14. Oct 2026

AI-guided workflow develops PU adhesive for untreated titanium

AI-guided workflow develops PU adhesive for untreated titanium

An NCMS-backed project generated more than 100,000 viable adhesive formulations and identified polyurethane systems capable of bonding difficult substrates during a five-week development program.

The National Center for Manufacturing Sciences (NCMS) has reported that a combined human and artificial-intelligence workflow developed several high-performance polyurethane adhesives in five weeks, including a formulation capable of bonding directly to grade 5 titanium without surface preparation.

The project investigated two-component PU adhesives intended for plastic, steel, aluminum and titanium. Titanium normally requires treatments such as chemical etching, anodizing or grit blasting before adhesive bonding.

Researchers began with conventional resins and varied parameters including raw-material molecular weights, curing agents, resin-to-curing-agent ratios and curing temperatures. They then trained a Bayesian machine-learning model using experimental results and existing chemical knowledge.

The resulting formulation generator produced more than 100,000 unique combinations within boundaries intended to ensure that the proposed materials were chemically plausible and could be manufactured.

Under the project’s human-in-the-loop approach, the model proposed 30 formulations each week. Researchers selected the 15 candidates expected to provide the most useful experimental data, prepared and tested them, and returned the results to the model.

Testing covered MEK-cleaned 6061 aluminum, MEK-cleaned hot-dip galvanized steel, IPA-cleaned PC-ABS and grade 5 titanium. The team evaluated lap-shear strength and cohesive failure, which measures whether a joint fails within the adhesive rather than at the bond with the substrate.

According to NCMS, the proportion of formulations classified as performing well increased substantially within three weeks. By the end of the five-week program, the workflow had produced multiple adhesives with acceptable direct-to-metal adhesion and lap-shear strengths above 5 MPa, equivalent to approximately 725 psi.

At least one model-generated formulation demonstrated strong adhesion to grade 5 titanium without prior surface treatment. The researchers also found that formulations containing aminic reactive resins consistently outperformed those without them.

NCMS said the work could shorten adhesive-development cycles by reducing reliance on extended rounds of mechanical testing. Potential applications include automotive and aerospace structures, medical and dental devices, electronics, construction and packaging.

The results represent an experimental demonstration rather than the launch of a commercial adhesive. NCMS has not disclosed the formulations or announced plans for their commercialization.

Picture caption: NCMS has published a technology brief describing a five-week human-AI program that developed PU adhesives for substrates including aluminum, steel, PC-ABS and grade 5 titanium. Graphic: NCMS

NCMS

Privacy settings

We use cookies on this website that are necessary for the operation of the website and therefore cannot be deselected. If you would like to know which cookies these are, you will find them listed individually in the privacy policy. Our website also uses external components that may set cookies. By loading external components, data about your behaviour can be collected by third parties, which is why we need your consent. Without your permission, there may be restrictions on content and operation. Detailed information can be found in our privacy policy.

Necessary cookies are always loaded