跳到主要內容 :::

年度	113	專案性質	實驗性質
專案類別	研究專案	研究主題	調查
申請機構	明志科技大學	申請系所	材料工程系
專案主持人	黃宗鈺	職等/職稱	副教授
專案中文名稱	結合斜向線超材料完美吸收體與分子官能基吸收達到土壤或水源汙染物檢測並利用機器學習最佳化其感測靈敏度
中文關鍵字	超材料完美吸收體, 分子功能基吸收, 汙染物檢測, 機器學習
專案英文名稱	Combining oblique wire bundle metamaterial absorber and molecular functional groups' absorption for contaminant detection of soil or water and optimization by machine learning
英文關鍵字	metamaterial perfect absorber, molecular functional groups' absorption, contaminant detection and machine learning
執行金額
執行期間	2024/12/1 至 2025/11/30
計畫中文摘要	在此次的期末報告中，我們針對銀斜向結構數個可能造成低吸收表現的因素去進行製程改動並重新製備試片進行量測。結果顯示吸收表現上雖有提升，但仍然不佳，最終決定改以鋁金屬取代銀金屬去進行試片的製備，並在後續量測中確認了吸收強度的顯著增加。接著我們將此鋁斜向試片分別滴入甲基環己烷和異戊二烯兩種待測物，從而探討我們試片對有機汙染物的檢測能力。結果顯示法諾共振的成功激發能讓我們有效對有機汙染物進行檢測，可惜有機汙染物的高揮發性使我們不能很好的進行有效的定量量測。接著我們改以滴入硫酸銅與4-MBA混合物，來確認對重金屬銅的檢測能力。結果顯示我們可以順利地對銅含量進行檢測，但待測物蒸發後溶質分佈不均以及特徵峰位置位於相對低吸收區段仍對我們在感測表現上造成了不良影響。在後續的機械學習中，我們以x極化吸收曲線積分為最佳化目標，結合粒子群最佳化演算法(particle swarm optimization, PSO)的全域探索與貝葉斯優化(Bayesian optimization, BO)的局部精修以尋找尺寸最優解。驗證證實最優解在可行域內、鄰域響應平滑且具重現性，並與資料分佈與可解釋性分析一致。此流程可有效縮短設計驗證迭代，並作為工程實務中的通用設計工具。
計畫英文摘要	In the final report of this project, we modified the fabrication process of our obliquely deposited silver structures to address several factors suspected of causing poor absorption, and we re-fabricated samples for measurement. Although absorption improved, the performance remained suboptimal. We therefore replaced silver with aluminum for sample fabrication and subsequently confirmed a marked increase in absorption strength. We then dispensed methylcyclohexane and isoprene onto the Al oblique samples to evaluate their capability for detecting organic pollutants. The results show that successful excitation of Fano resonances enables effective detection of organics; however, the high volatility of these analytes hindered reliable quantitative measurements. Next, we introduced a copper sulfate + 4-MBA (4-mercaptobenzoic acid) mixture to assess heavy-metal (Cu) sensing. While Cu content could be detected, post-evaporation non-uniform solute distribution and the characteristic peak falling in a relatively low-absorption region degraded sensing performance. In the subsequent machine-learning stage, we used the X-polarized absorption-curve integral as the optimization objective and combined particle swarm optimization (PSO) for global exploration with Bayesian optimization (BO) for local refinement to find the optimal geometric dimensions. Verification confirmed that the optimum lies within the feasible region, exhibits smooth neighborhood response, and is reproducible, consistent with the data distribution and interpretability analysis. Overall, the workflow effectively shortens the design-to-validation cycle and serves as a generalizable engineering tool for materials parameter design.