Page 38 - IJAMD-1-1

International Journal of AI
for Material and Design ML for quality improvement in L-PBF

Figure 9. The indicators and parameters for different aspects of quality improvement.

to extract valuable insights and patterns. In addition, it the classification and analysis based on diverse ML model
establishes the relationship between real-time monitoring algorithms.
of critical information and product quality, enabling
the rapid detection and correction of potential issues. 3.1. Parameters optimization
This capability of ML enhances production efficiency In the traditional process of parameter optimization,
and ensures that the part’s quality aligns with specified researchers have typically used the DoE and simulation to
standards. find the relationship between the input parameters (such
In the following section, the application of ML is as laser power, scanning speed, and material composition
31
delineated into two primary categories: parameter design. ) and the characteristics of the printed product
32
optimization and in situ process monitoring. We deeply (such as porosity and yield strength). However, despite
explore and analyze a range of common case studies the DoE method’s partial simplification of experiment
aimed at quality improvement. Parameter optimization times, the metal printing process remains complex and
33
is conceptualized as a foundational application of ML time-consuming. The accuracy of results cannot be
methods in the conventional process of metal printing guaranteed, leading to substantial waste of metal powder
manufacturing. In such applications, the employed and incurring unnecessary costs.
models are predominantly straightforward. Therefore, In contrast, ML methods can effectively address
the classification approach in the parameter optimization the optimization of multiple parameters, identifying
segment is selected based on current hot topics in the optimal combinations to enhance component quality.
research domain. On the other hand, in the realm of in Once an ML model is established, it can adapt to various
situ process monitoring, the tasks to be managed are printing conditions and materials, thus offering optimal
considerably more complex and involve significantly larger parameter settings across different scenarios and reducing
data volumes, necessitating the use of more sophisticated human errors and costs. In addition, different prediction
models and algorithms. The choice of algorithms in this models are suitable for different prediction parameters.
segment profoundly influences the outcomes, prompting Therefore, selecting the appropriate model requires

Volume 1 Issue 1 (2024) 32 https://doi.org/10.36922/ijamd.2301