Integrating Machine Learning Model and Digital Twin System for Additive Manufacturing

This chapter describes optical diagnostics employed for in situ, on-line monitoring of metal additive manufacturing (AM) processes. Diagnostics are sought that can be incorporated into a feedback loop that optimizes build parameters to minimize defects in the construction of high-quality components. Optical diagnostics have been utilized for real-time process monitoring in a variety of industrial processes to include laser cladding, laser beam welding, machining processes, and metal emission control. AM of metals employing solid free-form fabrication by plasma-transferred arc processing may also be used in larger format 3D printing. An optical technique devised for surface roughness characterization during a standard machining process might be adapted to AM monitoring. A brief overview of some basic relationships between typical AM process parameters is presented below in the context of Air Force Institute of Technology's work with Ti-6Al-4V and cobalt chromium molybdenum alloys in a powder bed, selective laser melting system.

Monitoring and controlling Additive Manufacturing (AM) processes play a critical role in enabling the production of quality parts. AM processes generate large volumes of structured and unstructured in-situ measurement data. The ability to analyze this volume and variety of data in real-time is necessary for effective closed-loop control and decision-making. Existing control architectures are unable to handle this level of data volume and speed. This paper investigates the functional and computational requirements for real-time closed-loop AM process control. The paper uses those requirements to propose a function architecture for AM process monitoring and control. That architecture leads to a fog-computing solution to address the big data and real-time control challenges.

Additive Manufacturing is increasingly used in the development of new products: from conceptual design to functional parts and tooling. However, today, variability in part quality due to inadequate dimensional tolerances, surface roughness, and defects, limits its broader acceptance for highvalue or mission-critical applications. While process control in general can limit this variability, it is impeded by a lack of adequate process measurement methods. Process control today is based on heuristics and experimental data, yielding limited improvement in part quality. The overall goal is to develop the measurement science 1 necessary to make in-process measurement and real-time control possible in additive manufacturing. Traceable dimensional and thermal metrology methods must be developed for real-time closed-loop control of additive manufacturing processes. As a precursor, this report presents a review on the additive manufacturing control schemes, process measurements, and modeling and simulation methods as it applies to the powder bed fusion process, though results from other processes are reviewed where applicable. The aim of the review is to identify and summarize the measurement science needs that are critical to real-time process control. We organize our research findings to identify the correlations between process parameters, process signatures, and product quality. The intention of this report is to serve as a background reference and a go-to place for our work to identify the most suitable measurement methods and corresponding measurands for real-time control.

Additive manufacturing (AM) has revolutionized the manufacturing industry by offering flexibility, customization, and rapid prototyping capabilities. Complex geometries and low volume parts can be produced via AM in much lower lead time and cost compared to traditional manufacturing methods. However, ensuring the quality and reliability of AM parts remains a significant challenge due to variations in material properties and process parameters. Traditional material property analysis methods, that require experimental testing and computational modeling, are often costly and time-consuming and often not applicable to the complex geometries of AM components. These challenges are further magnified when repetitive sampling is required. To address these challenges, this paper presents a novel approach that combines experimental findings with Finite Element Modeling (FEM) to construct a digital twin (DT) model of AM parts. The DT model will serve as a virtual representation along with real-time monitoring, simulation, and optimization capabilities. Numerical simulations are conducted to correlate the resonance frequency of parts at different process conditions with their material characteristics, represented as RF Z-score values. Hence, the results suggest that such an approach provides manufacturers with cost-effective and efficient means of analyzing material properties and optimizing AM processes. Moreover, by leveraging digital twin technology, manufacturers can achieve greater process control and quality assurance in additive manufacturing. The findings of this study contribute to advancing the understanding and implementation of digital twin technology in AM; therefore, paving the way for enhanced productivity and reliability in additive manufacturing processes.

Complex geometry, variable macrostructure, anisotropy, distributed defects, surface roughness, and residual stresses of additively manufactured (AM) components challenge nondestructive evaluation processes developed and validated for conventional manufacturing. Electromagnetic and eddy current (EC) technologies were identified as promising for in-process, real-time monitoring, and post-process examination of AM components. Powder and solid specimens were prepared for electromagnetic property measurements and technique development and optimization. The post-process specimens were fabricated with roughness representative of surface conditions during fabrication inside the AM machine. Artificial and natural discontinuities were fabricated with the electrical discharge machining process and by varying the laser fusion parameters. Extensive computed modeling was conducted to develop an optimized array EC (AEC) sensor and technique for AM real-time monitoring. The AEC sensor could withstand temperatures up to 100ºC. Computational modeling was conducted to investigate the likelihood of metal powder disturbance by the EC sensor electromagnetic field. A real-time data acquisition system was assembled and integrated with an open-architecture laser powder bed fusion (L-PBF) test bed. The EC system performance for real-time monitoring of the L-PBF process was demonstrated through layer-by-layer testing of Inconel 625 coupons, with AM built discontinuities and lack of fusion. The L-PBF process was representative of a typical AM process used for super alloys and steels with preheat of substrate plate and AM test specimens to 100ºC. Existing AEC equipment and techniques were used for post-process examination of AM solid specimens at room temperatures. The AEC technology demonstrated excellent sensitivity to seeded and natural surface and near-surface subsurface discontinuities and surface topography during real-time monitoring and post-processing. Electromagnetic and EC techniques were demonstrated to examine powder before fabrication. The application of modeling tools reduced developmental time and improved efficiency. Further research and development was recommended to improve data acquisition process and transition AEC monitoring technology to other AM processes.

Digital twin (DT) and additive manufacturing (AM) technologies are key enablers for smart manufacturing systems. DTs of AM systems are proposed in recent literature to provide additional analysis and monitoring capabilities to the physical AM processes. This work proposes a DT framework for real-time performance monitoring and anomaly detection in fused deposition modeling (FDM) AM process. The proposed DT framework can accommodate AM process measurement data to model the AM process as a cyber-physical system with continuous and discrete event dynamics, and allow for the development of various applications. A new performance metric is proposed for performance monitoring and a formal specification based anomaly detection method is proposed for AM processes. Implementation of the proposed DT on an off-the-shelf FDM printer and experimental results of anomaly detection and process monitoring are presented at the end.

In situ process monitoring includes any technologies that monitor or inspect during an additive manufacturing (AM) process. This article presents the types, process considerations, and challenges of in situ monitoring technologies that can be implemented during an AM process. The types include system health monitoring, melt pool monitoring, and layer monitoring. The article discusses data analysis, and provides an overview of the integration of sensors into AM machines.

Digital twins for additive manufacturing (AM) have drawn much research attention recently, thanks to advancements in artificial intelligence and machine learning. Machine learning takes the process and measurement data from the manufacturing process to build data-driven models instead of physics-based descriptive models. The latter are usually hard to obtain for complex AM processes such as laser powder bed fusion. This study proposed a digital twin framework for the laser powder bed fusion AM process control and optimization. The framework is created based on the recently developed advanced point-wise scan control method. It consists of four components: digital twin of process design, digital twin of process control, digital twin of process monitoring, and digital twin of printed part. Their construction is detailed, and potential applications are demonstrated/discussed.

Biopharma Is Going Digital … Bit by Bit

The one-step AM process is considered the goal many researchers seek in the field of Additive Manufacturing (AM) of high-technology ceramics. Among the several AM techniques, only Powder Bed Fusion (PBF) can directly print high-technology ceramics using one step. However, the PBF technique faces numerous challenges to efficiently be employed in the PBF of ceramics. These challenges include the formation of cracks, generated thermal stress, effective laser–powder interaction, and low acquired relative density. This study developed a new preheating mechanism for ceramic materials using two laser systems to surpass beyond these challenges and successfully print ceramics with a single-step AM method. One laser is used to preheat the powder particles before the second laser is utilised to complete the melting/sintering process. Both lasers travel along the same scanning path. There is a slight delay (0.0001 s) between the preheating laser and the melting/sintering laser to guarantee that the melting/sintering laser scans a properly preheated powder. To further facilitate testing of the preheating system, a numerical model has been developed to simulate the preheating and melting process and to acquire proper process parameters. The developed numerical model was shown to determine the correct process parameters without needing costly and time-consuming experiments. Alumina samples (10 × 10 × 6 mm3) were successfully printed using alumina powder as feedstock. The surface of the samples was nearly defect-free. The samples’ relative densities exceeded 80%, the highest reported relative density for alumina produced by a single-step AM method. This discovery can significantly accelerate the transition to a one-step AM process of ceramics.

The chemical synthesis of polysiloxanes from monomeric starting materials involves a series of hydrolysis, condensation and modification reactions with complex monomeric and oligomeric reaction mixtures. Real-time monitoring and precise process control of the synthesis process is of great importance to ensure reproducible intermediates and products and can readily be performed by optical spectroscopy. In chemical reactions involving rapid and simultaneous functional group transformations and complex reaction mixtures, however, the spectroscopic signals are often ambiguous due to overlapping bands, shifting peaks and changing baselines. The univariate analysis of individual absorbance signals is hence often only of limited use. In contrast, batch modelling based on the multivariate analysis of the time course of principal components (PCs) derived from the reaction spectra provides a more efficient tool for real-time monitoring. In batch modelling, not only single absorbance bands are used but information over a broad range of wavelengths is extracted from the evolving spectral fingerprints and used for analysis. Thereby, process control can be based on numerous chemical and morphological changes taking place during synthesis. “Bad” (or abnormal) batches can quickly be distinguished from “normal” ones by comparing the respective reaction trajectories in real time. In this work, FTIR spectroscopy was combined with multivariate data analysis for the in-line process characterization and batch modelling of polysiloxane formation. The synthesis was conducted under different starting conditions using various reactant concentrations. The complex spectral information was evaluated using chemometrics (principal component analysis, PCA). Specific spectral features at different stages of the reaction were assigned to the corresponding reaction steps. Reaction trajectories were derived based on batch modelling using a wide range of wavelengths. Subsequently, complexity was reduced again to the most relevant absorbance signals in order to derive a concept for a low-cost process spectroscopic set-up which could be used for real-time process monitoring and reaction control.

Additive manufacturing (AM) has recently been accorded considerable interest by manufacturers. Many manufacturing industries, amongst others in the aerospace sector, are already using AM parts or are investing in such manufacturing methods. Important material properties, such as microstructures, residual stress, and surface topography, can be affected by AM processes. In addition, a subtractive manufacturing (SM) process, such as machining, is required for finishing certain parts when accurate tolerances are required. This finish machining will subsequently affect the surface integrity and topography of the material. In this research work, we focused on the surface integrity of Ti-6Al-4V parts manufactured using three different types of AM and finished using an SM step. The aim of this study was to gain an understanding on how each process affects the resulting surface integrity of the material. It was found that each AM process affects the materials’ properties differently and that clear differences exist compared to a reference material manufactured using conventional methods. The newly generated surface was investigated after the SM step and each combination of AM/SM resulted in differences in surface integrity. It was found that different AM processes result in different microstructures which in turn affect surface integrity after the SM process.

Vibrational spectroscopy in the mid (MIR) and near infrared (NIR) range provides structural information on virtually any organic compound and thus presents many interesting perspectives in both qualitative and quantitative analysis. These two techniques are powerful analytical tools for the identification and molecular characterization of pharmaceutical raw materials, intermediates and final dosage forms. NIR spectroscopy being fast, noninvasive, multiparametric, robust and easy to interface with a process, perfectly matches the measurement system requirements within the process analytical technology (PAT) framework. This chapter focuses on modern pharmaceutical MIR and NIR applications and covers (1) basic principles of both techniques including recent advances in instrumentation, sampling and data processing, (2) characterization of solid drug molecules and excipients, (3) nondestructive analysis of solid oral dosage forms, (4) secondary structural analysis of protein therapeutics, and (5) real-time process monitoring and control in drug substance and drug product manufacture. Strategies of MIR and NIR signal processing and multivariate NIR method implementation are highlighted and practical challenges of specific applications discussed. Finally, theoretical and practical aspects of NIR in vivo applications are briefly reviewed. Overall, the analytical versatility of infrared spectroscopy to be applied in various fields of pharmaceutical development, production and quality control is demonstrated and potential benefits and limitations of the distinct wavelength regions are scrutinized.

Recent Advancements in Additive Manufacturing of Metals and Alloys

This paper reports on the design, implementation and real-time operation of advanced process monitoring techniques for wastewater treatment plants. The paper presents the development of a software platform and its implementation in a full-scale wastewater treatment plant. The software platform allows the real-time execution of advanced process monitoring techniques that are used for fault and process upset detection and identification. The statistical methods are part of a supervisory control level that allows for integration of the different facets of process monitoring and control, including fault detection, knowledge extraction and controller action. Results obtained by the real-time execution of such algorithms are presented.