Summarize this for me. Someone who understand the HEA and MEA perfectly and also the summary should be 2 or over 2 pages. Someone who understands this perfectly

A scrap-tolerant alloying concept based on high entropy alloys M.R. Barnett a, , M. Senadeera b, D. Fabijanic a, K.F. Shamlaye a, J. Joseph a, S.R. Kada a, S. Rana b, S. Gupta b S. Venkatesh b a Deakin University, Institute for Frontier Materials (IFM), Australia b Deakin University, Applied Artificial Intelligence Institute (A212), Australia A R T I C L E I N F O A B S T R A C T Article history: The present article advances the idea of compositional flexibility. The intent is to provide an avenue Received 17 July 2020 for value creation from waste alloys in cases where costs of separating co-mingled alloy flows are pro- Revised 3 September 2020 hibitively high. A compositionally flexible alloy has the potential to be produced solely from varying and Accepted 8 September 2020 Available online 12 September 2020 intermingled flows of end-of-life alloys. While the alloy composition may vary from day-to-day or from heat-to-heat, the intent is that it will still satisfy base performance targets. Using FeNiCr based high entropy alloys as inspiration, we exploited Bayesian Optimization to define a number of example compositionally complex alloys that display strengths within specified limits (150-250 MPa, 200-350 MPa and 300-350 MPa). The concept was motivated by the surprising number of high entropy alloys reported in the literature that display a similarity of properties despite their wide dispersion of composition. We propose a simple measure of compositional flexibility by combining the ranges over which each elemental addition is permitted to vary and suggest methods of alloy specification. A series of 68 prototype compositions were produced and characterized to demonstrate the concept and approach. We show that our model alloys could conceivably be produced using multiple different combinations of existing Ni alloys in combination with 316 stainless steel. Our hope is that the concept will find application to facilitate alloy recovery where it might not otherwise be feasible. (c) 2020 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved. 1. Introduction erant' alloying concept. A compositionally flexible alloy will have the capability to accept input from varying co-mingled flows of End-of-life alloy stocks and flows are frequently intermingled end-of-life alloys. While the alloy composition may vary (within and variable [1-5]. If new niche speciality materials [6-9] and limits), from day-to-day or from heat-to-heat, the intent of the alcomplex multi-alloy products continue to enter the market, the loy is that it should still be able to satisfy base performance tarproblem will exacerbate. Where basic comminution (e.g. shred- gets, thus providing a potential avenue for value creation via reding) followed by thorough physical separation of end-of-life prod- melting in cases where costs of physically separating different alucts is feasible, re-melting is a desirable pathway to alloy recov- loys are prohibitively high. ery and such an approach is ubiquitous in steel manufacture and we propose that for an alloy to be compositionally flexible it in the recycling of aluminium cans. Where the degree of inter- must excel in at least one of two measures. The first is the effecmingling prohibits ready physical separation, the end-of-life stock tive range over which its constituents are permitted to vary. This can be treated as an ore and subjected to the hydro- and/or pyro- can be efficiently captured by taking the geometric mean, c, of metallurgical processes applied to ores [10]. This is an approach each of the individual ranges of element addition specified for the that obviously retains less of the value obtained by incurring the alloy. The benefit of this definition is that a comparable parameoriginal costs in arriving at the alloyed metallic state - be they fi- ter can be established from a list of discovered material variants nancial, energetic, social and/or environmental. A strategy that will (as will be shown below). Another useful measure is the number facilitate increasing use of re-melting in the face of complex used of additions, n, that are allowed to vary to an appreciable level. alloy stocks is to create scrap tolerant alloys [11-13]. The present A related term is the number, n0, of optional additions that are article advances the idea of compositional flexibility as a 'scrap tol- permitted to vary significantly (where, obviously, n00.998 ). 2.1. Machine learning A Gaussian Process was then fitted to the entirety of the sampled points and overlaid onto a new search grid in atomic percentage We used Bayesian Optimization to search a Calculated Phase Di- space (see Appendix for grid specifics). This returned a series of agram (CALPHAD) framework (using 'Thermocalc') to find composi- additional compositions predicted to fall within our performance tions that satisfy a 'performance predictor'. Bayesian Optimization criteria. Re-running the performance predictor for variants selected [27] is a well-known sample-efficient method to find global opti- at random from these predictions (100 for alloy CFA4, 2,000 for mal black box functions (without the need to specify derivatives) CFA17) shows that the prediction accuracy for an fcc fraction of [28]. The method has theoretical guarantees on convergence to the >0.998 was 100% for CFA4 and 59% for CFA17. These values were global optimum values [27]. It thus becomes particularly advanta- employed to establish lower estimates for the total numbers of geous in respect to alternative material searching approaches (e.g. successful alloy variants within our search spaces. [22,29-31]) as the expense of the black-box function rises. Genetic Algorithm based approaches [31] for instance rely on random vari2.2. Performance predictor ations and become time inefficient as the performance predictor becomes more complex. The present approach maintains a proba- The likelihood of obtaining a solid solution fcc phase at elebilistic model of the function that directs the next sample point - vated temperature (with the intent that it be retained, via quenchwhere it assumes the chance of finding a favourable value is high. ing, at room temperature) was determined using the Thermocalc A Gaussian process is typically used as the probabilistic model and TCHEA3 database and seeking the equilibrium phase present at is fully defined by a mean function and a covariance function. The 800C (CFA4) or 1,100C (CFA17). The database is arguably the best covariance function provides a mechanism to define smoothness available for the present systems but its application to CFA17 with over the function i.e. how knowledge of the function value at one is large number of elemental additions, while shown to be acceptlocation influences the estimate of the function values around it. able [32] comes with a level of error. It is also true that the ability Here, the square exponential function was used [14], which admits to quench the alloys to completely lock in the fcc phase also deall smooth functions. We used an Upper Confidence Bound (UCB) pends on phase stability, which was not explicitly addressed here. acquisition function to find the most promising sample to be eval- Our aim is to demonstrate the concept and acknowledge that fuuated next [14]. ture developments and refinements of performance predictors will The Gaussian Process employs hyperparameters - the kernel pa- be needed to apply the approach more generally. It is nevertherameters: signal standard deviation, noise standard deviation, and less found that the 68x-ray spectrum measurements made on althe set of ARD (Automatic Relevance Determination length scales). loys produced in the course of the present work yield a dominant We start with a few random points and estimate the initial hy- fcc structure following quenching in each case. In three instances, perparameters of the GP. These hyperparameters are updated each the presence of a secondary phase was detected but this was not iteration by including the new observations [14]. Since all the hy- seen to create a major discontinuity in the strength prediction, nor perparameters are estimated from the data, it is the best fit that in the ductility. There is evidently a certain robustness of the meis guaranteed given the choice of the form of the kernel and the chanical properties of the present system to small inaccuracies in available data. phase prediction. Due to the high dimensionality of the problem (the large num- The model of Varvenne et al. [33] in its reduced elastic form ber of elements in the system), the Bayesian Optimization was was used to calculate the yield strength contribution due to solid completed in blocks. In each block, 50 compositions were ran- solution strengthening of the predicted fcc alloys. The model caldomly selected from prior samples to build the Gaussian Process culates the interaction energy between a dislocation and a solute model (in the first batch, these 50 compositions were randomly atom, which is then adjusted for thermal contributions. We employ sampled from the space). The parameters of the Gaussian Process, the simplified elastic distortion version of the model [33] based on length scale, signal standard deviation and noise variance, were es- descriptions of the Peierls Stress at 0K(y,0), and the energy bartimated from the function observations. Following this, Bayesian rier, Eb, to be overcome to move a dislocation: Optimization based sampling was applied to sample 100 more compositions with the function model being updated with all function observations available up to that point of time. y.0=0.0178531(1v1+v)34[b6ncnVn2]33 Table 1 m ) of Inconel 718 (AP\&C Advanced Powders \& Coatings Inc.) and stainless steel 316L (Australian Metal Powders Supplies Pty Ltd). These metal powders were fed from two different hoppers and different CFA17 alloy compositions were deposited by adjusting the flow rate of the hoppers. The depositions were homogenized at 1,100C for 5 hours in Ar atmosphere followed by water quenching. Grain sizes measured on 30 samples fell in the range 36-69 microns. Tensile testing of the alloy specimens was performed using a 30 kN load cell (Instron-5567, USA) equipped with a video extensometer. Tensile samples with a gauge length, gauge width and thickness of 16mm,4mm and 2mm respectively, were wire electrical discharge machined from homogenized alloy specimens. Three sets of tensile tests were performed for each composition at an initial strain rate of 103s1. Vickers indentation hardness testing was conducted on the polished surface of the homogenized alloys using a HWDV-7S Vickers Hardness Tester (TTS Unlimited, Inc., Japan) and a 5kg load applied for 15s. At least 20 indents were conEb=1.561831b3(1v1+v)32[b6ncnVn2]31 ducted for each alloy. Prior to tensile and hardness testing, the al- (2) loy surfaces were polished using 1200 grit SiC paper. X-ray diffraction measurements were performed on a Panalyt is a constant given as 0.123,b, is the mean Burger's vector, cn The elastic constants and v refer to the average shear modulus multiple whole profile fitting method. Fig. 1 a shows measured lattice parameters in support of the values given in Table 1 . The sured elastic constants of the predicted to be established. The general efficacy of the Varvenne strength model for 30 of the present compositions is illustrated in Fig. 1b. variability due to the use of an additive manufacturing process the critically resolved shear stress: [35] y(T,)=y,0(T)[1(Eb(T)kTln0)32] (3) 3. Results 3.1. Search results k is Boltzmann's constant, 0 is the reference strain rate (set as 104s1 after [33]), 1 was set to 103s1 and T at 300K. The Two example 'alloys' with flexible compositions were identicalculated yield strength depends substantially on Vn, the volu- fied. One alloy (CFA4) is based on a four element system and the metric misfit of each atom, calculated as the difference between other (CFA17) on a system containing seventeen elements. Both are the volume of the nth atom (Vn) and the atomic volume of the al- based on the FeNiCr system [15-21]. As described above, over their loy (V=ncnVn). The lattice parameters employed in the present entire compositional range, the alloys satisfy a computationally efstudy are listed in Table 1. The value for Ni was adjusted using a ficient and physically based 'performance predictor' that comprises least squares approach based on the data in ref. [33]. To approxi- two well validated modules (i) a Calculated Phase Diagram query mate the effect of carbon we added an increment of 75MPa per (using the Thermocalc TCHEA3 database) to ensure the existence atomic percent of addition though we acknowledge that there is of an equilibrium fcc phase at an elevated temperature from which an interaction with grain size that would need to be taken into the alloy could conceivably be quenched to lock in a stable or account in a fuller model [34]. meta-stable fcc structure, and (ii) a physically based solid solution strength prediction. 2.3. Experimental alloys and testing We aimed for a single fcc phase to ensure dominance of solution strengthening, which (amongst strengthening mechanisms) Prototype samples were produced by additive manufacturing is well described physically and is generally 'smooth' and slowly using an OPTOMEC LENS MR-7 system (USA) equipped with 1kW changing over composition space [36]. Although precipitation IPG Fiber Laser. This technique was chosen because it permitted a strengthening often provides an attractive pathway, such an aplarge number of samples to be made in a relatively short time - as proach is inherently sensitive to minor alloying additions, small a means to obtain uniform equiaxed single phase microstructures changes in composition and variations in production [37]. For our with minimal impact of processing history. Rectangular columns of example alloys, we sought a base level of structural performance 12mm width 10mm thickness 65mm length were deposited under static loading. The alloys are thus constrained to display on mild steel substrates using a laser power of 800W in Ar atmo- Young's moduli within +/- 15\% of the mean value. Strengths are sphere (layer thickness: 0.75mm, scan speed: 890mm/min, pow- constrained to the bands of 150250 MPa for alloy CFA4 and 200 der flow rate: 15g/min and cross-hatch scanning strategy). The 350 MPa for alloy CFA17. These bands are not excessively broad; CFA4 compositions were produced using spherical gas atomised it is not uncommon to find a variation in up to 100 MPa in the powders (99.9\% pure, 45-150 %m ) of Ni,Co, Ti and Fe (supplied yield strength of commercial structural alloys [38-41]. Although by TLS Technik GmbH \& Co, Germany). Powders were fed from the strength minimum is the key structural design parameter, setfour different hoppers and their flow rate was adjusted to produce ting an upper limit to strength helps avoid brittleness (or poor different CFA4 alloy compositions. CFA17 compositions were fabri- ductility), which becomes more likely at higher strengths [42]. The cated using spherical gas atomised powders ( 99.9 \% pure, 45-150 extent of compositional flexibility is linked to the extent of allowed Fig. 1. Predictions against experimental measurements. (a) Lattice parameter predictions made using Vegard's law based on the effective elemental parameters given in Table 1. (b) Yield stress predictions plotted against values measured in tension for 30 variants of the present alloys ( R squared =0.71 ). Fig. 2. 'Compositional flexibility' of the present alloys in terms of the allowable composition variation and (a) the number of significantly varying elemental additions and (b) the number of significant optional additions. The present alloys can be seen to display broad composition ranges and a high numbers of additions that are permitted to vary. performance variability so we further constrained alloy CFA17 to create CFA17' by imposing a higher and tighter strength band of 300-350 MPa and insisting upon at least 10at%Cr (which would dent and so a geometric mean can be calculated from them diimpart corrosion resistance). For alloy CFA4, a total of 61,000 vari- rectly. ants were counted within the search space and for alloy CFA17, The example alloys are seen in Fig. 2 to exceed the composithe count was 795,000; with 295,000 variants for constrained alloy tional 'flexibility' of common alloy classes, in terms of the extent CFA17:. Of allowable variation in the level of constituents (alloy CFA4) and A 'mean composition range' was established from the discov- in the numbers of constituents that are permitted to vary (alloy ered variants as follows. It is assumed that the alloy occupies a CFA17). As one might expect, such high levels of compositional negligibly small region of the total available composition space. flexibility makes an alloy challenging to specify. In Table 2a-b we The hyper-cube volume, v, for each grid point is established from present the compositional ranges spanned by the discovered varithe individual intervals employed to lay out the grid for all ad- ants but we hasten to add that this does not fully describe our ditions that vary over a total range >0.5 at\% (except for the addi- alloys. Rather it delineates a larger space within which we know tion that was assigned the role of the base or 'floating' element): many acceptable variants reside - a kind of upper limit for the ali.e. v=c1c2cn1 (where n=4 for CFA4 and 15 for loy specification. For alloy CFA4 there are no essential additions CFA17). This term was then multiplied by the number of discov- defined a priori. For CFA17, the only addition that turned out to be ered variants and raised to the power of the inverse of the num- essential is Ni (and CFA17' contains the extra specification that Cr ber of additions, to arrive at an effective geometric mean. As noted >10 at\%). In all, Alloy CFA17 occupies only approximately 6% of the above, a comparable term can be derived from alloy specifications compositions that satisfy the ranges given in Table 2. for commercial alloys. These specifications provide the upper and The present alloys can also be specified in terms of exemlower limits for each addition. The ranges are typically indepen- plar compositions, which sets a kind of lower limit to the alloy Table 2 Compositional ranges for our example compositionally flexible alloys, CFA4 and CFA17. In a) and b) the compositions are given in terms of alloying additions. In c) the composition is given in terms of binary combinations of a Ni alloy and SS316. In c) the ranges refer to the extent of permitted variation of Ni alloy additions in each binary combination. specification. Alloy exemplars (variants) discovered during the cessive strength or to secondary phases, leading to failure of our Bayesian Optimization search are illustrated in Fig. 3. In general, performance requirement of a single fcc phase at an elevated temit can be seen that the alloys span large swathes of composition perature. The main point is that we illustrate how a compositionspace. Fig. 3a maps the compositions back into the ternary Fe- ally flexible alloy (CFA17) could conceivably be produced from a CrNi space. The variation of composition for Alloy CFA17 reveals wide cross section of existing alloys making them tolerant to mixthe coarseness of the underlying grid which was imposed on the ing and variability in the input stream. search. This alloy occupies less of the ternary space than alloy The complex interactions present between the elements in the CFA4 due to the effect of the multiple other elemental additions present alloys - due to the different phases that can form and due in CFA17. to the different contributions of different elements to the strength Parallel composition plots showing alloy variants are given in _ - are not easily delineated from any of the ways we can conceive of Figs. 3b and c. In both cases variants with high iron contents are tabulating or pictorially representing the alloys. This will be a genhighlighted in red, to provide an example of how the composition eral feature of Compositionally Flexible Alloys. In specifying (and ranges allowed for each addition depend on the levels of other ad- producing) such alloys it will typically be necessary to make reditions. Obviously, when the level of one addition is high, the other peated reference back to a 'ground truth' supplied by the algorithadditions will be necessary low. For CFA4, the high iron compo- mic performance predictors employed to discover them. That is, sitions can tolerate high variations in Ni level whereas the addi- full definition of compositionally flexible alloys will require an altions of both Cr and Co are limited to the lower halves of their gorithm. We will return to this point in the discussion. total ranges. A number of other interactions can be detected in the plots, for example, hafnium additions of 0.5wt% cannot be toler4. Experimental results ated in CFA17 when the iron levels are high. While the extremes of the composition variations are readily discernible, it is not easy To begin to put our example alloys to the test, we manufactured to discern which combinations of additions actually fall outside of sixty-eight lab-scale variants using a direct metal deposition addithe 'alloy' performance specification. tive manufacturing technique. The compositions produced span a A practical solution to the challenge of specifying composition- large portion of the breadth of our discovered alloys after mapping ally flexible alloys is to provide variant exemplars based on the back to the FeNiCr ternary space (Fig. 4a). After manufacture, all end-of-life alloys from which the alloy could conceivably be made compositions were heat treated at 1,100 C and quenched into wa- rather than their elemental compositions. As an example, we ran ter. Strengths are presented in Fig. 4b. Critically, the strength (in 37 common Ni alloys through our performance predictor for dif- terms of proof stress - measured directly and inferred from hardferent additions of stainless steel grade 316. We see that the spec- ness) satisfy one or both of the strength ranges set for the two ifications of alloy CFA17 can be met using binary combinations of alloys, thus satisfying the core design criterion. Full property data alloy SS316 with any one of 32 of the Ni alloys we investigated. are given in the Appendix, where the variability is reported and The actual proportions of Ni alloy addition in these binary mixes where it can be seen that all samples bar three displayed a sinare given in Table 2c. To interpret the trends evident in this table, gle fcc phase in their x-ray diffraction profiles, as desired. Our use we note that the addition of a Ni alloy to SS316 raises the strength. of additive manufacturing introduced a variability to the measured For certain Ni alloys, higher levels of their addition give rise to ex- ductilities, which ranged from 8 to 50%, and this can be put down Fig. 3. The present compositionally flexible alloys expressed in composition space. Compositions discovered in the first stage of the search in weight percent: (a) CFA4 and CFA17 mapped back into the Fe-Ni-Cr ternary space, (b) parallel plot of alloy CFA4 showing high Fe variants in red (c) parallel plot of alloy CFA17 showing high Fe variants in red. The alloys discovered clearly straddle vast swathes of composition space. b Fig. 4. Plots showing the results of 68 alloy compositions produced using direct metal deposition additive manufacturing and subjected to heat treatment followed by quenching. (a), Experimental exemplar compositions mapped back into ternary space (circles CFA17, triangles, CFA4). (b) Distribution of experimentally determined yield strengths ( 30 tests measured directly, 38 converted from hardness readings)-all experimental alloys fall within our two performance target bands. See Data in the Appendix for uncertainties. Fig. 5. Plots showing the results of 68 alloy compositions produced using direct metal deposition additive manufacturing and subjected to heat treatment followed by quenching. (a) Impact of Ti addition on the strength of ternary compositions showing a similar impact for different base alloys and demonstrating how Ti "trimming" can be employed to tune strength. (b) Strength of quenched blends of Ni718 and SS316 showing how these fall within the new alloys and how Ni alloy additions act as a strengthening addition, which also allows for 'trimming'. See Data in the Appendix for uncertainties. to the introduction of an expected small but variable amount of that differ from the mean permit trimming of strength. One means porosity arising from the fact that we made large compositional of achieving this in the present examples is the addition of Ti variations on-the-fly without the capacity for repeated fine tuning (quite conceivably as low cost scrap or as Ferrotitanium). This is ilof the deposition conditions. We note, however, that measured val- lustrated in Fig. 5a, which shows a uniform strengthening effect of ues for the work hardening index fell at within the tighter range Ti on three ternary alloys that fall within the compositional ranges of 0.29 to 0.47 . As pointed out in relation to Fig. 1, the strengths of alloy CFA4. (These tests also allowed us to tune the effective latobtained in the present tensile tests are well explained by the cur- tice parameter of Ti in an fcc lattice - using a FeNiCo - see Fig. 1a.) rent strength model. The other useful strength relationship we identified is seen in the addition of higher levels of Ni base alloy to SS316 in the case of 5. Discussion alloy CFA17 (see Fig. 5b). In any given production of a compositionally flexible alloy, we Although we have claimed and illustrated a number of alloys envisage a production process mediated by computational perforthat display unprecedented flexibility of composition, we hasten mance prediction. One would have on hand two or more flows of to add that these are examples of the concept. Lower values of distinctly different alloy mixes. The flows may vary from heat-tocompositional flexibility than shown here may well still be ben- heat but the initial step would be to determine their actual comeficial outside of the high entropy alloy paradigm. And more de- position. A computer aided search would then be run in conjuncmanding performance criteria of course may be applied. Even in tion with a 'performance predictor' to identify what fractions of the regime of solid solution hardening there is scope for property each input stream would provide the requisite performance. Our enhancement (e.g. higher strengths). The most desirable alloy will approach clearly suits applications that exhibit a relatively small depend on the specific (local) availability of end-of-life alloys and number of performance criteria, which are amenable to computamarket demand. Large volume commodity metals such as iron and tionally efficient and accurate physically based mathematical dealuminium already display relatively high recycling rates (though scription. Fast, accurate performance prediction would be essential, there is still much room to improve) and the present approach will not only for production, but also for the discovery and description only be suitable for mixed, higher value and lower tonnage alloy of new compositionally flexible alloys. The dominating structureflows. A case-in-point is the variability in mixed metallic 'alloys' property relationships would also need to be engineered so as to produced from the different Li-ion battery recycling processes [10]. avoid high sensitivities to compositional variation. Energy and cost may be saved by diverting the intermingled and We believe the present conception of a compositionally flexible varying metallic products from these processes to a composition- alloy might prove a useful addition to our current choice of pathally flexible alloy in lieu of a series of costly separation and purifi- ways towards increased material circularity [2,43]. The two-stage cation processes. But the nature of the particular recycling process Machine Learning method adopted in the present work is straightand market demand will dictate the best solution. forward and readily applicable to the discovery of new compoProduction of conventional alloys typically involves a 'trimming' sitionally flexible alloys. The adoption of Bayesian Optimization step where the composition of the melt is adjusted to ensure de- means that optimality of solution for a given number of iterations sired specifications are met. Despite the ability of a composition- is guaranteed as the performance predictor becomes more compually flexible alloy to accept a varying stream of input, a similar tationally expensive. And more complex phase prediction will be need for trimming is still likely to arise in a commercial situa- needed than employed here because of kinetic effects, which can tion. But here we run into a paradox because the whole premise dominate in alloy manufacture. Searching for new alloys and maof the present alloys is that they would display a low sensitivity terials in multi-dimensional composition space is a challenge curto alloying additions. However, the reality is that the robustness rently exercising the field [15-19,21,22,29-31] and our hope is that of strength against changes in composition arises chiefly from the compositional flexibility might find a place in the design targets similarity of effective lattice spacings in the fcc lattice (Table 1). for such efforts. In doing so, it may prove advantageous to employ Addition of alloy mixes rich in elements with effective atomic sizes existing available alloys, as opposed to elemental additions, as the basis for the original search space. And detailed opportunity cost for assistance with additive manufacturing, as are Mr Alex Orokity analyses would be required. Evidently, if the present strategy is to and Mr Robert Lovett for assistance with rapid machining of tensile be pursued, quite a number of current paradigms will need to be samples. The authors are grateful to the Deakin Advanced Characchallenged and we acknowledge this will not be trivial. terisation Facility. The reviewers are also thanked for their contributions to improving the study. 6. Conclusion The present article advances the idea of compositional flexibility as an alloying concept that provides a potential avenue for value creation from waste alloys. It may prove of use in cases Appendix where separating co-mingled alloy flows is prohibitively costly or otherwise impractical. We exploited a well-known Machine Learn- In the present study, the search grids were defined to set the ing technique - Bayesian Optimization - and this enabled us to bounds and steps of the searches. Initially, the grids were laid out define classes of compositions that display plastic and elastic re- as follows. For alloy CFA4 (in weight percent with intervals of 1\%): sponses that fall within specified limits. We report example alloys Fe: 0-60\%, Ni:0-60\%, Cr:0-60\%, Co:0-60\% (total 1.31 105 indepenthat display flexibilities of composition significantly greater than dent variants). For alloy CFA17 (in weight percent with varying inthose seen in existing alloys. A series of 68 prototype composi- tervals): Fe: floating, Ni: 10-80\% (10\% intervals), Cr: 0-40\% (10\% intions were characterized to support the concept and approach. Our tervals), Co: 030% (10\% intervals), Ti: 0-4.5\% (1.5\% intervals), Al: approach is suitable for applications that exhibit few performance 012% (4\% intervals), Ta: 01.5% ( 0.5% intervals), W: 03% ( 1% intercriteria and which are amenable to computationally efficient and vals), Mo: 04.5% (1.5\% intervals), Mn: 09% (3\% intervals), Si: 02% accurate physically based mathematical description. Fast, accurate (1\% intervals), Nb: 01% ( 0.5% intervals), Cu: 02% (1\% intervals), C : phase and performance prediction is needed for commercial ex- 01%(0.5% intervals), Hf: 00.5%(0.5% intervals), V: 01% ( 1% inploitation of the idea. Our hope is that compositional flexibility tervals), Re:0-1\% (1\% intervals), (total 2.06108 independent varimight find a place in facilitating end-of-life alloy recovery. A subsequent search, based on a single Gaussian Process fit was Declaration of Competing Interest then performed on the following grids. For alloy CFA4 the grid comprised a total of 1.31105 independent variants (in atomic The authors declare that they have no known competing finan- percent with intervals of 1% ): Fe: 0-60\%, Ni:0-60\%, Cr:0-60\%, Co:0cial interests or personal relationships that could have appeared to 60%. For alloy CFA17 the new grid comprised a total of 2.08107 influence the work reported in this paper. independent variants (in atomic percent with varying intervals): Fe: floating, Ni: 10-80\% (10\% intervals), Cr: 0-40\% (10\% intervals), Acknowledgements Co: 0-30\% (10\% intervals), Ti: 0-4.5\% (1.5\% intervals), Al: 012% (4\% intervals), Ta: 00.5% ( 0.5% intervals), W: 01% ( 1% intervals), Mo: This research was partially funded by the Australian Research 03% (1.5\% intervals), Mn: 06% ( 3% intervals), Si: 02% (1\% interCouncil (ARC). Prof Venkatesh is the recipient of an ARC Aus- vals), Nb: 00.5% ( 0.5% intervals), Cu: 02% (1\% intervals), C: 01% tralian Laureate Fellowship (FL170100006). The financial support (0.5% intervals), Hf: 00.15%(0.15% intervals), V: 01% (1\% intervals, of the Australian government via the Department of Industry, In- only added when Re not present), Re:0-0.3\% (0.3\% intervals, only novation and Science Australia-India Strategic Research Fund on added when V not present). project AISRF53731 is also acknowledged, along with Accelerating The following tables provide full account of the mechanical Commercialisation grant No. AC74806. So are discussions with Pro- property measurements made in the present study (Table A1, fessor Gorsse and Dr Luhua Li. Dr Mohammad Imran is thanked Table A2, Table A3). Table A1 Merhanieal nennerty data for Ti enntainine ternary momnncitione Table A2 Mechanical property data for SS316L-IN718 alloy blends (yield strength estimated from hardness according to YS=2HV-85 MPa). The strain hardening index was averaged from a true strain range of 0.050.15. Table A3 Characterization of four-component CFA4 experimental exemplar compositions (yield strength estimated from hardness according to YS=2HV85MPa<>