建置適用於程序模擬器之第一原理熱力學性質計算平台

Development of first principle Thermophysical property Estimator for Process Simulators

蔡昌哲(Chang-Che Tsai)

指導教授 ： 林祥泰

國立臺灣大學/工學院/化學工程學研究所/博士(2022年)

https://doi.org/10.6342/NTU202201907

摘要

為了讓使用程序設計軟體Aspen Plus的使用者可以更方便的使用基於量子化學的第一原理熱力學模型，我們開發了一個名叫T.E.A.M. (Thermophysical property Estimation for Aspen Plus Modeling)的網頁版熱力學計算機，任何人都可以透過該網頁提交他們的物質清單，很便利地就得到應該在Aspen Plus裡面輸入的熱力學性質。對於使用者來說，只需要輸入分子的表示式簡化分子線性輸入規範SMILES (Simplified molecular input line entry specification)後，我們的系統會產生一個含有提交物質之相關資訊(CAS RN、化學名稱、別名)的總表，便於不熟悉SMILES的使用者再度確認物質資訊是否正確，而後即可成功提交計算任務。我們的平台會檢查物性是否已存在於Aspen Plus內建資料庫裡，並對有缺漏之參數進行基於量子化學的估計。最終，使用者會收到一個裝有所有在進行進一步程序模擬前應具備之熱力學性質的Aspen Plus bkp檔，並附上兩張總表完整揭露每樣物質之熱力學參數之來源。如此一來，使用者將不會對於使用此bkp檔去進行進一步模擬感到過多的不安。

在T.E.A.M.中包含的所有熱力學性質及其相關的相平衡計算對於程序設計及開發都至關重要。然而，一般來說熱力學模型的參數卻不是經常可得(尤其是對於包含特用化學品的製程設計來說)。在本研究中，我們使用了近期開發的、具有預測性的熱力學模型去提供所有必需的熱力學參數，並評估了該模型的可靠度。更具體地來說，我們使用了G3模型預測理想氣體熱容、生成熱與生成自由能; 利用PR+COSMOSAC 狀態方程式與COSMO-SAC活性係數模型來預測及估計純物質及混合物的熱力學性質與相行為。由於使用這些模型時，並不需要額外提供物性之實驗值，因此理論上可以應用於所有的物質。我們使用了一個具有972樣物質、含有九種純物質參數及二元交互作用參數的資料庫來驗證我們方法的可靠度。我們發現到模型預測具有一定的準確度(誤差小於10%: 臨界溫度[5%]、臨界壓力[10%]、臨界體積[5%]、定壓理想氣體熱容[5%]、氣化熱[10%]; 誤差大於10%: 偏心因子[33%]、蒸氣壓[73%])。此外，結果顯示預測值之整體系統性誤差不大，亦即當實驗值不存在時，我們可利用T.E.A.M.可靠的預測值來做為程序模擬時之熱力學參數。我們的分析也顯示，若想以預測值取代實驗值，必須將生成熱、生成自由能及蒸氣壓預測得更為準確。然而，我們以模型預測物質之正常沸點，並將物質之正常沸點之相對大小倆倆比較後，發現94%的沸點相對大小都與實驗一致，顯示我們目前的模型準確度仍能勝任如溶劑篩選這類的任務。

為了進一步提升蒸氣壓的準確度，我們加入了隨處可得的正常沸點實驗值來改善蒸氣壓預測值，並使用了近乎20000個物質來驗證這樣的改進是否有效。在之前的研究中我們發現到，使用PR+COSMOSAC來預測臨界性質及偏心因子，將其代入彭-羅賓遜(Peng-Robinson，PR)狀態方程式後，蒸氣壓預測結果與PR+COSMOSAC的預測結果類似。因此，我們嘗試使用20000個物質的正常沸點來修正臨界性質及偏心因子，想以此方式間接改善蒸氣壓的預測值。首先，我們發現到，臨界溫度預測值(9%→3%)相對於臨界壓力(13%→9%)及偏心因子(47%→58%)有較大幅度的改善。我們總共比較四個模型的蒸氣壓預測值：原始的PR+COSMOSAC狀態方程式、以PR+COSMOSAC預測的臨界性質及偏心因子為參數的PR狀態方程式、以正常沸點實驗值改善過的PR+COSMOSAC預測的臨界性質及偏心因子為參數的PR狀態方程式，及以實驗的臨界性質及偏心因子為參數的PR狀態方程式。我們在中、高溫下衡量預測值的準確度後，發現加入正常沸點改善的模型在中溫區及高溫區分別有16%及19%的準度。相較之下，原始PR+COSMOSAC在該溫度下的準度分別是114%及75%，而我們的標竿模型(使用臨界性質及偏心因子的實驗值)的預測準度分別是7%與3%。另外，我們雖然也對低溫區的蒸氣壓預測作了完整檢視，然而整體誤差進步並不顯著(PR+COSMOSAC:323%; 使用正常沸點實驗值:237%)。總體來說，在正常沸點實驗值的校正後，蒸氣壓預測值準確度在不同溫度下，或對於不同長度的分子都有所提升。引入正常沸點實驗值來改善PR+COSMOSAC蒸氣壓預測值可謂是一個簡單、不直接修改模型的有效而間接的解決方案。

最後，我們也展示了三個應用T.E.A.M.來估計參數的程序模擬的範例。這三個程序都與CO2減量有關。透過這些應用的範例顯示了T.E.A.M.在內建官能基貢獻法不能派上用場時特別有效。在這個方面來說，對於預測程序中缺失的參數來說，T.E.A.M.的確提供了一個簡單而可靠的解決方案。

關鍵字

PR+COSMOSAC ； COSMO-SAC ； G3 ； 量子化學計算 ； 第一性原理計算 ； 程序模擬 ； 參數預測

並列摘要

To facilitate the use of quantum mechanical (QM) based thermodynamic models to predict the essential thermodynamic properties for process design in Aspen Plus, an easy-to-use website called T.E.A.M. where anyone can readily send a request to our servers to compute the properties is developed. The only requirement for users is the SMILES of the chemicals of interest. Before sending the jobs, a summarized table of all molecular information such as CAS RN and chemical names along with the input SMILES are presented. This table prevents users from sending jobs with wrong compounds due to unfamiliarity of SMILES. After calculation, the computed parameters are fed into the Aspen Plus if the built-in experimental data are not available. Finally, users will receive an email with an Aspen file together with summary figure. Prioritizing the reliable built-in parameters maximally ensures the quality of parameters provided in the Aspen file. Also, a summary figure is attached to clearly show if the parameters are obtained from built-in databanks or predicted by our approaches. Hence, users would not hesitate to proceed the process design with the Aspen file attached.

The underlying thermodynamic properties and fluid phase equilibria are crucial for the design and development of a chemical process. However, such data may not always be available, particularly for fine or specialty chemicals. In this work, we evaluate the reliability of using modern computational chemistry combined with recently developed predictive thermodynamic models to provide all the thermodynamic properties required in process design with ASPEN PLUS. Specifically, the G3 method is used for the ideal gas heat capacities and properties of formation, and the PR+COSMOSAC equation of state and COSMO-SAC activity coefficient model are utilized for the properties and phase behaviors of pure and mixture fluids. These methods are chosen because they do not require any species-dependent parameters and can, in principle, be applied to any chemical species. For a set of 972 chemicals, it is found that most properties can be predicted with a satisfactory accuracy (less than 10%: critical temperature [5%], critical pressure [10%], critical volume [5%], constant pressure ideal gas heat capacity [5%], and heat of vaporization [10%], except for the acentric factor [33%] and vapor pressure [73%]). Furthermore, the predicted results show little bias suggesting that these theoretically based methods are reliable for new chemicals for which experimental data are not yet available. Our analyses show that better accuracy in the prediction of vapor pressure and formation enthalpy and free energy is necessary for the design of chemical processes without relying on any experimental input. Nonetheless, these methods often provide reliable relative property values (e.g., relative value of normal boiling temperature can be predicted with 94% accuracy), making it possible to screen for new chemicals for improving existing processes.
To further improve the accuracy of prediction on vapor pressure, we include easily accessible experimental normal boiling points from a large dataset of nearly 20000 compounds to correct the critical points and acentric factor predicted by PR+COSMOSAC. A stronger linear relationship between the deviation of critical temperature and that of normal boiling points is observed, and thus critical temperature is significantly improved (9%→3%) compared to the critical pressure (13%→9%) while the accuracy of acentric factor is lowered (47%→58%). The large dataset is used to test the prediction of vapor pressure by several models, including original PR+COSMOSAC, PR EOS with critical properties and acentric factor from PR+COSMOSAC, PR EOS with corrected critical properties and acentric factor, and, the benchmark, PR EOS with experimental data. Predicted values are compared to the experimental data at mid-, high-temperature region. It is found that the corrected one shows remarkable accuracy with AARD at mid- and high-temperature region being 16% and 19%, respectively. In comparison, the AARD of predictions from original PR+COSMOSAC are 114% and 75% whereas the best one, predictions from benchmark models, are 7% and 3%. On other hand, the prediction at low-temperature region have been thoroughly examined as well but limited improvement is observed (PR+COSMOSAC: 323%; the one improved by experimental boiling point: 237%). In general, the vapor pressure predictions are shown to be comprehensively improved at different temperature regions and for compounds with different size. Correcting vapor pressure prediction with the help of experimental boiling point manifest itself as a simple workaround to the refinement of theory of PR+COSMOSAC.

Three process design works related to CO2 reduction has been demonstrated to be benefited from T.E.A.M. for evaluating the missing parameters. It is shown to be especially useful for process involving the compounds which cannot be dealt with built-in group contribution methods. In this regard, the developed online property calculator indeed provides an easy but reliable way of completing the missing parameters preceding the process design.

並列關鍵字

PR+COSMOSAC ； COSMO-SAC ； G3 ； quantum chemical calculation ； first principles calculation ； process simulation ； property prediction

參考文獻

(1) Pistikopoulos, E. N.; Barbosa-Povoa, A.; Lee, J. H.; Misener, R.; Mitsos, A.; Reklaitis, G. V.; Venkatasubramanian, V.; You, F.; Gani, R. Process systems engineering–the generation next? Computers Chemical Engineering 2021, 147, 107252.

(2) Vogel-Heuser, B.; Hess, D. Guest editorial Industry 4.0–prerequisites and visions. IEEE Transactions on automation Science and Engineering 2016, 13 (2), 411-413.

(3) Vogel-Heuser, B.; Bauernhansl, T.; Ten Hompel, M. Handbuch Industrie 4.0 Bd. 4. Allgemeine Grundlagen 2017, 2.

(4) Lasi, H.; Fettke, P.; Kemper, H.-G.; Feld, T.; Hoffmann, M. Industry 4.0. Business information systems engineering 2014, 6 (4), 239-242.

(5) Reis, M. S.; Saraiva, P. M. Data-centric process systems engineering: A push towards PSE 4.0. Computers Chemical Engineering 2021, 155, 107529.