This literature review maps artificial intelligence and machine learning across the composite-material lifecycle. A bibliometric analysis counts 1,474 publications from 2010–2025 in seven journals, complemented by a technical review of over 200 studies in the supplied author-submitted manuscript. Mechanics and failure dominate research volume, and the review connects methods to design, manufacturing, performance, monitoring and sustainability tasks. Its synthesis emphasizes physics integration, uncertainty quantification and sensing to address limited generalization from simulations or small d

An atlas of learning · Composite materials

Strong fibres. A surrounding resin. A complex life. Explore how machine learning is helping researchers connect a composite’s design to its performance.

01Follow the material

One composite. Four different questions.

A composite combines reinforcing fibres with a surrounding material. Its layers, processing history and hidden damage all matter. Choose a task, then look inside.

Mechanics & performance

What happens under load?

Conceptual scene Not a physical simulation

The outside hides the story. Look inside to reveal a conceptual damage plane between the layers.

The learning task

Predict the response.

Tests and simulations reveal how a composite deforms and fails. Learning models can turn these examples into faster estimates of its response.

A surrogate is a fast stand-in for a more expensive calculation. A neural network learns relationships from many connected computational units.

A conceptual view of the questions mapped in this literature review.

02The evidence atlas

Different questions. Different learning methods.

Mechanics & Failure accounts for roughly half of the mapped AI/ML publications (§3, p. 6). Choose a topic to explore the methods used, then tap a bar to learn how it works.

Mechanics & performance

Top listed methods

Counts are publications using a method, not model accuracy. A publication can use more than one method; categories overlap. Bars show the leading listed methods, not every method in the table. All topics share the same scale, making differences in research activity visible.

03The next panel

Would you trust the next panel?

A coupon is a small piece cut from a larger material panel for testing. Coupons from the same panel share a manufacturing history. How should we split examples when testing a learning model?

Familiar panels reach both sets.

A test coupon can share its manufacturing history with the learning examples. The review explains why this can make performance on a new panel harder to judge.

Schematic exploration, not reported experimental data. Based on the paper’s synthesis of panel and batch variation, validation and sparse sensing: §8.1–8.3, pp. 27–29. Reserving a panel tests a new manufacturing history.

Where this could lead

Fast predictions. Connected to real materials.

The review points toward workflows that combine learned relationships, physical knowledge, uncertainty and new measurements.

Explore more designs

Fast surrogates can help screen fibre architectures and layer arrangements before costly detailed analysis.

Read the process

Temperature histories and images can help connect manufacturing settings with curing and defects.

Know when to measure

Estimates of uncertainty can signal where more testing or inspection is needed. Sparse sensors provide a few fresh measurement points.

Understand a second life

Learning methods are being used to characterize recycled composites and screen bio-based formulations.

Artificial Intelligence and Machine Learning in Composite Materials: A Comprehensive Literature Review and Bibliometric Analysis

Huilong Fu, Amirali Eskandariyun, Paulina Portales Picazo, Kendall A. Johnson, Aric Morton, Mathew Wynn and Navid Zobeiry.

This experience was developed from the University of Washington author-submitted manuscript (2026). The journal article appears in Composites Part A: Applied Science and Manufacturing, volume 210, article 110092. Page and section references here refer to the manuscript.

The bibliometric analysis covers 1,474 AI/ML-enabled publications in seven composite materials journals from 2010 to 2025 (§2, pp. 4–5); the technical review discusses more than 200 selected papers. Method counts are recreated from Tables 1–4 (pp. 13, 17, 21 and 26). They are overlapping research categories, not measurements of model performance.

The supplementary material contains four tables: learning-method definitions and composite examples (ST1), foundational papers (ST2), a tool guide (ST3), and additional studies outside the seven core journals (ST4). ST1 informs the plain-language method explanations in the atlas.

The graphical abstract is an AI-generated conceptual illustration informed by the lifecycle taxonomy (Figure 2) and topic–method map (Figure 6). The moving specimen and coupon experiment are original teaching illustrations. Numerical charts are recreated from the cited manuscript tables.

AI-generated conceptual illustration, informed by the paper’s lifecycle taxonomy.