Slide 1

Discovering good parameters for good gravity casting simulations! June 13-14, 2013113th FLOW-3D European Users Conference in Madrid, Spain Stefano Mascetti XC Engineering srl Ngadia Taha Niane PSA Peugeot-Citroen Slide 2 XC Engineering presentation June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain2 XC Engineering srl is a small company that works as a Flow Science associate All engineers in XC Engineering are highly qualified fluid dynamics engineers It is responsible for FLOW-3D sales and technical support in Italy and France Located 30km from Milan XC Engineerings strengths are in building custom FLOW-3D solutions, through customizations of the FLOW-3D solver. It has a growing cluster, to provide better support service also to FLOW-3D/MP users. Slide 3 Thanks to PSA for this test case! June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain3 This work has been developed with PSA, in particular thanks to: Mr. Ngadia Taha Niane Specialist in numerical modeling of casting processes Slide 4 The question June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain4 This work started by trying to answer the following recurrent questions: How accurate can my simulations be with FLOW-3D? What to set for numerical parameters like the HOBS? Sometimes when tuning numerical parameters it is difficult to know what to set and represent like a barrier between the simulation and potentially accurate results. fluid properties physical models HOBS1 HOBS1S FSCO TSDRG FSCR HFLV1 bad results excellent results Slide 5 The test case June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain5 Thanks to Peugeot-Citroen we have available a good experiment to compare with: A gravity casting pouring, with well known data and a mold fully filled of thermocouples. Having several thermocouples in the mold is helpful to track temperature history along the process time, as well as to give an idea of the filling dynamic. n 15 thermocouples time available from t=0 to t=1000 s Slide 6 The goal June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain6 The goal of this study is to find the best numerical parameters to set in FLOW-3D to match both the filling dynamic and the solidification behavior. HOBS = ? FSCO = ? FSCR = ? HFLV1 = ? Slide 7 The tools June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain7 To perform the best at this job, we need to couple FLOW-3D with an optimization tool, to allow the automatic calling of FLOW-3D simulations and a certain strategy to converge to the optimal solution. The physics involved and the amount of tuning parameters that we want to change are in fact too complex to be managed simply through human experience. Pre-simulation improvements [1] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain9 Optimization tools, especially if good, have the natural tendency of cheating! To push the solution at really high extremes, these tools often make use of the smallest software deficiencies or weak points, not always correspondent to the real deficiencies of the system. Then, before running the job, we carefully improved the setup trying to make it: 1.accurate -> we want search for a good comparison with thermocouples data 2.robust -> to prevent the optimizer from cheating 3.faster -> to run the highest number of simulations in a shorter time Slide 10 Pre-simulation improvements [2] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain10 To make the simulation more accurate we decided to introduce the following parameters: 1.roughness factor between metal and mold. (mold is sand, so it can have a significant roughness!) (ROUGH) 2.heat transfer between metal and air (HFLV1) and an external air domain + coefficient (HOBVT1) 3.temperature-dependent heat transfer coefficients (HOBS1 -> HTC vs. T table) 4.reducing the pressure convergence (EPSADJ = 0.1) 5.all temperature-dependent properties (from JMatPro) 6.in deep analysis of fluid motion, to correctly identify when it is possible to switch between the filling simulation and the solidification simulation (rapid model): before turbulent dynamic filling up to 16 seconds, then restart with the rapid shrinkage model turbulent dynamic filling up to 16 seconds, then restart, laminar fluid motion up to 200 seconds restart, with the rapid shrinkage model after Slide 11 Pre-simulation improvements [3] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain11 To make the simulation more robust we decided to work on some numerical parameters: 1.introducing a value for fluid fraction cleanup (FCLEAN = 0.025) 2.reducing the f-packing coefficient (CFPK = 0.01) 3.all external outflow boundary conditions to allow any fluid droplet exiting from the mold to be discarded from the domain Slide 12 Pre-simulation improvements [4] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain12 To make the simulation faster we decided to work on: 1.eliminating the initial ladle pouring, by defining an initial fluid region in the mold with a certain initial mean kinetic energy (TKEI) 2.using symmetry 3.using domain removing regions 4.using thermal penetration depth (LTHOBS) With all of these improvements we believe we were able to generate a setup file that was at the same time more accurate and faster. Simulation speed was reduced from ~30 hours to 5 hours, while additional physics were involved in this case. Slide 13 Optimization parameters filling stage June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain13 For the filling stage, we decided to pick-up from the setup as design parameters: Prepin variableMeaningRange (min-max) HFLV1heat transfer between metal and void5-80 CLHT1latent heat of solidification(nominal) 10% TSDRGsolidification drag coefficient0 250 FSCOcoherent solid fraction0 0.6 ROUGHroughness for solid component0.01 1 *e-3 TEMPIinitial metal temperature720 750 HOBS(750)htc between metal and mold at T=750C100 16000 HOBS(650)htc between metal and mold at T=650C100 16000 HOBS(613)htc between metal and mold at T=613C100 16000 HOBS(580)htc between metal and mold at T=580C100 16000 10 independent variables Slide 14 Optimization parameters solidification stage June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain14 About the solidification stage, a similar (but different) set is chosen: Prepin variableMeaningRange (min-max) HFLV1heat transfer between metal and void8-25 CLHT1latent heat of solidification(nominal) 10% TSDRGsolidification drag coefficient100 250 FSCOcoherent solid fraction0 0.49 FSCRcritical solid fraction0.5 0.9 HOBS(613)htc between metal and mold at T=613C100 16000 HOBS(580)htc between metal and mold at T=580C100 100000 HOBS(560)htc between metal and mold at T=560C100 100000 HOBS(540)htc between metal and mold at T=540C100 100000 HOBS(490)htc between metal and mold at T=490C100 100000 HOBS(450)htc between metal and mold at T=450C100 100000 HOBS(20)htc between metal and mold at T=20C100 100000 12 independent variables Slide 15 Definition of optimization objectives [1] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain15 x15 Problem: How to translate in an efficient mathematical equation the qualitative objective minimize the error between the curves? from experiment there are n15 independent curves of temperature vs. time each curve is composed by > 2000 temperature points, to be compared with their respective temperature point coming from simulation. Slide 16 Definition of optimization objectives [2] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain16 The actual solution is based on a two objective solution: Objective #1: minimize the difference between the temperature (numerical experiment) recorded at a certain time of interests, or better: Objective #2: minimize the difference between the integral of each temperature curve, or better: for t = (ex.: 16 s) Slide 17 Uncertainties on the experimental data June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain17 About experimental data, we know there can be uncertainties both in the temperature value, and in the time. Hence, for the purposes of this work it is still considered good a value that is inside a range of 10 C, and an uncertainty on the time up to 1 second. Furthermore, observing the experimental curves, it seems that 3 thermocouples over 15 can be discarded. Slide 18 Running the optimization task June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain18 The two phases (filling and solidification) have been optimized separately. Each of them, brought to a Pareto curve, show all combinations of optimal points. Concerning solidification, the most interesting phase for this work, for example, was that about 50 calls to FLOW-3D simulation were necessary to get our best solution. Slide 19 Analyzing the optimal solutions June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain19 The optimizer provided a specific set of values that give the optimal solution, but identifying what ranges are good is not an easy task. Some parameters exhibit a certain dependency on the output, while some others did not: FSCOHOBVT1 examples: This plot clearly shows a certain dependency on the optimal solution (y-axis) by the coherent solid fraction value. On contrary, the htc between mold and air seems to be quite arbitrary. Slide 20 Results Comparing temperature plots [1/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain20 By comparing the temperatures at the thermocouples point, between numerical (optimal filling and solidification) and experiment we get 10/12 excellent matching! (before using optimal parameter, the matching was quite poor) Slide 21 Results Comparing temperature plots [2/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain21 [1/2] not best match Slide 22 Results Comparing temperature plots [3/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain22 [2/2] not best match Slide 23 Results Comparing temperature plots [4/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain23 Slide 24 Results Comparing temperature plots [5/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain24 Slide 25 Results Comparing temperature plots [6/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain25 Slide 26 Results Comparing temperature plots [7/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain26 Slide 27 Results Comparing temperature plots [8/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain27 Slide 28 Results Comparing temperature plots [9/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain28 Slide 29 Results Comparing temperature plots [10/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain29 Slide 30 Results Comparing temperature plots [11/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain30 Slide 31 Results Comparing temperature plots [12/12] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain31 Slide 32 Results The filling video June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain32 Slide 33 Results Comparative video June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain33 Slide 34 Results Optimal parameters [1] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain34 okay, but what are the optimal parameters to use then?? Remember that optimal parameters may be linked to the case of a gravity casting of aluminum in sand molds! Slide 35 Results Optimal parameters [2] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain35 Prepin variable MeaningDecisiveSuggested values HFLV1heat transfer between metal and void(15-25) CLHT1latent heat of solidification!(nominal) +2-4% TSDRGsolidification drag coefficient~150 200 FSCOcoherent solid fraction!!> 0.4 FSCRcritical solid fraction!0.65 < x < 0.9 ROUGHroughness for solid component!! (filling)0.55 0.8 *e-3 TEMPIinitial metal temperature(735 745) HOBVT1htc between void and mold~(15-25) ~ = objective is a little sensitive from this value ! = objective is quite sensitive from this value Slide 36 Results Optimal parameters [3] June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain36 Prepin variable MeaningDecisiveSuggested values HOBS(750)htc between metal and mold (T=750 C)(3000 - 6000) HOBS(650)htc between metal and mold (T=650 C)~3000 9000 HOBS(613)htc between metal and mold (T=613 C)~3000 8000 HOBS(580)htc between metal and mold (T=580 C)(5000 < x < 20000) HOBS(560)htc between metal and mold (T=560 C)~< 10000 HOBS(540)htc between metal and mold (T=540 C)~< 25000 HOBS(490)htc between metal and mold (T=490 C)~50000 < x < 90000 HOBS(450)htc between metal and mold (T=450 C)(50000 < x < 80000) HOBS(20)htc between metal and mold (T=20 C)~70000 < x < 90000 ~ = objective is a little sensitive from this value ! = objective is quite sensitive from this value Slide 37 Results Final remarks June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain37 With a good tuning of numerical/physical parameters it is possible to reach state-of-the-art filling and solidification behaviors, with an excellent matching with experimental thermocouples During filling stage, for sand molds, it seems crucial to set up a proper roughness factor (quite different from the default) Also it seems important to take into account for heat transfer with void, giving an average range for ht coefficient Default values for coherent and critical solid fractions seems also reasonable, and their sensitivity to final objective is present but minimal (using values around the default ones) During solidification, heat transfer coefficients seem important, but only as a range, not through a high accuracy Latent heat given by JMatPro database seems also an excellent number, with an optimal value in a range of < +4% of the nominal value. Slide 38 Results Open questions June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain38 In the present work, HTC coefficients are given through a long tabular input (temperature- dependent): it would be of great interest to try to condense the tabular input into a single input. More user-friendly! Also: the initial question of setting optimal HTC is partially answered, because it seems that the fitting curve is not so sensitive from HOBS coefficients. What about then leaving the default HOBS=-1 value? To perform accurate HPDC simulation: would these coefficients work well as well? (need for a test-case to compare with) What about using the rapid shrinkage model after the filling? How large is the error? As often happens, this work opened new questions other than closing the initial one: Slide 39 End of presentation - thanks for your attention - June 13-14, 201313th FLOW-3D European Users Conference in Madrid, Spain39