Pusat Sains Angkasa, Institut Perubahan Iklim, Universiti Kebangsaan Malaysia (UKM), Bangi, Malaysia
Abstract: This work presents a high-performance metamaterial (MTM)-based sensor system optimized through machine learning for real-time detection and localization of insoluble particles in multiphase oil flow. The sensor integrates four split-square resonators (SSRs) of varying dimensions into a compact microstrip design, each operating at distinct resonant frequencies (3.36, 3.66, 4.26, and 4.69 GHz). Fabricated on a Rogers RT5880 substrate (εr = 2.2, loss tangent = 0.0009, thickness = 1.6 mm), the SSRs exhibit high quality factors and strong field confinement in their capacitive regions, enabling sensitive dielectric contrast detection. When exposed to a flowing medium—purified palm oil (ε = 2.8) mixed with insoluble particles such as air (ε = 1) and distilled water (ε = 78)—each SSR demonstrates distinct frequency shifts corresponding to local dielectric variations. The sensor achieves average sensitivities of 0.75 GHz/ε for air and 0.025 GHz/ε for water. To enhance reliability and account for environmental and material variations, five machine learning classification models were evaluated. Among them, the decision tree classifier achieved the highest performance, delivering an average classification accuracy of 99.25%. The proposed system demonstrates strong potential for integration into smart industrial monitoring platforms for multiphase flow diagnostics.