******************************************************** * * * OVERVIEW OF LS-OPT * * * ******************************************************** Contents: 1. General information 2. Running with LS-DYNA 2.1 Input template 2.2 Command file 2.3 More on LS-OPTui 2.4 Cleaning the directories during an optimization run 2.5 Preparing for and starting an optimization run 2.6 Testing and installation 2.7 What happens during an optimization run ? 2.8 Post-processing 2.9 Running LS-DYNA MPP 2.10 Job distribution through load sharing facilities NOTE: ---- If you want to go directly to the Installation instructions, please find the README.Installation file on the lstc.ftp.com site. 1. General information ------------------- LS-OPT is a standalone optimization and probabilistic analysis program which interfaces with simulation and preprocessor codes. A comprehensive interface has been provided for the LS-DYNA code. LS-OPT conducts optimization by doing a number of parallel runs in each iteration and processing the results from these runs. The LS-OPT solver interface allows distribution of the simulation runs through queuing systems. Automatic result extraction on the remote node and data transfer to and from the node is supported. LS-OPTui is a graphical user interface for LS-OPT. It allows input specification, monitoring of distributed simulation runs and post-processing of both the optimization and simulation results. The postprocessor can be executed separately or from LS-OPTui. LS-OPTui produces an ASCII command file called "com". (See "Command file" below). 2. Running with LS-DYNA -------------------- To be able to conduct an optimization, LS-OPT typically requires the input template and any files that are to be appended to the final run input file. It also requires an optimization command file. 2.1 Input template: -------------- This file is a standard input file, but it contains labels to indicate which parameters are variables. An example of an input line in a LS-DYNA input file is: *MAT_PLASTIC_KINEMATIC 1,1.000E-03,<>,0.300,<>,10.0,0.000E+00 LS-OPT now imports the <<>> parameter names from the keyword files. The *PARAMETER keyword is supported from Version 2.2. The final run input file is constructed by LS-OPT by making the necessary substitutions in the template and appending the necessary files. Each run is conducted in a SUB-SUB-directory opened by LS-OPT so that all the files pertaining to a particular run are kept in a unique directory. These directories are referred to as the run directories. The run directories are also kept for distributed runs to assemble input for and collect output from the remote node. There is a different SUB-directory for each solver (just above the run directories). A typical LS-DYNA optimization run involving only one simulation case will therefore have only one SUB-directory. Multiple SUB-directories are only defined when doing multi-disciplinary optimization or multi-case optimization in which there is more than one distinct Finite Element model required by the design. The described logistics do not require user intervention. 2.2 Command file: ------------ The optimization command file can be set up by hand or produced by the interface program LS-OPTui. The commands are described in the LS-OPT User's Manual. The syntax is reasonably intelligible, without reference to the Manual. LS-OPTui produces a neatly formatted command file. 2.3 More on LS-OPTui: ---------------- The selection of various input phases in LS-OPTui has been designed in a logical sequence in the main bar above the LS-OPTui window. These are Info, Solver, Variables, Point Selection, Responses, Objective, Constraints, Run and View. The "Run" selection is for starting and monitoring the optimization process. "View" is for post-processing. The other selections relate to specifying the simulation program data and the optimization or robustness formulation. 2.4 Cleaning the directories during an optimization run: --------------------------------------------------- The user can define a file named "clean" which contains a script to erase simulation files from each run directory during an optimization run. This file should be created in the main work directory and will be executed by the LS-OPT job extractor (or wrapper in the case of distributed runs). 2.5 Preparing for and starting an optimization run ---------------------------------------------- The most convenient mode in which to run LS-OPT is through the graphical user interface, LS-OPTui. LS-DYNA Version 950d or later (960 or later for MPP) is recommended for simulation as only these versions provide sufficient data for LS-OPT monitoring. This progress monitoring shows up as a % progress when LS-OPT is run in batch mode or as a green thermometer when LS-OPT is run through the user interface LS-OPTui. To start the user interface, type: lsoptui if the command file already exists, or lsoptui to create a new command file. LS-OPT is then executed and monitored from the interface by simply selecting the [Run] button in the [Run] window. Optimization can also be run in batch mode by typing: lsopt e.g. lsopt com Therefore LS-OPTui and LS-OPT can be run on different platforms, the former being used to create the input and the latter to run the input command file. LS-OPTui has not been secured. 2.6 Testing an installation: ----------------------- A good initial test is to run example CLASS_EXAMPLES/EX4/EX4a (simple crash problem). This model will test optimization (one iteration) with LS-DYNA and requires a few minutes. The example contains most of the necessary logistical tests for running a general program, but it does not test the running of a preprocessor. For this purpose, EX5/EX5a can be used (also runs in a few minutes). 2.7 What happens during an optimization run ?: ----------------------------------------- LS-OPT opens a series of run directories and starts a series of simulation runs according to the chosen point selection scheme. These can be run in parallel. After completion of each run, the results are extracted from the simulation output in the various run directories and gathered in the "AnalysisResults" files in the solver directories. These files are then used for further processing such as to construct the response surfaces for optimization. If more than one iteration is allowed, a new set of run directories will be created and a new set of runs will start. The run directories are numbered as .1, .2, etc. The runs can be monitored using LS-OPTui. A series of progress bars automatically appears after selecting the "Run" button. LS-OPTui can also be started up during a batch LS-OPT run and will immediately show the progress. At the end of the process, one final run is made to validate the last proposed optimum. If the baseline simulation run fails, the user is referred to a lognnnnn (where nnnnnn is the process ID) file in the run directory which provides the log of the run. If the space between the '=====' lines is empty, LS-OPT probably could not find the executable program. 2.8 Post-processing: --------------- The optimization process produces a variety of output files for studying and plotting: lsopt_input Echoes the input data lsopt_output Contains detailed output data for all iterations. Experiments.* Contains the design parameters for each simulation. AnalysisResults.* Contains the simulation results for each simulation DesignFunctions.* Contains the metamodel definitions. Net.* Contains the metamodel definitions. ExtendedResults.* More simulation results StatResults.* Contains statistical results, e.g. standard deviations of responses. OptimizationHistory Optimization history for all variables For other output files refer to the Manual. Post-processing using LS-OPTui can be done to study the following: 1. Variable screening results computed using the analysis of variance method (ANOVA). Confidence levels. 2. The accuracy of intermediate response surfaces. 3. The optimization history of all the variables and responses. 4. The trade-off qualities of the design. 5. The reliability of a design. 6. The degree of predictability of a design when changing the specified design parameters. Design sensitivities are also available from the lsopt_output file. The lsopt_input file echoes all the input data and reports which input files and include files were read together with the parameters found in those files. 2.9 Running LS-DYNA MPP ------------------- LS-DYNA MPP can be run using the LS-Dyna option in the "Solver" window of LS-OPTui. However, the run commands must be specified in a script, e.g. the UNIX script "runmpp": mpirun -np 2 lsdynampp i=DynaOpt.inp cat dbout.* > dbout dumpbdb dbout The solver specification in the command file is as follows: solver dyna 'Anyname' solver command "../../runmpp" solver input file "car5.k" solver insert file "rigid2" Note: ---- 1. DynaOpt.inp is the reserved name for the LS-DYNA MPP input file name. This file will be created in the run directory by LS-OPT after substitution of the variables or creation by a pre-processor. The original template file can have a different name and is specified as the input file in the "Solver" window. 2. "lsdynampp" is the name of the MPP executable. 3. The file "dumpbdb" for creating the ASCII database must be executable. 4. The script must be specified in one of the following formats: - path relative to the run directory: two levels above the run directory (see example above). - absolute path, e.g. "/origin/users/john/crash/runmpp" - in a directory which is in the path. In this case the command is: solver command "runmpp" 5. The script must be executable. 6. LS-DYNA MPP will only give continual progress feedback through LS-OPT for version 960 and later. 2.10 Job distribution through load sharing facilities ------------------------------------------------ Details can be found in the User's Manual. ----------------------------------- E N D ----------------------------------