Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 50, 1, pp. 231-250, Warsaw 2012 50th anniversary of JTAM PIECEWISE RELIABILITY-DEPENDENT HAZARD RATE FOR COMPOSITES UNDER FATIGUE LOADING ADJUSTMENT Chung-Ling Chen Kuo-Shong Wang National Central University, Department of Mechanical Engineering, Jhongli, Taiwan, Republic of China; e-mail: sysengrg@hotmail.com.tw Based on the derived transition period and reliability drop, this paper proposes amethod of piecewise combination of the reliability-dependent hazard rate function named (eocp) model to describe the dynamical re- liability in a two-stage fatigue loading process. First, the parameters eo, c, p are fitted through simulated failure data under various constant- amplitude cyclic stresses. The reliability of the high-low loading process is described piecewisewith the corresponding values of (eo,c,p) for each respective stress level, and maintains Ra in the transition period while Ra denotes the reliability at which the stress level changes. The reliabi- lity of the low-high process is determined by subtracting the portion of reliability drop at Ra from the piecewise fitted curves. The proposed re- liability behavior is verified successfully.The linear damage sum is found to be larger than unity for the high-low loading, and on the contrary for the low-high cases. A larger difference between the stress level changed results in larger deviation of damage sum from unity, especially when Ra near 0.9. Key words: fatigue loading adjustment, hazard rate function, dynamical reliability, Monte Carlo simulation, linear damage sum 1. Introduction The dynamical reliability of composite laminates when subjected to fatigue loading adjustment is a fundamental issue in evaluating these materials for practical applications. Several researchers (Broutman and Sahu, 1972; Yang and Jones, 1980, 1981, 1983; Gamstedt and Sjögren, 2002; Found andQuare- simin, 2003) have reported thatwhen composites are no longer able to sustain 232 C.-L. Chen, K.-S. Wang the fatigue load,Miner’s damage sumwill be larger than unity in the high-low sequence and smaller than unity in the low-high sequence. In contrast, others (Han and Hamdi, 1983; Hwang and Han, 1986) have reached the opposite conclusion for other types of constituentmaterials. Regardless, little attention has been focused on an explanation of the load sequence effect based on the dynamical reliability of composites under varied stress-level fatigue situations. As for the dynamical reliability of materials subjected to two-stage cyclic stresses, only limited research has been done successfully in this area. This is mainly because the sample size of most two-stage fatigue tests is too small to verify statistical analysis accurately. Tanaka et al. (1984) used theB-model to analyze the probability distribution of fatigue life of a large size of nickel-silver samples. However, it is difficult to apply this model to predict the behavior in a two-stage loading process when only results of a single-stage fatigue test are available. After the development of several hazard ratemodels as reviewed by Wang (2011), a two-parameter reliability-dependent hazard rate function h(R)= eo+c(1−R) is used to dealwith the dynamical reliability of amaterial concerning fatigue loading adjustment (Wang et al., 1997). When the stress level of fatigue loading is adjusted, the hazard rate right before the adjust- ment becomes the intrinsic weakness at the beginning of the following stage loading. This relation has been verified by the data given by Tanaka et al. (1984). Later, Ni and Zhang (2000) presented a two-stage fatigue reliability method based on two-dimensional probabilistic Miner’s rule. The results are also verified by the data of Tanaka et al. (1984), but the application of this method is restricted by some assumptions. The composites are inhomogene- ous and anisotropic materials, and more complicated in the fatigue behavior and failuremechanisms than those of homogeneous and isotropicmetallic ma- terials. The above methods have not proven to be valid for composites yet. Wang et al. (2002) modified the above two-parameter hazard rate relation to a three-parameter form of h(R)= eo+c(1−R) p, the so-called (eocp) model, to depict the dynamical reliability of several types of engineering components and devices. Thismodel has been verified to describe the dynamical reliability of composite laminates under simulated single-stage fatigue loadingwith good results (Chen et al., 2009). In the region of high cycle fatigue of composites, it is found that eo and p can be considered as a fixed value; c can be a power function of the stress level. Recently, Chen and Wang (2011) defined two parameters, the transition period n2a and reliability drop |∆R| (see Appendix), respectively, to describe the effect of high-lowand low-high fatigue loadingadjustmenton the reliability degradation of compositematerials. Figure 1 shows a typical expression of the Piecewise reliability-dependent hazard rate... 233 reliability degradation of composite laminates under two-stage fatigue loading processes. Denote the reliability at the instant of loading adjustment by Ra. In the high stress section of both the high-low and low-high loading processes, the strength of composite laminates degrades at a relatively higher speed than that under a low level stress. Consequently, the higher rate of fatigue failure causes the reliability to degrade relatively steeply. At the instant the stress is adjusted fromhigh to low level, the residual strength of the survivals becomes larger than the low-levelmaximumcyclic stress.Duringaperiodof n2a, named the transition period, no failure occurs until the minimum residual strength degrades to the low-level maximum cyclic stress. Thus, the reliability remains unchanged in n2a. Analogously, at the instant of low-high adjustment, those specimenswith a residual strengthwithmagnitude between the two levels fail right away and the reliability drops sharply by |∆R|. Fig. 1. Typical expression of reliability degradation of composites under two-stage loading The purpose of this paper is to extend the application of the (eocp)model for single-stage fatigue loading to two-stage cases, using a piecewise combina- tion with n2a or |∆R| to describe the whole picture of dynamical reliability. The reliability in the high-low loading process can be divided into three sec- tions: a high stress section, a transition period, and a low stress section. A modification equation of the parameter c for the low stress section of the high-low loading is proposed to get better fitting of themodelwith the fatigue failure data. In the low-high case, it initially follows the behavior of low-level stress situationuntil the stress adjusting, thenwith a simultaneous drop |∆R|, it degrades as the case at high-level stress afterwards. Miner’s rule provides a simple way to predict the fatigue life of materials under a staged fatigue loading; nevertheless, it does not address the effect of the load sequence on the fatigue life of the composite. Here, based on the dynamical reliability, we 234 C.-L. Chen, K.-S. Wang present a way to estimate the linear damage sum in large populations of com- posites under various two-stage fatigue loading processes. Thepresent study is the first to describe accurately the dynamical reliability of composites under a two-stage fatigue loading and explains the effect of stress level, instant of ad- justment and load sequence on the linear damage sumof compositematerials. 2. Piecewise hazard rate function and linear damage sum 2.1. (eocp) Model By definition, the hazard rate h(t) is related to the reliability R(t) as follows h(t)=− 1 R(t) dR dt (2.1) In a deteriorating system, the reliability R(t) degrades monotonically with time t, thus R corresponds to t in a one-to-one relationship. This leads the time-dependenthazard rate function h(t) tobeexpressed in termsof reliability R as h(R). Wang et al. (2002) proposed a reliability-dependent hazard rate function, named the (eocp) model, in the form of h(R)= eo+ c(1−R) p eo > 0, c> 0, p> 0 (2.2) where eo is defined as the imbedded decay factor which takes account of the intrinsic defects during themanufacturing of themechanical elements. Thepa- rameter c represents the process-dependent decay factor which is concerned with the rate of damage accumulation ofmaterials under loading. A larger va- lue of c represents a larger hazard rate resulting from the higher fatigue stress level or other types of heavier mechanical loading. The parameter p denotes the beginning of noticeable degradation in reliability, referring to thememory characteristic of the damage. Assume the static strength of composite mate- rials to have a two-parameter Weibull distribution, as in the widely accepted cases.When the composites are subjected to a constant-amplitude maximum cyclic stress S at a certain stress ratio and a certain frequency, the correspon- ding values of (eo,c,p) can be obtained by fitting Eq. (2.2) with the fatigue failure data. It is found that in the region of high cycle fatigue of composites that eo and p can be taken to have a fixed value while c is correlated as a power relation for the ratio S/β as follows (Chen et al., 2009) c= ε (S β )λ (2.3) Piecewise reliability-dependent hazard rate... 235 where β is the scale parameter of the Wiebull static strength distribution; ε and λ are related to the initial material characteristics. To express the reliability of a composite under constant-amplitude cyc- lic stress as a function of fatigue cycles n, R(n), the mean cycles to failure (MCTF) of composite specimens can be calculated by integrating R(n) N = ∞ ∫ 0 R(n) dn (2.4) Replacing t with n and substituting Eq. (2.1) into Eq. (2.4) allows N to be N =− 0 ∫ 1 1 h dR=− 0 ∫ 1 1 eo+ c(1−R)p dR (2.5) Let 1−R=F,−dR= dF. It leads the above integration to be N = 1 ∫ 0 1 eo+ cFp dF = 1 eo 1 ∫ 0 1 1+ c eo Fp dF = 1 eo 1 ∫ 0 ∞ ∑ k=0 ( c eo Fp )k dF = 1 eo ∞ ∑ k=0 1 ∫ 0 ( c eo )k Fpk dF = 1 eo ∞ ∑ k=0 ( c eo )k 1 pk+1 +Ci (2.6) where Ci is the constant of integration. To save the work of integrating the above equation, an approximated equation of fatigue life (Shih, 2000) is pro- posed in terms of c/eo as cµ= ρ1 ( c eo )υ1 +ρ2 ( c eo )υ2 (2.7) where µ is the approximated mean fatigue life of the composite under the maximum cyclic stress S; the other parameters ρ1, ρ2, υ1 and υ2 are given in tables. 2.2. Modification of parameter c in high-low loading Consider a two-stage fatigue loading process in compositematerials, where S1 represents the first stage maximum cyclic stress, and S2 the second stage. Let the reliability at the instant of load adjusting be Ra. Denote eo, c1, p as the parameters fitted in the (eocp) model for S1, and eo, c2, p for S2. Thus the variation of the hazard rate under various stress levels can bemainly 236 C.-L. Chen, K.-S. Wang determined by the ratio S/β which appears in the representation of c, as shown in Eq. (2.3). In the high stress section of the high-low loading process, the residual strength of the survivals will degrade at the same rate as in a single-stage loading process at high-level stress, in other words, the same as the reliability does. In this section, the process-dependent decay factor c1 is decided by the high-level stress S1. Right after the high-low adjustment, the reliability remains at Ra during the transition period n2a. After the transition period, c2 is basically decided by the low-level stress S2. However, the survivals after the high-stress section and the transition period shouldhave experiencedmore cumulative damage than those specimens under a single-stage loading at low- level stress.Thus the residual strengthwill degrade further after the transition period. As amatter of fact, c2 is replaced by c ′ 2 as in c′2 = η(n2a,S1,S2)c2 (2.8) where c2 is given for a single-stage loading at low-level stress S2, η is a function of S1,S2 and n2a formodifying c2 in the low stress section of a high- -low loading process.Themodification for c2 indicates the hiddendegradation which exists in composites under the load S2 in the free-failure period n2a. Thus, η should be larger than unity; a longer n2a implies a larger η. It can be seen in Eqs. (A.1)-(A.4), for certain composite laminates with specific values of α,β,K, b,ω, d and αf,n2a is a function of S1,S2 and Ra. For fixed values of S1 and S2, n2a increases monotonically with the decreasing Ra, thus c ′ 2 can be further reduced to c′2 = η(Ra)c2 (2.9) To obtain a better fit for the low stress section of a high-low loading process, η(Ra) is proposed to modify c2 as in η(Ra)= 1+ζ (1−Ra Ra )γ (2.10) where ζ and γ are related to the material characteristics of composites. 2.3. Piecewise combination of hazard rate function For the low-high situation, the reliability in the first section (R ­Ra) is described by the hazard rate with (eo,p,c1) under low-level stress conditions. The moment the stress level is increased from S1 to S2, failure occurs right away in those survival specimens of which the residual strengths are between S1 and S2 inmagnitude.Thus, the reliability instantlydropsby |∆R| (seeEqs. Piecewise reliability-dependent hazard rate... 237 (A.5)-(A.7)). The remaining specimens after the reliability dropare considered to have experienced nearly the cumulative damage as those specimens having experienced a single-stage process at high-level stress. Thus, the reliability after the reliability drop follows the hazard rate as described by (eo,p,c2) for a single-stage under high-level stress. Thus, the hazard rate in the first stage is h1 = eo+ c1(1−R) p 1>R>Ra (2.11) and in the second stage it is h2 = eo+c2(1−R) p R S2. As can be seen in Fig.1, themean fatigue cycle for the process includes three parts: for the high stress section n1,HL = n1,HL ∫ 0 R(n) dn (2.15) where n1,HL is the number of applied cycles in the first stage of the high-low loading process; for the transition period n2a,HL =Ran2a (2.16) The mean fatigue cycle of low-level stress loading after the transition period is n2b,HL = ∞ ∫ n1,HL+n2a R(n) dn (2.17) 238 C.-L. Chen, K.-S. Wang The total mean fatigue cycles of the complete process becomes n1,HL + +n2a,HL+n2b,HL. (b) For the low-high loading process, S1 < S2. As shown in Fig.1, the mean fatigue cycle of the process includes two parts. Themean fatigue cycles of the first part is n1,LH = n1,LH ∫ 0 R(n) dn (2.18) where n1,LH is the number of applied cycles in the first stage of the low- -high loadingprocess.Right after the low-highadjustment, the reliability drops by |∆R|. In the second part we have n2,LH = ∞ ∫ n1,LH R′(n) dn (2.19) where R′(n) is the part of R(n) in the range of (Ra − |∆R|,0). The total mean fatigue cycles of the low-high loading process is n1,LH +n2,LH. (c)For composites ina two-stage fatigue loadingprocess, the lineardamage sum is Dm = n1 N1 + n2 N2 (2.20) where n1 and n2 are the mean fatigue cycles for the periods under the stress levels S1 and S2, respectively; N1 and N2 are the corresponding mean cyc- les to failure. Substituting Eqs. (2.15)-(2.17) into Eq. (2.20) yields the linear damage sum for the high-low loading process DHL = n1,HL N1 + n2b,HL N2 + n2a,HL N2 (2.21) According to Miner’s rule, the sum of the first two terms becomes unity; the third term yields the total sum that is larger than unity. Similarly, the linear damage sum for the low-high loading process is DLH = n1,LH N1 + n2,LH N2 (2.22) where n2,LH, Eq. (2.19), is smaller than the integral ∫ ∞ n1,LH R(n) dn due to the existence of a drop in the reliability. Thus, Miner’s damage sum for this case is smaller than unity. Piecewise reliability-dependent hazard rate... 239 3. Curve fitting with failure data in simulation Based on the residual strength equations by Yang and Jones (1980, 1981, 1983), this study uses MATLAB package to carry out Monte Carlo simula- tions of the residual strength degradation and fatigue failure for ISO standard [±45]S glass/epoxy laminates under single-stage and two-stage loading. There are 16 loading cases as shown in Table 1. Table 1. Cases of Monte Carlo fatigue loading simulation for G1/Ep[±45]S laminate Case Constant- Case High-to-low Case Low-to-high No. -amplitude S No. S1 S2 No. S1 S2 1 75.5 7 75.5 56.6 12 56.6 75.5 2 70.8 8 70.8 56.6 13 56.6 70.8 3 66.6 9 66.6 56.6 14 56.6 66.6 4 62.9 10 62.9 56.6 15 56.6 62.9 5 59.6 11 59.6 56.6 16 56.6 59.6 6 56.6 units: MPa The stress ratio of cyclic loading is set to be 0.1 for various stress levels. The loading frequency is assumed to be proportional to 1/S2 so that over- heating of the specimens is avoided. The associated parameters used in the simulations are α=59.8, β =113.26, K =1.2E-25, b=11.1806, ω =4.9633 and r = 12.9238 (Philippidis and Passipoularidis, 2007). The values of para- meters (eo,c,p) for a single-stage fatigue loading under S = 75.5, 56.6, 45.3 and37.8MPa (i.e., the ratios β/S =1.5, 2.0, 2.5 and 3.0), respectively, are ob- tained inChen et al. (2009). The specific parameter values are eo =1E-12 and p=0.84. Also, the parameters in Eq. (2.3) are ε=0.079246 and λ=11.378. Since the range of themaximumcyclic stress S =75.5-56.6MPa considered in this paper is within the range S = 75.5-37.8MPa considered in the previous paper of the authors, thus the values of eo, p, ε and λ are the same as above. The simulation procedure of strength degradation and reliability decay in each two-stage fatigue loading case is: (1) Generate randomly a total of 104 samples with the static strengths ha- ving a two-parameter Weibull distribution. (2) Compare each sample strength with the maximum cyclic stress S. The specimens with strength >S are deemed as survivals, and the others as 240 C.-L. Chen, K.-S. Wang failures. The value of S is fixed in each stage loading process. The value of S is adjusted at the specified loading cycles (or specified reliability). (3) Calculate the reliability and hazard rate of composite versus the number loading cycles according to the associated definition in engineering. (4) Calculate the residual strength XS(n) of the survivals individually by Eq. (A.4) after each time of simulation with specified additional loading cycles. Repeat the steps (2)-(4) until all specimens fail. 4. Results and discussion It can be seen in Fig.2 that the fitted curves of the (eocp) model correspond to the simulated data for stress at 75.5, 70.8, 66.6, 62.9, 59.6 and 56.6MPa, respectively. The fitted values of (eo,c,p) andMCTFunder these stress levels are summarized in Table 2. e0 and p remain unchanged and c increases with decreasing β/S. Fig. 2. Curve fitting of the (eocp) model for simulated fatigue data for G1/Ep[±45]S laminate under various constant-amplitudemaximum cyclic stresses S Figure 3 shows that the comparison between the predicted mean fatigue cycles in the transition period n2a,HL and the simulated data under various high-low loading conditions is satisfactory. As shown in Eq. (A.1), for fixed values of S2 and Ra, the larger S1 the larger value of n2a. For fixed values of S1 and S2, n2a increases monotonically with the decrease of Ra to a finite value. Thus, as shown in Fig.3, n2a,HL, the product of n2a and Ra, increases steeply at the beginning, and quickly approaches a peak near Ra =0.9, then decreases gradually afterwards. Piecewise reliability-dependent hazard rate... 241 Table 2.Fitted eo, p, c andmean cycles to failure forG1/Ep[±45]S laminates under various stress conditions S β/S eo p c N by (2.4) [MPa] [cycle] 75.5 1.5 1E-12 0.84 7.90E-4 7742 70.8 1.6 1E-12 0.84 3.78E-4 15984 66.6 1.7 1E-12 0.84 1.88E-4 31896 62.9 1.8 1E-12 0.84 9.83E-5 60797 59.6 1.9 1E-12 0.84 5.33E-5 1.1029E+5 56.6 2.0 1E-12 0.84 2.94E-5 1.9627E+5 Fig. 3. Comparison between the predictedmean fatigue cycle in the transition period and the simulation data under various high-low fatigue loading adjustments Figure 4 shows that the typical piecewise fitting of the (eocp) model for the simulated data for the hazard rate versus reliability in a high-low process, adjusted from S1 = 66.6MPa to S2 = 56.6MPa at Ra = 0.5 is satisfactory. It is evident that the hazard rate rises at a relatively higher rate in the high stress section anddrops suddenly to zeroat the instant of high-lowadjustment, Ra = 0.5. After the transition period, the reliability degrades from 0.5 and hazard rate continues to increase from a value lesser than that right before Ra =0.5. The slop of the hazard rate appears lesser in the low stress section than in the high stress section. Figure 5 depicts the step-by-step piecewise fitting of the reliability for the corresponding conditions in Fig.4. Figure 5a shows the fitted curves under single-stage S =66.6MPa,where the shaded area represents themean fatigue cycles n1,HL in 1­R> 0.5. Figure 5b shows the fitted result under single- stage S=56.6MPa, where the shaded area indicates themean fatigue cycles 242 C.-L. Chen, K.-S. Wang Fig. 4. Typical piecewise fitting of the (eocp) hazard rate model for simulated fatigue data for G1/Ep[±45]S laminate under high-low loading conditions, from S1 =66.6MPa to S2 =56.6MPa at Ra =0.5 Fig. 5. Typical piecewise fitting of the (eocp) model for high-low simulation data for G1/Ep[±45]S: (a) under S=66.6; (b) under S=56.6; (c) adjusted from S1 =66.6MPa to S2 =56.6MPa at Ra =0.5, with c2 =2.94E-5 n2b,HL in 0.5>R­ 0. Figure 5c shows the over-all picture for the high-low loading process including the transition period. The fitted reliability curves agree with simulation data except for the tail of the low stress section, say Piecewise reliability-dependent hazard rate... 243 0.2>R.Obviously, there is an increase ofmean fatigue cycles in the transition period, i.e. n2a,HL, but a decrease in the low stress section. The fitted curve of reliability is little higher than data in the tail part, thus it needs an additive modification in the parameter c2 for better fitting. Figure 6presents thedegradation of themean residual strengthof survivals in thehigh-low loadingprocess, byEq. (A.4), over the reliability. It canbe seen that the mean residual strength in the high-stress section (R> 0.5) complies with that in the single-stage process with S =66.6MPa. The zoom-out view around the adjustment shows that the mean residual strength is smaller in the low stress section than that in the single-stage process at low-level stress S = 56.6MPa. Thus, a modification of cumulative nature is needed as the loading process is adjusted from high-level to low-level stress. Fig. 6. Variation in the mean residual strength of survivals over the reliability for composites under constant-amplitude cyclic stresses and the high-low loading process shown in Fig.4 Figure 7 depicts the even better piecewise fitting for the same case as in Fig.5. It results from the increasing modification of c2 in Eq. (2.10) with ζ = 0.167 and γ = 2, which are obtained by fitting the simulated fatigue failure data for every Ra, 10% apart, in 0.9­Ra ­ 0.1. The increase ofmean fatigue cycles in the transition period appears larger than the decrease in the low stress section. The damage sum DHL calculated by Eq. (2.18) is 1.031. Figure 8 shows the typical piecewise fitting of the (eocp) hazard rate func- tion as given by Eqs. (2.11) and (2.12), for the low-high simulation data, adjusted from S1 = 56.6MPa to S2 = 66.6MPa at Ra = 0.5. As shown in this figure, except for the abrupt rise at the instant of low-high adjustment, the piecewise fittings are satisfied. It is obvious that the hazard rate is higher in the section of S2 =66.6MPa than that in the section of S1 =56.6MPa. 244 C.-L. Chen, K.-S. Wang Fig. 7. Piece-wise fitting of the (eocp) model for high-low simulated data for G1/Ep[±45]S, from S1 =66.6MPa to S2 =56.6MPa at Ra =0.5, with c ′ 2 =3.5E-5 Fig. 8. Typical piecewise curve fitting the (eocp) hazard rate function for simulated fatigue data for G1/Ep[±45]S laminate under low-high loading adjustment, from S1 =56.6MPa to S2 =66.6MPa at Ra =0.5 Figure 9 displays the piecewise representation of the reliability for the corresponding conditions in Fig.8. Figure 9a shows the fitting under S1 = 56.6MPa. The shaded area indicates the mean fatigue cycles n1,LH for 1­R> 0.5. Figure 9b shows the fitting under S2 =66.6MPa, where the shaded area indicates the mean fatigue cycles n2,LH for R < (0.5− |∆R|). The area under the fitted curve from R = 0.5 to (0.5− |∆R|) denotes the decrease of the mean fatigue cycles at the low-high loading adjustment. As shown in Fig.9c, the comparison between the piecewise fitted curves and the simulation data is satisfactory. Thedamage sum DLH calculated byEq. (2.22) is 0.975. Piecewise reliability-dependent hazard rate... 245 Fig. 9. Typical piecewise fitting of the (eocp) model for low-high simulation data for G1/Ep[±45]S laminate: (a) under S=56.6MPa; (b) under S=66.6MPa; (c) adjusted from S1 =6.6MPa to S2 =66.6MPa at Ra =0.5 Figure 10 depicts the variation in the damage sums when all composite specimens fail under two-stage fatigue loading with various values of Ra. As shown in Fig.10a, the damage sum DHL obtained from Eq. (2.21) is gre- ater than unity under high-low fatigue loading. This value approaches a peak when Ra is near 0.9.With S2 =56.6MPa, the larger S1 the larger DHL. As commented on Fig.7, the positive deviation from unity is mainly due to the term n2a,HL/N2 in Eq. (2.21). Hence, the trend of variation of DHL over Ra complies with that of n2a,HL, as shown in Fig.3. It can be seen in Fig.10b that DLH is smaller than unity for composites experiencing the low-high fa- tigue loading process. As commented on Fig.9b, the negative deviation from unity results from the decrease in the mean fatigue cycles from R = Ra to (Ra−|∆R|). This deviation decreases to the lowest level when Ra around 0.9. With S1 =56.6MPa, a larger S2 leads to a smaller DLH. This paper presents an easymethod to describe accurately the overall dy- namical reliability of composites under two-stage fatigue loading processes by 246 C.-L. Chen, K.-S. Wang Fig. 10. Variation in the linear damage sum from the (eocp) model for simulation data over the reliability at loading adjustment for: (a) high-low cases; (b) low-high cases a simplemethod of piecewise combination of the (eocp)model. The derivation of the transition period and reliability drop is a pioneer research concerning the effect of fatigue loading adjustment on the dynamical reliability and linear damage sum of composites. The transition period can also be applied in the stress screening of newly developed products of composite materials. The po- sitive and negative deviation of the linear damage sum from unity in high-low and low-high loading, respectively, correspondswith the results ofmostprevio- us researches of the load sequence effect (Broutman andSahu, 1972; Yang and Jones, 1980, 1981, 1983; Gamstedt and Sjögren, 2002; Found andQuaresimin, 2003). Furthermore, this paper shows how and howmuch the stress level and instant of adjustment affect the linear damage sum of composites. The above results can be helpful for the designing and maintenance of the structure of composite materials. 5. Conclusions Basedon the (eocp)model forfinding thehazard rate, thefitted reliabilities for a single-stage loading process are successfully extended to cases of two-stage loading in combinationwith the predicted transition period or reliability drop. A better fit can be obtained for the process-dependent decay factor c2 when c′2 is replacedwith amodification for the second stage, especially for ahigh-low fatigue process.Although the failure does not occur during n2a, the imbedded strength degradation still continues. Piecewise reliability-dependent hazard rate... 247 As all specimens fail, the linear damage sum is observed to be larger than unity in the high-low loading process, and smaller than unity in the low-high cases. The sums always rise to a peak near Ra =0.9 for high-low cases, and fall to a low value for low-high cases. With a fixed low-level maximum cyclic stress, the deviation of the fatigue damage sum from unity becomes larger as the high-level stress increases. Appendix The transition period at the high-low fatigue loading adjustment is expressed as n2a = (Sω1 −S ω 2 ) [ Sr2 −β r(− lnRa) r α ] βrKSb2 [ Sω2 −β ω(− lnRa) ω α ] (A.1) where α and β are the shape parameter and scale parameter of the Weibull static strength distribution of composites. K and b are the parameters in the S-N curve equation, KSbN∗ = 1, where N∗ is the characteristic fatigue life associated with S. r = α/αf is the ratio of α to the shape parameter αf of the distribution function of the fatigue life N (Yang and Jones,1980, 1981, 1983) P[N ¬n] =      1−exp { − [ n N∗ + (S β )r]αf } for n­ 0 0 for n< 0 (A.2) ω is the degradation rate parameter in the residual strength equation XωS(n)=X ω(0)− Xω(0)−Sω Xr(0)−Sr βrKSbn (A.3) where X(0) is the random static strength, and XS(n) is the random residual strength after n cycles under S. For a two-stage fatigue loading process, the equation of residual strength is XωS1+S2(n1+n2)=X ω(0)− Xω(0)−Sω1 Xr(0)−Sr1 βrKSb1n1− Xω(0)−Sω2 Xr(0)−Sr2 βrKSb2n2 (A.4) where XS1+S2(n1 +n2) is the random residual strength after n1 cycles un- der S1 plus n2 cycles under S2. 248 C.-L. Chen, K.-S. Wang The reliability drop at the low-high fatigue loading adjustment is |∆R|=exp [ − (x1 β )α] −exp [ − (x2 β )α] (A.5) where x1 is the static strength of the specimens the residual strength of which degrades to S1 at n1,LH cycles under S1; and x2 the static strength of the specimens the residual strength of which degrades to S2 at n1,LH cycles un- der S1. The static strength x1 is in the form x1 =(n1,LHβ rKSb1+S r 1) 1 r (A.6) and x2 can be obtained by solving the following equation numerically xr+ω2 −S ω 2x r 2− (S r 1 +Kβ rSb1n1,LH)x ω 2 +S ω 2S r 1 +Kβ rSω+b1 n1,LH =0 (A.7) References 1. BroutmanL.J., SahuS.A., 1972,Anew theory topredict cumulative fatigue damage, [In:] Fiberglass Reinforced Plastics, Composite Materials: Testing and Design (Second Conference), ASTM STP 497, 170-188 2. Chen C.L., Tasi Y.T., Wang K.S., 2009, Characteristics of reliability- dependent hazard rate for composites under fatigue loading, Journal of Me- chanics, 25, 2, 117-126 3. ChenC.L.,WangK.S., 2011, Effects of loading adjustment on the reliability degradation of composites, Science and Engineering of Composite Materials, 18, 1/2, 61-67 4. Found M.S., Quaresimin M., 2003, Two-stage loading of woven carbon fi- ber reinforced laminates, Fatigue and Fracture of Engineering Materials and Structures, 26, 17-26 5. Gamstedt E.K., Sjögren B.A., 2002, An experimental investigation of the sequence effect in block amplitude loading of cross-ply composite laminates, International Journal of Fatigue, 24, 2/4, 437-446 6. HanK.,HamdiM., 1983,Fatigue life scattering ofRP/C,38thAnnualRP/CI 7. Hwang W., Han K.S., 1986, Cumulative damage models and multi-stress fatigue life prediction, Journal of Composite Materials, 20, 125-153 8. Ni K., Zhang S., 2000, Fatigue reliability analysis under two-stage loading, Reliability Engineering and System Safety, 68, 153-158 Piecewise reliability-dependent hazard rate... 249 9. Philippidis T.P., Passipoulardidis V.A., 2007, Residual strength after fa- tigue in composites: theory vs. experiment, International Journal of Fatigue, 29, 12, 2104-2116 10. Shih Y.C., 2000, Study of the relationship between the cumulative failure and system reliability,Master thesis, National Central University, Taiwan, R.O.C. 11. Tanaka S., Ichikawa M., Akita S., 1984, A probabilistic investigation of fatigue life and cumulative cycle ratio,Engineering Fracture Mechanics, 20, 3, 501-513 12. Wang K.S, Shen Y.C., Huang J.J., 1997, Loading adjustment for fatigue problem based on reliability consideration, International Journal of Fatigue, 19, 10, 693-702 13. Wang K.S., Hsu F.S., Liu P.P., 2002, Modeling the bathtub shape hazard rate function in terms of reliability,Reliability Engineering and System Safety, 75, 397-406 14. Wang K.S., 2011, Study of hazard rate function on the cumulative damage phenomenon, Journal of Mechanics, 27, 1, 47-55 15. Yang J.N., Jones D.L., 1980, Effect of load sequence on the statistical fati- gue of composites,AIAA (American Institute of Aeronautics andAstronautics) Journal, 18, 12, 1525-1531 16. Yang J.N., Jones D.L., 1981, Load sequence effects on the fatigue of unnot- ched composite laminates, [In:] Fatigue of Fibrous Composite Materials, Lau- raitis K.N. (Edit.), ASTM STP 723, Philadelphia, 213-232 17. Yang J.N., Jones D.L., 1983, Load sequence effects on graphite/epoxy [G35]2s, [In:]Long TermBehavior of Composites, O’BrienT.K. (Edit.), ASTM STP 813, Philadelphia, 246-262 Konstrukcja funkcji ryzyka uszkodzeń kawałkami zależnej od niezawodności dla kompozytów poddanych różnym scenariuszom obciążenia zmęczeniowego Streszczenie W oparciu o wyznaczony okres przejściowy i spadek niezawodności, artykuł pre- zentujemetodę określania funkcji ryzyka uszkodzenia kawałkami zależnej od poziomu niezawodności, zwanej (eocp) i służącej do modelowania dynamicznej niezawodności dla dwustanowych procesów obciążania zmęczeniowego. Na poczatku, parametry eo, c, i p dopasowano do danych otrzymanychw drodze symulacji uszkodzeń pod wpły- wem działania cyklicznych naprężeń o kilku stałych amplitudach. Niezawodność dla 250 C.-L. Chen, K.-S. Wang obciążeń przechodzących od dużej amplitudy do małej opisano kawałkami zależny- mi od poziomuprzykładanych naprężeń i odpowiadającymi imwartościamieo, c, i p. Wynosi ona Ra wokresieprzejściowym,gdzie Ra jest niezawodnością,przyktórej po- ziom naprężeń jest zmieniany. Niezawodność przy obciążeniu rosnącymwyznaczono, odejmując część jej spadku przy Ra od kawałkami dopasowanych krzywych. Zapro- ponowany sposób opisu niezawodności sukcesywnie weryfikowano. Zaobserwowano, że liniowa suma uszkodzeń przekracza jedność dla scenariusza obciążeń stopniowo malejących i nie osiąga tej wartości w przypadku przeciwnym.Większe różnice w po- ziomach obciążeń skutkowaływwiększych odstępstwach liniowej sumy uszkodzeń od jedności. Szczególnie duże zauważono dla Ra =0.9 Manuscript received January 20, 2011; accepted for print April 3, 2011