應用水溶液電漿光譜技術於重金屬檢測之參數研究與優化演算法

Abstract

工業濕製程與廢水排放常含有多種重金屬，如銅 (Cu)、鎳 (Ni)、鋅 (Zn)、鉛 (Pb)。若濃度控制不當，不僅影響製程品質與反應速率，亦可能對環境造成長期污染。而傳統電感耦合式光譜法 (ICP-OES) 與 原子吸收光譜法(AAS) 等離線分析受限於實驗室檢測與回報延遲，難以因應取樣頻率高之製程控制與排放警戒。本研究整合脈衝式水溶液電漿光譜 (solution plasma–OES) 與嵌入式微電腦 (Raspberry Pi) 控制單元，建立一套連續、快速之線上多金屬自動監測平台，並透過系統化參數研究與梯度導向參數優化，提升檢測靈敏度與長期穩定度。 第一部分為討論脈衝式水溶液電漿光譜系統不同參數（脈衝開啟時間、脈衝關閉時間、電壓值）對於不同金屬之光譜強度響應，於1 wt % 硝酸溶液中進行系統性參數研究（Parametric Study）的定量分析。並針對四種金屬的最佳參數條件個別建立檢量線，得到低於 ppm 等級的檢測能力。隨後進行線上長期測試，於過程中加藥並觀察光譜強度之變化，並評估其長期穩定度，驗證其相對標準差低於5 %，濃度漂移控制在 ±3 % 以內。 第二部分為開發結合梯度資訊之參數優化演算法，並針對其超參數（步伐調整方式、各金屬之迭代次數）進行微調與討論，並針對使用者需求提出三種搜尋策略，其中以全局梯度搜尋 (GSS) 表現最佳。透過整合式的多金屬優化，於270個參數的多維空間中，其實驗操作步數 < 80步即可得到最佳的參數條件；在此條件下 Cu、Ni、Pb、Zn 之 偵測極限 (LOD) 分別達 0.116、0.234、0.029、0.287 ppm。最後針對未知溶液（pH = 2 硝酸溶液）進行演算法的可行性驗證。

Industrial wet-chemical processes and wastewater effluents often contain multiple heavy-metal ions such as Cu, Ni, Zn, and Pb. It’s obvious that inadequate concentration control not only degrades product quality and reaction rates but also poses long-term environmental risks. However, Conventional off-line techniques such as ICP-OES and AAS are limited by laboratory analyses and delayed reports, making them unsuitable for high-frequency process control and real-time discharge alarms. This study integrates pulsed solution-plasma optical emission spectroscopy (solution plasma–OES) with an embedded micro-controller unit (Raspberry Pi + Arduino) to construct a continuous, high-speed, on-line platform for multi-metal monitoring. A combined approach of systematic parametric study and gradient-guided parameter optimization algorithm is proposed to enhance detection sensitivity and long-term stability. The first part includes parametric study and Longterm qualification test. The influences of pulse on-time, off-time, and voltage on the emission intensities of Cu, Ni, Zn, and Pb were quantified in 1 wt % nitric acid solution (HNO3). Calibration curves are established under the element optimum conditions and yield sub-ppm limits of detection (LODs). Furthermore, A continuous longterm test is evaluated with dosing qualified metal ion in the process, the result shows high stability and accurcy with relative standard deviations below 5 % and concentration drift within ± 3 %. In second part, a gradient-based parameter-search algorithm was developed. Hyperparameters (step-adjusted method, number of iteration per element) algorithm is tuned and discussed, and three search strategies were introduced based on user’s purpose. The result shows that GSS out-performed the other stragedy in fewer than 80 experiments within a 270-point parameter space. Under this optimized condition for each metal, LODs of 0.116 ppm (Cu), 0.234 ppm (Ni), 0.029 ppm (Pb), and 0.287 ppm (Zn) were achieved. The algorithm was further validated on an unknown sample (pH = 2 HNO₃), demonstrating capability and feasibility.