Computer models reduce need for real-world testing

• Accurate simulations can replace many experiments and plant field tests
• Multi-scale and multi-physics models bridge science and engineering
• Open source software continues to make inroads

As engineers and scientists strive to do more with less, computer modelling has become essential to cut costs, speed development and reduce uncertainty when designing everything from processes to molecules. Flowsheet simulators, a defining tool for every chemical engineer since the 1990s, have seen incremental improvements in power and usability over recent years. Computational fluid dynamics and molecular modelling, in contrast, have had more room to advance, and are now able to replace a great deal of experimental work. Open source simulators offer a serious alternative to commercial software in several areas, while powerful general-purpose modelling tools and “multi-scale” models are blurring the boundaries between different types of simulation.

Prediction is a vital part of the scientific method. Only when we can forecast how a process or a molecule will behave, independently of experiment, can we claim real understanding. Accordingly, mathematical models of physical, chemical and thermal phenomena lie at the core of engineering, many branches of chemistry, and increasingly the biological sciences too.

Much of the mathematics underlying heat and mass transfer dates back to the 18th and 19th centuries. But since it is often easier to write differential equations than to solve them, practical solutions to many engineering design problems had to wait until computers could provide brute-force solutions (“numerical methods”).

Since then, advances in computer power and mathematics have enabled both highly complex time-independent optimisations such as protein folding, and also dynamic simulations of gas flows, combustion modelling, and advanced process control.

Multi-purpose simulators

For many decades, libraries of numerical methods were available on mainframes and minicomputers to anyone with the skill to write their own mathematical models. But this was difficult stuff, often left to departmental experts and not for one-off use.

In the early 1980s the first spreadsheets – VisiCalc and Lotus 1-2-3 – made microcomputers a practical everyday tool for chemical engineers. Spreadsheets made it much easier to solve the complex sets of simultaneous equations that characterise plant flowsheets, and enabled design improvements through a series of “what-if?” calculations.

Spreadsheets have now been joined by multi-purpose modelling environments such as Mathematica/Wolfram SystemModeler (Wolfram Research, USA), MATLAB/Simulink (MathWorks, USA), and dozens of alternatives. Combining flexibility with great power, these can model mathematical functions, process plants, mechanical devices and electrical systems.

The open source Modelica modelling language, for instance, allows users to create and link blocks of equations describing, say, individual items on a flowsheet. In turn, Modelica can be used with a number of commercial front-ends including Wolfram SystemModeler and SimulationX (ITI, Germany).

One reviewer wrote recently that for modelling a fuel cell application, the open source Scilab/Xcos environment has 80–90 percent of the power of the commercial package MATLAB/Simulink, which costs several thousand dollars.[1] As with much open source software, Scilab/Xcos offers an active user community but documentation is in short supply.

For the most difficult tasks, multi-purpose simulators can run on high-performance computing (HPC) clusters, often taking advantage of each server’s graphics processing unit (GPU) as well as the main floating point processor (CPU). At the other end of the scale, capable equation solvers and graphing packages are available on smartphones.

The most powerful multi-purpose simulators offer ideal platforms for the new trend towards multi-scale modelling (see below).

Molecular modelling

Harnessing computing power to a knowledge of atomic properties and chemical bonding can help chemists predict the shapes and chemical properties of complex molecules. This has many applications in the life sciences, from fundamental research to the development of new drugs. “Computational chemistry” is also increasingly used in materials research to help design new products including catalysts, polymers, electrodes for high-performance batteries, and thermal insulators, and to understand reaction kinetics.

Dozens of programs are available in this demanding field. Some model atoms alone (“molecular mechanics”), while others (“quantum models”) take account of electrons too. Some are preferred for investigating existing molecules; others target the design of new substances. Some can handle a large range of structures, while others are more specialised: for instance, a whole class of software is available for modelling nanostructures such as carbon nanotubes and graphene. Some use HPC to target cutting-edge problems, while others yield useful results on modest PCs.

Examples of the many commercial molecular modelling packages for life sciences are Biologics Suite (Schrödinger, Japan) and Lead Finder (Molecular Technologies, Russia). Open source software includes DelPhi (USA) and Ascalaph (Russia/Sweden). For engineers, one example of a chemical simulation package is CHEMKIN/CHEMKIN-PRO (Reaction Design, USA), which is aimed at combustion processes and especially engines.

Accelrys (USA; a subsidiary of French 3-D systems specialist Dassault Systèmes) offers Materials Studio for studying catalysts, polymers, metals and electrical materials, alongside software for chemistry and the life sciences. According to Accelrys, simulation has allowed some customers to cut by 90 percent the number of experiments needed to launch a new product.

Though modelling at quantum or atomic level is often important in its own right, especially in materials research we are often more concerned with bulk properties or the challenges of manufacturing new materials. As a result there is growing interest in multi-scale modelling, which seeks to combine knowledge at quantum, atomic, intermediate and bulk or continuum levels.

Academic work on multi-scale modelling is taking place at institutions including the universities of Manchester[2] and Oxford (UK), the University of Basel (Switzerland) and the Fraunhofer Ernst-Mach-Institut (Freiburg, Germany). The design and manufacture of materials and physical products via multi-scale modelling is also known as integrated computational materials engineering (ICME).[3]

Computational fluid dynamics

At the opposite end of the size scale from molecular modelling, computational fluid dynamics (CFD) uses equations describing turbulence and heat transfer in bulk fluids to model engineering problems involving fluid flow. Applications include aerodynamics, complex flows in reactors and packed beds, dryers and heaters, and combustion processes, including explosions.

20 years ago the first commercial CFD programs were time-consuming to set up and took days or weeks to solve practical problems. As a result, CFD was used only to confirm final designs or as a troubleshooting measure. Today, software advances and affordable HPC allow CFD to provide useful input much earlier in the design process, and to optimize designs via repeated simulations, with minimal input from engineers.

Hygienic processing specialist GEA Process Engineering (Denmark), for example, uses CFD to design and troubleshoot spray dryers and mixers for the food and pharmaceutical industries. The company’s DRYNETICS modelling technique, introduced in 2008, combines CFD with real-world measurements on actual droplets and particles. Simulation is done on a new HPC cluster with 512 cores, 90 TB of disk space and 2 TB of RAM.

Just as with molecular modelling, CFD has now expanded its reach to multiple engineering disciplines and size scales. In fact, the lines between CFD and structural mechanics – finite element analysis (FEA) – are now so blurred that it makes little sense to refer to “CFD” at all, claimed Bill Clark, Executive Vice President of simulation company CD-adapco (USA) recently.[4]

As simulation increasingly replaces physical testing, simulation specialists face a great deal of responsibility to come up with the right answers. At the same time their jobs are becoming harder as the problems get bigger. “Customers want to see the big picture, with whole systems rather than individual components, and there are really no easy problems left to solve,” Mr Clark said.

CFD codes such as Star-CD and Star-CCM+ from CD-adapco, and Fluent and CFX from ANSYS (USA), the largest commercial CFD supplier, combine good performance with an all-in-one approach that can make them a good choice for firms new to CFD, notes aerospace CFD expert Dr. Chris Nelson.[5] On the other hand, solutions based on separate components — grid generator, flow solver and post-processor – can be more powerful.

Also to consider are the many excellent open source CFD codes, of which OpenFOAM (ESI Group, France) is possibly the best known. Dr. Ma Shengwei of the Institute of High Performance Computing, Singapore, says that open source CFD can be just as good as the commercial version (“there are almost no secret recipes”), but depends on skilled personnel and so is not necessarily cheaper.[6]

Flowsheet simulation

Flowsheet simulation lies at the heart of chemical engineering. Its foundations are mass balances, energy balances, mass transfer, heat transfer, phase equilibria, and reaction modelling.

Compared to molecular modelling or CFD, steady-state simulators are relatively undemanding in terms of computing power. Combined with their smaller market, this means that vendors are more likely to differentiate themselves on the basis of industry focus, ease of use, customer service and licence costs than on pure technical performance.

For the oil, gas and chemical industries the traditional market leaders are Aspen HYSYS (hydrocarbons) and Aspen Plus (chemicals) from AspenTech; UniSim (developed from the same code base as HYSYS) from Honeywell; and SimSci PRO/II from Schneider Electric. ProMax from Bryan Research and Engineering is a strong challenger to HYSYS and UniSim, especially among smaller customers. Other important players include CHEMCAD (Chemstations), DESIGN II (WinSim), ProSimPlus and the Simulis family (ProSim), and VMGSim (Virtual Materials Group).

Alongside their flowsheet simulators, all the large vendors supply packages and tools aimed at specialist industries (e.g. fuel cells), processes (e.g. crude units), equipment items (e.g. heat exchangers), and design techniques (e.g. heat recovery networks and financial analysis). Since flowsheet modelling depends on accurate characterisation of individual feedstocks and products, databases of physical properties and predictive “equations of state” are key to every simulator. The gap between physical property data within any database and the models within the simulators can either be closed by using specialized software tools like DECHEMA’s Data Preparation Package DPP (DECHEMA e.V.) or in most cases also with inbuilt tools from the different vendors.

Several of the original flowsheet simulators, notably Aspen Plus, have their roots in publicly funded research projects, and open source competitors are available, though not to the same extent as in CFD. A recent review of the open source DWSIM simulator rated it comparable in some ways to Aspen HYSYS, ProSim and VMGSim.[7] Both DWSIM and another open source simulator project, EMSO, originate in Brazil.

Different from open source software, but with a similar aim of promoting transparency, is the veteran CAPE-OPEN project that sets standards for the interchange of data in chemical process modelling. A simulation package that complies with CAPE-OPEN standards, for instance, can draw on different physical property databases and add in third-party unit operations such as novel reactor types, as long as these too meet CAPE-OPEN standards.

Process plants rarely operate entirely under steady-state conditions. For complex processes, dynamic effects may dominate operability and safety, especially during startup and shutdown. Many vendors therefore offer dynamic modelling capabilities through either their standard flowsheeting tools or dedicated products. An example of the latter from AspenTech is Advanced Process Control, part of the company’s aspenONE suite, which aids the design of complex control strategies to keep processes running under optimum conditions. Operator training is another important sector for dynamic simulation.

The multi-physics and multi-scale modelling discussed above has direct application to process plant modelling, too. A leading proponent of this “advanced process modelling” approach is Process Systems Enterprise (PSE; UK), with its gPROMS product. Through system-wide optimisation based on first-principles models at multiple scales, PSE claims that gPROMS can create benefits beyond the scope of traditional flowsheet simulators.

References
1. http://www.engineeringnewworld.com/?p=292
2. http://www.ceas.manchester.ac.uk/our-research/themes-challenges/themes/multi-scale-modelling/overview/
3. Committee on Integrated Computational Materials Engineering, National Materials Advisory Board, Division on Engineering and Physical Sciences, National Research Council (2008). Integrated Computational Materials Engineering: A Transformational Discipline for Improved Competitiveness and National Security. National Academies Press. p. 132. ISBN 9780309178211
4. http://insidehpc.com/2014/09/past-present-future-engineering-simulation/
5. http://www.innovative-cfd.com
6. http://caewatch.com/why-open-source-cfd-solution-is-not-for-everyone/, http://caewatch.com/cfd-in-2014/
7. http://www.engineeringnewworld.com/?p=168