跳到主要內容 :::

年度	111	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	東海大學	申請系所	環境科學與工程學系
專案主持人	陳鶴文	職等/職稱	特聘教授
專案中文名稱	電子鼻與人工智慧整合分析技術在油品污染偵測上的應用
中文關鍵字	電子鼻,氣味成分分析,油污偵測,人工智慧
專案英文名稱	未填寫
英文關鍵字	未填寫
執行金額	770,000元
執行期間	2022/9/1 至 2023/8/31
計畫中文摘要	目前台灣有許多運作超過十年的老舊加油站，油品洩漏問題嚴重，易危害健康與環境，且眾多的老舊加油站導致環保署監測工作效率低落，而傳統的光譜與基於氣相色譜的分析方法無法有效且即時分析出油品污染之土壤的組成成分與污染程度。本研究利用電子鼻（e-nose）來檢測土壤污染的程度，並利用LSTM預測發生污染事件後所經過的時間的影響。 本計畫預計結合電子鼻與人工智慧技術，建立一個低成本、高準確性、快速的檢測系統，協助判定受汙染土壤中揮發性有機物的組成。 為了測試土壤內之VOC濃度，設計一組VOC微型感測器測試腔體，並使用手持式PID進行校正，在進行實場VOC感測器偵測受VOC污染之土壤時，可有一台可攜式比對校正儀器，確保VOC感測器偵測數據的準確性及穩定性。 本計畫氣體感測器會對多種氣體有反應，後續採用三組氣體感測器進行測試，針對乙醇氣體進行實驗數據之收集與分析，接著使用丙酮氣體進行實驗數據之收集與分析，但因氣體感測氣無分離氣體之功能，在測試乙醇和丙酮的實驗過程中，無法辨別氣體的種類和濃度，需要再增加數種氣體感測器的種類，進行交叉比對後，再進行大數據分析。 未來希望將測試腔體的容積縮小，以加快氣體在腔體內平均擴散的速度，並提供攜帶的功能，在氣體感測器的部分，增加不同種類感測器之數量，透過氣體感測器之不同性質，收集氣體數據，再經由大數據分析和神經網路訓練，實踐分辨氣體之功能。
計畫英文摘要	Currently, there are many old gas stations in Taiwan that have been operating for over a decade. The issue of fuel leakage is severe, posing risks to health and the environment. The large number of these old gas stations has led to low efficiency in monitoring by the Environmental Protection Administration. Traditional spectroscopy and gas chromatography-based analysis methods are ineffective and not real-time in analyzing the composition and degree of soil contamination caused by oil products. This study utilizes an electronic nose to assess the level of soil pollution and employs LSTM (Long Short-Term Memory) to predict the temporal impact following a pollution event. The project aims to integrate electronic nose technology with artificial intelligence to establish a low-cost, high-accuracy, and rapid detection system. This system will aid in determining the composition of volatile organic compounds (VOCs) in contaminated soil. To measure VOC concentrations in the soil, a set of micro-sensors for VOCs is designed for testing chambers. Handheld PID (Photoionization Detector) devices are used for calibration. During the on-site detection of soil contaminated with VOCs using the VOC micro-sensors, a portable reference calibration instrument is employed to ensure the accuracy and stability of the VOC sensor data. The gas sensors in this project respond to multiple gases. Subsequently, three sets of gas sensors are tested. Experiments are conducted for ethanol gas data collection and analysis, followed by experiments with acetone gas data. However, due to the lack of gas separation functionality in the sensors, the type and concentration of gases cannot be distinguished during the testing process. It is necessary to incorporate various types of gas sensors and conduct cross-comparisons before performing extensive data analysis. In the future, the aim is to reduce the volume of the testing chamber to accelerate the average diffusion rate of gases within it. Portability will be enhanced. In terms of gas sensors, the variety of sensors will be increased to collect gas data based on the different characteristics of the sensors. Through extensive data analysis and neural network training, the capability to differentiate between gases will be achieved.