207
A highly simplified model is presented of the erosion process under liquid impact. The theory outlined allows the mathematical expectation of the erosion rate as a function of time (erosion curve) to be predicted. This simplified theory of erosion is based on the statistical nature of the repetitive loading due to liquid impact. It employs a classification of individual impacts according to their quality of or effectiveness in removing material. The specific loss of material (per impact) is considered to be a random variable, which assumes different, discrete values depending on the order of repetitions of effective impacts at the same location on the target (surface). The probabilities of producing any of these repetitions are calculated. For particular applications, the theory requires characteristic data for the specific loss of material as well as typical data concerning the hydrodynamic loading conditions. For the case of a ductile erosion process, a very approximate solution is presented. Nevertheless, it exhibits the typical features of erosion curves as presented in the literature.
" A s long as the surface is smooth, it offers no hold for the impinging drops of water and the water flows off on all sides. Therefore erosion does not occur for some time. H o w e v e r , as soon as any roughness forms, erosion develops rapidly because the water penetrates the unevenness of the surface at a high pressure due to the impact, and acts very violently. Finally, when the erosion has attained a considerable depth, a layer of water adheres to the now completely roughened surface. This water dampens the impact of subsequent drops so that their destructive action is diminished. The specific erosion consequently decreases after a certain depth has been reached." E. H o n e g g e r (1927)
1. Introduction
Since H o n e g g e r ' s experiments in 1927 it has been known that erosion by liquid impact does not develop at a constant rate. H o n e g g e r [1] presented erosion curves which exhibited, after an initial incubation period, a phase of increasing erosion rate (acceleration), followed by a phase of decreasing rate (deceleration). His explanation is essentially the same as that put forward in recent publications on erosion under drop impingement [2,3], Fig. 1. Even today, opinions differ about which stage in the erosion process is the most important. The quantitative prediction of the erosion curve could definitely be an important step towards finally establishing not
overly conservative, admissible hydrodynamic loading conditions. The particular shape of the erosion curve depends, of course, on the material and (geometric) surface conditions of the solid structure, as well as on the type of (thermo and) hydrodynamic impact loading. It is obvious that realistic analytical solutions to this problem cannot be expected and an approximate numerical solution is extremely timeconsuming and expensive, even with the simplest of assumed material behavior and failure mechanism. Experimental studies also are difficult, timeconsuming and expensive. What remains is then the hope that, on the basis of preliminary experimental evidence, a
208
I ~b b
o=
i
1T Trr J~
UJ
i/g
m ~"
V
Time of Exposure
V i
Time of Exposure
(b
o=
T i m e of
Exposure
Time of Exposure
Fig 1. Typical erosion rate vs. time curves according to different investigators [2].
highly simplified approach will contain the essential features and provide some basic insight, and an understanding of the controlling parameters. With a consistent set of concepts it is then possible to guide the experimental work needed; to check, in principle, the results of computer calculations; to advise on the choice of appropriate materials and finally to define the admissible hydrodynamic loading conditions. The problem of erosion under liquid impact involves a wide variety of material, solid mechanics and fluid dynamics parameters. When fatigue and brittle failure can be excluded, the erosion process is ductile and during an incubation period shows complex details of grain boundary delineation and plastic depression of individual grains below the original surface level, finally leading to a general, fairly uniform undulation of the surface and the formation of small, smoothedged pits. This has been described in detail by Vyas and Preece [4]. Towards the end of the incubation period, the general undulations which may be considered as resulting from the combined pressure pulses of a large number of collapsing bubbles, or from impinging drops, develop into craterlike depressions with large smooth lips. Material loss is considered to occur from the lips of the craters by ductile rupture and again those pressure pulses (effective impacts) may be
considered responsible for it. A marked timedependence in the material loss rate is observed. The change of surface topography as erosion proceeds and its feedback to the hydrodynamic loading (e.g. trapped gas and/or liquid at the bottom of deep craters; changes in impact angle) are considered to be major factors. Another factor may be the change in material behavior due to repetitive impact loading, i.e., work hardening may play a role. For example, an average increase in surface hardness by a factor of two has been reported and almost a cubic dependence of erosion resistance on hardness [3]. All these changes are likely to make removal of material far more effective while erosion proceeds through the layers very close to the surface, i.e., at the early stage (after a possible incubation period). The purpose of this paper is to present a highly simplified model of the erosion process under liquid impact. A theory is outlined which allows the mathematical expectation of the erosion rate as a function of time (erosion curve) to be predicted. This theory is based on the statistical nature of the repetitive loading by liquid impact. It employs a classification of individual impacts according to their quality of or effectiveness in removing material. To this end, the specific loss of material (per impact) is considered to be a random variable which assumes different values depending on the order of repetition of impacts at the same location on the target (surface). In the case of a ductile erosion process, we assume that the early repetitions of the impact loading are most effective for the removal of material. Nevertheless, the very approximate solution exhibits the typical features of erosion curves as presented in the literature[
The theoretical approaches advanced so far for the quantitative prediction of the erosion rate timebehavior are essentially based on ad hoc assumptions characterising overall effects. None of them examines or models the actual physical occurrences in a more specific way, so they do not lend themselves to further physical refinement. Heymann [5] considers the lifetimes of the top surface and subsurface layers as random variables with assumed probability densities. These densities are supposed to reflect all statistical aspects of the erosion problem. The approach is selfconsistent; there is, however, no direct way of making a physical refinement, Continued on p. 209
209
N=__eN{m}=Em,p N,
i
pN=_eU{m=rn,},
N=1,2,3,...
(i)
From the physical point of view, we need a classification of the individual impacts according to their quality of or effectiveness in removing material. Various types of classifications are conceivable. We shall adopt a particularly simple one which appears to be supported by experimental evidence. Moreover, this approach should lend itself to straightforward extensions and refinements.
or drops (liquid impact). The actual situation is extremely complex. Overlapping craters of various sizes m a y continuously cover the entire surface (Fig. 2). We adopt a simplified model where only nonoverlapping craters of the same size m a y repeatedly cover the surface layers of the target. Thus, successive craters are considered either not to overlap at all or to overlap completely, and partial overlapping is excluded. Let the probability of hitting an existing crater by liquid impact be denoted by p. Assume this probability to be constant and equal to the ratio of the area of a crater to the area of the total surface exposed to liquid impact. The smaller the impinging jet or drop c o m p a r e d to the size of the crater, the better this assumption is fulfilled. It further contains the idea that wherever the crater is on the surface, the chances of hitting it are the same. Let us consider, at the N t h impact, the N separate events: "producing a new crater", "producing a first repetition", "producing a second repetition", . . . . "producing a ( N  1)th repetition". These events together form the certain event. It is convenient to define the production of a 0th repetition as the formation of a new crater. We may then base the classification of the individual impacts on the repetitive nature o ( impact loading and we may choose the probability p U equal to the probability of producing an ith repetition, at the N t h impact,
Pi =Previ,
Footnote 1 Cont. The fundamental reasoning in Thiruvengadam's theory [6] is, in a sense, a vicious circle. His 'efficiency' of erosion, introduced as a premise, is very closely connected with his 'intensity' (or rate) of erosion for which he draws his conclusion. He thus gives as proof the assumption from which he starts. Despite references to concepts such as fatigue and Weibull distribution, Springer [7] assumes from the outset that the erosion rate is constant with time. His model comes down to the assumption that shorter incubation periods correspond to higher subsequent erosion rates. This may, for certain empirical data, be an approximation for the acceleration phase if an incubation period exists. The mathematical approach recently advanced by Noskievic [8] definitely takes us outside the discipline ot physics. Notwithstanding numerous allusions to concepts of rigid body dynamics, viscoelasticity and damped vibrations, the whole approach is still a curve fitting exercise in elementary geometry.
N
i=O, 1 , 2 , . . . , ( N  1 ) ,
N
(2)
where Prep 0  Pnew" We thus have to determine N the probabilities Prep i" Let us consider a preliminary problem first.
Fig. 2. Surface under repetitive impacts. The surface at the center corresponds to the superposition of major craters (left) corresponding to effective impacts and weaker craters (right). In the simplified model, craters of one size only cover the surface, and we assume that the successive craters are either not at all or totally overlapping. Weak impacts affect, first of all, the material behavior.
210
O
2 P2
OO
p2 1
Fig. 4. Probabilities P~ and P~2 of having obtained, after the second impact one first repetition C22, and two new craters C~.,, respectively.
p~ =pp22 =p2.
The formulation for higher values of N is straightforward. For N = 47 the corresponding probabilities are given in Appendix 1.
2p)P:1+(1
p)P~=(1
p)2
(6)
(4)
After the third impact, N = 3, according to the
and so on. See Appendix 2 for the corresponding probabilities at impacts N = 47. Similarly, we can determine the probabilities N Prep1 of producing, at the Nth impact, a first repetition. Thus, for N = 1, 2, 3 . . . . .
P)ep 1 = 0 , 2 2 P r e p 1 = P2 =
P ,
(7)
(I) 0
3 2
0
2 1
0
1
pN
3;2,2
and so on. See Appendix 2 for the corresponding probabilities at impacts N = 47. N We note the probabilities P r e p 2 of producing, at the Nth impact, a second repetition, again for N = 1,2,3, P~ep 2 = Prep22
=
Prep 2 = PP2 = P
(8)
Fig, 3. Configuration c3N,2,2,1 ..... l with one second repetition (3 overlapping craters) and two first repetitions (2 overlapping craters each) after the Nth impact. The probability of having obtained this configuration after the Nth impact is denoted by pN 3,2,2"
See Appendix 2 for the corresponding probabilities at impacts N = 47. The probabilities P rNp 3 , P rNp 4 and P rNp 5 f o r e e e producing, at the Nth impact, a third, a fourth
211 2.0
and a fifth repetition, respectively, are given in Appendix 2. Let us finally note the probabilities pNp ~ 2 of producing, at the Nth impact, a second or higher repetition. It follows immediately N N __ pS (9) Preps>2  1  P r e p I new Thus we have
P~ep~>2 2 : Preps2 = 0 , 3 3 p2 Preps>2 = P r e p 2 =
;/m6
(lO)
1,0 // / // /
and so on.
1.0
~ep
/ / P r ep>_.3 N
0.5
To bring out the essential picture of the dependence of the probability of a first repetition on the number of impacts, and for this purpose only, let us choose for the probability of hitting an existing crater the value p = 0.5. The correspondin~ values for the probabilities P~new, prNepl and P~ep~2 for the first 7 impacts, are given in Fig. 5. At this point it is interesting to observe that a N weighted superposition o f prNpl and Preps2 results in the typical shape of an erosion rate vs. time curve if the number N of effective impacts is taken as a measure of time (which should be a reasonable assumption), Fig. 6, curve (a).
l
3
~r'rep 2/ ~
0.0 /,, 5 6
w
7 ,'~.m.N
Fig. 6. Erosion rate vs. time (specific loss of material vs. number N of impacts) relative to its stationary value. The different curves (a), (b), and (c) correspond to three different influences of early repetitions, cf. Table 1.
This observation is no surprise. In fact the expected value of the specific loss of material, at the Nth impact, can be written, eq. (1),
~1 ~
N
E~{m}=
N N m l P r e p 1 "~ m2Prep~>2 ,
1.0
Pr~p i
'
'
'
i //
i j.~
N=1,2,3,...
(11)
X N XPnew
N Prep > 2
o.s i~
0.0 0 1 2 3 4 S 6 7 ~i~' N
Fig. 5. Probabilitiesfor producing, at the Nth impact, a new crater, a first repetition, or a second or higher repetition. For the purpose of illustration, the probability of hitting an existing crater has been assumed to be p = 0.5.
Then all we have to assert is that the specific loss of material m 1 for a first repetition is much greater than the average specific loss m 2 for the second and all higher repetitions; this is amply justified throughout the literature. We note that curve (a) in Fig. 6 exhibits all the stages usually found in erosion curves: an incubation period, an acceleration phase, a deceleration phase and a final stationary stage with constant erosion rate. It is natural to ask how the shape of the erosion curve will be influenced by the higher order repetition probabilities. The values for the . N N N N probabllmes P r e p 2, Prep 3, P r e p 4 a n d P r e p 5 a r e given in Fig. 6. Again, we observe that weighted superpositions of various P r e p / result in shapes which are typical for erosion rate vs. time curves as reported in the literature. Note that the par
212
ticular curves (a), (b) and (c), Fig. 6, represent the relative expected value (i.e., measured in units of the corresponding stationary expected value) of the specific loss of material, at the Nth impact, hence at time t~,
k N__ N N k~k = Ek {m} : ~] m i P r Np i + l n k + i P r e p ~ k + 1 , e
i1
N=1,2,3
....
(12)
where m i denotes the specific loss of material for an ith repetition, and m,+~ denotes the average loss for the (k + 1)th and all higher repetitions; cf. eq. (1). The curves in Fig. 6 were generated by using the numerical values for the relative effects of early repetitions as given in Table 1. Finally, we note that eqs. (1) and (2) also allow for the general case where m o ~ O, and m o denotes the specific loss of material when a new crater is produced. It has been observed in certain, less frequent, cases that the erosion curve starts with a maximum value, without an acceleration phase; this could correspond to a significant material loss even when a new crater is
formed [5].
5. Conclusions It appears that the following conclusions can now be drawn. (1) On the basis of the statistical nature of the repetitive loading by liquid impact a theory has been outlined which allows the mathematical expectation of the erosion rate as a function of time (erosion curve) to be predicted. The specific loss of material (per impact) may assume different discrete values, with probabilities corresponding to: the formation of a new crater; a first repetition of impacts; a seconds repetition; or a repetition of even higher order, at the same
Table 1 Relative effect of early repetitions (normalized by stationary effect of later repetitions), cf. eq. (12) and Fig. 6 mI m6 Erosion rate curve (a) (b) (C) m2 m6 m3 m6 m4 m6 m5 m6
location on the target (surface). For a ductile erosion process and on the assumption that the early repetitions are the most effective for removing material, a solution is presented. Although this is a very approximate solution it exhibits the typical shape of erosion curves usually found in the literature: an incubation period, the phases of acceleration and deceleration, as well as the final stationary stage of a constant erosion rate. (2) We note that the specific loss of material enters the expression for the expectation of the erosion curve only in the form of a discrete weighting function, in a weighted sum of probabilities. We might therefore think that the typical shapes of erosion curves are to a large extent statistical in nature. They should then be relatively insensitive to particular loading conditions, material behavior, and failure mechanisms. (3) The above should be true although the number N of impacts may be interpreted in several ways. In principle, N may refer to the very large number of individual impacts (of average intensity). It may equally well refer only to a small number of effective impacts (also of average intensity). Consequently the specific loss of material values for either interpretation must be adjusted. (4) For particular applications, the theory requires characteristic data for the specific loss of material as well as typical data for hydrodynamic loading. Further systematic effort is needed to classify the wide ranges of material behavior and failure mechanisms and the various hydrodynamic loading conditions. In particular, it is recommended that approximate calculations concerning the formation of craters and ductile failure, as well as for brittle failure and fatigue, all caused by liquid impact, be performed. (5) The simplified theory presented allows straightforward indepth refinements and direct generalizations on various levels to be made. The concepts presented in this paper for a ductile process of erosion can also be applied, with minor modifications, to the case of erosion by fatigue and brittle failure.
3 3 3
1 0 0.5
1 3 1
1 0 2
1 2 3
Appendix 1: Probabilities for the configurations In Section 3.1 the probabilities of the configu
213
rations of the erosion process have been presented, for the first 3 impacts, N = 13, eqs (3)(5). We may write, in general,
p N = pNp~.~,
m(N)
pN=I,
i=1
j=l,2,...,m(N1).
(A.2)
N = 1, 2, 3 . . . .
N= 1,2,3,...
(A.1)
We finally have p~ = (p2p3... pN)Tp~ , (A.3) where T denotes the transposed matrix, and
N where the column pN = (pN., p 2 N , . . . , Pro(N)'} is formed of the (unconditional) probabilities that, at the Nth impact, the erosion process will pass to the configurations C N, i = 1, 2 . . . . , re(N), respectively. We denote by P the matrix of the transition probabilities pq(N)= pN, p N = IIp~ll. The transition probability pN is the conditional probability, at the Nth impact (or a time tN), that after the Nth impact C N is obtained under the assumption that cN.~ was obtained at the ( N 1)th impact. We note that in each line of the matrix pN there is at least one element different from zero, N and the transition probabilities pq, for any N, satisfy the relation N=5
P~  {P~} = 0}.
We present in the following the probabilities pN of the configurations C N, i = 1,2 . . . . . re(N), for N = 47. N=4
P31
p l4
(1 
P32
3p)
(1 
p3
2p) (1 p) P
e~
P~
3p
p P
(A.4)
P
e~,2
e~
P51 Ps2 P53 p54 (1  4p) 4p
e~
(1  3p) p
e~
P~
eL
(1  2p) p (1  p)
(A.5)
p
P~ P~.2
P~.2
2p
p
(1  2p)
2p
ts3
pS 4
p5 5
p52,2
P53, 2
P~
(1 
e~
P~
5p
3p)
(1 
e~ e~ P~
e~.2 P~.2 P~,2
P 3p 2p P
3p) (i2p)
P
2p
P
P~,2.2
P~,3
214 N=7 p~ p~ p6 3
6 P4
p6 5
6 P6
. P2.2
6 P3,2
p. . 4,2
P2,2,2
P3,2
e; e; e; e~ e~ P2 P;
P~,2 P~,2 P~.2 P~,2
(1 
6p)
(i  5p)
6p
(1 
4p)
(1 
3p)
(1 
2p) (1  p ) P 4p)
(1  3p) p (12p)
P 4p 3p
2p
(1 
2p
p 2p p P
p (1  3p) 3p
(1  2p) 2p
P~,2,2
P~,2,2 P~,3 P~,3
(A.7)
Appendix
2: Probabilities
In Section 3.2 the probabilities for repetitions of various o r d e r have b e e n given, for the first three impacts, N = 1  3 , eqs. ( 6 )  ( 8 ) . N F o r N = 4  7 , the probabilities P,e,~ of p r o d u c ing, at the N t h impact, a new crater are as follows: Pnew = (1 5 Pnew = (1 4
3pe~ + pe32
3p)P~ + ( 1
2p)P3~ + (1  p ) p 33 ,

4p)P~ + ( 1  3p)e 4 + (1
2p)e~
2p)P~ + ( 1
p)P~
(A.8) 
(A. 10)
+ (1  3p)PS2,: + (1  2p)P~.:,
7 e.ew = (1  6p)P 6 + (1  5p)P~ + ( 1
4p)P 6 p)e~
T h e probabilities Prep 2 o f producing, at the N t h impact, a second repetition are, for N = 4  7 , as follows:
p r 4 p 2 =
+ (1  3p)P~ + (1  2p)P 6 + (1 + (1 
pP~,
4p)p62 + ( 1
3p)P~,2
+ (1  2p)p6,2
I(1 6  3p)P2,2, z + (1 __
2p)p6,3
215
We note
Prep>~3 = Preps>2  Prep 2
7 Prep 5 =
PP5
5 .
(A. 12)
We note
N : pN N Preps>6 rep~5  Prep
N The probabilities Prep3 of producing, at the Nth impact, a third repetition are, for N = 17, as follows:
P~ep 3 = Prep 3 = Prep 3 = 0 2 3 4 4 3 Prep 3 ~ P 4 = P , P~ep 3 = P P ~ ,
(A.18)
Acknowledgment It is a pleasure to acknowledge valuable discussions with Professor I.L. Ryhming, and his very encouraging interest throughout the preparation of this work.
(A.13)
p6o. 3 = pp5 + p P L ,
Prep3  2pP63,3 + P P 3
7 
6 + PP3,2
We note
N N N Preps>4 " Preps>3  Prep 3
References (A.14)
[1] E. Honegger, "Essais d'6rosion des ailettes de turbines vapeur", Revue BBC 14(4) 95104 (1927). [2] C.M. Preece, "Cavitation Erosion", in: C.M. Preece, ed., Erosion, Treatise on Materials Science and Technology, Vol. 16, Academic Press, New York (1979). [3] J.H. Brunton and M.C. Rochester, "Erosion of Solid Surfaces by the Impact of Liquid Drops", in: C.M. Preece, ed., Erosion, Treatise on Materials Science and Technology, Vol. 16, Academic Press, New York (1979). [4] B, Vyas and C.M. Preece, "CavitationInduced Deformation of Aluminium", in: Erosion, Wear, and Interfaces with Corrosion, ASTM STP 567, Philadelphia (1974). [5] F.J. Heymann, "On the Time Dependence of the Rate of Erosion Due to Impingement or Cavitation", in: Erosion by Cavitation or Impingement, ASTM STP 408, Philadelphia (1967). [6] A. Thiruvengadam and S.L. Rudy, "Experimental and Analytical Investigations on Multiple Liquid Impact Erosion", N A S A CR1288, Washington, DC (1969). [7] G. Springer, Erosion by Liquid Impact, Wiley, New York (1976). [8] J. Noskievic, "The extended mathematical model of cavitation and erosion wear", in: Proc. 6th Int. Conf. on Erosion by Liquid and Solid Impact, Cambridge, England (1983).
The probabilities Pr~p4 of producing, at the Nth impact, a fourth repetition are, for N = 17, as follows:
P~ep 4 = Prep 4 = Prep 4 : Prep 4 2 3 4 e~ep 4 ~ P55 = p 4 , p6ep4=pp54,
= 0 ,
(A.15)
Prep4
~
6 6 P P 4 + PP4,2
We note
N N N erep>~5 = Preps>4  Prep 4 . N
(A.16)
The probabilities Prep5 of producing, at the Nth impact, afifth repetition are, for N = 17, as follows:
P~ep 5 = Prep 5 = Prep 5 = Prep 5 : Prep 5 ~ 0 , 2 3 4 5
pr6p5 = p6 = p 5 ,
(A. 17)