Every engine, data center, and power plant on Earth is haunted by the same invisible enemy: waste heat. The fluids we rely on to carry that heat away—water, ethylene glycol, and their mixtures—have barely changed in decades, and their modest thermal conductivity places a hard ceiling on how efficiently our machines can run. Now, a team of researchers at Universiti Malaysia Pahang Al-Sultan Abdullah has reported a strikingly modern way forward, combining exotic carbon-and-cellulose nanoparticles with one of the most powerful pairings in machine learning: an artificial neural network tuned by a genetic algorithm. Their study, published in Neural Computing and Applications, demonstrates that a carefully engineered hybrid nanofluid can boost thermal conductivity by more than half over a conventional coolant, and that an evolutionarily optimized AI model can predict that performance with near-perfect fidelity.
The fluids at the heart of the study belong to a class of materials called nanofluids—base liquids seeded with nanoparticles so small that thousands would fit across the width of a human hair. The idea traces back to a landmark 1995 proposal by Stephen Choi, who argued that suspending metallic or carbon-based particles in conventional fluids could dramatically raise their ability to conduct heat. In the decades since, engineers have tested a zoo of particle combinations, but a persistent challenge remains: single-particle nanofluids often deliver disappointing gains, suffer from instability, or thicken so much that pumping them costs more energy than the improved heat transfer saves. Hybrid nanofluids, which blend two or more nanoparticle species, attempt to capture the best properties of each ingredient while compensating for their individual weaknesses.
The Malaysian team chose an unusual and intriguing pairing: graphene nanoplatelets, ultrathin sheets of carbon with exceptional intrinsic thermal conductivity, and cellulose nanocrystals, rod-like particles derived from plant matter. The logic is elegant. Graphene brings extraordinary heat-carrying capacity, but flat platelets tend to clump together and settle out of suspension. Cellulose nanocrystals, being hydrophilic and mechanically robust, can act as dispersing agents and structural spacers, helping to keep the graphene sheets separated and stabilized within the liquid. Before any fluid was made, the researchers characterized their raw nanoparticles using field-emission scanning electron microscopy and transmission electron microscopy, confirming the morphology of both particle types and ensuring that the building blocks of their coolant were exactly what they intended.
Preparing the fluids themselves followed a well-established two-step method, in which nanoparticles are first procured or synthesized as dry powders and then dispersed into the base liquid—in this case, a mixture of ethylene glycol and water, the workhorse formulation of automotive cooling systems. The team prepared both single-particle GNP nanofluids and hybrid GNP/CNC formulations across a range of volume concentrations from 0.02 to 0.2 percent, then measured thermal conductivity and viscosity across temperatures spanning 30 to 80 degrees Celsius. That experimental matrix matters, because the two properties pull in opposite directions in any engineering trade-off: higher particle loading improves heat conduction but raises viscosity, which increases the pumping power needed to circulate the fluid and can promote unwanted deposits and erosion in cooling channels.
The headline experimental result is remarkable. At the highest tested concentration of 0.2 volume percent and the highest temperature of 80 degrees Celsius, the hybrid GNP/CNC nanofluid exhibited a thermal conductivity 52.77 percent greater than the plain base fluid. That is not an incremental improvement; it is the kind of jump that, if translated into a real cooling loop, could allow smaller radiators, lower fan power, and tighter thermal margins in everything from car engines to concentrated solar collectors. Equally important is the temperature trend: the enhancement grew with heating, meaning the fluid performs best precisely in the hot conditions where cooling demand is most severe. The researchers attribute the synergy to the complementary geometry of the two particle types, with the platelets and rods forming networks that create more efficient pathways for phonon transport through the liquid.
Viscosity told a more complicated story, and the team mapped it carefully. As expected, viscosity increased with particle concentration and decreased as temperature rose—the standard rheological behavior of most suspensions. Even at the relatively low concentration of 0.2 volume percent, the hybrid nanofluid’s viscosity at 30 degrees Celsius was approximately 27.5 percent higher than that of the base fluid. That penalty is real but modest, and because viscosity falls steeply with temperature, the fluid becomes easier to pump exactly when an engine or thermal system is running hot. Analyzing this rheological behavior is essential for any practical deployment, since a nanofluid that conducts heat brilliantly but flows like honey would be a laboratory curiosity rather than a commercial coolant.
What elevates the study beyond careful measurement is its computational core. Artificial neural networks have become the tool of choice for predicting nanofluid properties, because the relationships between particle loading, temperature, and thermophysical behavior are nonlinear and resistant to simple equations. But a neural network’s accuracy depends heavily on its internal settings—the number of neurons in its hidden layers, the choice of activation functions, the learning parameters that govern how it fits the training data. Traditionally, engineers tune these hyperparameters by trial and error or grid search, a slow process that often leaves significant accuracy on the table. The team instead deployed a genetic algorithm, an optimization technique inspired by biological evolution, in which candidate solutions compete, mutate, and recombine across generations until the fittest configuration survives.
The marriage of the two algorithms paid off handsomely. The resulting GA-ANN model achieved a coefficient of determination of 0.998, meaning it explained 99.8 percent of the variance in the experimental data, alongside a root mean square error of 0.3231 and a mean absolute error of 0.683. Statistical testing confirmed that the optimized model was significantly more accurate than conventional neural network approaches, with the improvement reaching the stringent threshold of P less than 0.01. In practical terms, this means an engineer could input a desired temperature and particle concentration and receive a trustworthy prediction of the fluid’s thermal conductivity and viscosity without running a single new experiment. Such a predictive framework compresses months of laboratory work into seconds of computation, and it provides a consistent, reusable tool for designing coolants tailored to specific applications.
The implications ripple outward across the energy landscape. Better coolants translate directly into lower fuel consumption in vehicles, denser and more efficient battery packs in electric vehicles, improved performance in electronics cooling, and higher output from solar thermal plants. The choice of cellulose nanocrystals also carries a sustainability signal: unlike many synthetic nanoparticles, cellulose is abundant, renewable, and biodegradable, which softens the environmental footprint of nanofluid production and disposal. And because the AI model was trained on a modest, well-characterized experimental dataset, the approach is accessible to research groups without access to massive computing infrastructure, lowering the barrier to entry for nanofluid innovation worldwide.
Challenges remain before GNP/CNC coolants flow through production engines. Long-term colloidal stability, cost at industrial scale, compatibility with seals and metals in real cooling loops, and behavior under freeze-thaw cycling all demand further study, and the authors note that their data are available upon request for groups wishing to extend the work. Yet the study marks a genuine convergence of two technological currents: the materials science of engineered nanoparticles and the computational power of evolutionary machine learning. As data-driven design tools grow more sophisticated, the slow craft of mixing and measuring fluids one batch at a time is giving way to a faster paradigm in which algorithms propose, experiments verify, and each result feeds back into smarter models. For a field as old as heat transfer, the future suddenly looks very young indeed.
Subject of Research: Machine learning prediction of thermal conductivity and viscosity of graphene nanoplatelet–cellulose nanocrystal hybrid nanofluids
Article Title: Engineering the future of heat transfer: artificial neural network and genetic algorithm synergy to model next-generation hybrid nanofluid performance
Article References: Hasan, M. M., Rahman, M. M., Bakar, S. A., & Ramasany, D. (2026). Engineering the future of heat transfer: artificial neural network and genetic algorithm synergy to model next-generation hybrid nanofluid performance. Neural Computing and Applications, 38(17), Article 716. https://doi.org/10.1007/s00521-026-12438-9
Image Credits: AI Generated
DOI: 10.1007/s00521-026-12438-9
Keywords: hybrid nanofluid, thermal conductivity, artificial neural network, genetic algorithm, graphene nanoplatelets, cellulose nanocrystals, viscosity, heat transfer, machine learning, ethylene glycol, nanotechnology, thermal management
News Source: Cassandra Pierce. (October 6, 2026). AI Meets Nanotech: Neural Networks Crack the Code of Next-Generation Engine Coolants. Scienmag.
