A central group effort within SoftLab addresses the full range of issues in the future paradigm of computational science. These include collaboration with researchers in chemical engineering, mechanical engineering, biology, and biochemistry. In this paper, we describe the design and implementation of two prototype virtual laboratories, BioSoftLab and MicroSoftLab. Each is a hybrid problem-solving environment which supports the seamless and intelligent interaction of the experimental and computational models in a single, unified environment. We will show how these specialized environments attempt to fulfill the SoftLab research objective.
FIGURE 1. The SoftLab gateway
The virtual laboratory is a hybrid PSE that encompasses all research activities occurring the scientific laboratory environment, both "wet" and "dry". It is a software layer above the experimental and computational processes which exist in the physical laboratory side by side. But the virtual laboratory attempts to bring these two models of scientific research together in a way that allows them to interact with each other, so that feedback from one can enhance and improve the methodologies applied in the other.
This software layer should support interaction with remote instruments for control of the experimental process. It must also support simulation of the experimental process - which means that it must be able to access and execute the PDE code which models the process. It must act as a training simulator for the scientific methodology, and it should provide scientific visualization of output data resulting from either research model. Since the software layer must handle both models in a unified way, it is clear that they must be tied together from a theoretical viewpoint and from an operational viewpoint. The input and output associated with both models must appear in a uniform way to the scientist using the virtual lab.
It is not clear, a priori, that the two research models are coupled in this manner. Generally, the computational code that models the experiment is a fortran program driven by a fortran input file, where the input data bears little resemblance to the experimental input and process that it is supposed to represent. In addition, when the experimental input or process is varied, it is often unclear how to modify the input data to reflect the changes. The uncoupled state of these models is a significant problem for the implementation of a SoftLab laboratory.
Once this problem has been solved for a particular scientific environment, the design of the virtual laboratory can begin. In order to support the various activities that take place in an physical laboratory, scientists must be able to use the virtual laboratory to
FIGURE 2. Four scenarios of the virtual laboratory
The graphical interface which is then presented to the user is a software representation of the physical laboratory. Each important physical device is present, in particular, all instruments and equipment used during the experimental process must have a visual representation in the virtual laboratory. These software devices will henceforth be called virtual instruments.
Each scenario has a specific functionality that relates to a basic activity occurring in the lab. In the physical experimentation scenario, scientists can set up and control the laboratory instruments and experimental process from a remote location, and they can monitor the experiment via animation of the virtual instruments. The animation may be driven by state reports from the physical devices, or by calculations based on the input data. They can control data collection and extraction remotely during the experiment, and they can visualize the experimental results at its conclusion.The virtual lab can also act as an "experiment database", where the experiment's input configuration and output results can be saved for later analysis. The option of naming and saving experiments is truly a benefit for the scientists, since information defining the experiments will now automatically be available on-line to browse through, catalog, and analyze.
In the virtual experimentation scenario, scientists will set up the virtual instruments and experimental process just as in the physical scenario. Afterwards, the physical setup is transformed to the input required for the computational model. Additionally, parameters that are strictly numerical will be specified via special interfaces, and an expert system will be on hand to query for process characteristics or computational parameters. During the processing of the computational code, the virtual instruments will receive intermediate results so that they can be animated to show the progression of the simulation process. Results can be visualized exactly as in the physical case. The simulated experiment may also be saved to the experiment database. Here, too, the input configuration and output results will be used to define the experiment in the database. Since the physical and simulated experiments now reside together, computations on their associated data can easily be done for comparison or analysis.
The playback scenario uses the experiment database, from which physical and simulated experiments can be retrieved. Scientists can review or compare experiment input configurations and selectively display output results. Physical and simulated experiments can be compared or combined, and multiple data sets can be visualized concurrently. Physical results can be used to improve the computational model via feedback processes, and simulation results can be used to design new physical experiments. Finally, scientists can "playback" an experiment on the virtual instruments. The functionality of the playback scenario links the two research models in new and important ways, and thus fulfills one of the main objectives of the SoftLab project.
The training scenario has two distinctly different audiences. First, it can be used by novice scientists to learn about the physical layout of the laboratory, the instrument characteristics and operation, and the experimental methodology. Novice scientists can also learn about the computational model associated with the experimental process, and how the experimental process and its simulation interact. Second, it can be used by more experienced researchers to learn how Softlab can be used as a virtual laboratory. The eventual use of SoftLab's virtual laboratories in engineering classrooms is a major goal of the Softlab project.
The infrastructure that supports the four scenarios is shown in figure 3. The core functionality of the SoftLab laboratory is the representation and operation of the virtual instruments. They can be set up and controlled just as their physical counterparts are. They can be animated to show the progression of a process, physical or simulated. Access to the experiment database for storage, retrieval, and analysis is another core operation in the virtual lab. These functions are required throughout all four scenarios. Some functionality is common to a restricted subset of the scenarios, such as visualizing results in real time. Other functionality is specific to one scenario, such as the multi-media annotations used in the training scenario. The look-and-feel of this environment is uniform across all scenarios. Each scenario presents the virtual laboratory with working virtual instruments. The functionality of each scenario is described at the top level, and help is available at every level.
The remainder of this paper is as follows: a detailed description of the implementation of the virtual environment for the Bioseparation Laboratory, a brief description of the implementation of the Micro-electronics Laboratory, a summary of the functionality of SoftMedia - the multimedia component of SoftLab, and the conclusion.
One final note: the system described here was implemented using LabView [2]. This software is commercially available from National Instruments, Inc. The infrastructure and design of this system are not tied in any way to a LabView implementation.
FIGURE 3. The SoftLab infrastructure : integrating physical experiments and numerical simulation in a single, unified environment
Bioseparation is the process of separating chemical components by passing solution mixtures through an adsorbent column. The column is packed with sorbents selected for their adsorption properties with respect to the incoming solutions. Solutions pass through the column and elude from the bottom. Since each solution component adsorbs to the surface of the sorbents in a unique way, components elude at different times. Reactions within the column may also produce new components. The primary goal of bioseparation is to control what happens in the bioseparation column, so that one can predict when and at what concentration certain components will elude from the column. This process is used for final purification of proteins, chemicals, and biochemicals used in the manufacture of pharmaceutical and food products, for water treatment, and for many other biochemical processes. [3]
FIGURE 4. The Bioseparation laboratory
The physical instruments required to support this research are from the Waters Chromatography Division of Millipore. The bioseparation column is the focal point of the experimental process. Solution mixtures enter the column from two sources. One source is an injection syringe mounted on a tube inserted in the column. The other is a Waters 600E pump controller which regulates flow into the column from a combination of four attached reservoirs. When solutions elude from the column, component concentrations are measured by the Waters photodiode array detector. The measurements are collected by an attached NEC 80286 computer which Waters provides with custom hardware and software for communicating with the detector.
Because bioseparation experiments are both expensive and complex, computer simulation of the process is an attractive alternative to performing experiments. VERSE is a custom implementation of a numerical simulator which models the process, and this code is used by bioseparation scientists to simulate experiments. [4]
FIGURE 5. Diagram of the experimental process and its relationship with the simulation
For us to better understand and implement this common interface, we had to acquire a plethora of information relating the experimental process and the simulation process. By grouping actions together and relating them to the virtual instruments, the bioseparation scientist can now see how different parameters of the simulation input relate to the physical input. This correlation, which will become obvious in later sections, was not obvious prior to the development of BioSoftlab.
During the implementation of this PSE, it became obvious that the new environment could offer more than a common interface to the two environments. BioSoftlab could offer a means of communication between the two: data could be passed back and forth and information could be stored for later access and comparison. To support this functionality, the playback and analysis scenario was added.
FIGURE 6. Entering the physical experimentation scenario
In order to setup and run a physical experiment, the user has to set up the physical lab: this includes filling up the reservoirs and the injection with the solutions used in the experiment. Now scientists may perform the rest of the set up remotely using the BioSoftlab interface to the physical laboratory. There are various types of information required by BioSoftlab:
Given this input, BioSoftlab has all the information required to run a physical experiment. After the experiment is started, BioSoftlab displays an animation of what is happening in the physical lab: the reservoirs empty, liquid flows through the column, etc. Currently, this animation information is calculated given the controller flow rate information. As the next step, we would like to communicate with the controller to acquire this information and display it in the same way on the BioSoftlab virtual instruments.
When the experiment is complete, BioSoftlab can be used to connect to the Waters PC and access the data which was collected. This data is displayed on BioSoftLab's virtual PC. After reviewing the data, the scientist can choose to record the experiment in an experiment database for later retrieval. The information saved in the database includes both the experimental input and output. For the information to be complete (so that we can re-create the experiment or compare it to some other experiment) we ask the user to provide the following information:
During the implementation of this scenario, we encountered various problems related to the passing of information between the controller and BioSoftlab. This was particularly hard, since the Waters equipment was not designed to allow for remote control. Therefore, control via the virtual instruments was possible only by translating the user's input to the actual controller language. The ease of implementing a physical scenario, therefore, is directly related to the design and functionality of the actual physical equipment.
Another problem was more specific to interaction with the user. The virtual environment must be able to request the necessary information in such a way that even novice bioseparation scientists could supply the required input. The interaction between physical equipment is automatic, and we had to simulate the same inter-connection. If the user inputs a certain piece of information then we should be able to derive from it as much as possible, without asking the user to supply more than is absolutely necessary.
FIGURE 7. Entering the virtual experimentation scenario
To simulate an experiment numerically using the VERSE simulation code, select BioSoftLab's virtual mode. BioSoftLab presents you with a software view of the physical laboratory and its equipment. The instruments will be set up and programmed exactly as if for a physical experiment; the functionality of each virtual instrument is identical to that of its physical counterpart. However, in the virtual mode, the physical input is then transformed to the input required by the computational model.
Since there are additional numerical parameters and system characteristics associated with simulation, special interfaces built into BioSoftLab's virtual mode allow you to control the additional specifications. This information consists of the following:
In order to implement the above scenario, we encountered some major difficulties. We had already built the software representation of the physical instruments, and these virtual instruments were fully "functional". However, there existed no mapping between input/setup for the physical experiment and input to the computational model which was used to simulate it. We had to identify and separate the kinds of information required by the simulation code. A clear understanding of the differences between the following categories of data enabled us to devise a stable and consistent mapping.
FIGURE 8. Instrument setup: programming the virtual pump controller
Therefore, users must specify the contents of the solutions and their physical
characteristics. If a reservoir is filled with a "taxol broth"
solution, the user must name (or tag) the solution and describe exactly what
is contained therein. Clearly, users should not be required to identify the
contents of tagged input more than once. Therefore, this information
required the introduction of a database.
FIGURE 9. Solution input: filling reservoir A with tissue culture broth
FIGURE 10. Numerical properties of a component: describing taxol numerically
the column is to be monitored by the simulation process. Special information
about these components is needed (for example, isotherm coefficients,
diffusivity, etc.). This again requires a database interface for storing and
retrieving such data.
FIGURE 11. Reaction specification: identifying a dimerization reaction
FIGURE 12. Packing the column
There were also differences in the way physical and virtual experiments were visualized in BioSoftLab. The simulated experiment is monitored and visualized directly on the BioLabView instruments, just as in the physical experiment. But in this case, the data used to "animate" the instruments is generated by the simulation instead of by the actual devices.
FIGURE 13. Monitoring the column during simulation
This again allows the experimenter greater flexibility in visualizing the experiment, since the simulation generates intermediate data which the physical devices cannot produce. Column profiles generated by VERSE produced three-dimensional time-dependent component concentration data along the length of the column, which we visualized on special column-graphs within BioSoftLab. We also animated this data using scientific visualization tools in PDELab. [5] Graphs such as these were not possible before the advent of BioSoftLab.
FIGURE 14. Three-dimensional view of a reactant's concentration as it moves down the column In this case, the concentration is uniform across the column width.
In order to be able to re-run experiments, we had to construct a database where bioseparation experiments could be stored, and we had to define a format for the stored data. As a result, the user can now choose to save an experiment from either the physical or virtual modes. After the user specifies the input to the experiment, BioSoftLab collects all the information necessary to identify the experiment and records it in a file with predefined format. This includes physical characteristics of the solutions, physical dimensions of the column and the tube, etc. This data file is stored in the database where it can be retrieved at a later time. If the user is performing a virtual experiment, the datafile generated by BioSoftLab (and used as input to the simulation module, VERSE) will also be stored in the database.
Running the physical experiment or the simulated experiment generates various mode-dependent data files. In the case of a physical experiment, if the user opted to retrieve the detector-generated data from the PC, a file showing the concentration at the end of the column is the only file output during the experiment. In the case of a simulated experiment, the user specifies whether a history file (time-dependent end of column concentration) and/or a profile file (time-dependent concentration along the entire length of the column) should to be generated by VERSE, as well as whether PDELab format files should be generated. These output files are all stored into the database together with the input file.
With all the files required to define an experiment now stored within the database, the user is able to re-play an experiment. Selecting the Playback mode from the BioSoftLab menu, presents users with the following choices:
After the user selects an experiment for playback, BioSoftLab displays the input to the experiment: physical or virtual. In the case where users are comparing physical data to simulated data, both the physical and simulation input will be displayed. The virtual input is displayed in a format similar to that of the physical input, except that information about the components and reactions is also included.
If the user is simply reviewing the output data corresponding to a virtual scenario, the column profile data will be displayed (if it has previously been saved). Next, in both the physical and the virtual scenarios, an animated histogram of the concentration at the end of the column will be displayed on the virtual PC monitor. In the hybrid case, when the user is comparing physical data to the simulation results, both the detector-generated histogram and the corresponding VERSE-generated histogram will be displayed on the virtual PC, so that the results can be compared.
FIGURE 16. Comparison of the input to the physical and virtual scenarios
Even though the actual input and output is displayed as part of the Playback tool, there is currently no information saved which shows how an experiment progresses over time. A next step would be to enhance the database so that additional intermediate results could be stored which would allow bioseparation scientists to "run" an actual animation of the experiment, instead of viewing a sequence of static presentations of the data.
If the Teaching Tool is selected, users are presented with the following options:
For the bioseparation novice:
If the user is already familiar with the actual bioseparation experiment, he or she can use BioSoftLab's Teaching Tool to learn or practice setting up a physical or virtual experiment and better understand the interaction between the physical and computational models.
Clearly BioSoftlab can be used in an engineering classroom as an instructional aid for teaching
The advantages of having this capability in the classroom are obvious. It increases expertise in laboratory instrument operation on a large scale and without use of the costly lab. It allows access to laboratories that may not be locally available. It allows beginners to perform "dangerous" experiments. It uses a combination of recorded actual and simulation data to understand how these two models interact. It uses simulation to visualize and explain phenomena not measurable by real-time experimentation. These and many other benefits can result from a successful implementation of BioSoftlab.
The experiment involves the chemical vapor deposition (CVD) of silicon from silane on Si(100) substrate surfaces (wafers). In situ infrared spectra emissions are collected real-time to characterize the silicon surface and study reaction intermediates during the growth of a thin film of Si(100) at 750 degrees C and 10.0 Torr.[6]
The microelectronics laboratory consists of a reactor chamber (cell) that is placed inside a Nicolet 800 FTIR spectrometer. The cell was purchased by SpectraTech and modified to allow for better temperature control vacuum capabilities. Auxiliary equipment needed to supply the emission cell with N2, SiH4 and cooling water along with all temperature, pressure and mass flow instruments remain on the outside of the spectrometer.[7]
FIGURE 18. The MicroSoftlab environment
specified input, and then supply the Nicolet with the necessary information every 10 to 15 minutes. This allows the microelectronics scientist to perform a physical experiment without being present in the lab throughout the entire process.
Only part of the design for MicroSoftLab has been implemented. Remaining tasks include the integration of the computational model and animation of the virtual instruments.
SoftMedia is a general framework for integrating multimedia conferencing tools to any problem solving environment requiring multimedia facilities. An instance of SoftMedia is a collaborative platform which supports audio conferencing, video conferencing, and whiteboard capabilities.
SoftMedia supports dynamic re-configurability, so that a single interface can be presented to the user across machines with varying low-level audio and video software. These off-the-shelf multi-media tools are used to configure SoftMedia properly, according to the communication requirements of the PSE.
SoftMedia has been integrated into Softlab, and provides a multi-media platform for collaboration between the physical and virtual laboratories. This allows users of Softlab to look at and monitor the physical laboratory via video conferencing while running a remote experiment using SoftLab's virtual laboratory. Scientists using the virtual environment can also communicate with experimental scientists in the actual lab using audio conferencing and whiteboard facilities.
[2] National Instruments Coporation, LabView User Manual, Autin, Texas, 1992.
[3] K. E. Van Cott, R. D. Whitley and N.-H. L. Wang, ``Effects of temperature and flow rate on frontal and elution chromatography of aggregating systems'', Separat. Technol.1, 1991, pp. 142-152.
[5] Sanjiva Weerawarana, Elias N. Houstis, John R. Rice, Ann Christine Catlin,Cheryl L. Crabill, Chi Ching Chui and Shahani Markus, PDELab: An Object-Oriented Framework for Building Problem Solving Environments for PDE Based Applications, Proceedings of the 2nd Annual Object-Oriented Numerics Conference, 1994, pp 79-92.
[6] I-H. Oh and C.G. Takoudis, "Modeling of Epitaxial Silicon Growth from the SiH2Cl2-H2-HCl System in an RF-Heated Pancake Reactor," J. Appl. Phys. 69, 1991, pp. 8336-8345.
[7] I-H. Oh and C.G. Takoudis, "Modeling of Chemical Vapor Depostion in Pancake Reactor Systems," J. Electrochem. Soc., 90-12, 1990, pp. 75- 81.
For further information contact softlab@cs.purdue.edu