Infrared spectroscopy with machine learning detects early wood coating degradation

From Japanese cypress to ponderosa pine, wood has been used in construction for thousands of years. Materials such as steel and concrete have taken over the construction of large-scale buildings, but timber has been revived and is increasingly being used in public and multiple buildings due to environmental benefits.

Of course, wood is easily damaged by sunlight and moisture when used outdoors, so it is often passed in favor of other materials. Wood coatings are designed to protect the surface of the wood for this reason, but damage to the coating often begins before it is visible. If the deterioration is visible to the naked eye, it is already too late.

To solve this problem, a team of researchers at Kyoto University are working to create a simple and effective method of diagnosing this almost invisible deterioration before the damage is irreparable.

“Being able to ‘see’ what’s not visible can help extend the lifespan of wooden structures and improve the sustainability of the construction industry,” says corresponding author Yoshikuni Teramoto.

The team strives to bring data-driven tools to traditional wood maintenance by combining mid-infrared spectroscopy and machine learning. They began by testing artificially weathered wood coatings, along with coatings containing cellulose nanofiber, a plant-derived additive that can improve the durability of these coatings.

Their machine learning components use a technique called partial minimal squares that they used to construct the model to predict the extent of degradation. Genetic algorithms were also used to identify the most beneficial infrared signals, improving both accuracy and interpretability.

“We were surprised that infrared spectroscopy captured and predicted by the model,” Tiramoto said.

This approach allows researchers to detect subtle chemical changes and estimate the level of degradation with high accuracy. By quickly diagnosing early coating degradation and allowing wood to be diagnosed without damaging, the method can also reduce the need for costly visual inspections by detecting early warning signs of degradation and preventing further decay.

Through the research, researchers demonstrated how chemistry and data-driven modeling techniques can work together to support smarter maintenance of sustainable buildings. “We hope this technology will help bridge the gap between traditional craftsmanship and modern data science,” continues Terramoto.

The research team is currently testing on real wooden buildings and plans to improve the model of new paint and coating product development applications.

Beyond wood, team methods can also be applied to materials such as concrete and metals to diagnose other types of early material failures and unlock new possibilities for improving the sustainability of other applications and industries in the process.