Open Material Property Library With Native Simulation Tool Integrations -- MASTO
aa r X i v : . [ c ond - m a t . m t r l - s c i ] J a n Open Material Property Library With NativeSimulation Tool Integrations – MASTO
Antti Stenvall and Valtteri Lahtinen
Abstract —Reliable material property data is crucial for trust-worthy simulations throughout different areas of engineering.Special care must be taken when materials at extreme condi-tions are under study. Superconductors and devices assembledfrom superconductors and other materials, like superconductingmagnets, are often operated at such extreme conditions: at lowtemperatures under high magnetic fields and stresses. Typically,some library or database is used for getting the data. We havestarted to develop a database for storing all kind of materialproperty data online called Open Material Property LibraryWith Native Simulation Tool Integrations – MASTO. The datathat can be imported includes, but is not limited to, anisotropiccritical current surfaces for high temperature superconductingmaterials, electrical resistivities as a function of temperature,RRR and magnetic field, general fits for describing materialbehaviour etc. Data can also depend on other data and it canbe versioned to guarantee permanent access. The guiding idea inMASTO is to build easy-to-use integration for various program-ming languages, modelling frameworks and simulation software.Currently, a full-fledged integration is built for MATLAB to allowusers to fetch and use data with one-liners. In this paper webriefly review some of the material property databases commonlyused in superconductor modelling, present a case study showinghow selection of the material property data can influence thesimulation results, and introduce the principal ideas behindMASTO. This work serves as the reference document for citingMASTO when it is used in simulations.
Index Terms —experimental data, material property database,numerical modelling, simulations
I. I
NTRODUCTION I N all simulation work a guiding principle is garbagein means garbage out . The inputs of typical simulationsinclude the device under study, the operation conditions andthe material properties characterizing the components of thedevice. The model of the device under study can typicallybe taken, and often simplified, from the design drawings, orfrom the constructions. Operation conditions may be knownor certain conditions are sought. It can be difficult to findreliable material property data for special materials or typicalmaterials at extreme operation conditions. Characterization isoften possible but can be very time consuming. Therefore,modellers often rely on known sources – material propertylibraries or databases. In principle they are the same thingwith different names.Three larger material property libraries, databases or sourcesare well-known in the superconducting magnet community, among others. The data from the Cyogenics TechnologiesGroup at the Material Measurement Laboratory, NationalInstitute of Standards and Technology (USA) is often calledNIST data in modellers’ jargon [1]. It includes thermal con-ductivities, electrical resistivities and specific heats, amongother properties, for materials commonly used in constructingsuperconducting magnets such as different aluminum alloys,stainless steels and copper. Fits as a function of temperature,and in some cases for different material purities and as afunction of magnetic flux density, are provided with the data.The data are available via web pages of NIST in HTML formatfrom which one can copy the fits and the coefficients to one’ssimulation tool. The data are originally from measurementsat NIST. The NIST database represents data from a nationallevel public organization.MATPRO – A Computer Library of Material Property atCryogenic Temperature represents another such library [2].MATPRO was developed in a collaboration between CERNand University of Milano, Italy. Typical materials (insulators,metals, alloys) utilized in constructing superconducting mag-nets are included in MATPRO including the most common lowand high temperature superconductors. The data in MATPROare restricted to densities, specific heats, electrical resistivi-ties and thermal conductivities of materials. In the electricalresistivity and thermal conductivity data, the magnetic fieldand RRR dependence are also included in addition to thetemperature dependence. MATPRO is written in Fortran77 andit also has a Windows compatible executable that one canuse via the command prompt. The executable can be used toextract numerical data in ASCII format. The data in MATPROare collected from different sources, all documented in [2].CryoComp is a commercial material property library fromEckels Engineering Inc. [3]. A Windows compatible exe-cutable is provided for a license buyer. Data can be accessedvia an index and can be represented with tables, which can bedirectly saved or copied and pasted into, for example, Excel.This database also includes typical materials for superconduct-ing magnet design which are grouped by type in the index. Forexample, seven different nickel alloys are included. One canalso add private data into the database and use it via the samegraphical user interface as the built-in data. Data in CryoCompcomes without references and without any warranty or evena guarantee of suitability for the condition for which it isprovided. One should, however, note that even though thismay sound strange, this is standard license agreement text.Common to the presented three material property librariesis that they are standalone ones, not integrated to modellingsoftware, and not extendable by people not working directlywith these software. Data from the presented databases, as wells others, has been partially collected and implemented intosome special software, like for performing quench analyses inROXIE [4]. This integration, however, is not straightforwardas it is merely an export, and requires implementation of theproperties to the software, rather than a dependency whichupdates itself when the original software is updated.This paper introduces an ongoing effort to build a materialproperty database with the same principles as social codingnetworks like GitHub [5] and tools utilizing shared softwarelike Composer [6] or Bower [7], but with the possibility offeatures following scientific practices. This includes optionaldata review to promote their reliability. The database is calledThe Open Material Property Library With Native SimulationTool Integrations – MASTO. A domain name (masto.eu.com)for MASTO has been reserved and an early prototype isaccessible there via a web browser. MASTO will not belimited to just hosting for material property data, but alsofor networking among all the different stakeholders aroundmaterials and their properties. The completely new way ofacting with material property data that MASTO representsaims to establish a sustainable, open, extendable and reliablematerial property network from which one can find materialsand their properties, find experts to characterize materials andwhich one can use as a store window to make materialsavailable for possible customers.In this paper we first compare selected data from the threeintroduced material property databases – NIST, MATPRO andCryoComp – and study how the variation in the data influenceson modelling results. The aim of this comparison is not toquestion the reliability of the databases, but to emphasize howimportant it is to select the material properties for simulationsand how to get error estimates for the results by using differentdata sources. After this we introduce the fundamental ideasin MASTO and present its MATLAB integration. Finally,conclusions are drawn.II. I
NFLUENCE OF M ATERIAL P ROPERTY D ATA ON S IMULATION R ESULTS – A C ASE S TUDY
A simple way to estimate hot spot temperature in a super-conducting magnet follows the so-called adiabatic MIITs [8]approach in which the heat conduction is neglected and thehot spot temperature can be directly estimated from the currentdecay curve – either measured or simulated. The adiabatic heatbalance equation: c ∂T∂t = ρ (cid:18) IA (cid:19) , (1)where c is the volumetric effective specific heat, ρ is the sta-bilizer resistivity, I is the current and A is the cross-sectionalarea of the stabilizer, can be separated into material propertiesand current-dependent terms, and integrated separately as: Z T max T op A cρ d T = Z ∞ I d t, (2)where T op corresponds to the operation temperature at whichheat starts to be generated and T max corresponds to the upperlimit of the hot spot temperature, typically at the origin of the quench, at which the integrals on the right and left handsides are equal. For the material properties, those averagedover the magnet’s unit cell are used. Therefore, by measuringthe current decay curve in a quench experiment, which isstraightforward to measure, one can find T max . One shouldnote that because the heat conduction, and any cooling, isneglected, T max does not represent the hot spot temperature,but an upper limit for it – within the error of material propertydata and assumption of isothermal unit cells. MIITs abbrevi-ation comes from the scaled units: mega, current, current andtime and (2) is called the MIITs equation.The computation of the material property integral in theMIITs equation to different values of T max can reveal thesensitivity of the estimated hot spot temperature to changes inmaterial properties. To study this, we use the three presentedmaterial property sources: NIST, MATPRO and CryoComp.To simplify, we use only material properties of copper. Weconsider RRR of 100 and magnetic flux densities of 0 T and12 T to take into account also the magnetoresistivity which atlow temperatures is significant.The databases use different definitions for RRR. Here, wedefine RRR as the ratio ρ (273 K) /ρ (4 . at 0 T. The RRRvalues for the ratio of 100 defined accordingly correspondingto the RRR values in the different databases are the following:NIST 99.7, MATPRO 99 and CryoComp 100.We tabulate the material property values from 4 to 300 Kwith a spacing of 1 K and perform the integration with thetrapezoidal method. Our T op is 4.2 K. We normalize our resultsto the NIST data. Because the material properties changeby orders of magnitude as a function of temperature, thedifferences are better visible in the normalized graphs.Fig. 1 compares the resistivities from the three differentdatabases. As can be seen, CryoComp data at 0 T is almostthe same as NIST data, differing by less than 0.5%. At 12 T thelargest difference is still less than 4%. This characterizes thedifficulty of describing resisitivity as a function of RRR andmagnetic flux density. NIST and MATPRO data differ substan-tially both at low temperatures and intermediate temperatures.At 0 T field, the MATPRO data plummets to 22% below theNIST value around 35 K. At 12 T this peak is reduced to 6%at the same temperature. Another local maximum in relativedifference between the MATPRO and NIST data can be foundat 147 K and 142 K for 0 T and 12 T, respectively. In the caseof 12 T, the difference is 13%.Fig. 2 displays the difference in the values of specific heat.At low temperatures the relative variation oscillates whereasat higher temperatures it changes less rapidly. Above 160 KCryoComp and MATPRO data fit very well together and differat most 2% from the NIST data. One should keep in mind thata small absolute variation at low temperature results in a largervariation in relative error than at high temperatures. The ratioof copper’s specific heat at 300 K to 4.2 K is more than 4000.To study how these variations in material property valuesinfluence modelling, it is important to consider their inte-grals. For example, the typical quantity of interest in quenchmodelling is the hot spot temperature at the end of quench.Because the ratio of the specific heat and the resistivity playa crucial role in the temperature increase, the integral of that
100 200 300
Temperature [K] N o r m a li z ed r e s i s t i v i t y MATPRO 0 TCryoComp 0 TMATPRO 12 TCryoComp 12 T
Fig. 1. Normalized copper resistivities from the MATPRO and CryoMatdatabases for RRR=100 and 0 T and 12 T. Normalization is done accordingto the corresponding data from NIST database.
Temperature [K] N o r m a li z ed s pe c i f i c hea t MATPROCryoComp
Fig. 2. Normalized specific heat of copper from MATPRO and CryoMatdatabases. Normalization is done to the NIST data. is the most visible parameter showing the difference. Further,variation of this at low temperatures can have an influence onpredicting, for example the quench protection system delay.Such sensitivity analysis has been done in [9].Fig. 3 displays how different current decay curves, i.e.MIITs, predict different hot spot temperatures when materialproperties are taken from different libraries. We do not presentactual MIITs values but only compare how the predicted T hs changes. At 0 T, MATPRO is more optimistic than NISTwhich is still slightly more optimistic than CryoComp. Ifthe MIITs correspond to such a value that NIST data gives300 K for T hs , the MATPRO data predicts T hs of only 279 K.The corresponding value for CryoComp data is 304 K. Inparticular, the very low resistivity of MATPRO below 60 Kinfluences its considerably lower value. Interestingly, at 12T the situation is reversed. With the MIITs corresponding to
200 220 240 260 280 300 T hs based on NIST data [K] T h s w i t h c o rr e s pond i ng M II T s [ K ] MATPRO 0 TCryoComp 0 TMATPRO 12 TCryoComp 12 T
Fig. 3. T hs predicted by material property data from MATPRO and CryoMatdatabases as a function of T hs predicted by the data from NIST database at0 T and 12 T, when the operation temperature was 4.2 K. NIST data of 300 K, MATPRO predicts 334 K and CryoComp296 K. Because as a function of field both of the datasetsintersect NIST data, a field value which matches at a giventemperature for NIST and MATPRO or NIST and CryoCompcan be found. However, real situations are seldom like that,usually the field changes during the current decay and isdifferent in different parts of the system.There exists large variation in the literature data for materialproperties of even the most common material, and possibly themost important, from the stability point of view, copper. Thisvariation can have a significant influence when deciding ifdevices designed on their limits are feasible or not. Further, toinclude this data into a simple simulation tool, we needed toimplement relevant functions to our simulation tool manuallyby extracting the data from the databases and compiling to anappropriate format. In the case of a database with simulationtool integrations, it would be easy to change the data sourceand re-run the simulations to get an estimate for the reliabilityof the results.III. P
RINCIPAL IDEAS BEHIND
MASTO
AND ITS
MATLAB
INTEGRATION
MASTO is a new kind of effort for constructing a central-ized database for material property data that is also integratedto simulation tools. The level of integration will depend onthe particular tool at hand.Critical questions in a centralized effort for building amaterial property database are • How to ensure the reliability of data? • How to ensure that no user is blocking another one fromentering similar data? • How to ensure the persistence of data?In a fully open system aimed at distributing information,such as arXiv [10], GitHub or any social media platform likeTwitter and Instagram, in principle anyone can make anythingvisible for all web users. This is contrary to the scientificeer reviewing practice, in which experts check the submittedmaterial beforehand and communicate their findings to editorswho decide if revision is needed, material can be accepted orif it must be rejected. Still, arXiv is popular among scientists,people upload a lot of software to GitHub and social mediais used as the main information source by many people. Toincrease the reliability of data, MASTO features an optional peer review that one submitting a material can ask for. Inthis case, the foreseen MASTO editorial and developmentoffice finds experts to blind review the data. One can usethe advanced search functionality to only search among the reviewed materials. Other options for one browsing MASTOin estimating the reliability of data include a star ranking from 1 to 5 (similar to systems for evaluating e.g. movies) orsearching for materials marked as confirmed . Naturally, all theuploaded data can include citations to literature and descriptionof the characterization mentioning the standards that have beenfollowed.Stainless steel, for example, means different things to dif-ferent people; and even a standard such as AISI 316L couldbe in question. The microstructure and manufacturing historyhas a non-negligible effect on the yield strength. Therefore,a stainless steel expert must be able to identify her steelsin detail while a typical user can import data that is namedjust as stainless steel. This leads to the requirement of non-blocking namespaces for materials. To allow this, materials areorganized under communities . The name of each communityis unique in the system, but different communities can have materials with the same names. A material is an entity inwhich a user has composed her data in a way she wants to.For example, a material can be the yield strength of a stainlesssteel having a certain manufacturing history, or it can be alist of densities of all solid materials at standard temperatureand pressure conditions (STP) [11], [12]. The communities arethe data sources that the editorial and development office cantag as trusted in a similar way as in Twitter one can identifythat the account @realDonaldTrump really belongs to the45 th president of the United States and is not one of themany parody accounts. The administration of communities canbe distributed and therefore many researchers can contributeto the same community. A unique identifier for a materialproperty, data, fit etc, can be experts/aisi-304 , withformat community/material . This community idea isalso in use in GitHub for sharing code. Communities can alsohave private material properties for internal use.The International Digital Object Identifier Foundation [13]has developed a wide spread way to identify publications andtheir locations in the internet allowing publishers to change theactual links in case of content management system upgradesetc. In the MASTO system there is an additional dimensionfor this, as material property data can have various versions .Consider one entering electrical resistivity of copper as afunction of temperature there. Because some code can dependon this, this cannot be removed from the system any more, inthe same way as one cannot undo a published publication.Therefore, when the data owner updates this material toalso include magnetic field dependence and the effect ofRRR, a new version is created from the data. The versioning system follows the schema x.y.z , where z means a smallfix, or a bug fix, e.g. in data or its description; y meansminor improvement, or addition of new simulation softwaredependency, to the data that kees the same interface; and x isa major upgrade. Because the guiding idea is to make MASTOcompatible with various simulation software, one can detail theversion to use. Therefore, persistence means identifying datalike experts/aisi-304/1.0.5 or in a simulation toolto have a dependency experts/aisi-304/1.ˆ , whichmeans the latest with major version 1. This allows one toalways find the data from MASTO, no matter if the internallinking system is changed.MASTO offers a Representational State Transfer (REST)architecture [14] based application programming interfaceto fetch, import and update data via HTTP. Currently,an utility package is built to MATLAB to allow fullMASTO integration to in-house software [15], but thedevelopment of other integrations is underway. Theutility package can be found from MASTO community stenvala [16] and package utils . With this utilitypackage (that an initialization script will install), for exampledensities of basic elements can be fetched with command masto.stenvala.utils.latest.require(’stenvala’, ’element-densities’); . To get thedensity of solid copper at STP conditions, one uses command masto.stenvala.elementDensities.latest(’cu’, ’solid’) .IV. C ONCLUSIONS
Several material property databases or libraries for materialsutilized in cryogenic environments exist. Some of these areopenly available on the internet, some meet the definition ofproperietary software and some are developed in collaborationbetween research institutes. Typically, common to all of theseare that when material property data for a given materialat given conditions are fetched, it may differ, people cannotcontribute to the libraries and any software integration mustbe done manually.We studied three different databases called NIST, MATPROand CryoComp and considered the electrical resistivity andspecific heat of copper. We considered a simple adiabatic Jouleheating case and showed that at 0 T when NIST data predicteda temperature increase from 4.2 K to 300 K, MATPROdata gave only 279 K. With CryoComp data the result was304 K. At 12 T the corresponding numbers were 334 K forMATPRO and 296 K for CryoComp. We presented an ongoingeffort to construct a new online material property database,Open Material Property Library With Native Simulation ToolIntegrations – MASTO, to which anyone can contribute withone’s own material property data, the credibility of data can beassessed in multiple ways, where persistence of data as wellas its versioning is guaranteed, and the material propertiescan be linked as dependencies to external software with noprogramming effort. We aim to make MASTO a new researchinfrastructure connecting different people: modellers, exper-imentalists, material providers etc. around material propertydata.
EFERENCES[1] Cryogenics Technologies Group, Material Measurement Laboratory,National Institute of Standards and Technology USA [Online]. Available:http://cryogenics.nist.gov/, Accessed on: Aug. 18, 2017.[2] G. Manfreda, L. Rossi, and M. Sorbi ”MATPRO upgraded version 2012:a computer library of material property at cryogenic temperature”
INFNTechnical Report
INFN-12-04/MI, 2012.[3]
CryoComp . LHC-Note354, CERN, Geneva.[5] GitHub, A Git repository hosting service. [Online]. Available:https://github.com/, Accessed on: Jan. 22, 2018.[6] Composer, A tool for dependency management in PHP. [Online]. Avail-able: https://getcomposer.org/, Accessed on: Jan. 22, 2018.[7] Bower, A package manager for the web. [Online]. Available:https://bower.io/, Accessed on: Jan. 22, 2018.[8] E. Todesco, ”Quench limits in the next generation magnets”
CERNYellow Report
CERN-2013-006, 10-16, 2013.[9] T. Salmi, D. Arbelaez, S. Caspi et al. , ”A Novel Computer Code for Modeling Quench Protection Heaters in High-Field Nb Sn AcceleratorMagnets”