應用超聲波於監控與診斷膜分離程序

在薄膜分離精製植物原油系統中，隨著過濾時間增加，薄膜上會產生結垢，導致滲透端流量的降低，過去有超音波射頻訊號監控結垢程度，但需要相當程度專業知識，且不易得知結垢的原因。此外，過去僅單點量測薄膜的觀測數據亦無法代表整張薄膜的結垢行為。在本研究中，首先使用區塊式掃描的方式觀測薄膜結垢，以改善以往單點觀測數據代表性不足的缺陷。為易於直觀過濾系統中結垢的情形，我們使用希爾伯特轉換來取得射頻訊號能量變化的趨勢，後進行成像，讓使用者可以清楚的觀察到薄膜結垢型態隨著過濾時間的變化。其次，由於薄膜結垢產生的原因相當複雜，操作條件或進料濃度的改變皆會影響薄膜結垢的生成。因此，開發過濾診斷系統有助於過濾效能的維持。然而，在以往超音波監控薄膜結垢的研究中，皆未能以有效的量化方式對過濾系統進行診斷。由於超音波訊號具有不同的頻率變化，本研究以小波包轉換以獲得超音波訊號在各頻帶的資訊，而後我們將以頻帶能量作為樣品的特徵。由於超音波訊號為間接式觀測數據，並非系統操作狀態之觀測值，雖可萃取出特徵，仍不易判讀，且為了降低診斷模式訓練上的負擔，本研究對各頻帶下的觀測數據以Gaussian smoothing的方式進行定性的分組，為了建立系統診斷，採用決策樹C4.5演算法產生診斷規則。我們對2個典型的過濾系統，(1)死端過濾，及(2)掃流過濾，利用所提的方法分別測試。比較過去的診斷方式，本研究所提的數據推論方式確實較佳。

關鍵字

錯誤診斷；監控；超聲波；膜過濾；膜分離

並列摘要

In the membrane separation filtration system of refined vegetable oil, fouling of membrane will occur with the increase of filtration time, which will lead to lower permeate side flow rate. In the past, membrane fouling monitoring using ultrasound not only required professional knowledge to interpret of membrane fouling from the ultrasonic reflection signals (RF), but also the measurements from single point on membrane can not represent the fouling behavior of the whole membrane. In the first part of this research, we used c-mode scanning to observe the membrane fouling on the local membrane, to improve the defects of single-point observed data. In order to understand the fouling behavior of membrane filtration system, we use Hilbert transform on each RF signal, and take images to visualize the fouling change with filtration time. The causes of membrane fouling are complex, which can result from operating conditions or feed concentration changes. Therefore, the development of a diagnostic system of filtration helps to maintain filtration efficiency. However, the research using ultrasonic to monitoring membrane fouling in the past can not effectively quantify filtration system for diagnosis. The ultrasonic signal has different frequency, and the information in each bandwidth by can be obtained using wavelet packet decompose. The calculated energy of each bandwidth is used to characterize the sample. The measurements of ultrasound are not a direct observation of the operational status of system, although the features can be extracted, is still not easy to interpret. In order to reduce the load of training diagnostic model, Gaussian smoothing on each bandwidth is used for qualitative classification, and C4.5 algorithm is used to build diagnosis systems. Verification of the proposed method on two typical filtration systems, dead end filtration and cross flow filtration shows the better performance of the proposed data driven method.

並列關鍵字

fault diagnosis ； membrane filtration ； ultrasonic ； monitoring

參考文獻

[26] 江衍澄, “HSMM為基礎的錯誤診斷於批次操作系統”,台灣, 中原大學化學工程碩士學位論文(2009)

[2] Mulder, “Basic principles of membrane technology”, second edition, 2004

[3] Pagliero et al. “Fouling of polymeric membranes during degumming of crude sunflower and soybean oil”. Journal of Food Engineering, 78, (2007) 194-197

[4] Yao et al. “Non-invasive observation of flow profiles and polarisation layers in hollow fibre membrane filtration modules using NMR micro-imaging”, Journal of Membrane Science, 99, (1995) 207-216

[5] Altmann and S. Ripperger, “Particle deposition and layer formation at the crossflow microfiltration”, Journal of Membrane Science, 124, (1997) 119–128.

國際替代計量

應用超聲波於監控與診斷膜分離程序

全文下載

主題瀏覽