Laser-based additive manufacturing (LAM) offers the ability to produce near-net-shape metal parts with unparalleled energy efficiency and flexibility in both geometry and material selection. Despite advantages, these processes are inherently, as they are characterized by multiphysics interactions, multiscale phenomena, and highly dynamic behaviors, making their modeling and optimization particularly challenging. Artificial intelligence (AI) has emerged as a promising tool for enhancing the monitoring and control of additive manufacturing. This paper presents a systematic review of AI applications for real-time control of laser-based manufacturing processes, analyzing 16 relevant articles sourced from Scopus, IEEE Xplore, and Web of Science databases. The primary objective of this work is to contribute to the advancement of autonomous manufacturing systems capable of self-monitoring and self-correction, ensuring optimal part quality, enhanced efficiency, and reduced human intervention. Our findings indicate that 62.5 % of the 16 analyzed studies have deployed AI-driven controllers in real-world scenarios, with over 56 % using AI for control strategies, such as Reinforcement Learning. Furthermore, 62.5 % of the studies employed AI for process modeling or monitoring, which was integral to the development or data pipelines of the controllers. By defining a groundwork for future developments, this review not only highlights current advancements but also hints future innovations that will likely include AI-based controllers.

Schlieren monitoring techniques were applied to Laser Directed Energy Deposition (L-DED) and Laser Cladding (LC) processes to evaluate their performance in capturing thermal dynamics and spatter formation under varying process parameters. Two schlieren setups were compared based on their ability to detect changes in temperature, pressure and material state within the process zone. The first, using a background-oriented design, exhibited higher resolution and sensitivity to local thermal gradients and plume dynamics, while the second, with a broader field of view, demonstrated enhanced stability in monitoring cumulative thermal buildup. Optical flow analysis revealed a strong correlation between line energy and the magnitude of optical turbulence, particularly in regions where vaporization occurred, with a clear plateau observed between 110.6 J/mm and 243.9 J/mm, corresponding to optimal melt pool conditions. Beyond 243.9 J/mm, a significant increase in optical flow was observed, indicating plasma formation and enhanced turbulence. A dome-like schlieren structure consistently formed above the melt pool, expanding with higher energy input, offering insights into the balance between thermal buoyancy and vapor pressure. Additionally, the quadratic relationship between line energy and the schlieren dome volume of enclosed optical flow provided a means to identify energy-efficient and stable process conditions. The findings underscore the potential of schlieren-based monitoring for precise control and optimization of additive manufacturing processes, with implications for improving process stability and minimizing defects like spatter and porosity.

Real-time process monitoring is essential for ensuring stability and defect control in directed energy deposition - laser beam/metal (DED-LB/M). Schlieren imaging has proven to be a valuable tool for detecting refractive index variations in the process zone, providing insights into gas flow behaviour, shielding gas efficiency and process plume dynamics. However, schlieren imaging typically requires specialized optical setups, making integration into industrial systems challenging. This study explores an artificial intelligence-driven cross-modality transfer approach that enables the prediction of schlieren-induced refractive index variations from coaxial imaging data, eliminating the need for a dedicated schlieren setup. A background-oriented schlieren system was used to capture reference data, while a coaxial camera recorded the melt pool and surrounding process zone during DED-LB/M. A machine learning model was trained on the combined dataset, establishing correlations between schlieren activity and intensity variations in the coaxial images. The model successfully predicted schlieren-induced disturbances, allowing for the indirect detection of gas flow instabilities and shielding gas deficiencies. The results demonstrate that artificial intelligence-based analysis of coaxial imaging can provide schlieren-equivalent process information, making it possible to monitor refractive index variations, detect process deviations and improve defect prediction in real time. This approach enhances process monitoring capabilities in DED-LB/M, enabling cost-effective, scalable and easily integrable monitoring solutions for industrial applications.

A schlieren system, adapted for Laser Directed Energy Deposition, was used to monitor and analyze the process zone under various conditions, including deliberate contamination and parameter limits. This approach enabled the identification and correlation of process-induced defects with schlieren phenomena. Events and zones were characterized and qualitative categorized to validate schlieren monitoring as a diagnostic tool. Notably, a highly active and spatially confined schlieren formation was consistently observed above the melt pool. Using a tailored schlieren optical setup and simulations, schlieren patterns were linked to refractive index changes in process gases, enabling quantitative analysis. The refractive index within the hot gas dome over the molten pool was observed to range from 1.00000712 to 1.00875126, with fluctuation speeds reaching up to 210 m/s. As a result, a model was developed to describe the impact of refractive index dynamics on the performance of coaxial monitoring systems in laser processes. A case study using an exemplary imaging monitoring system demonstrated that schlieren phenomena can cause wavelength-dependent lateral geometric shifts of up to 228 µm, significantly affecting the accuracy of object-based monitoring outcomes. The findings offer critical insights into the complex interplay between refractive index variations and monitoring results, paving the way for refined monitoring strategies that enhance reliability and precision in laser cladding applications.

Bringing AI models from digital to real-world applications presents significant challenges due to the complexity and variability of physical environments, often leading to unexpected model behaviors. We propose a framework that learns to translate images into control actions by modeling multimodal real-time data and system dynamics. This end-to-end controller offers enhanced explainability and robustness, making it well suited for complex manufacturing processes. This end-to-end framework differs from traditional approaches that rely on manually engineered features by learning complex relationships directly from raw data. Labels are only required during training to define the observable feature to be optimized. This adaptability significantly reduces development time and enhances scalability across varying conditions. This approach was tested in the Directed Energy Deposition (L-DED) process, a laser-based metal additive manufacturing technique that produces near-net-shape parts with exceptional energy efficiency and flexibility in both geometry and material selection. L-DED is inherently complex, involving multiphysics interactions, multiscale phenomena, and dynamic behaviors, which make modeling and optimization difficult. Effective control is crucial to ensure part quality in this dynamic environment. To address these challenges, we introduce Joint Embedding Multimodal Alignment with Sparse Identification of Nonlinear Dynamics for control (JEMA-SINDYc). It combines an image-based JEMA monitoring model, which predicts the melt pool size using only the on-axis sensor with labels provided by the off-axis camera, and dynamic modeling using SINDYc, which acts as a World Model by capturing system dynamics within the embedding space. Together, these components enable the development of an advanced controller trained via Behavioral Cloning. This approach improves part quality by minimizing porosity and reducing deformation. Thin-walled cylindrical parts were produced to validate and compare this approach with other control strategies, including both open-loop and JEMA-PID. This framework improves the reliability of AI-driven manufacturing and enhances control of complex industrial processes, potentially enabling wider adoption of the process. 

This thesis investigates optical monitoring of laser-based additive manufacturing processes. The main focus is on spectral and Schlieren-based process monitoring for real-time defect detection and process optimization.

The research involves two new monitoring strategies for powder bed fusion - laser beam/metal, analysing energy absorption and process emissions in the first three papers. In paper A, the spectral signal of the laser-material interaction zone in powder bed fusion – laser beam/metal is analyzed using a coaxial and quasi-coaxial measurement setup. The study demonstrates that the detected spectral intensity distribution strongly depends on the angle of incidence between the measuring beam and the process zone. High-speed recordings and optical simulations enabled the development of a correction model for solid materials, which accounts for the numerical aperture of the measuring optics and laser intensity distribution across the working field. However, when measuring powders, strong signal fluctuations were observed, preventing a direct transfer of the correction model. This variation was attributed to differences in powder absorbance, which is further explored in paper B. This paper systematically investigates the absorbance behaviour of metal powders used in laser-based additive manufacturing. A high-precision spectrometer was used to measure 39 powders over a broad spectral range, examining the influence of aging, grain size, contamination and usage conditions. The study derives 20 technically relevant laser wavelengths, identifying those with improved process efficiency and stability. The resulting dataset provides a valuable foundation for laser parameter optimization by estimating the energy coupling efficiency for different materials. To enable in-situ absorbance determination in powder bed fusion - laser beam/metal, paper C introduces a method for high-resolution coaxial imaging of the powder bed at multiple wavelengths. This technique enables spatially resolved absorbance mapping across the entire processing plane, allowing for the detection of impurities, oxidation and foreign particles. The concept was experimentally validated using 20 different powders and further confirmed through optical simulations, ray tracing and comparative spectrometer measurements.

In three further papers, the laser material interaction and its effect on the refractive index variation of the gaseous process media is investigated for laser directed energy deposition. In paper D, a Schlieren imaging setup was applied to real-time monitoring of laser directed energy deposition, classifying different Schlieren phenomena and linking them to process instabilities and parameter deviations. A previously unknown recurring Schlieren structure was identified and its influence on coaxial imaging accuracy was analyzed using optical simulations that assigned a precise refractive index distribution to the observed Schlieren object. Paper E expands on these findings by implementing a background-oriented Schlieren system alongside shadowgraphy to analyze gas flow and refractive index variations from multiple orientations. A quantitative image processing method was developed to extract Schlieren intensity gradients and directional vectors, allowing for the correlation of Schlieren activity with process parameters. This approach enables the derivation of process boundaries based on optical flow measurements, offering a novel method for assessing process stability. Finally, paper F explores whether Schlieren-induced refractive index variations can be inferred from coaxial imaging data using an artificial-intelligence-based approach. A machine learning model was trained on background-oriented Schlieren data and coaxial imaging artifacts, demonstrating the feasibility of an indirect Schlieren analysis without requiring a dedicated Schlieren setup. By linking Schlieren structures to melt pool behaviour and process instabilities, the study contributes to real-time process monitoring and adaptive control strategies in laser directed energy deposition.

Overall, this thesis provides new insights into laser-material interactions, spectral absorbance properties and advanced optical process monitoring techniques. By combining spectroscopy, Schlieren imaging and artificial-intelligence driven analysis, this research advances the digitalization of laser-based additive manufacturing processes, improving process control, defect detection and manufacturing efficiency in powder bed fusion - laser beam/metal and laser directed energy deposition applications.

This study presents an approach for in-situ monitoring of laser powder bed fusion. Using standard laser optics, coaxial high-resolution multispectral images of powder beds are acquired in a pre-objective scanning configuration. A continuous overview image of the entire 114 × 114 mm powder bed can be generated, detecting objects down to 20 µm in diameter with a maximum offset of 22–49 µm. Multispectral information is obtained by capturing images at 6 different wavelengths from 405 nm to 850 nm. This allows in-line determination of the absorbance of the powder bed with a maximum deviation of 2.5% compared to the absorbance spectra of established methods. For the qualification of this method, ray tracing simulations on powder surfaces and tests with 20 different powders have been carried out. These included different particle sizes, aged and oxidized powders.

Absorbance is often used for simulations or validation of process parameters for powder-based laser materials processing. In this work, the absorbance of 39 metal powders for additive manufacturing is determined at 20 laser wavelengths. Different grain sizes and aging states for: steels, aluminum alloys, titanium alloys, Nitinol, high entropy alloy, chromium, copper, brass and iron ore were analyzed. For this purpose, the absorbance spectrum of the powders was determined via a dual-beam spectrometer in the range of λ = 330 - 1560 nm. At the laser wavelengths of λ = 450 nm, 633 nm and 650 nm, the absorbance averaged over all materials was found to increase by a factor of 2.4 up to 3.3 compared to the usual wavelength of λ = 1070 nm, with minimal variations in absorbance between materials. In the investigation of the aged or used powders, a loss of absorbance was detectable. Almost no changes from the point of view of processing aged and new AlSi10Mg powders, is expected for laser sources with λ = 450 nm. The resulting measurements provide a good basis for process parameters for a variety of laser wavelengths and materials, as well as a data set for improved absorbance simulations.

Process monitoring is becoming increasingly important in laser-based manufacturing and is of particular importance in the field of additive manufacturing [e.g. Laser Powder Bed Fusion (LPBF)]. Process monitoring enables a reduction of production costs and a lower time-to-market. Furthermore, the data can be used to create a digital twin of the workpiece. There are already many established processing head-integrated monitoring systems for such applications as the multispectral analysis of process radiation. However, the monitoring of complex signals in systems with F-Theta scanner lenses is very challenging and requires specially adapted optics or measuring sensors.

In this paper a potential arrangement for spectroscopy-based process monitoring in pre-objective scanning is presented. The process radiation was monitored using a coaxial and a quasi-coaxial observation system. The measurements were carried out on both a solid and a powder coated sample of 2.4668 (Inconel 718) to show the potential use of these systems in laser-based additive manufacturing. In order to obtain comprehensive data about the process signal, the process zone was analyzed at different angles of incidence (AOI) of the laser using a high-speed camera (HSI) and a spectrometer. The connection between the HSI and the spectral measurements is discussed. The ionization of the material and the formation of a plasma was observed and found to lose intensity as the angle of incidence increases. A model of the system that demonstrates the intensity of the emitted radiation of the plasma was created. It enables the measured values to be corrected. The corrected measurement data can be used to detect impurities or a non-ideal energy input across the entire processing field, which is a move towards robust process monitoring.

This study investigates the impact of powder aging on the degradation of AlSi10Mg powder during processing in laser powder bed fusion. Powder aging as result of handling, continuous storage and recycling is a fundamental concern for aluminum alloys as it introduces oxygen to the feedstock material. In this work, the analysis of the powder properties, affected by laser exposure and the aging procedure, showed a change of chemical and morphological characteristics of the powders in virgin and aged conditions. The oxygen content in the powders appeared to have a significant effect on the powders' surface appearance and light absorbance, gradually deteriorating the processability of the powders with the increase of oxygen level. Optical microscopy and X-ray computed tomography were used to analyze the porosity distribution in the printed part samples, identifying the origin, size and location of the pores. A direct relationship between the pore occurrence in final parts and the oxygen content in the powders was observed, revealing a higher degree of porosity in the aged powder sample (6.5%) in comparison with the virgin state (3.16%). The evolution of mechanical properties in the part samples after laser processing and powder aging was also studied, demonstrating a rapid decrease of ultimate tensile strength and elongation from virgin condition to aged.