Automated Installation Verification of COMSOL via LiveLink for MATLAB

Transcription

1 Autoated Installation Verification of COMSOL via LiveLink for MATLAB Michael W. Crowell Oak Ridge National Laboratory, PO Bo 2008 MS6423, Oak Ridge, TN Abstract: Verifying that a local software installation perfors as the developer intends is a potentially tie consuing but necessary step for nuclear safety related codes. Autoating this process not only saves tie, but can increase reliability and scope of verification copared to hand coparisons. While COMSOL does not include autoatic installation verification as any coercial codes do, it does provide tools such as LiveLink for MATLAB and the COMSOL API for use with Java through which the user can autoate the process. Here we present a successful autoated verification eaple of a local COMSOL 5.0 installation for nuclear safety related calculations at the Oak Ridge National Laboratory s High Flu Isotope Reactor (HFIR). Keywords: autoation, software verification, software quality assurance. 1. Introduction In the past, installation verification of COMSOL for HFIR nuclear safety related calculations has been perfored by hand and in an ad-hoc anner. This has liited the precision of the individual verifications, the consistency of the verifications over tie, the nuber of versions of COMSOL verified, and the nuber of achines on which any version of COMSOL has been verified. In this current effort to verify the installation of COMSOL v on our local coputer cluster, we also seek to autoate and standardized the verification process soewhat. We epect this autoated process to yield increased precision and consistency, and draatically speed up the verification process such that updated COMSOL releases can be verified ore frequently, and on as any local achines as desired, with inial personnel effort. 2. Methods We have chosen to work via the LiveLink for MATLAB functionality offered in COMSOL, which allows full access to the COMSOL Application Prograing Interface (API): creating, editing, running, and etracting results, through the MATLAB coand line and scripting interfaces. Identical access to the COMSOL API is provided through the Java prograing language with the current COMSOL API for use with Java support. Although this verification effort could have been copleted with either the MATLAB or Java interfaces to COMSOL, MATLAB was selected due to the author s failiarity with it, and its ease of use for technical coputing. In order to verify the installation of COMSOL, it is necessary to have odels with results already provided by COMSOL ( included results) which can be rerun on the local achine where the COMSOL installation is being verified. This requireent is satisfied by the odel library that is included with a COMSOL installation wherein any of the odels are provided with already coputed results. Once these odels are rerun on the local achine, these local results can be copared with the included results and if the differences are sufficiently sall the COMSOL installation will be considered verified. Note that due to differences in achine architecture and atheatical libraries, the liitations of achine precision, and the relative error tolerance convergence criteria ipleented in COMSOL, the included and local results are not epected to be identical. We ust therefore decide which differences are sufficiently sall and which are not. To do this, we ust choose both a difference etric and a threshold. For the etric, we have defined a relative difference per u dependent variable, d i r, that is the aiu difference between the local and included results, noralized by the largest value in the included results for that particular dependent variable: Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston

2 d r u i = a ( u i loc u iinc a( u iinc ) ) Here u i is a particular dependent or solution variable, u i is the vector of the values for that variable at all the odel node points for each paraeter sweep, if any, and the inc and loc superscripts refer to the included solution and locally coputed solution respectively. Then we take the aiu relative difference over all the dependent variables in a odel, and that is the relative difference for the odel, d r : d u r = a(d i r ) i Our approach was to define the threshold for sufficiently sall fro a practical standpoint as being less than typical easureent uncertainty for our probles of interest. We deterined this to be a per odel relative difference of less than This equates to the aiu difference of any dependent variable at any node point being less than 1/10 th of 1 percent of the aiu or full scale value for that variable across the odel. Now this per odel relative difference ust obviously be bounded by the relative error tolerance ipleented in COMSOL. The eact relationship is nontrivial to deterine due to differences in forulation of the two etrics, but if the relative error tolerance is set too large we could be unable to eet our per odel relative difference threshold even with a correctly functioning COMSOL installation. Therefore part of this verification process will necessarily be to deterine if the default COMSOL relative error tolerance of is sufficient to guarantee that the per odel relative difference, as we have defined it here, is less than our prescribed threshold of Finally note that calculating d r as we have defined it requires calculating the differences between the included and locally coputed odel results for all dependent or solution variables at all odel node points for all odel paraeter sweeps. This approach is desirable in that even very inor discrepancies will be identified, but it would not be feasible without an autoated approach such as the one we have eployed. However, with autoation this approach can be both faster and ore reliable/repeatable than previous verification efforts. 3. Models In selecting which odels to consider, we first searched the odel library included with our local COMSOL installation for odels with included results. Fro the odels with included results we deterined what physics features were being used and copiled a aster list, organized by COMSOL odule, of the physics features available for verification (Table 1). Fro this list we down-selected to the physics features ost relevant/needed for HFIR related calculations, which can be found in Table 2, again organized by COMSOL odule. Finally, fro the list of odels with included results we deterined a inial set which utilized the desired physics features fro Table 2. The resulting odel/physics atri is found in Table 3. One odel, in addition to the 11 found in Table 3, was also considered due to a known error in the current COMSOL release. This error is an incorrect ipleentation of the periodic boundary condition for fluid doains where there are wall functions applied to the periodic boundary. To establish that this error does not occur when there are no wall functions on the periodic boundary, the odel circuit board forced 3d was also considered. 4. Results For the 12 odels considered in this verification study, the range of d r values calculated was fro a iniu of for the disk stack heat sink odel to a aiu of for the naca0012 airfoil odel. Machine precision for double precision floating point nubers is , so the iniu d r value was on the order of achine precision while the aiu was still well below our practical threshold of In general, odels with turbulent fluid flow (naca0012 airfoil, ahed body, displaceent ventilation, circuit board forced 3d) or nonlinear solid echanics (arterial wall echanics) had the highest d r values while odels with only heat transfer and/or linear solid echanics had significantly lower d r values. Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston

3 The d r values for all 12 odels can be found below in graphical for in Figure 1 and in tabular for in Table 4. The relative difference calculation was also perfored on the esh coordinates of each odel, and the aiu value across all 12 odels was which is on the order of achine precision, as epected. This value is shown graphically by the green horizontal line in Figure Conclusions All d r values were well below our threshold of and thus we ay conclude that the current COMSOL installation on our local coputer cluster is verified for the physics of interest as listed in Table 2. We ay further conclude that the default COMSOL relative error tolerance of is sufficient for eeting our current verification etric/threshold cobination. Finally, with the MATLAB scripts now in place, the above verification study can be repeated on any local coputer desired with inial user input and with the tie required governed only by how long the odels take to run on the given achine (~1/2 day on our local coputer cluster). The above study can also be epanded to include any or all of the additional physics features available in COMSOL (by using ore of the available odels) in a straightforward anner, with the tie required again governed only by how long it takes to run all the odels on the given local achine. 6. Notes As the conference paper length restrictions preclude inclusion of the MATLAB autoation scripts herein, please feel free to contact the author if interested in the scripts theselves. Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston

4 Table 1. Available physics in COMSOL database odels with included results. AC/DC: Charged Particle Tracing Electric Currents Electrostatics Magnetic Fields Acoustics: Pressure Acoustics, Frequency Doain Cheical Species Transport: Reacting Flow Transport of Diluted Species Heat Transfer: Bioheat Transfer Heat Transfer Heat Transfer in Fluids Heat Transfer in Pipes Heat Transfer in Porous Media Heat Transfer in Solids Heat Transfer in Thin Shells Heat Transfer with Radiation in Participating Media Heat Transfer with Surface-to-Surface Radiation Radiation in Participating Media Fluid Flow: Brinkan Equations Creeping Flow Euler-Euler Model, Lainar Flow Fluid-Structure Interaction Free and Porous Media Flow High Mach Nuber Flow Lainar Flow Lainar Two-Phase Flow, Level Set Lainar Two-Phase Flow, Phase Field Miture Model, Lainar Flow Miture Model, Turbulent Flow Non-Isotheral Pipe Flow Particle Tracing for Fluid Flow Pipe Flow Rotating Machinery, Lainar Flow Rotating Machinery, Turbulent Flow, k-ε Thin-Fil Flow, Edge Thin-Fil Flow, Shell Turbulent Bubbly Flow Turbulent Flow, SST Turbulent Flow, k-ε Turbulent Flow, k-ω Water Haer Matheatics: Boundary ODEs and DAEs Coefficient For Boundary PDE Coefficient For PDE Convection-Diffusion Equation Curvilinear Coordinates Defored Geoetry General For Boundary PDE General For PDE Global ODEs and DAEs Laplace Equation Matheatical Particle Tracing Moving Mesh Optiization Phase Field Microfluidics: Slip Flow RF: Electroagnetic Waves, Frequency Doain Structural Mechanics: Bea Bea Cross Section Mebrane Shell Solid Mechanics Truss Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston

5 Table 2. Subset of available physics desired for HFIR related calculations. Fluid Flow: Fluid-Structure Interaction Turbulent Flow, SST Turbulent Flow, k-ε Turbulent Flow, k-ω Matheatics: Boundary ODEs and DAEs Curvilinear Coordinates Defored Geoetry Global ODEs and DAEs Heat Transfer: Heat Transfer Heat Transfer in Fluids Heat Transfer in Solids Heat Transfer in Thin Shells Shell Solid Mechanics Structural Mechanics: Table 3. Model/desired physics atri. Physics: Fluid Flow Heat Transfer Matheatics Struct. Mech. Models: Fluid-Structure Interaction Turbulent Flow, SST Turbulent Flow, k-ε Turbulent Flow, k-ω Heat Transfer Heat Transfer in Fluids Heat Transfer in Solids Heat Transfer in Thin Shells Boundary ODEs and DAEs Curvilinear Coordinates Defored Geoetry Global ODEs and DAEs Shell Solid Mechanics oscillating fsi naca0012 airfoil ahed body displaceent ventilation heat sink surface radiation disk stack heat sink cohesive zone debonding arterial wall echanics electrocheical polishing diffuse double layer bracket shell Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston

6 Figure 1. COMSOL provided versus local coputer cluster coputed results. Table 4. Model aiu relative differences. odel oscillating_fsi naca0012_airfoil ahed_body displaceent_ventilation heat_sink_surface_radiation disk_stack_heat_sink cohesive_zone_debonding arterial_wall_echanics electrocheical_polishing diffuse_double_layer bracket_shell circuit_board_forced_3d e e e e e e e e e e e e-08 d r Ecerpt fro the Proceedings of the 2015 COMSOL Conference in Boston