=== PAGE 1 === A review of basic snow mechanics Malcolm Mellor Abstract. In an introductory section the subject is defined, research goals are identified, a historical perspective of snow mechanics is outlined, and requirements for theoretical analysis, in the form of constitutive equations and failure criteria, are specified. A section on deformation opens with a review of idealized rheology for nondestructive loading, giving general differential equations and pertinent special solutions for one-dimensional models. Simple elastic properties, as derived from quasistatic and dynamic tests on snow, are summarized, and effects of rate, frequency, density, temperature, etc. are considered. Viscoelastic properties are described in terms of complex moduli, viscosity for creep transients, loss factor, etc., and independent variables such as temperature, density and frequency are taken into consideration. Viscous behaviour in sustained creep is introduced, and complications of characterization are emphasized during presentation of data obtained by measurements involving uniaxial stress and uniaxial strain. Simple stress analysis is introduced to facilitate further discussion; points covered include separation of stress and strain tensors into deviatoric and dilatational components, definition of stress and strain invariants, shear and bulk viscosities, viscous analogue of Poisson's ratio, and determination of stress components from equilibrium equations. Volumetric strain and irreversible compressibility are discussed, and 'equation of state' characterization is proposed. Pressure/density data for uniaxial stress states are compiled, and a plot of bulk stress against specific volume is developed for various rates and temperature conditions. Deviatoric stress/strain-rate relations for conditions of negligible volumetric strain rate are considered, drawing on data derived from unconventional sources, including uniaxial compression strength tests and creep profiles for settled snow slopes. The section closes with a discussion of the problem of formulating general constitutive equations for multiaxial stress states. A section on failure starts with discussion of failure, strength, and failure criteria, especially as the terms apply to snow. Strength measurements in uniaxial tension and compression are considered, and available data are compiled. Definition of shear strength is considered, the merits and limitations of c—0 characterization of snow are discussed, and some of the more reliable shear strength data are compiled. Physical and statistical failure theories based on structural defect concepts are mentioned, empirical relationships between the strength of snow and its porosity are described, and the possibility of deriving a relation between 0 and porosity is introduced. Attempts to develop phenomenological failure criteria for snow are mentioned briefly. A cautionary note on discontinuous collapse of snow under compression is included, and correlation of arbitrary strength indices with basic strength properties is discussed. Correlation of strength and elastic modulus is explored as a possible method for estimating strength from nondestructive in situ tests. A section on boundary friction introduces the physical theory relating to kinetic friction of coherent snow surfaces, and summarizes experimental findings that give variations of kinetic friction with contact pressure, dimensions of contact areas, sliding speed, ambient temperature, and slider materials. Static friction and adhesion are distinguished and discussed, and methods for measuring or estimating true static friction are suggested. Boundary friction of fluidized snow is considered, and drag induced by inertial effects in a well-diffused turbulent boundary layer is treated in an exploratory calculation. The review concludes with an appraisal of the major difficulties in snow mechanics and with some suggestions for research direction. INTRODUCTION Scope and trends Narrowly defined, snow mechanics deals with forces and displacements in all forms of snow, i.e. with the kinematics, dynamics and energetics of snow in both condensed and dispersed states. More broadly defined, it also embraces the underlying physics of processes relevant to mechanical behaviour, and the useful but disconnected empiricism === PAGE 2 === 252 Malcolm Mellor associated with snow engineering, avalanche prediction, etc. Since snow mechanics evolved as an identifiable field of study about 40 years ago, a good deal of attention has been given to the latter topics, but work on basic mechanics has tended to be more sporadic. Systematic application of formal mechanics to problems concerning stress/strain/time relations in complicated physical situations has been rather slow to develop, which is hardly surprising, since snow has very complex properties, exhibiting nonlinear viscoelasticity and undergoing large strains, both volumetric and deviatoric, under typical loading conditions. Because snow normally exists at homologous temperatures above 0.9, its properties are highly temperature-sensitive, and even its basic structure is thermally unstable. As a consequence of these things, it has been impractical to formulate general constitutive equations and failure criteria that are both realistic and tractable. Some current trends suggest that this situation might soon change for, in principle, the powerful formalism of modern rational mechanics is capable of dealing with the complicated behaviour of snow. However, it seems unrealistic to expect early solution of significant boundary value problems if very complex constitutive equations are employed. While attempts at highly rigorous theoretical analysis are obviously worthwhile, it is necessary to keep sight of the fact that snow mechanics is essentially a practical science, in which useful results are of overriding importance. Thus, the main thrust of research ought to be concerned with progressive upgrading of methods needed for the accomplishment of practical tasks. Such an approach demands some compromise between the almost impossible demands of rigorous theoretical analysis and the confusions of wholly empirical procedures. Perhaps the most troublesome single characteristic of snow is its high compressibility, which makes it behave very differently from most solids and granular materials. As long as the bulk stress in a multiaxial stress field is below a certain critical value, the defor­ mation and rupture of snow can be treated in much the same way as for low compressibility solids, but when this critical value is exceeded the snow undergoes large and irreversible volumetric strain, acquiring vastly different mechanical properties. Simple treatment of compressibility could well be one of the major goals for theoretical research. In this short review a complete coverage of snow mechanics is not feasible, and so emphasis is given to the basic mechanics of dry coherent snow (the major omissions are wet snow and boundary value problems). The chief aim is to summarize the mechanical properties of snow within the context of established continuum mechanics, and to explore the formulation of manageable constitutive equations and failure criteria. Historical perspective Systematic study of snow mechanics began in the 1930s, and in a remarkably short time most of the main features of the stress/strain/time relations for coherent snow were broadly defined by Swiss investigators. The constitutive relations and failure criteria considered for solution of boundary value problems were necessarily simple ones that did not fully reflect the known behaviour of snow, but basic body-force problems in the avalanche field tended to be of a type in which internal stresses are determined largely by the boundary conditions and the self-weight of the material, almost irrespective of rhéologie properties. Simple failure criteria of the Coulomb type were adequate for treating planar slip. More detailed knowledge of snow properties was used to refine results by accounting for creep effects and variations due to temperature, density, snow type, etc. Interest in snow mechanics widened considerably about 20 years ago, stimulated largely by international research in Antarctica and by military activity in Greenland. === PAGE 3 === A review of basic snow mechanics 253 The main emphasis shifted from stability and rupture of snow slopes to the long-term creep of snow under relatively steady sustained loadings, as in densification of ice cap snow, foundation settlement, tunnel closure, and structural loadings. In the latter type of problem initial transients of deformation could often be ignored, and simple viscous analyses became useful, although it was always accepted that assumptions of newtonian viscosity were gross oversimplifications. In some cases, complexities introduced by the rheology of snow were sidestepped by replacing continuum mechanics with special analogue models, e.g. in foundation studies. Another class of problems involved the bearing capacity and shear resistance of snow surfaces, especially as they affect the sinkage and traction of vehicles. In some of these problems simple elastic analyses sufficed (e.g. Boussinesq analysis for a high-speed wheel on dense snow) and ample data on elastic moduli were obtained from field seismic studies and laboratory tests. However, in the 'trafficability' studies there was little attempt to define limit conditions by formal failure criteria; the emphasis was on arbitrary strength indices that could be measured easily in the field. In the past decade the general scope of snow mechanics was further expanded by formal hydrodynamic consideration of dispersed snow as a two-phase flow (as in blowing snow, dust avalanches, snowplough discharges, etc.), and by study of intense compression and shock propagation (as in 'equation-of-state' problems involving explosions, impacts, and projectile penetration). One important hydrodynamic aspect that has not yet received much attention is the problem of snow as a fluidized solid (e.g. snow under slipping vehicle tracks, on a snowplough blade, or in a dense avalanche). The current situation, at least as it appears from the United States, is that interest in the mechanics of deposited snow has waned somewhat with the decline in operations on polar ice sheets, and most work is again motivated by avalanche protection requirements. In both the US and Europe there are signs of rising interest in applying modern continuum mechanics, which permits maintenance of a high degree of analytical generality for complicated constitutive equations, but at the same time practical problems are still being tackled largely by empirical methods. In some respects it seems that present-day activities are concentrated at opposite ends of the research spectrum, with a tendency for neglect of the middle ground. Field investigators and experimentalists are sometimes forced to operate without much framework of theoretical guidance, while theoretical mechanicians may be frustrated by lack of suitable comprehensive data on material properties. There seems to be a definite need for orderly assessment of snow properties, for realistic but simplified mechanical characterization of snow, and for sound and workable theory that can be applied to boundary value problems. Requirements for theoretical analysis In its simplest form, coherent deposited snow can be regarded as a quasihomogeneous sintered compact of equant ice grains. As such, it possesses most of the mechanical properties of solid ice, i.e. it displays temperature-sensitive nonlinear viscoelasticity under typical loading conditions, and rupture may occur by viscoplastic yielding, by brittle fracture, or by a transitional process. However, snow is highly porous and it has high, and largely irreversible, compressibility. It is this high irreversible compressi­ bility of snow that distinguishes it from ice and most other engineering materials. The basic requirements for formal mechanical analysis are constitutive equations that relate stress, strain and time for multiaxial stress states, and failure criteria that give the critical relationships of principal stresses at yield or rupture. Ideally, these should be general equations that permit the tracing of stress or strain history from an initial reference state, and they should be capable of accounting for temperature variations. === PAGE 4 === 254 Malcolm Mellor In the case of solid ice, there is now sufficient information on which to base constitutive equations for elastic deformation and constant-rate flow, but it is not yet possible to derive failure criteria that are applicable outside the range of viscoplastic yielding (ice is apparently anomalous with respect to 'Griffith-type' materials). However, the rheological properties of ice can already be expressed in forms that are becoming too complicated for useful application in practical continuum mechanics. Looking at snow, which is a much more complicated material than solid ice, it appears that presently available data are inadequate for defining general constitutive equations or failure criteria. Once this deficiency is remedied it will be possible to define the required relationships, but in general form these equations will almost certainly be too complicated for useful application in boundary value problems. While general description of rhéologie properties and failure conditions is obviously a desirable research goal, short-term needs in applied mechanics will probably have to be met by piecemeal simplifications appropriate to particular problems. In some cases these simplifications are straightforward (e.g. purely elastic problems), but in others there are still difficulties in arriving at valid representations (e.g. in problems involving simultaneous volumetric and deviatoric strains). DEFORMATION Idealized rheology for nondestructive loading Standard rheological models (Fig. 1) are widely used for describing the mechanical behaviour of snow. They are convenient for qualitative descriptions, and they provide a standard terminology, but because the elasticity and viscosity of snow are strongly nonlinear they have very limited quantitative value. Nevertheless, since some of the reported viscoelastic properties of snow are derived by assuming certain models and analysing test data accordingly, it is necessary to understand these models. Burgers model The deformation of snow and ice under nondestructive loading has long been described in terms of the four-element Burgers model. The general equation giving the rhéologie response of this model is: (Eu EM EK\ . EMEK .. EMEK . a+ — + — + — 0+ o = EMe + e (1) \*?M VK VK' VÎAVK VK where a and e are stress and strain (' and " denoting first and second time derivatives), EM and r/M are modulus and viscosity for the Maxwell unit, and EK and VK are modulus and viscosity for the Kelvin—Voigt unit. Solutions of special interest are those for abrupt application or removal of constant stress, and for application of constant strain rate or constant stress rate. For constant stress a0 applied abruptly at t = Q, the required solution is: (2) Equation (2) gives a strain—time relation, or creep curve, for the standard creep test. The curve shows instantaneous elastic strain, followed by decelerating (primary) creep, which in turn is succeeded by creep at constant rate (secondary creep). It does not simulate tertiary creep. The various terms of equation (2) are easily identified with the elements of the model; in the exponential term (VK/EK) is the relaxation time, i.e. the time required for strain to decay to 1/e of its initial value when load is removed. 1 t 1 — + — + — Eu VM EK r, / 1 — exp1 - EK \1 - — t\ VK n === PAGE 5 === A review of basic snow mechanics 255 gers But «_, lemen w l CO Û •Voi vin- I rt £ •3 ^ o S 1 i> «*< F 1 »? 3 1 ."[ii [ < s , 1 <^ fc^ / 1 j =1 •? 1 -> CO S 1 1 L «5 p | 3 =• 1 j_ H—i l-j-l =" « £0 «-Ï* :»l* ^-^ «s|^ + ufl^ + :0 ^ 1 « 1£jJ + t*r b «" +__ « c? •0 'S- £ c C -. ibï II c .2 I ai 1 c QJ O ___ £T# -—L~-- s 1 „l * + -1,3 + -ks -—_!, © to 1! ^T^, U 1 s? t~ •1 ~* -luT -k t? jp D. ? • -"• °!M c!S- f c 1 = » " II 0) II c ,-, O .-? o £ f j j* »? t! •Ç "t "-& ^ V w j- •-- - H? «SL* ik d,™ <5p «si^ + 1 i :« * fï J^ I ia t? "¥" Ï "T^ «" to —£_ _ UJ 4 II 0 îp ! ' ~ ' A: II 0 O K-îa <- ~ a « 11 W Q. n C M O [l O ^ Jp f -- . g\g d?|^ i -K T2~ -|w tk *" •u ^ ^~j—-- i i 'ïïïr i M -lu j — Ci^ .*" 'Jjv a. s -=- -IW ^?' ^ °i- l<5" dS- -*" S a " lo­ ir t- cO S ^ II w « If" c: M o f! O 3 1 UJ CO 3 3 % :^ ] 1 % 3 3 i ^" W :-3 1 uz — jtoa ,,„, LO '3 -(•7! 'CO — Juj CO > 3 V^ -,|CO -> S .^ u ST c °g €* 3 ^ •JZ -x ~\» \ 3 V1 '§' "'-. i" ~„ ? ~\è' Î Ir 3 u, -II ~=~ i 3 : r« j • it z 3 2 — î ^ ai O === PAGE 6 === 256 Malcolm Mellor For constant strain rate é = ku applied at t = 0, CT = 0, the required solution is: a = k, E* (e,'t-«'.() + VM r2 (r2er'f-/-le,''f) + T?M (3) where 1 /£M 2 \T?M £~M .ffK VK VK 2 E*A ^M ^K Î?M 1?K T?K' /jgM EK\ In effect, equation (3) gives the stress/strain curve for the standard uniaxial loading at constant strain rate, since t can be replaced throughout by ejkx. For constant stress rate à = k2, initiated at f = 0, e = 0, the required solution is: e = k-, 1 Eu + 1 > EK, \t I + 1 2T?M VK VK VM exp VK (4) The Burgers model is very convenient for qualitative description of deviatoric straining in the absence of bulk stress (pure shear), but its quantitative use is severely restricted by the inherent nonlinearity of ice viscosity and by the effective nonlinearity of both elasticity and viscosity which is caused by volumetric straining. Three-element model In elementary rheology a three-element model consisting of a spring in series with a Kelvin—Voigt unit is usually considered. The relevant equations are given in Fig. 1 ; they represent a degeneration of the equations for the Burgers model with r\m = °°, although in the solution for constant strain rate the correspondence is not immediately obvious. An alternative three-element model can be obtained by retaining the Maxwell dashpot and eliminating the Maxwell spring. Two-element model The simple Maxwell and Kelvin—Voigt models give representations that are qualitatively incomplete for snow under typical loading conditions, but they may sometimes provide useful simplifications for analytical purposes. It is usual to assume a two-element model for analysis of vibrational tests designed to measure dynamic modulus, dynamic viscosity, loss tangent, relaxation time, etc. The Kelvin—Voigt is the usual choice. Sinusoidal oscillation applications for models With sinusoidal oscillation of stress or strain there is viscous energy dissipation and consequently a phase lag between stress and strain. For alternation at angular frequency co with stress and strain amplitudes of a0 and e0, stress/strain relationships are expressed as a = Y(ico)e0eiœr (5) e = J(iu>) a0e/ajr (6) where Y(ioj) and J(ioo), the complex modulus and complex compliance respectively, are defined as: Y(icj) JQOJ) '- -- Y, + iY2 Jl + iJ2 (7) (8) y, and Ju the real parts, are the storage modulus and storage compliance respectively; these express elastic properties and give a measure of stored strain energy. The imaginary parts Y2 and J2 are the loss modulus and loss compliance respectively; they === PAGE 7 === A review of basic snow mechanics 257 3 3(K- 2 3K- X) -2G are related to energy dissipation by viscous damping, which is given by the loss factor, or loss tangent, tan 5, where 5 is the lag angle between stress and strain: Y2 J2 tan5 = — = — (9) Expressions for Y\, Y2,J\ and/2 according to the various rheological models are given in Fig. 1. When elastic modulus (Young's modulus) and loss tangent have been measured, dynamic viscosity according to a certain model can be found from equation (9). The properties discussed above are all frequency-dependent. Simple elastic properties of coherent snow Unless otherwise stated, the elastic modulus of snow is identified with the modulus of the Maxwell unit in the Burgers model. Data are commonly presented in the form of Young's modulus E, from which other elastic constants (bulk modulus K, shear modulus G, Lamé's constant X) can be derived by the standard identities if Poisson's ratio v is known: E 2G{\+v) \(1+I0 3X + 2G K = = — = = (10) 3(1 -2f) 3(1 -2v) 3v E 3K(\ - 2v) X(l - 2v) _ ,_. .., G = = _^ i = -1 i = _^ '. (11) 2(1 + v) 2(1 + v) 2v vE 2Gv 3Kv X = = = = (12) (\+v){\-2v) {\-2v) (l+i>) 3 The most consistent data for dense snow are derived from field measurements of seismic velocity and laboratory measurements of pulse velocity or high frequency (102—103Hz) flexural vibrations. Figure 2 gives a general impression of E as a function of snow density for cold snow, together with a summary of corresponding values of Poisson's ratio, as determined by dynamic methods.* Traditional quasistatic uniaxial loading tests are inherently deceptive when applied to snow, since they produce stress/strain curves that represent the response of all elements of the Burgers model. If the test is run at high strain rates, the initial tangent modulus from a uniaxial stress/strain curve should give a good approximation to the 'Maxwell spring' modulus, provided that good experimental technique is employed. However, with low compliance specimens (dense and cold), contact imperfections at the loading platens will depress the apparent value of initial tangent modulus if displacement is measured across the platens rather than across the specimen itself. Figure 2 gives values of initial tangent modulus derived from careful tests on cold snow. The results for high density snow are lower than the dynamic values by a factor of 2; since the tests were run at high strain rates on 'stiff snow, the discrepancy may well be attributable to the fact that strain was calculated from platen displacement. Extrapolation of the dynamic data suggests that uniaxial values for low density snow may be low by a factor of 5, probably because of the low strain rates and low viscosity of the snow. The true effect of temperature on E is hard to determine. In uniaxial tests where apparent initial tangent modulus is rate-dependent, a change of temperature will have the effect of changing the rate or frequency by virtue of its strong influence on viscosity. The best estimates of true temperature effect are obtained from high * In this and subsequent figures where a mechanical property is related to a single independent variable, the width of data bands partly reflects variation of secondary parameters (e.g. structural characteristics or temperature conditions). === PAGE 8 === 258 Malcolm Mellor io5 10'r ~T "i i i i r // *S y i i > i „>*** a 0.4 0.6 0.8 p. Density (g/cm3. Mg/m3) I _L J_ —• 1.0 _L -10 -100 8 Temperature (°C) I I I I I 0 0.2 0.4 0.6 0.8 p, Density (g/cm3, Mg/m3) FIGURE 2. Young's modulus for dry, coherent snow. (A) Pulse propagation or flexural vibration at high frequencies, -10° to -25°C (Smith, 1965; Nakaya, 1959a, b; Bentley et al, 1957; Crary étal, 1962; Lee, 1961; Ramseier, 1963). (B) Uniaxial compression, strain rate approximately 3 X IO-3 to 2 X 10~V, temperature -~25°C (Kovacs et al, 1969), (C.) Uniaxial compression and tension, strain rate approximately 8 X 10~6 to 4 X 10~V, temperature -12° to ~25°C. (C2) Static creep test, -6.5° to -19°C (Kojima, 1954). (D) Complex modulus, 103Hz,-14°C (N.Smith, 1969). frequency vibration experiments, which suggest that E does hot vary much over typical temperature ranges for dry snow (Fig. 2). The data summarized in Fig. 2 refer to dry snow that is well bonded. Values for granular snow with low cohesion are smaller, and E increases exponentially with time during sintering, finally tending asymptotically.to a limit. When dense snow is artificially disaggregated and then allowed to sinter at approximately ^10°C, the initial value of dynamic modulus is typically about 40 per cent of the value reached after 3 weeks, and about 30 per cent of the value reached after 1 year or more. Perhaps the most important conclusions that could be drawn for the theoretical mechanician concern the linearity of elasticity in snow. If the problem involves bulk stresses that are insignificant (e.g. pure shear or, more practically, small bulk stress in dense snow), then it may be justifiable to assume that snow is linearly elastic, permitting application of simple elastic theory. However, if bulk stresses are significant, === PAGE 9 === A review of basic snow mechanics 259 density will increase irreversibly in response to increases of bulk stress, and the elastic modulus is a very strong function of density, varying by three orders of magnitude or more within the range of densities commonly encountered in field situations. Viscoelastic properties of coherent snow At first sight it appears quite feasible to determine separately the moduli and viscosity coefficients of the Burgers model from quasistatic tests, e.g. by analysing constant-stress creep curves according to equation ('2j. However, there are practical difficulties in making tests that do not involve significant bulk stress, and since elastic and viscous properties are highly sensitive to density change, simple tests such as the uniaxial compression test can be misleading. One alternative is to use vibrational techniques, such as have been described by Nakaya (1959a, b), Lee (1963), N. Smith (1969), and Chae (1967). A snow specimen is set in flexural, torsional or longitudinal oscillation, and viscoelastic properties are calculated from test measurements, usually by assuming either a Kelvin—Voigt or a Maxwell model for analysis (see equations (5)—(9) and Fig. 1). Values for the real part of the complex modulus have already been discussed and given (Fig. 1); a plot of the complex modulus as a function of density has been added to Fig. 2 for comparison. Variation of the complex modulus with frequency is indicated in Fig. 3. ,1,1 " - cl '«^ Jo.9liJJ ' ' -IO°C (Nakoya, 1959) -|4°C (Smith, 1969) il ,1 , r ' yO(g/cm3) . 0.88 0J2__- 068 OJ55^ 0.48 Cv48__ 1 ' I * 1 1 1 1 MM. ~ - __-4°C — (Chae, 1967) ! 1 ! 1 1 10" 10' 10° Frequency (Hz) FIGURE 3. Dynamic Young's modulus as a function of frequency. 10" Variation of the loss factor tan S with frequency does not appear to be very well defined, although there is clearly a decrease with increasing frequency (Fig. 4). The available data for loss tangent as a function of density (Fig. 5) are confined mainly to densities higher than those normally found in seasonal snowpacks, but there are indications that tan 5 rises to high values, say 0.4, at densities representative of surface snow layers. Nakaya's (1959a, b) data indicate that tan S is a very strong function of temperature between —5° and —20°C, although rough interpolation from other data suggests a less strong temperature-dependence in this range (Fig. 6). Between ^50° and —90°C, maxima in tan S occur; these are reminiscent of similar features that relate to dielectric processes. Dynamic viscosity TJ can be calculated from the loss factor when a suitable rheological model is assumed. For a Maxwell model, r\ = E/(co tan 5), and for a Kelvin—Voigt model, rj = (E tanS)/co. In Fig. 7 dynamic viscosity values are plotted against frequency, indicating a decrease with increasing frequency. Dynamic viscosity is plotted against density in Fig. 8, showing a rapid increase of viscosity with increasing === PAGE 10 === 260 Malcolm WleUov U. lb 0.12 0.08 0.04 1 - - 1 ' 1 i 1 ' 1 ' 1 p e (g/cm3)(°C) 0.48, - ' 3.63,-9 0.91,-510-10, i 1 i 1 i 1 i 1 1 i 1 i l e, 1967 Nakayo, 1959 1 i 1 , 1 i 1 i - - - ™ - i 1 i 10' 10° Frequency (Hz) FIGURE 4. Loss factor plotted against frequency. 10' 0.20 0.16 00 0.12 -I 0.08 — FIGURE 5. 1 1 1- •9°C 500 Hz fa, 1959 "i'i, "r'J J_ 0.2 0.4 0.6 p. Density (g/cm3, Mg/m3) Loss factor as a function of density. "1 r density. Variation of dynamic viscosity with temperature does not appear to have been investigated systematically for snow, but Nakaya's (1959a) results for ice indicate variation according to an Arrhenius-type relation with activation energy values ranging from 12.7 to 18.7k.cal/mole. === PAGE 11 === A review of basic snow mechanics 261 oes-JDq o J I ! I T - - ~ ~ - z. - - i, i i i "E 0.91 g/c 1 CD 1 o I'M 1,1, o 1 1 ' 1 111 i 1 1 = 0.48 ex. y y y y y y y y y y 1,1, 1 l,i, i , i CB 1 y y h- 196 a JZ o 1 "O c O |'M | ' ! en m o> o a a 1,1,1, ! O 1 |M '- y' — yS _ - ~z — - ^ - - 1 ,1 o o (ssiod) A|,soos|A U, ' U Q w D O _ _ - — - - — ~ — -_ _ - — - 1 1 X O eu CM u M 1 1 1 1 *> ~ £ ~ en m — d _ l N 0) O O x o — _ —.. *t ai ***** / v< ~ >v "*'••. \\ "**• -\ \ } — -iy"'* — i i o d o d e w ci O O UDl 'JOPDJ SSO") === PAGE 12 === 262 Malcolm Mellor 10'" I09 — I07 - r — -_ r _ Iini i i i i lihi i l - i -7 1 to-|( ' 1 1 1 ïtati c J, 1954) 1 1 1 -9° 1 1 1 1 ( 71 Dynamic 'M 100-600 Hz (Nokaya, 1959) 1 1 1 1 1 lllll — --1 1 11 III i — 1 -\o' 10° FIGURE I 0.2 0.4 0.6 D, Density (g/cm3, Mg/m3) Viscosity as a function of density. Practical experimental difficulties have inhibited application of dynamic methods to low density snow, so that data are lacking in this region. Kojima (1954) undertook quasistatic creep tests on low density snow in order to determine the moduli and viscosities of the Burgers model, and his values of T?M have been plotted in Fig. 8. One possible source of trouble in dynamic measurements at high temperatures might be temperature rise in the specimen due to internal energy dissipation. Relaxation times for high-frequency vibration (given by -qK/EK, or tanô/co) appear to be in the range 10~5 to 10"3s, which is suggestive of molecular processes within the individual ice crystals (dielectric relaxation times are of the same order). However, relaxation times derived from quasistatic measurements (Kojima, 1954; Shinojima, 1967) are very much greater (5-2000 s),* and Shinojima (1967) finds r decreasing with strain rate (approximate inverse proportionality) for the range 8 x 10~6 to 4 x 10~4s_1. The latter suggests a bulk property that may be dependent on structural arrangement and connection of constituent crystals. Viscous behaviour in sustained creep Relaxation times for the Kelvin—Voigt unit are short, and so for sustained creep processes the stress/strain/time relationships ought to be controlled by the Maxwell viscosity T7M. The most direct way to study this property is to make constant-stress creep tests and thus obtain relationships between stress and strain rate for secondary creep. However, such results tend to be complicated by volumetric straining if bulk stress is significant; this complication can be avoided by employing a pure shear test (torsion of a hollow cylinder; application of equal-magnitude, opposite-sign principal stresses), or by applying stresses that are small relative to the creep resistance (density). * Ramseier and Pavlak (1964) report values up to 8 X106 s. === PAGE 13 === A review of basic snow mechanics 263 Unfortunately, adequate precautions have been lacking in some test programmes, and the resulting stress/strain-rate data could be somewhat misleading. Nevertheless, creep test data show quite clearly that for snow of a given density the stress/strain-rate relation for secondary creep is strongly nonlinear. The latter fact has an important bearing on the interpretation of viscosity data obtained by less direct means. 'Uniaxial stress viscosity' Very simple techniques have often been used for making creep tests under uniaxial compressive stress. For example, a cylinder or prism of snow is subjected to constant axial deadload and allowed to deform until a plot of displacement against time appears to become linear, at which time final strain rate is recorded against initial stress and initial density. This method works well as long as the stress level is low and total strains are small, but otherwise it is unsatisfactory unless special data analysis methods are adopted. It can be particularly misleading when stress/strain-rate relationships are being investigated. 10'"^" < 10 - — - ; - - - - — ^ Ë _ - ¥" -1 — r 1 - r i - zr- _ " i i / i i i I \ „v •-I to Comp i 1 C S^\ i i i / J i i i 1 1 1 1 ( -22 to-48 °C 0.08 to 0.5 bar A B -1 to-35 0.5 bar -1 to-40°C 0.01 to 0.07 bar -35°C ress on and Extension 1 1 1 1 1 ' L A) 12 bar, Temp C A-61 .= - Û-463 . û-30^ _ 4-22- E _ Û-IO - A-5_ °c 2 à 0 - / -2to-IO°C - 0.08 to 20 bar - - — "I - - 1 -1 i i hill - 1 1 1 -^IO8 I0b 0 0.2 0.4 0.6 0.8 p, Density (g/cm3, Mg/m3) FIGURE 9. 'Axial viscosity' (principal stress/strain-rate ratio for creep under uniaxial stress) plotted against density. (A) Ramseier and Pavlak, 1964. (B) Mellor and Smith, 1967. (C) Bûcher, 1948. (D) Shinojima, 1967. (E) Mellor and Smith, 1967; Mellor and Testa, 1969. === PAGE 14 === 264 Malcolm Mellor Uniaxial test results are not normally analysed to separate dilatational and deviatoric effects, and data are usually in the form of axial stress/strain-rate relationships, or in terms of an 'axial viscosity'. This 'uniaxial stress viscosity' is the ratio of axial stress to axial secondary strain rate; it is analogous to Young's modulus, and here is designated TJE. Figure 9 gives a general impression of the magnitude of r?E, and it is immediately evident that values fluctuate wildly. For a given material, r;^ can vary by 4 to 5 orders of magnitude as temperature and stress level vary within conceivable practical ranges. Variation of creep viscosity with temperature is usually represented by an Arrhenius relation (see Fig. 10). Values of the appropriate activation energy found by direct experiment lie in the range 11 — 18 kcal/mole (Mellor and Smith, 1966), while other test data (Bûcher, 1948) suggest values from about 9—20 kcal/mole. More careful experiments give a value of 16.4 kcal/mole for polycrystalline ice, and show quite clearly that the Arrhenius relation breaks down at temperatures above — 10°C with ice of typical chemical purity (Mellor and Testa, 1969). Although it has a basis in chemical rate theory, the Arrhenius equation gives a poor mathematical description of temperature-dependence for the viscosity of snow (Mellor and Smith, 1966) and for some other properties that are controlled by more than one activated process or mechanism. Alternative empirical equations (e.g. simple exponential) could be considered for application to constitutive equations. FIGURE 10. Variation of viscosity with temperature according to Arrhenius relationship. === PAGE 15 === A review of basic snow mechanics 265 ov = pgdz dz dz dr da dV ev = —- = -dz = Ats da À dr pg À dp p2g dz 'Compactive viscosity ' About 20 years ago, Bader (1953) drew attention to Sorge's (1935, 1938) observations and deductions relating to the time-invariance of Greenland snow density profiles; he went on to formalize steady-state relationships between density p and accumulation rate A (Bader, 1960, 1962), and defined a 'compactive viscosity' 7?c as the ratio of overburden pressure ov to vertical strain rate év at any given depth z in a flat-lying deposit of dry snow: ffv Vc = - (13) (14) (15) 1 dp - -f (16) p dr where rs is the elapsed time since the layer in question was deposited at the surface and V is the velocity at which it sinks below the surface. Viscosity values obtained from this ingenious scheme were used with considerable success in engineering analyses of foundation settlement and tunnel closure in Greenland and Antarctica. The concept was also applied to settlement of seasonal snowpacks by other investigators. Compactive viscosity has been widely regarded as characterizing continuous steady creep, Bader himself assuming newtonian viscosity for stresses less than 0.8 bar, but in view of the demonstrated nonlinearity of viscosity and the implicit relation between av and p in determinations of T?C, a careful appraisal seems called for. Sequential measurements of density in snow deposits yield diagrams known as 'layer profiles', in which the density of a given layer is plotted against time, and increments of new snow accumulation are recorded. Layer profiles for seasonal snowpacks show that when new snow is added to the surface, each underlying layer densifies rapidly for a few days, but gradually the vertical creep decays to a relatively insignificant rate. The same behaviour is found in uniaxial-strain compressive creep tests. Thus densification occurs largely by a quasiplastic collapse of limited duration rather than by a steady creep, and apparent values of 77c are partly dependent on the interval between snowfalls. This type of densification is in the nature of an equilibration process. After each event the stress is at a level which produces only very slow creep, but any significant addition to that stress gives disproportionately rapid creep, a situation similar to the equilibration of maximum deviator stress in glaciers. Thus ?7C, which must be nonlinear with respect to stress or strain rate, is automatically determined for a stress level that is critical for the prevailing snow density. Figure 11 gives values of r/c calculated from field observations, together with some results from uniaxial-strain creep tests. The discrepancy between values for polar snowfields and for seasonal snowfields seems too great to be explained by temperature difference alone; it is possible that the seasonal snow values reflect basic shortcomings of the concept when large discrete snowfalls occur at frequent intervals. In view of the uncertainties surrounding the determination and interpretation of T?C, it may be safer to work directly with stress/strain-rate data. Sustained creep is of overwhelming importance in snow mechanics, but in view of the complications arising from nonlinearity of viscosity with stress and density, it is necessary to have some systematic framework for analysis of viscous deformation in multiaxial stress fields. === PAGE 16 === 02 0.4 p. Density (g/cm3, Mg/m3) 0.6 FIGURE 11. 'Compactive viscosity' (major principal stress/strain-rate ratio for creep under uniaxial strain) plotted against density. (A) Greenland and Antarctica, -20° to -50°C (Bader, 1963). (B) Seasonal snow, Japan, 0° to -10°C (Kojima, 1967). (C) Alps and Rocky Mts (Keeler, 1969). (D) Uniaxial-strain creep tests, -6° to -8°C (Keeler, 1969). (F) Uniaxial-strain creep tests, -23° and -48°C (Mellor and Hendrickson, 1965). Simple stress analysis Further discussion of mechanical behaviour is facilitated by separating volumetric, or dilatational, effects from pure shear, or deviatoric, effects. The general stress tensor ay can be split into deviatoric components a[-v which tend to change the shape of an element without change of volume, and a dilatational (isotropic) component, or bulk stress 5, which tends to change only specific volume: °\i = fflj+V = ffij+35ijOkk (17) (/, /, £ = 1,2,3, or x, y, z; Sy = 1 when / =/, Sy = 0 when / ¥= /; repeated suffix denotes summation). It is also convenient to express stresses in such a way that they are independent of the directions of coordinate axes. Stress invariants /, ,I2,h are expressed in terms of cartesian components axx, axy, etc., or principal stresses a,, a2, o3, as: ayxavv + avva„ + ar,a *2 "xxvyy ' vyy"zz ' vzz"xx ~~ "xy CTyz~ 0Îx = 0lO2 + O203 + Oi0l CTxx°Vz ^yy^zx ^zz^xy "*" ^xy^yrPz h ~ crxxCTyyCTzz The bulk stress, or octahedral normal stress, a is ° = 1 Oxx + ffyy + ffzz) = i(o.l + a2 + 03) = \l °\ a2 a3 (18) (19) (20) (21) === PAGE 17 === A review of basic snow mechanics 267 The octahedral shear stress roct, which is also independent of coordinate axes, is root = 1 [(ai - o2f + (a2 - a3)2 + (a3 - a,)2]* = V(2)/3 (/? + 3/2)'/2 (22) Corresponding relationships exist for strain e^, and for strain rate èjj. A 'bulk viscosity' fj, analogous to the elastic bulk modulus K, and a 'shear viscosity' u, analogous to the elastic shear modulus G, can be defined such that (23) (24) i°i a'n and there °ij i = rth = 2uè'Vi fore, if isotropy is assume = 2uen + Sjjfjékk :d (25) Under this scheme, test data can be analysed to yield constants that are potentially applicable to general multiaxial stress states. However, there are still serious problems to be solved, since fj and u are nonlinear with respect to stress, and they vary greatly with density (or specific volume), strain rate, and temperature. Salm (1967) has approached these problems by assuming f\ and u to be functions of the invariants of strain rate, but other approaches can be taken. This is the essence of the problem in deriving a viscous, or viscoplastic, constitutive equation for snow. From equations (23) and (24), compactive viscosity r?c and 'uniaxial stress viscosity' % can be expressed in terms of fj and u: Vc = V+^U (26) 9 vu % = ~f (27) With a matched pair of tests giving T?C and r/£ for the same snow under similar conditions, values of fj and u can be determined from equations (26) and (27). However, data sets of this kind are not normally available, and it may be necessary to resort to more devious methods, such as estimation of the viscous analogue of Poisson's ratio vv, which may be defined as 3fj — 2u , ^ (28) 2(3fj+M) Figure 12(a) gives a summary of experimental data for vv. These results are widely scattered, but a suggested envelope of probable values for compression is given in Fig. 12(b). For most quasistatic problems in snow mechanics, gravity body forces are significant or even dominant, and they have to be incorporated into the equilibrium equations, which for negligible acceleration are: -y + pg-t = 0 (29) The equation representing resolution normal to the free surface of an extended snow mass can usually be integrated directly to give one component of normal stress (which may or may not be a principal stress), as in the case of densification theory. Other components of normal stress can sometimes be estimated with acceptable accuracy by assuming an appropriate value of vv. Volumetric strains, irreversible compressibility, and equations of state With typical engineering materials (e.g. metals, solid plastics, dense ceramics, rock, concrete) there is usually an obvious reference state for the unloaded material; === PAGE 18 === 268 Malcolm Mellor M. 0.4 0.2 i 1 I 1 I *7~T*1 Calculated from data by DeQuervain, I'll 1966 (•) Yosida,l963 Incompressible^) Uniaxial Tension* ' ' A o •» — o •1 " Uniaxial Compression* < 1 i I i / - - ider et al., I95I . Roch, 1948 "(Shinojima, 1967) 1 i 'i i 0.2 0.4 0.6 O, Density (g/cm3 , Mg/m3) 0.8 FIGURE 12. Summary of data for viscous analogue of Poisson's ratio (a), and speculative envelope for viscous analogue of Poisson's ratio (b). volumetric strains tend to be small and substantially reversible up to the point of incipient failure of the material, and it is quite realistic to describe the volumetric stress/strain behaviour by a simple rate-dependent bulk modulus. Typical snow is very different, in that compressive volumetric strains are large and substantially irreversible, resistance to deviatoric straining is highly sensitive to volumetric strain, and in many problems it is unrealistic to interpret irreversible volumetric strain as 'failure'. Thus it is not sufficient to characterize the mechanical behaviour of snow solely in terms of the original physical state (density, etc.) and the state of stress; it is necessary to also account for the stress or strain history. In principle, the last statement applies to virtually all materials, and in rational mechanics there exist formal procedures that account for stress or strain history. However, the practical difficulties of treating stress history are formidable, and with snow the difficulties are accentuated by the time-dependent structural changes that can occur almost independently of stress (sintering, grain growth, etc.). In those branches of solid mechanics where compressibility and large volumetric strains are major considerations, it is usual to define .'equations of state', which give === PAGE 19 === A review of basic snow mechanics 269 relationships between hydrostatic pressure and specific volume. Since temperature and strain rate are controlling parameters, there are innumerable pressure/volume relations for any given material; of special interest are the relations for slow isothermal compression and for adiabatic shock compression. For materials with finite yield stress, the 'isotherm' in a pressure/volume plot sets a lower limit to the family of compressibility curves, while the shock adiabat sets an upper limit. The latter relation, which represents the locus of final states for shock compression, is known as a Hugoniot, or sometimes as the Rankine—Hugoniot equation of state; the actual stress/strain path between the initial state and any given final state is known as the Rayleigh line. When snow is compressed hydrostatically, density increases quite readily until the grains are 'close-packed', after which the snow becomes more resistant, and there is more gradual compression to porous but impermeable ice, which in turn can be compressed to solid bubble-free ice. If compression is continued, the Ice I-h formed from snow undergoes successive transformations to high-pressure polymorphs. Since density is the reciprocal of specific volume in dimensional form, any graph that gives a systematic relationship between pressure and density can be interpreted as an equation of state. As already mentioned, the upper limit to the system of pressure/ density curves is set by the high-rate shock adiabat, but because snow does not appear to have a finite yield stress there is, in principle, no lower limit to the curves representing isothermal compression. However, for practical purposes there does appear to be a lower limit; it represents the maximum density that can be reached under a given sustained pressure in time periods of the order of months or more. This has already been referred to in the discussion of compactive viscosity. True hydrostatic tests are rarely made, and compressibility characteristics have to be deduced mainly from uniaxial strain experiments, results of which are summarized in Fig. 13. The same data are replotted in terms of bulk stress a, or first stress invariant Iu in Fig. 14, the transformation being effected by assuming appropriate values of vv (see Fig. 12) and calculating secondary principal stresses for the uniaxial strain conditions. In attempting to define the adiabatic upper limit to the compressibility character­ istics, only one set of Hugoniot data (Napadensky, 1964) can be drawn upon; the only other known attempt to measure these characteristics (J. Smith, 1969) involved shock tube experiments that produced volumetric strains far too small to yield meaningful Hugoniots. The original results by Napadensky clearly contain gross errors (densities in excess of 1.1 g/cm3 at pressures of the order of 100 bar), but for the range plotted in Figs. 13 and 14 the data are probably reasonably valid. No Hugoniot data are available for snow of typical densities, but an effort has been made to fill the gap by drawing on calculations for high-speed impact stress (Mellor, 1968). These are based on mass and momentum conservation, and assumption of a limiting density for compaction under plane-wave impact. Kinosita's (1967) compression data, which were generated at rather low rates with a somewhat ill-defined stress state, converge with the Napadensky Hugoniot data, and this suggests that it may be possible to obtain a good approximation to the high-rate compressibility characteristics by simple uniaxial strain compression tests on a modern testing machine (the high-speed actuator of CRREL's closed-loop servo machine has a maximum controllable speed of 0.42 m/s, which is almost 4 orders of magnitude higher than the speeds used in Kinosita's experiments). However, for impact at very high speeds it seems likely that there could be stresses significantly higher than any shown in Figs. 13 and 14. Dr G. K. Swinzow of CRREL finds that when inert projectiles of steel and aluminium are fired at velocities around 103m/s into snow of density 0.4-0.5 g/cm3, the metals (which have yield stresses of the order of 3 x 103 to 5 x 103 bar) are deformed plastically, suggesting stresses of the order found in Hugoniots for ice. This condition === PAGE 20 === 270 Malcolm Mellor p, Density (g/cm3, Mg/m3) FIGURE 13. Compilation of data relating major principal stress to bulk density for compression in uniaxial strain at various rates and temperatures. (A) Natural densification of snow deposits, -1° to -48°C. Approximate average loading rates 1CT10 to 10"8 bar/s (data from depth/density curves for many sites). (B) Slow natural compression of dense firn and porous ice. Approximate loading rates 10~10 to 10-9 bar/s (from depth/density curves for polar ice caps). (C) Slow compression of solid ice. (D) Isothermal compression of high pressure polymorphs of ice, -10°C (Bridgman, 1-911, 1937). (E) Calculated values for plane wave impact at 20-40 m/s (Mellor, 1968). (F) Hugoniot data for explosively generated shock waves (impact velocities 1-12 m/s), -7° to — 18°C (data selected from Napadensky, 1964). (G) Hydrostatic compression under constant stress at -10°C. Strain rates 10~6 to 10" V (Landauer, 1957). (H) Hugoniot data for solid ice at -10°C (Anderson, 1968). (I) Hugoniot data for solid ice at -10°C (Nakano and Froula, 1973; Larson et al., 1973). (J) Compression at approximately constant strain rate, -7° to -18°C. Strain rate « 10"" s"* (Kinosita, 1967). (K) Compression in uniaxial strain - incremental loading to collapse, -2° to -3°C (Bûcher and Roch, 1946). of virtual incompressibility is probably reached when the velocity of the plastic 'wave' (density jump) approaches the elastic wave velocity of the snow. Compressibility curves for slow isothermal straining can be obtained from depth- density relations. Integration of unit weight with respect to depth gives a relation between major principal stress and density for uniaxial strain, and a conversion to bulk stress can be made with the aid of estimated values of vv, as mentioned above. === PAGE 21 === A review of basic snow mechanics 271 io" = = 0.48 D I I liliul "Tir C — I I I Mill 10 10 10 10 10 Deviaîoric {octahedral) Stress (bar) FIGURE 16. Deviatoric stress/strain-rate relationships deduced indirectly from various data sources. (A) Snow, =» -10°C (data from Butkovich, 1956; Kovacs et at, 1969; Smith, 1965;Mellor and Smith, 1965; Bûcher and Roch, 1948). (B) Snow, «0° to -7°C (calculated and extrapolated from data by Haefeli, 1939; Martinelli, 1960; Frutiger and Martinelli, 1966; Judson, pers. comm.). (C) Ice, «0.9 g/cm3, -2° to -10°C (data from Mellor and Smith, 1967; Mellor and Testa, 1969; Hawkes and Mellor, 1972). (D) Snow, 0.49 g/cm3, -1° to -10°C (Mellor and Smith, 1967). Constitutive equations for multiaxial stress states In seeking a general constitutive equation it seems advisable to first consider separately the relationships for deviatoric and dilatational effects. Presently available data are inadequate for this purpose, but some speculation may be helpful for guiding new experimental investigations. In states of pure shear, snow of constant density seems to be quite similar to ice, in that elasticity may be taken as linear while strain rate for secondary creep is proportional to some power of stress in the ranges that are of most practical interest. For the complete stress range,probable boundary conditions suggest that a simple power law for steady creep is less suitable than a polynomial such as the hyperbolic sine, but in practice this refinement is not of overriding importance. Parameters of the secondary creep stress/strain-rate relation are temperature and density. The temperature parameter has already been covered in the previous discussion of creep, but the earlier presentation of viscosity data perhaps gives a misleading impression of density effects. === PAGE 25 === A review of basic snow mechanics 275 From the very skimpy data of Fig. 16, it appears that strain rate for a given stress varies enormously with density; for typical densities the relation can be approximated by a simple exponential of the form e~ap, in which a * 30 cm3/g. For more general shear situations, in which elasticity and viscoelastic creep transients are involved, the stress/strain relations for pure shear become enormously complicated, as can be seen if nonlinear viscosity is considered in the Burgers model. However, there may be cases where an assumption of linear viscosity in a four-element model is supportable if the problem involves a fixed time duration with small creep strains (the situation which gave rise to spurious indications of newtonian viscosity in low stress creep tests on ice). The volumetric stress/strain relation is even more complicated than the deviatoric one and, of course, volumetric strain strongly influences deviatoric strain rates. The elastic bulk modulus is very strongly dependent on density and therefore on volumetric strain: for snow of density less than 0.6 g/cm3, bulk modulus increases with the sixth power of density. Bulk viscosity has a particularly complicated dependence on strain, since it is controlled by the interplay of two opposing stress effects. As the stress on typical snow increases, strain rate tends to increase corre­ spondingly, and because volumetric strain is initially achieved by moving grains relative to each other, just as in a deviatoric process, strain rate tends to increase with some positive power of the stress. At the same time, resistance to volumetric straining is increasing with strain (and therefore stress) according to some sort of exponential trend. If viscoelastic creep transients were not involved, this effect might be described qualitatively by something similar to é = aane-ba (31) in which a and b are 'constants' related to initial density and other factors, and n is an exponent (>1) identifiable with the power law exponent for deviatoric stress/ strain-rate relations. The prospects for a rigorous volumetric constitutive equation are very discouraging, and it seems inevitable that drastic simplications will be needed for problem solving. In cases involving rapid compression and large strains the answer might be a rate- sensitive equation of state, or a quasiplastic compression modulus. In other cases where total volumetric creep strain is small, it may be feasible to combine bulk viscosity and shear viscosity through simple proportionality, in effect assuming that the viscous analogue of Poisson's ratio remains constant. The final step in formulating a complete constitutive equation is to combine the dilatational and deviatoric relations into a single stress/strain/time relationship that takes account of viscoelastic behaviour, including nonlinearity of elasticity and viscosity, and is spatially three-dimensional. This development is nowhere in sight, and greatly simplified approximations will have to be devised in accordance with the conditions of each individual problem. However, continued efforts along the lines pioneered by Salm (1967, 1971) seem important as a means of ensuring that simplified constitutive equations are compatible with more general forms. FAILURE Failure of snow The definition of failure in engineering materials is largely arbitrary, since it merely refers to the conditions under which performance ceases to be satisfactory. It may involve rupture, loss of load-bearing capacity, excessive strain, or excessive strain rate. In snow, failure could be actual separation under shear or tension, acceleration of deviatoric strain rate by onset of tertiary creep or by stress increase, volumetric collapse or acceleration of volumetric strain rate, or other effects. This uncertainty carries over into the definition of strength. === PAGE 26 === 276 Malcolm Mellor For present purposes strength will be regarded as time-dependent, and related to the maximum deviatoric stress that can be reached at a given strain rate, irrespective of whether rupture actually occurs within a limited period of observation.* This does not completely correspond to Salm's (1971) definition, but it permits unified discussion of both brittle and ductile strength failures. In multiaxial stress fields it is necessary to consider how the interaction of stress components affects failure and strength, and this necessitates a failure criterion, which may be broadly described as a critical functional relationship between the principal stresses for the conditions producing failure. Failure criteria, which may be either empirical or derived from physical models, have variously postulated constancy of maximum principal stress, maximum shear stress, maximum strain energy, and maximum shear strain energy. For snow, Salm (1971 ) proposed addition of critical power as a criterion, taking into account total elastic strain energy and dissipation rate for viscous work. Brown et al. (1973) were more ambitious, attempting to define the conditions for failure by integration over the entire strain history for isothermal deformation according to a fairly general constitutive equation. The Drucker—Prager criterion, a generalized form of certain classical criteria (von Mises, Tresca, Mohr— Coulomb), has attracted interest in snow mechanics (Mellor, 1968; Salm, 1971), since it separates out the effect of the isotropic bulk stress. The Drucker—Prager criterion reduces to the Mohr—Coulomb criterion for the case of plane deformation, and the latter has been used in snow mechanics since the earliest days of the subject. Failure under uniaxial stress The most unambiguous determinations of strength are obtained from direct uniaxial stress tests in tension or compression, which do not call for any knowledge of rheological properties in their interpretation. However, while these tests are simple in principle they can be complicated in practice, especially where high strain rates and brittle fracture are concerned (Hawkes and Mellor, 1970). There are other tests which purport to measure uniaxial strength indirectly by invoking rheological properties for the material, usually an assumption of linear elasticity. They include the ring tensile test, which has been shown to be invalid for most engineering materials (Mellor and Hawkes, 1971), and beam bending tests, which also seem dubious for highly compressible viscoelastic materials. A general impression of the uniaxial strength of bonded dry snow is given by Fig. 17, which shows tensile and compressive strengths as functions of density for relatively high strain rates (10~4 to 10~2s-1). Broadly speaking, tensile and compressive strength are equal at low densities, while at the density of solid ice the ratio of compressive strength to tensile strength is about 5 on this plot. Brittle fracture of porous solids (rocks, ceramics, concrete, etc.) is often analysed in terms of Griffith fracture theory and its later derivatives, using failure criteria that imply compression/ tension strength ratios ranging from the classical Griffith value of 8 to values between 10 and 20. The data of Fig. 17 might be regarded as characterizing brittle fracture, i.e. creep strain is small relative to elastic strain for the loading that leads to failure. As strain rates drop (or temperatures increase) so that failure becomes ductile, i.e. creep strain becomes large relative to elastic strain, then strength (as defined earlier) can be expected to decrease significantly for very dense snow. This relationship has been given for a very wide range of strain rates in ice (Hawkes and Mellor, 1972), but a different situation prevails for slow compression of typical snow, which increases in density and effectively work hardens. * This implicitly identifies the secondary creep inflection point of a complete classical creep curve with the peak of a conventional stress/strain curve. In either case the time taken to reach this point is what is sometimes called 'time-to-failure', which could be effectively infinite in some practical problems. === PAGE 27 === A review of basic snow mechanics 277 S 10" I- - Ë~ -_ - zT — - - r 1 -Il ! 1 1 1 1 1 1 1 1 1 1 1 UNIAXIAL STRESS / / / 1 Compression --~~Z~--^^^^^^^^ ^ ^^ s / / X / Tension / / / / / i i i i i i i i : - z 1 - — - 10 1^ 10 10 10" 10 < fcr 0.2 0.4 0.6 y3, Density (g/cm3, Mg/m3) FIGURE 17. Strength of dry, coherent snow under rapid loading in uniaxial stress states. (Data from Bûcher, 1948; Butkovich, 1956; Haefeli, 1939; Hawkes and Mellor, 1972;Keeler, 1969; Keeler and Weeks, 1967; Kovacs era/., 1969; Mellor and Smith, 1965; Ramseier, 1963; Smith, 1963; Smith, 1965.) One interesting and somewhat controversial point about the rate sensitivity of strength concerns the apparent small drop in compressive strength as strain rate increases through the ductile/brittle transition range. In the case of ice it has been shown that, at least in some cases, the drop was caused by poor testing technique for the brittle range (Hawkes and Mellor, 1972). However, during the course of a test there is viscous energy dissipation which must produce temperature rise in all but very slow tests, and in principle viscous heating should lower the appâtent strength. This effect, which is an intrinsic part of Salm's (1971) theory, does not appear capable of producing significant homogeneous temperature increases in typical ice test specimens, but if temperature rise is concentrated at active microdeformation sites (grain boundaries, defect structures) then the effect may well be significant, especially in snow. Many experimenters, including the writer, have attempted to define the effect of temperature on strength, but presently available data are not very convincing. It seems likely that data for high temperatures (above — 10°C) actually reflect temperature/ strain rate interactions that are not adequately defined. In the brittle fracture range, strong temperature dependence is not to be expected, whereas in the creep rupture range the temperature relationship should be similar to that which controls sustained creep. Shear strength The term 'shear strength', which is subject to varied interpretations, tends to generate confusion at the best of times, but in the case of highly compressible snow it can be very confusing indeed. One measure of shear strength can be obtained from uniaxial === PAGE 28 === 278 Malcolm Mellor stress tests, in which maximum shear stress is aj2 and octahedral shear stress is aJy/3, Ox being the axial stress. However, the general practice has been to define and determine shear strength according to the Mohr-Coulomb failure criterion for biaxial stress fields: {o\ - 02) ~ {o\ + o2) sin0 = 2c cos (32) or s = c + p tan0 (33) where Oi and a2 are principal stresses, c is cohesion,