What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to establish a compact virtual test platform through a data - driven method in the laser powder directed energy deposition (LP - DED) process, so as to accurately predict the observable variables in the process and improve the understanding of the relationship between the input process parameters and the final product. Specifically, the research objectives include: 1. **Establish a virtual test platform**: Utilize the dynamic mode decomposition and control (DMDc) method, combined with uncertainty quantification techniques, to develop a virtual test platform that can accurately handle any process input and report the observable variables in the process. 2. **Improve prediction accuracy**: Through data - driven methods, reduce the problem of accurate prediction that is difficult to achieve in traditional physical models due to high computational costs, and ensure the robustness and reliability of the model when dealing with complex systems. 3. **Understand the relationship between input and output**: Through the virtual test platform, better understand the relationship between the input process parameters (such as laser power, scanning rate, etc.) and the final product quality, so as to optimize the process parameters and improve the reliability and performance of the product. ### Paper Abstract Digital twins play an important role as a bridge connecting the industrial age and the digital age in the manufacturing industry, providing great value. The current additive manufacturing processes can generate a large amount of process data, and if these data are effectively processed, they can be transformed into valuable decision - making information. Data - driven methods, such as reduced - order modeling and system identification, show great potential in managing these data. This research focuses on the laser powder directed energy deposition (LP - DED) process equipped with in - situ process measurement, aiming to develop a compact virtual test platform. This platform can accurately handle any process input and report the observable variables in the process. The virtual test platform is constructed using the dynamic mode decomposition and control (DMDc) method and combined with uncertainty quantification techniques to ensure the robustness of prediction. ### Research Background Additive manufacturing (AM) has attracted wide attention in the past few decades, and its advantage lies in enabling designers and engineers to create lighter and stronger products. However, unlike traditional manufacturing methods, additive manufacturing introduces new challenges, especially the significant uncertainty between the input process parameters and the final product. These uncertainties may include changes in material properties, shape defects, and geometric non - compliance, etc. Although these problems are manifested to different degrees in different additive manufacturing technologies, this research focuses on directed energy deposition (DED). ### Directed Energy Deposition (DED) Directed energy deposition (DED) is an additive manufacturing technology that melts metal powder by a laser heat source and deposits it onto a substrate. Sensors are commonly used in the DED process to record information such as the size, temperature, and light emission intensity of the molten pool. The DED machine used in this study is equipped with multiple sensors, including sensors for coaxial measurement of the molten pool size and temperature and an off - axis camera for measuring the distance between the nozzle and the center of the molten pool. ### Data - Driven Model The data - driven model used in this study relies entirely on empirical data for prediction without the need to understand the internal working principles of the system. These models use algorithms to identify patterns and relationships in the data and are suitable for dealing with complex nonlinear systems. The dynamic mode decomposition and control (DMDc) method is used in the study, which simplifies the processing of multiple control input points by incorporating control inputs and evaluating the impact of these inputs on the system state. ### Methods and Materials The FormAlloy X1/L1 DED machine was used in the study, and the material was Inconel 625. The experimental data set contains 26 individual experiments, each of which aims to investigate various process parameters in the DED process. The data were divided into a training set and a test set, with 25 experiments used to train the model and 1 experiment used to test. The data generation frequency ranges from 10 Hz to 500 Hz, usually 100 Hz. ### Virtual Test Platform Workflow 1. **Generate G - code program**: Generate a G - code program and convert it into a time - series representation. 2. **Standardize data**: Standardize the input data to zero mean and unit standard deviation. 3. **Predict state - space model**: Use the state - space model trained by the DMDc method to iteratively predict the observable variables in the next time step. 4. **Inverse transformation**: Inverse - transform the prediction results back to the original units. 5. **Uncertainty quantification**: Combine uncertainty information to provide the confidence of the model prediction. ### Mathematical Basis Dynamic mode decomposition (DMD) analyzes in dynamic systems.

Data-driven Virtual Test-bed of the Blown Powder Directed Energy Deposition Process

Probabilistic Digital Twin for Additive Manufacturing Process Design and Control

Towards a digital twin framework in additive manufacturing: Machine learning and bayesian optimization for time series process optimization

High-throughput, in Situ Imaging of Multi-Layer Powder-Blown Directed Energy Deposition with Angled Nozzle

Statistical Parameterized Physics-Based Machine Learning Digital Twin Models for Laser Powder Bed Fusion Process

Digital twin inference from multi-physical simulation data of DED additive manufacturing processes with neural ODEs

Toward the digital twin of additive manufacturing: Integrating thermal simulations, sensing, and analytics to detect process faults

Digital Twin in Process Planning of the Additive and Subtractive Process Chain for LMD and Micro Milling of Stainless Steel

Process Optimization and Distortion Prediction in Directed Energy Deposition

Deep Neural Operator Enabled Digital Twin Modeling for Additive Manufacturing

Data-Driven Prediction and Uncertainty Quantification of Process Parameters for Directed Energy Deposition

Application of Deep Transfer Learning and Uncertainty Quantification for Process Identification in Powder Bed Fusion

Inconel-718 processing windows by directed energy deposition: a framework combining computational fluid dynamics and machine learning models with experimental validation

Development of a Digital Twin for Additive Manufacturing

A Data-Driven, Non-Linear, Parameterized Reduced Order Model of Metal 3D Printing

Advancements in 3D Printing: Directed Energy Deposition Techniques, Defect Analysis, and Quality Monitoring

Numerical Simulation Development and Computational Optimization for Directed Energy Deposition Additive Manufacturing Process

Accurate Inverse Process Optimization Framework in Laser Directed Energy Deposition

IoT-Based Data Mining Framework for Stability Assessment of the Laser-Directed Energy Deposition Process

Investigation of Effect of Processing Parameters for Direct Energy Deposition Additive Manufacturing Technologies