Preface to the DDB in Springer Materials

The year 2010 has proved to be an historic occasion for both the Dortmund Data Bank (DDB) and the Landolt-Börnstein Database (SpringerMaterials).

The Landolt-Börnstein data collection was started in 1883, and has always been dominated by scientists in a quest to structure, evaluate, and verify the mutual consistency of the vast amount of experimental results published in the open literature. With individual topics always being maintained by respected specialists for the respective property, Landolt-Börnstein has always been recognized as an important and respected source of data.

Nevertheless, when computers became widely available, and significant effort went into the simulation and optimization of the different processes (e.g., unit operations) carried out in the chemical industry, a strong need arose for the highly specific information required for this purpose. Process simulation requires reliable model parameters regressed to the most complete set of data, whereby these data, in addition to published results, will also often include proprietary information relating to the company involved.

The DDB, which was started in 1973 for the development of predictive methods, exactly suited this need by providing the data in a simple computer-readable format that could be directly interfaced to all forms of data regression and visualization software.

Over the years, there have been many parallel developments on both sides, such that the DDB now comprises nearly 4.2 million data tuples, and is growing by 7…8% every year. As a commercial product maintained by DDBST GmbH, the focus of the work must always be the needs of the customers; consequently, this data bank is very strong with respect to data of practical relevance.

With the release of its 2010 version, a subset of these data, limited to 50 common components and their mixtures, is now also available as part of the SpringerMaterials Online Data Bank, in the form of graphics and data tables. To find out about the availability of further information stored in the complete DDB, the user must still employ the free online DDB search at www.ddbst.com (Online).

DDBST GmbH hopes, that this cooperation is well received and can be further extended in the future.

Oldenburg, April 27, 2010

Prof. Dr. Jürgen Gmehling, President DDBST GmbH

About the Dortmund Data Bank

The Dortmund Data Bank (DDB) was started in 1973 by Dr Jürgen Gmehling, while working in the research group of Prof. Dr Ulfert Onken at the Department of Chemical Engineering at the University of Dortmund, Germany. The aim of the DDB was to build the comprehensive database that was required for the development of reliable predictive thermodynamic models (group contribution methods) for phase equilibria, with a wide range of applicability.

Because of the initial major importance of distillation processes, vapor-liquid equilibrium (VLE) data of non-electrolyte mixtures for compounds with a normal boiling point > 0°C were mainly stored. In its initial phases, this work was strongly supported by Dr. Hermann Stage (Stage-Schmidding-Heckmann), who had for a long time been collecting such data for the design of separation processes [ 1].

During the development of the DDB, several rules were strictly applied:

• Data were only entered from the primary publication.
• Copies of the primary publications were stored for data verification.
• Only experimental data were considered.
• The development of individual data banks was performed by specialists for the respective types of data; these usually were scientists who, at the same time, were in charge of model development.

Besides the easily accessible thermophysical properties from well-known scientific journals, the DDB contains large amounts of data that are not available via the open literature — for example, from systematic measurements performed for the development of predictive thermodynamic models, private communications, confidential data from industry, and unpublished experimental data from BSc, MSc and PhD theses … from all over the world.

With the appointment of Prof. Dr. J. Gmehling to the Chair of Industrial Chemistry at the University of Oldenburg in 1989, the spin-off company DDBST GmbH (www.ddbst.com), which was founded by Prof. Gmehling together with Jochen Menke and Jürgen Rarey, took over the further development and update of the DDB. Since this time, a continuous data input has been realized at DDBST GmbH in Oldenburg by long-term employees with many years of experience in thermophysical properties, and with comprehensive support being provided by colleagues in Estonia, China, Japan, and Korea. The present DDBST team from Oldenburg, Germany is shown in Figure 1.

In addition to making continuous updates of the different data banks, DDBST is also engaged in the development of software tools for data handling, retrieval, correlation, estimation, and visualization, as well as process synthesis tools (DDB Software Package DDBSP).

While, initially, mainly VLE data for non-electrolyte mixtures and compounds with a normal boiling point > 0°C were considered, later on also:

• vapor-liquid equilibria of low-boiling substances
• liquid-liquid equilibria (LLE)
• solid-liquid equilibria (SLE)
• azeotropic data
• adsorption equilibrium data
• polymer data
• activity coefficients at infinite dilution (γ^∞)
• excess enthalpies (h^E)
• excess heat capacities (c_P^E)
• excess volumes (ν^E); and
• mixture viscosities

not only for non-electrolyte, but also for electrolyte systems, were stored. Starting in 1991, the DDB was extended to all types of pure component properties.

Today, the DDB contains nearly all worldwide available phase equilibrium data, excess properties, transport properties and pure component properties, even for polymer systems, and continues to grow by almost 8% per year in terms of data volume. The current status of the individual mixture data banks in the DDB data pool is shown in Table 1. Due to the vast amount of pure component data published in the literature, this data bank is now the largest single sub-file of the DDB. The current status of the pure component data bank is shown in Table 2.

Currently, the DDB contains more than 4.2 million data points for more than 35 250 components, including salts, adsorbents, and polymers from more than 50 000 references from approximately 1 500 journals. The references cover worldwide sources in more than 20 languages. Worldwide, the DDB is the largest factual data bank for thermophysical properties, and is therefore an ideal tool for the development of predictive thermodynamic models and the synthesis, design, and optimization of the different processes in the chemical, petrochemical, gas processing, pharmaceutical, food, and environmental industries.

For almost 28 800 compounds, besides the experimental data the molecular structures are also stored in the form of connectivity matrices, and more than 49 000 sets of model parameters were derived by regression of vapor pressures, liquid densities, heat capacities, viscosities, thermal conductivities, etc., of pure components as well as simultaneous regression to VLE, h^E, γ^∞, … data for mixtures. For the use of modern methods employing results from quantum chemistry, more than 2850 COSMO σ-profiles have already been derived, using a high level of theory.

In particular, the intensive use of data for the development of predictive methods, the publication of data compilations, and the fitting of model parameters, led to the identification of a large number of errors in the primary literature (wrong components, wrong component order, wrong units, typographical errors, etc.).

The DDB is available under several different software platforms (DDBSP, DETHERM, …). DDBST GmbH offers various in-house versions, together with an integrated software system for data retrieval, storage, correlation, visualization and property estimation. The software package DDBSP also contains highly developed process synthesis tools for the selection of entrainers for extractive and azeotropic distillation, extraction, and absorption processes, the synthesis and design of separation processes, and the fitting and critical examination of model parameters used prior to process simulation.

DDBSP interfaces to the Aspen Plus^®, EPCON^® and ProSim^® process simulators for parameter and data interchange (IK-CAPE, as well as proprietary formats). In addition, Aspen Plus^® and Pro/II ^® property packages can be accessed within DDBSP in order to verify physical property parameters prior to process simulation.

A free explorer edition (www.ddbst.com ) allows the availability of data in the DDB to be checked, and to try out a large variety of DDBSP software.

In collaboration with the research group of Prof. Aa Fredenslund (Instituttet for Kemiteknik, Danmarks Tekniske Højskole, Lyngby, Denmark), the DDB was used for the further development of the group contribution method UNIFAC [ 2]. As early as 1977, a large parameter matrix was available, and the book Vapor-Liquid Equilibria Using UNIFAC was published, which included the exact calculation procedure, the available group interaction parameters, numerous applications of UNIFAC, and an extensive software source code [ 3].

As a result of the great success of the UNIFAC method, the development of predictive methods was later extended to further types of data in order to develop an improved UNIFAC version (mod. UNIFAC (Do) [ 4, 5]) as well as group contribution equations of state for the description of high-pressure phase equilibria (PSRK [6, 7], VTPR [8]), and electrolyte models (LIQUAC [9, 10, 11, 12], LIFAC [13]). Whilst, initially, the development was supported by research projects, today the further development of the group contribution methods of UNIFAC, modified UNIFAC and PSRK is sponsored by a company consortium [ 14]. Within the consortium, the quality of the results was improved and the range of applicability of the afore-mentioned group contribution methods largely extended. While, for example, group interaction parameters for 653 group combinations were published for the modified UNIFAC, group interaction parameters for 1379 group combinations are available for the consortium members. At the same time, the published group interaction parameters were revised for 255 group combinations.

With the availability of a vast amount of experimental data stored in the DDB, and the required software to minimize tedious and both time-consuming and error-prone tasks, work was started on the test of literature models and the development and regular extension and revision of new pure component property estimation methods for:

• normal-boiling temperature (T_b) [21, 22]
• vapor-liquid critical data (T_c, P_c, ν_c) [23]
• liquid-saturated vapor pressure as function of temperature (P ^S(T)) [24, 25]
• liquid-saturated viscosity as function of temperature (η^S(T)) [26]

The most important types of data used for the development of the different estimation methods mentioned are summarized in Table 3.

Due to the great importance of VLE data and g^E-model parameters, a large proportion of the VLE data were published in the form of a printed data series, starting in 1977 [ 27]. Data representations in these books focused on the raw experimental data and regression results for the most commonly used g^E-models, and the results of two thermodynamic consistency tests are given where applicable. This form of publication provided exactly the information required for the modeling of industrial distillation processes. Due to the success of the published Vapor-Liquid Equilibrium Data Collection, further data stored in the DDB were published, for example, for liquid-liquid equilibria [ 28], while excess enthalpies and activity coefficients at infinite dilution [29] were published within the DECHEMA Chemistry Data Series [30], and azeotropic data by VCH [ 31] (see Figure 2).

In order to educate chemical engineers to be capable of better exploiting the possibilities of modern factual data banks, software, and estimation methods, regular graduate courses in applied thermodynamics for the synthesis, design and simulation of thermal separation processes are carried out by the authors at various places worldwide ( 1).

Since work on the DDB was started 37 years ago, its primary mission has always been the development of reliable estimation methods for pure component and mixture properties. An overview of the work on mixture property estimation was given in the 2008 Rossini Lecture of Jürgen Gmehling (ICCT 2008, in Warsaw) [ 32].

The development of estimation methods for pure component properties, process synthesis tools and further applications of the DDB were discussed in an IUPAC overview article on the development and current application of physical property data banks [ 33].

For mixture properties and a wide variety of pure component properties, a significant number of estimation methods is available in the open literature. The application of these methods usually requires fragmentation of the molecules into a list of structural groups. In addition, many methods require additional information such as the number of rings, topological indices, or symmetry numbers. Fragmentation rules are sometimes not trivial, and may differ significantly between the different methods, resulting in tedious manual group fragmentation.

To improve this situation, and to be able to rapidly test the different models for a large number of components, several important software tools were developed [34]:

• An editor program to draw and display molecular structures (ARTIST).
• A molecular structure data bank for all components in the DDB (ChemDB).
• An automated procedure to correctly identify the different structural groups in a molecule (AutoInkr).
• Separate group structure definitions for each group contribution method, defining the group and its valid context inside a molecule.

The program package ARTIST integrates these software tools, and interfaces to commercial quantum chemical software, basic structure optimization, substructure search, …. Approximately 100 different group contribution methods are available for a large variety of different properties.

With the ability to rapidly estimate physical properties using different methods, combined with the vast store of experimental data in the DDB, a thorough test of estimation methods is possible. In order to assist the engineer to select the best method(s), the software ARTIST automatically analyzes the structure of the molecule and determines the chemical families to which it belongs (e.g., aliphatic alcohols, aromatic hydrocarbons). The results of the method tests are then presented in form of the typical deviations of each method for the respective chemical families.

Besides providing reliable experimental information for process simulation and the development of modern property estimation methods, the DDB also serves as a valuable tool for process synthesis. Due to the different advantages of distillation, this separation process is also employed in case of unfavorable separation factors or for azeotropic systems. Different alternatives are available if ordinary distillation is not feasible:

• In some cases, the azeotropic behavior disappears at lower or higher temperature (pressure), and ordinary distillation is possible at these different conditions.
• In the case where the azeotropic composition varies strongly with temperature (pressure), pressure-swing distillation is often used. This situation arises mostly if the enthalpy of vaporization — and, consequently, the slope of the vapor pressure curve of the two components to be separated — differ significantly, as for water plus organic components.
• In the case where the mixture forms a light-boiling heterogeneous azeotrope, the distillate streams of two columns are condensed and the liquid phases separated in a decanter and recycled to the columns. The pure compounds are obtained as the bottom products of the columns.
• If neither of the above options solves the problem, separation can often be achieved by using an entrainer.
• In extractive distillation, a high-boiling solvent is added slightly below the top of the column, its task being to alter the ratio of the activity coefficients of the two components to be separated. In the case of an aliphatic-aromatic separation, the interaction between NMP and the aromatic electrons of benzene lowers the latter's volatility; at sufficiently high NMP concentrations, this results in a separation factor of less than 0.4 between benzene and cyclohexane. The selective solvent is then recovered in a second column and recycled to the extractor column.
• In the case of azeotropic distillation, a suitable solvent is added which forms the lowest-boiling azeotrope with one or both of the components to be separated. The light-boiling azeotrope is obtained as the top product. In the case of an heteroazeotrope, separation of entrainer and product is usually simple, although in other cases a similarly simple separation of this mixture must be found.

One simple criterion to test the applicability of a component as an extractive solvent is to calculate the activity coefficients at infinite dilution of the components to be separated in the entrainer. These should be as different as possible, although high values of one of the activity coefficients can lead to limited miscibility and therefore lower the capacity of the solvent.

Potential entrainers for azeotropic distillation can be found by examining available data on binary or ternary azeotropes, or by estimating these data using predictive models.

In order to help the engineer identify optimal entrainer candidates, a special program has been developed as part of the package "DDB Process Synthesis" [ 35]. This software first examines the behavior of the binary mixture that needs to be separated, and verifies the existence of an azeotrope. It will then analyze whether this separation problem can be solved without introducing a further component (at reduced or elevated pressure, heteroazeotropic distillation or pressure-swing distillation). If this is not possible, the program proceeds by searching for a suitable entrainer by direct access to the DDB (56 660 γ^∞ values, 51 560 azeotropic or zeotropic data), or searching the results from predictive methods for a large number of potential solvents (up to 28 800). The detailed procedure is shown in Figure 3 for both options.

In case of the selection of suitable solvents by DDB access, information concerning the zeotropic/azeotropic behavior is taken from the data base for azeotropic data (51 560 data points). The data base on activity coefficients at infinite dilution which contains 56 660 data points is used to account for information on selectivity and separability.

As this procedure can only detect entrainers for which data with the components to be separated are in the DDB, a second complementary approach was implemented, in which all 28 800 components can be used as potential suitable solvents. In this case, the required activity coefficients at infinite dilution ( γ^∞) are estimated using models such as UNIFAC, mod. UNIFAC, and ASOG. Whether an azeotrope exists in a binary mixture can be decided from the pure component vapor pressures P₁^s and P₂ ^s and estimated γ^∞-values of the two components (2), while calculation of the exact azeotropic concentration requires iteration.

Both approaches are complementary, in that the selection by DDB access is able to yield only components for which the required experimental information is available, while the predictive approach can only cover components for which predictive models are applicable. It is, therefore, important to use both approaches to select the most suitable solvent.

Factual data and data estimation methods are of great and increasing practical importance, whereby the development of reliable predictive methods would not be possible without the vast amount of experimental information stored in factual data banks. While the different data compilations developed up to about 1970 focused on supplying information to the user directly, modern electronic data bases such as the DDB rely heavily on integrated software packages that feature complex data visualization, regression techniques and export formats to other software products, such as process simulation software. In the future, an ever-stronger integration of data banks, integrated software and applications can be expected. If experiences from other fields can be applied to sources for thermophysical properties, more and diverse forms of delivery and presentation (portals) will be developed, while maintenance of the data collections and the development of methods will remain the most cost- and labor-intensive tasks (content generation).

In the 21st century, simulation models are predominantly important for scientific and technical developments, and form the basis of economic and political decisions. Experiences from chemical process simulations show, that inadequate basic data such as thermophysical properties are the major reason for unrealistic results of simulation models leading to customer frustration and erroneous business and political decisions.

Therefore, factual data banks and predictive methods, combined with appropriate data mining tools carry a huge potential for economic and technological progress.

For a growing number of applications, thermophysical data banks and integrated reliable property estimation methods are of great importance, and lead to improved processes with a more efficient use of natural resources and less environmental hazard, as well as rapid technological progress. New developments and technologies in many fields would not be possible without the vast amount of knowledge in these data compilations and methods, and it can be expected that the use of these tools will not only intensify in core areas but also expand to fields that still employ more empirical approaches to problem solution.

This should generate sufficient interest to sustain the past 37 years of continuous development of the Dortmund Data Bank well into the future.

Endnotes

References