Uranium contamination in water may soon become easier to detect, thanks to a reusable magnetic sensor that combines gold-enhanced Raman spectroscopy, selective chemical capture and machine learning. Developed by researchers at North China Electric Power University, the platform can identify trace amounts of uranyl ions—the most common soluble form of uranium found in environmental waters—while using a magnetic field to collect, concentrate and recover the sensing material. The approach is designed to address a persistent challenge in environmental monitoring: finding extremely low concentrations of uranium in chemically complex water without relying entirely on expensive laboratory instruments or lengthy sample preparation.

The new material, named FA@tPF, is a hybrid sensing platform built around Fe₃O₄@SiO₂@Au microspheres and a polyarylene ether-based covalent organic polymer called tPF. Each component contributes a specific function. The iron oxide core gives the particles magnetic properties, allowing them to be rapidly separated from water with an external magnet. A silica layer helps stabilize the structure, while the outer gold component creates the intense electromagnetic “hot spots” required for surface-enhanced Raman scattering, or SERS. The tPF polymer forms the chemically active interface, helping capture uranyl ions and position them close to the gold surface, where their molecular vibrations can be amplified and measured.

SERS works by illuminating molecules with a laser and recording the light they scatter. Under ordinary conditions, Raman scattering is weak, but metallic nanostructures such as gold nanoparticles can dramatically amplify the signal through localized electromagnetic fields generated by surface plasmons. This makes it possible to recognize chemical fingerprints at very low concentrations. In the FA@tPF system, the polymer does more than provide a coating: it enriches uranyl ions near the gold nanoparticles, increasing the probability that their characteristic vibrational signature will be detected. The result is a sensing material that combines selective adsorption with optical signal enhancement rather than treating those functions separately.

Under a 20-minute enrichment condition, the platform detected uranyl ions at concentrations as low as 1 × 10⁻⁷ mol·L⁻¹. The researchers tracked a characteristic Raman feature near 850 cm⁻¹, a spectral region strongly associated with the vibrations of the uranyl group. As the concentration of uranyl increased, the intensity of this band showed a clear relationship with the amount of analyte present. This type of correlation is essential for converting a Raman spectrum into a quantitative measurement. In additional flow-through experiments, designed to mimic moving water rather than a stationary laboratory sample, FA@tPF maintained the same detection limit after 20 minutes of enrichment, suggesting that the design could eventually be adapted for more dynamic monitoring conditions.

Environmental water rarely contains only the substance a sensor is intended to measure. Natural waters and industrial effluents may contain large amounts of sodium, calcium, magnesium, potassium, zinc, manganese, nitrate and sulfate, among other dissolved species. These ions can compete for binding sites, alter the chemical environment or generate background signals that make target compounds more difficult to recognize. The researchers therefore tested FA@tPF in the presence of several common interferents. The uranyl-associated Raman signal remained stable, indicating that the polymer-functionalized surface could selectively enrich uranyl ions even when other ions were present. This selectivity is particularly important for uranium surveillance, where a sensor must distinguish a weak target signal from a crowded chemical background.

The material’s magnetic behavior also gives the platform a practical advantage. After the particles capture uranyl ions, a magnet can pull them out of the sample and concentrate them for Raman measurement. Once analysis is complete, the researchers used a sodium carbonate solution to remove the adsorbed uranyl and regenerate the sensing surface. The platform continued to produce clear SERS signals after six adsorption and desorption cycles, while largely retaining its original structure and functional groups. Reusability could reduce material consumption and operating costs compared with single-use sensors, although longer-term testing will be needed to determine how the platform performs after many more regeneration cycles and in real environmental samples.

The team added another layer of technology by using machine learning to interpret the Raman spectra automatically. Principal component analysis, a statistical method that reduces complex datasets into their most informative patterns, produced consistent clustering for the FA@tPF measurements. The researchers then trained a convolutional neural network to classify spectra recorded before and after uranyl adsorption. In the study dataset, the model achieved 100% classification accuracy. While such a result does not by itself guarantee identical performance outside the laboratory, it demonstrates how artificial intelligence can help distinguish subtle spectral changes that may be difficult to evaluate manually, especially when large numbers of samples must be screened rapidly.

Crucially, the researchers did not treat the machine-learning model as an opaque black box. They used gradient-weighted class activation mapping, or Grad-CAM, to visualize which portions of the Raman spectra influenced the network’s decisions. The analysis indicated that the model relied heavily on the uranyl-related feature near 850 cm⁻¹ rather than on unrelated fluctuations or background noise. This interpretability is more than a technical detail: it links the computational prediction to a chemically meaningful signal. If a model identifies a sample as positive because of the expected uranyl band, scientists can have greater confidence that the classification reflects the target compound rather than an accidental pattern in the training data.

The researchers say the platform could form the basis of faster and more sustainable uranium monitoring in aquatic environments. Its combination of magnetic enrichment, selective polymer chemistry, SERS detection and interpretable artificial intelligence could support portable testing or distributed monitoring networks near nuclear facilities, mining areas and sites affected by industrial contamination. The system is not yet a replacement for established laboratory methods, which remain essential for regulatory confirmation and comprehensive chemical analysis. Nevertheless, FA@tPF illustrates how multifunctional materials can compress several stages of environmental testing—capture, concentration, detection, classification and regeneration—into one reusable system. As Raman instruments become smaller and machine-learning tools become more accessible, platforms of this kind could make trace-level chemical surveillance faster, more automated and easier to deploy beyond conventional laboratories.

Subject of Research: Reusable magnetic SERS sensing platform for trace-level uranyl-ion detection using covalent organic polymers and machine learning.

Article Title: Reusable magnetic SERS platform functionalized with covalent organic polymers for trace-level and machine-learning-assisted uranyl detection

News Publication Date: 27-Jul-2026

Web References: https://doi.org/10.48130/scm-0026-0023

References: Zhang W, Ma J, Zhang Y, Wang S, Wakeel M, et al. 2026. “Reusable magnetic SERS platform functionalized with covalent organic polymers for trace-level and machine-learning-assisted uranyl detection.” Sustainable Carbon Materials 2: e027. DOI: 10.48130/scm-0026-0023.

Image Credits: Wentao Zhang, Jing Ma, Yiyan Zhang, Suhua Wang, Muhammad Wakeel and Zhenli Sun

Keywords

Uranyl detection, uranium contamination, surface-enhanced Raman scattering, SERS, magnetic sensor, covalent organic polymers, Fe₃O₄@SiO₂@Au microspheres, machine learning, convolutional neural networks, Raman spectroscopy, environmental monitoring, nuclear safety, reusable materials

Tags: covalent organic polymers for heavy metal detectiongold-enhanced SERS sensors for water analysishybrid Fe₃O₄@SiO₂@Au sensorslow-cost uranium contamination testing methodsmachine learning in environmental sensingmagnetic microspheres for contaminant recoverynanomaterial-based uranium detectionrapid water testing for uranium pollutantsreusable magnetic sensing platformselective chemical capture of uranyl ionssurface-enhanced Raman spectroscopy for environmental monitoringtrace uranium detection