This review synthesizes artificial intelligence applications in additive manufacturing across lightweight design, in situ monitoring, energy optimization, and cost-effective manufacturing. It reports results from separate prior studies and proposes a four-stage closed-loop framework connecting design, process planning, production, and continuous model updating, with an infrastructure stack and implementation roadmap. The integrated framework is proposed; many reviewed models remain limited to particular machines, materials, or simulated datasets, and cross-machine closed-loop validation remain

A better part begins before the first layer—and keeps improving after the last.

How artificial intelligence could connect the whole journey of 3D printing: from a lighter design to a machine that responds as it builds.

A review of existing studies, with a proposed connected workflow.

Four stages. One learning loop.

Design

Choose a shape that does its job with less material.

Plan

Choose its direction, supports, and printing settings.

Print

Read the machine’s signals. Adjust the next pass.

Learn

Carry the build’s information into the next design.

The review’s proposal: connect these decisions so that information travels forward—and useful feedback travels back. Paper §4.5 ↗

Can a warning become an adjustment?

Start a print. Change its conditions. Then give the machine a way to respond.

The sensor sees a steady print. No adjustment is needed.

Let the sensor change the next pass.

A shape can hold heat unevenly. Introduce extra heat accumulation.

Use the signal to adjust input during the next passes.

Too little input can leave layers poorly joined. Too much can overheat the build.

The twin follows the current build.

Illustrative process model · based on §4.2–4.5. Qualitative states, not measured performance. Models are assumed to be adapted to the same machine and material.

Monitoring sees a problem. Feedback changes what happens next. Previously affected material stays marked; an adjustment acts on later passes. Saving the build’s information illustrates the proposed next-build learning mechanism.

A lighter shape. A different build.

Design and manufacturing are coupled. Remove material and the part may need new supports. Turn it, and the support problem changes again.

Schematic comparison of one bracket geometry. Support needs depend on the part and printing process.

In this upright example, the arch spans empty space. Temporary columns carry it while printing.

Small wins. Across the workflow.

Examples reported in the reviewed studies. Separate experiments, materials, and machines.

Different questions. Different kinds of learning.

The review maps many learning methods onto practical engineering tasks. Here is how the pieces fit.

Cameras, heat sensors, and acoustic signals reveal what is happening during printing. A model can turn these signals into a process-state estimate: is the build behaving as expected? Fast local computing can support an adjustment while a cloud system trains and updates models. §4.5.2 ↗

Supervised learning learns from examples with known answers, such as an image labeled with a defect. Unsupervised learning looks for patterns or unusual signals without those labels. Reinforcement learning learns a sequence of actions from the feedback those actions produce. The review explores where these approaches fit the printing workflow. §3 ↗

A digital twin is a virtual model connected to information from the real machine. It can carry sensor histories and quality records into simulation, planning, and model updates. In the proposed framework, the current build informs the next one. §4.5 ↗

The review’s roadmap calls for shared sensor datasets, models guided by physical laws, and validation across machines and materials. These steps support the move from individual demonstrations toward a connected production workflow. §5 ↗

From the paper.

A Review of Engineering Applications in Additive Manufacturing Enhanced by Artificial Intelligence

Alireza Yarmohammad Tooski, Ehsan Kargar, Mehrnegar Foratinejad, Mohammad Sadegh Javadi, Amin Mirgheisari, Mohammad Hossein Alizadeh Roknabadi, Alireza Soleimani, Anna Pinnarelli, and Goran Strbac.

AI · 2026 · 7(9), 372

Figures 9–10 reproduced without alteration from Yarmohammad Tooski et al. (2026), doi:10.3390/ai7090372, CC BY 4.0.

The lab and planning sandbox are conceptual teaching models. Reported values appear only in the research examples above. The article provides no separate supplementary media or public dataset.