Additive manufacturing, also known as 3D printing, builds parts through a layer-by-layer process. This approach is in contrast to conventional metal removal operations that are subtract in nature by taking a substrate and conducting machining techniques such as drilling to manufacture a part.

One of the benefits of doing additive manufacturing is its ability to produce parts with complex shapes and designs. But progress in developing additive manufacturing into a useful method has been slow due to such issues as slow turnaround times and cost. In addition, using additive manufacturing to process metals such as aluminum has been challenging.

In a previous TLT article,¹ a new aluminum alloy also containing titanium, iron, cobalt and nickel that consists of three major phases exhibits greater mechanical strength after undergoing the widely used 3D printing process, selective laser melting. When the metal is processed, solid precipitates known as intermetallic rosettes are formed in the molten metal matrix. Interactions among these hard phases with soft regions represent the source of the strength of this aluminum alloy. Compression values in excess of 900 megapascals were found in certain regions of this alloy.

The success of this development still has not allowed researchers to overcome the problem in how to achieve process optimization when 3D printing specific metal alloys. Xiao Shang, graduate student in the department of materials science and engineering at the University of Toronto in Toronto, Ontario, Canada, says, “Figuring out a systematic approach for optimizing the 3D printing of a specific metal alloy into a particular shape remains as a major challenge preventing the wider adoption of additive manufacturing. Currently, industries and researchers are relying on trial-and-error which is unfortunately time consuming and costly.”

This strategy is used in both offline empirical and analytical techniques, and online control methods where adjustments are made during the printing process. The latter are typically more accurate and rapid but still are detrimental from time and economic standpoints. Shang says, “One example of the expense in an online process is the need to use high-speed cameras each of which can cost in excess of $200,000 to conduct a proper evaluation.”

Shang indicates that machine learning is becoming a useful tool for working to develop additive manufacturing processes due to its ability to quickly learn about process-structure-property relations, swift inference time and the capability to transfer the knowledge obtained to other applications using different metal alloys. But only limited work has been done with additive manufacturing that uses laser directed energy deposition.

A new machine learning technique has now been developed that facilitates the optimization of additive manufacturing using laser directed energy deposition. This method has been evaluated on two metal alloys and been found to be transferable to a third. 

Accurate Inverse process optimization
Shang and his colleagues developed a framework known as Accurate Inverse process optimization to demonstrate how machine learning can be used for optimizing laser directed energy deposition. He says, “In the Accurate Inverse process, the user first determines the specific target, which in the case of 3D printing is the specific metal part to be manufactured from a particular alloy. Once identified, the framework will work backward to figure out autonomously the proper process parameters to achieve the objective of producing the metal part. This is known as an inverse approach because typically, a series of process parameters are tried experimentally in an effort to move forward and produce the metal part.”

The researchers developed Accurate Inverse process optimization through the use of machine learning and a genetic algorithm. Initially, they conducted a series of experiments to produce a vast series of data sets that were used to train the machine learning models. Figure 1 shows the experimental setup used and a metal part produced during the study is seen in Figure 2. 


Figure 1. This experimental setup was used with the machine learning framework to assess how Accurate Inverse process optimization was able to facilitate 3D printing of specific parts with two metal alloys. Figure courtesy of the University of Toronto.


Figure 2. Laser directed energy deposition was the 3D printing process used in this study to better understand how artificial intelligence (AI) can optimize additive manufacturing of specific metal alloys. Figure courtesy of the University of Toronto.

Shang says, “The alloys used in the study were 316-L and Ph17-4 steels. These are both stainless steels that have been widely used in 3D printing applications. Ph17-4 exhibits twice the strength of 316-L and is widely used as a bearing material.”

Various operating conditions were used by the researchers to produce results from single-track, multi-track and multi-layer printing. Shang says, “Single track represents a one-dimensional line, multi-track is a two-dimensional figure such as a geometric face or flat surface and multi-layer is the process that leads to the manufacture of a three-dimensional object.”

With this data in place, the researchers uploaded it to the machine learning model which Shang characterizes as a “black box.” Of particular interest was an analysis of the melt pools formed during the many experiments. These are regions of superheated metal near the laser/alloy interface. 

By using fabrication parameters including laser power, scan speed, power feed rate and hatch spacing, the researchers were able to train the machine learning model to determine optimal process parameters from customizable objectives in as short a period as one hour. Shang says, “We also found that the machine learning model is translatable to a third alloy, nickel.”

Shang indicates that the researchers did not consider the thermal history of the metal in predicting operating parameters. He says, “We will need to include a thermal model as part of the Accurate Inverse process optimization. In addition, we will work with other metal alloys such as aluminum and titanium to prepare a comprehensive data set.”

Ultimately, the goal of the research is to set up a closed loop control system to use machine learning to assist users with improving process parameters as they are in the midst of a specific additive manufacturing process. 

Additional information on this work can be found in a recent article² or by contacting Shang at xiao.shang@mail.utoronto.ca or Professor Yu Zou at the University of Toronto, mse.zou@utoronto.ca. Further details are available at the website for Professor Zou’s research group (www.zou-mse-utoronto-ca.net/). 

REFERENCES
1. Canter, N. (2024), “New high-strength aluminum alloy suitable for additive manufacturing,” TLT, 80 (10), pp. 14-15. Available at www.stle.org/files/TLTArchives/2024/10_October/Tech_Beat_II.aspx.
2. Shang, X., Talbot, A., Li, E., Wen, H., Lyu, T., Zhang, J. and Zou, Y. (2025), “Accurate inverse process optimization framework in laser directed energy deposition,” Additive Manufacturing, 102, 104736.