviews with the winners of awards that are presented at the annual SciX conference, which will be held this year from October 2 through October 7, in Covington, Kentucky.

Your paper (1) examines possibility of detecting surface defects, such as cracks, cavities, and thuds, with optical emission spectroscopy (OES), both theoretically via simulations and experimentally via measuring the optical signal from a sample with artificial defects. How effective is OES in finding defects and do you have good data to support your conclusion? Have you found a way to estimate the percentage of defects that will be found using OES?

OES was proposed as one of many diagnostic tools for detection of defects during additive manufacturing (AM). We did this work in a framework of a large interdisciplinary project initiated in our Institute to help improving the AM process. The other diagnostic techniques employed were thermography, optical tomography, eddy current testing, laminography, X-ray backscattering, particle emission spectroscopy, and photoacoustic methods. Among this group, the OES seemed to be the simplest and easiest to try, so we set a proof-of-principle experiment and ran it under rather artificial conditions. We have not yet come to a stage when we could detect real defects; we only established the validity of the method in detection of artificial defect. The natural defects appear randomly and have unpredictable nature; to find the functionality between the optical signal and printing irregularity, special reference samples would be required that had defects of known types situated at known spatial position. We did not have such samples, therefore, we prepared them by carving artificial defects on a metallic plate. We then verified if OES is able of detecting them. The artificial defects were a strongly exaggerated version of real ones; having samples with only artificial defects, it was difficult to estimate the efficiency of OES for detection of real defects.

Answering your question, yes, with OES, we were able to reliably identify all artificial defects during the printing, although the detection sensitivity was decreasing with each next overlaid printing, because it smoothed underlying irregularities. More work would be needed to answer your question about the percentage of identified real defects.

What inspired you to consider optical emission spectroscopy as a means for online defect measurements?

It is a simplicity of OES plus our previous experience in classification of materials by various chemometric techniques, like using correlation or principal component analyses. The problem of detection of printing defects is alike to a calibration or classification one: in requiring a training set. The calibration requires a set of spectra from a defectless surface; having that, one should find a subset of spectra that are “spoiled” by defects. If a training set contains spectra that are characteristic to known defect types, such as cracks, cavities, or thuds, even more advanced classification is possible that identifies the specific character of a defect.

What benefits did you anticipate by using OES over other analytical techniques for your intended analysis?

Here we speak about diagnostic rather than analytical techniques, because we did not (and could not) use OES for elemental analysis. In comparison to the above-mentioned methods, OES is probably the best suited for the continuous online monitoring of the AM process; the optical head can easily be attached to any printing machine.

Briefly discuss your overall findings and their implications. Were you surprised by the results?

We developed a thermal model of the printing process and combined it with a geometric model of collection of light from a molten metal pool that forms during the printing. This allowed us to calculate emission signals produced by different types of printing defects (grooves, holes, thuds, ripples, and so forth) and compare them to those measured experimentally. By “emission signals,” we understand one of three metrics: integral intensity, temperature, or correlation coefficient, which we plot versus the printing distance during laser scanning. As we initially expected, the signals showed strong variation in the vicinity of defects and that variation turned out to be very specific to a defect type. For example, a surface thud produced the emission signal that looked like a spectral line, and a hole produced the signal that looked like a self-reversed spectral line. Other signals were also very characteristic to defect types. From that, we concluded that OES was capable of not only detection of defects but also identification of their types. That was quite a surprising observation.

What are the biggest challenges that you have encountered in developing this method? What options or alternative developments are available to overcome these challenges or to improve your approach?

The first challenge was a necessity of working with continuous rather than discrete spectra. Under the best printing conditions, the laser created a pool of molten metal on the printed surface with a bright hallo above it. These two produced a broadband emission spectrum, and even though plasma spectra randomly appeared, they were rare and stochastic and not useful for diagnostics. The next problem was strong instability of the emission signal; it was caused by fluctuations in the powder supply rate, appearance of fumes and particulates above the printed spot, and sporadic ignition of plasma. It was a challenge to distinguish the signal change due to the surface defect from that caused by other reasons. Another challenge was a limited spectral range of our spectrometers for the detection of thermal radiation. Our available range was 200–1650 nm, and this was not enough to measure quality blackbody spectra. The extension into further infrared (IR), for instance, to 3 μm, would be beneficial. Yet another challenge was a limited spatial resolution of our instrument. We collected light from a spot that was bigger than even a biggest artificial defect. To improve that, the Czerny Turner spectrometer with an 1D detector can be replaced by the imaging spectrometer with a 2D CCD detector; this could be realized by using a circular-to-linear fiber bundle attached to a slit of the imaging spectrometer.

In the computational domain, the challenge was a need for using a very fine spatial discretization to accommodate small defects; that resulted in many hours of computational time. So, optimization of the software is needed to make computations more economical. Beside the optimization, new features must be added in the model to make it more realistic—these are the flow of a liquid phase and supply of a metallic powder into a molten pool of metal on the printed surface.

What sort of feedback did you receive from your paper and the study results?

Even though the paper is recent (available since September 2021), it has been noticed and received three citations in both theoretical (2) and experimental (3,4) studies. Besides, the colleagues from our Institute, who work with 3D metal printing, rely on our results to continue using OES for monitoring the process.

Are there any additional analytical applications or processes where your findings might be beneficial?

Sure! OES can be used for monitoring and analysis of any process accompanied by optical or thermal emission. Regarding the material processing by lasers, OES is beneficial in laser welding, where plasma is created, and plasma spectrum can provide information about a composition of the welded metal and solder and the heat affected zone. In applications that combine a 3D metal printing and laser induced breakdown spectroscopy (LIBS) (several works are available on this topic), OES is the natural choice. In both the welding and 3D printing, the thermal-emission model, which we developed, can be used to roughly predict process parameters and desirable outcome without doing tedious optimization experiments.

What are your next steps regarding this research?

In theory, we will extend the model to accommodate the liquid motion and powder supply (these features are currently absent). In further experiments, we will do spatially resolved measurements with both the thermal camera and imaging spectrometer. In addition, we plan to use a combination of OES with absorption and light scattering measurements; the correlation of signals from these techniques will improve the detectability of printing defects. We also plan using LIBS for online monitoring the 3D printing; this will allow us to control compositional uniformity of a printed workpiece. Finally, we will implement the data fusion and machine learning methods to make the detection of printing defect a more robust procedure.


  1. I.B. Gornushkin, G. Pignatelli, and A. Straße, Appl. Surf. Sci.570, 151214 (2021).
  2. Murer et al., Int. J. Adv. Manuf. Techn.120, 3269–3286 (2022).
  3. Feng et al., Addit. Manuf. 54,102760 (2022).
  4. Pignatelly et al.,Mater. Test.64, 24–32 (2022).

Igor Gornushkin is a senior scientist at the BAM Federal Institute for Materials Research and Testing in Berlin, Germany. His major expertise is in fundamental and applied spectroscopy including LIBS, emission, absorption, and fluorescence with a focus on modeling and simulation of laser induced plasma and development of spectroscopic methodsfor environmental, industrial, and laboratory applications.

He received his PhD from the University of Florida in 1997 where he worked until 2008, when accepted a position at BAM. Over the last two decades, he and his colleagues developed a comprehensive theoretical model of laser induced plasma, numerous methods for plasma diagnostics and LIBS-analyses. Recent works include application of spatial heterodyne spectrometry for LIBS and Raman qualitative and quantitative analyses, laser induced plasma for chemical vapor deposition and texturing materials, and optical spectroscopy for controlling metal 3D printing. The instrumental works are always being accompanied by theoretical modeling and numeric simulations of relevant processes.

As of now, Igor’s research activity has resulted in ca. 150 publications and several book chapters, which collected over 3500 citations, several patents, and copyrights. He is a member of the editorial board of Spectrochimica Acta part B and recipient of the 2001 Elsevier SAB best paper award. Over the years, Igor has been a mentor to many PhD students from USA, German, and other Universities.