Researchers at North China Electric Power University have engineered a reusable magnetic sensor that combines covalent organic polymers, gold nanoparticles, and artificial intelligence to detect trace uranium contamination in water down to 1 × 10-7 moles per liter.
Detecting uranium in aquatic environments has long required costly laboratory instruments, extensive sample preparation, and skilled operators. To address those barriers, investigators at North China Electric Power University developed a hybrid sensing platform known as FA@tPF, which merges selective chemical capture, magnetic concentration, surface-enhanced Raman scattering, and automated machine-learning classification into a single reusable cycle, according to research published in Sustainable Carbon Materials on 27 July 2026.
How the FA@tPF Sensor Captures and Concentrates Uranyl Ions
The material is built upon a structured microsphere core that gives each component a distinct physical task. Iron oxide forms the magnetic center, letting particles be rapidly gathered from a sample using an external magnet. A silica layer stabilizes the underlying architecture, while exterior gold nanoparticles generate the intense electromagnetic hot spots needed for surface-enhanced Raman scattering, or SERS. Wrapped around this metallic foundation is a covalent organic polymer called tPF.
This polymer layer selectively binds uranyl ions—the most common soluble form of uranium found in environmental waters. When tested with standard uranyl solutions, the team mixed the sensor with water, extracted the particles magnetically, and read their spectral signatures using a portable instrument. The sensor registered a characteristic Raman signal near 850 inverse centimeters that remained detectable down to 1 × 10-7 moles per liter after a 20-minute exposure.
Testing Performance Against Competing Contaminants and Flow Conditions
Natural waters and industrial effluents rarely contain a single target compound. They typically carry high concentrations of sodium, calcium, magnesium, potassium, zinc, manganese, nitrate, and sulfate. These background ions can interfere with chemical binding or distort spectral readings. During interference evaluations, however, the uranyl-associated Raman signal remained stable, proving that the polymer-functionalized surface successfully isolates target ions amidst crowded chemical environments.
The team also evaluated the platform under dynamic conditions. In flow-through experiments designed to mimic moving water rather than a static container, the system achieved the same detection limit after a 20-minute enrichment period. By contrast, control particles built without the polymer layer failed to produce an identifiable uranyl peak even at higher concentrations.
Machine Learning Automates Interpretation of Reusable Sensor Signals
Reusability remains a major hurdle for environmental sensors that otherwise degrade after a single deployment. After capturing uranyl ions, the sensor material is exposed to a sodium carbonate solution that strips away the adsorbed contaminants and regenerates the active surface. The platform successfully maintained clear SERS signals across six adsorption and desorption cycles while preserving its fundamental structure.
To automate interpretation, the researchers fed the gathered spectra into principal component analysis and a convolutional neural network. The trained model achieved complete classification accuracy on the reported dataset, with interpretive methods confirming that the algorithm relied primarily on the uranyl-associated peak near 850 inverse centimeters.
“Reusable magnetic SERS platform functionalized with covalent organic polymers for trace-level and machine-learning-assisted uranyl detection.”
Sustainable Carbon Materials
While the system offers a sustainable path toward distributed monitoring, investigators note that performance across a wider array of real-world environmental samples remains to be established before it can replace standard regulatory laboratory methods.