June 2009/ John Wægter Note 5.1 Stress range histories and Rain Flow counting Introduction......................................................................................................................................2 Stress range histories........................................................................................................................3 General .........................................................................................................................................3 Characterization of irregular fatigue loading ...............................................................................4 Stress range histogram .................................................................................................................5 Long-term exceedance diagram ...................................................................................................5 Stochastic processes.....................................................................................................................7 Rain flow counting...........................................................................................................................9 Cycle counting .............................................................................................................................9 Basic principle in Rain Flow Counting......................................................................................10 References......................................................................................................................................12 1 Stress range histories and Rain Flow counting Introduction In the previous sections we have considered most of the elements of a fatigue design. The starting point has been a given structure and a known loading from which we are able to construct a relevant stress history as basis for the fatigue design. When we use the nominal stress approach, a particular type of detail is assigned to a particular fatigue class with a given S-N curve. The design stress is determined as the nominal stress adjacent to the weld under consideration and can often be determined using beam theory. Alternatively, we may determine the nominal stress based on the results from a coarse mesh FE analysis. In some cases the effects of macro-geometric features (e.g. a hole in a plate) must be considered so that the relevant stress for fatigue design is σ local = SCFmacro ⋅ σ nom (1) where SCFmacro is the stress concentration from macro-geometric effects. When we use the hot spot stress approach, we determine a local stress at the hot spot that considers the influence of the structural discontinuities due to the geometry of the joint, but excludes the stress raising effects (notch) from the weld itself. Hot spot stresses are in principle structural stresses (geometric stresses) based on the theory of shells determined on the surface at the hot spot of the component which is to be assessed. When the stress concentration is captured in simple tubular joints using parametric stress concentration formulas we apply the relationship given in Equation (2) taking σ nom as the beam stresses in the brace considered. σ hotspot = SCF par ⋅ σ nom (2) where SCF par is the stress concentration from parametric stress concentration formulas. Alternatively, we may determine the hot spot stress using FE analysis, based on shell or solid elements. The nominal stresses in Equation (2) are then the stresses without considering the stress raising effect due to the joint geometry, e.g. the rightmost stresses in Figure 1. Since the SCF in Equation (2) in case of membrane stress σ m differs from the SCF in case of bending stress σ b , the total stress is in principle given by a combination of stresses as indicated in Equation (3). σ hotspot = SCFm ⋅ σ m + SCFb ⋅ σ b (3) 2 Figure 1 Determination of hot spot stress For some structural details with pronounced plate bending, load shedding may justify a reduction in the contribution from the bending stresses to the hot spot stresses in Equation (3). Typically the hot spot stresses (Figure 1) will be taken as either the extrapolated principal stresses, or the extrapolated resulting normal to weld stresses found from the FE analysis. In the above considerations we have assumed that a relevant stress history for the fatigue design was available. In the following sections we will focus on the details of establishing a representative stress history and use the Rain Flow Counting Method as a basis for the calculation of the fatigue damage. Stress range histories General Real life structures prone to fatigue will experience stresses that vary with time, often in a very complicated manner. These stress histories are generally the results of an irregular load history and the dynamic response of the structure considered. In some cases the dynamic response is negligible and the response becomes quasi-static and the stress history closely follows the external loading. In other cases the dynamics influence the stress history greatly. In both cases the basis for the fatigue calculation is the long-term stress distribution in the detail considered. The methods considered apply for high cycle fatigue design, which for convenience may be defined as a fatigue design for more than 104 cycles. The distinction between high cycle fatigue and low cycle fatigue is reflected in most design standards as high cycle fatigue design standards normally give S-N curves starting with N = 104 cycles. For low cycle fatigue design the stress range concept is not immediately valid, since typically the low cycle fatigue strength is governed by (large inelastic) strains. 3 Characterization of irregular fatigue loading For a typical offshore structure the fatigue load history spans a period of 20 years corresponding to about 108 wave load cycles (assuming an average wave load period of 6 sec.). Figure 2 shows the terminology used for irregular loading histories, and the most important concepts are explained below. For practical design the time history must be reduced to a manageable format still retaining the characteristics of the loading, and the stress history must be broken down into individual stress ranges with an associated number of stress cycles. Different approaches for this purpose will be discussed in the following. Figure 2 Definition of terms related to irregular load histories The most important concepts are Reversal the derivative of the load-time history changes sign. Peak the derivative of the load-time history changes from positive to negative sign. Valley (or trough) the derivative of the load-time history changes from negative to positive sign. Range is the algebraic difference between successive valley and peak loads (positive range) or between successive peak and valley loads (negative range). Sometimes the range may have different definitions depending on the counting method used. Mean crossing or zero crossing, is the number of times that the load-time history crosses the mean load level during a given duration of the history. Normally only crossings with positive slopes are counted. 4 Stress range histogram The stress range histogram also denoted the stress range spectrum, is a representation of stress ranges and the associated number of cycles. It may directly be used as the basis for the fatigue damage calculation, see Figure 3. In practice the spectrum is approximated by a manageable number of stress range blocks characterized by ( Δσ i , ni ) . For 108 cycles or more, 20 stress levels will normally be sufficient, while a lower number of stress range blocks may be sufficient for a smaller number of cycles. All cycles in a stress range block should be associated with the mean of the stress ranges in the block. Figure 3 Simplification of stress range spectrum The damage induced by each stress range block is determined as its contribution to the Miner’s sum and is found from ⎛n⎞ ΔDi = ⎜ ⎟ ⎝ N ⎠i (4) based on the appropriate S-N curve and Δσ i . Long-term exceedance diagram In the following the distribution of stress ranges from wave load will be derived based on a simplified long-term exceedance diagram for wave heights. It has been found that the distribution of the individual wave height is close to linear if H (wave height) is plotted as a function of logN (N is the number of waves exceeding H), see Figure 4. 5 Wave height H100 log N log N100 Figure 4 Simplified long-term exceedance diagram for wave heights In Figure 4, H100 is a wave height with a return period of 100 years, which means that H100 on average is only exceeded once every 100 years. N100 is the total number of waves in 100 years. The relationship between H and N can be found to be H= − H100 ⋅ log N + H100 log N100 ⎛ log N ⎞ H = H100 ⋅ ⎜ 1 − ⎟ ⎝ log N100 ⎠ (5) Alternatively the following relationships apply H − H100 = − H100 ⋅ log N log N100 ⎛ H ⎞ log N = log N100 ⋅ ⎜ 1 − ⎟ H100 ⎠ ⎝ N = N100 ⎛ H ⎞ ⎜ 1− ⎟ ⎝ H100 ⎠ (6) In Equation (6) above N is thus the number of waves exceeding a given wave height H during 100 years. If we now assume that the wave force is proportional to the wave height (true for inertia load dominated wave load), and the corresponding wave induced stress is proportional to the wave force, then the distribution in Figure 4 could be turned into a long term exceedance diagram for wave load induced stress range simply by scaling the values of the Y-axis to produce Δσ instead of H. These 6 considerations may be generalized to consider a general long-term exceedance diagram for stress range Δσ , as shown in Figure 5. Δσ ref is a stress range only exceeded once during the reference period, and N ref is the total number of stress ranges during the reference period. In analogy with Equations (5) and (6) the following relations apply ⎛ log N Δσ = Δσ ref ⋅ ⎜ 1 − ⎜ log N ref ⎝ N = N ref ⎛ Δσ ⎜ 1− ⎜ Δσ ref ⎝ ⎞ ⎟⎟ ⎠ (7) ⎞ ⎟ ⎟ ⎠ (8) In Equation (8) N is the number of stress ranges exceeding a given stress range Δσ during the reference period. Stress range Δσ ref 40 20 log N 0 N= Figure 5 1 1 N 40 2 1 10 10 N 20 log N ref 2 Long-term exceedance diagram for stress range From Figure 5 it can be seen, as an example, that N 20 stress range values exceed a stress range of 20, while only N 40 stress range values exceed a stress range of 40. We can therefore conclude that there are ( N 20 − N 40 ) stress range values between a stress range of 20 and 40. Continuing along these lines, we can turn the information given in Figure 5 into an equivalent stress range histogram with blocks similar to Figure 3. Stochastic processes In practical design work the stress history is often found to be irregular and in some cases also random. The stress range history can therefore be considered a stochastic process, where we describe the process by its statistical properties, i.e. mean value and standard deviation. 7 A process is said to be stationary if the statistical properties do not vary with time. Many processes may be considered stationary provided the time period considered is short enough. This is e.g. true for the sea surface elevation, which normally is considered stationary within time intervals of three to six hours. In the following we only consider stationary processes, and the value of the process at time t is denoted x(t). The probability density function p(x) is given by p( x ) ⋅ dx = prob( x ≤ x (t ) ≤ x + dx ) (9) The cumulative distribution function P(x) is given by x P ( x) = prob ( x(t ) ≤ x) = ∫ p( x) ⋅ dx (10) −∞ The expected value is given by ∞ E ( x ) = ∫ x ⋅ p ( x ) ⋅ dx (11) −∞ The expected value of the process is equal to the mean value of the process. Often the coordinate system is chosen so that the mean value is equal to zero. An example would be measuring the sea surface variation relative to the mean sea water level. The energy spectrum of the process S (ω ) can be found directly from sampled values of the process using Fast Fourier Transform (FFT). When performing FFT on a time signal we transform the process given by x(t) in the time domain into its equivalent representation in the frequency domain X (ω ) , from which we can derive the energy spectrum S (ω ) . It is normal to distinguish between narrow band processes and broad band processes, and to characterize a process with its energy spectrum S (ω ) as shown in Figure 6. S (ω ) S (ω ) ω Narrow band process Figure 6 ω Broad band proces Stochastic processes 8 A stationary stochastic process may be considered composed of infinitely many harmonic components, each of different frequency. Let us as an example consider a wave spectrum derived from the surface elevation x (t ) . The energy of a harmonic wave is proportional to the square of its amplitude, and the energy spectrum (wave spectrum) shows how this energy is distributed on the various frequency components as given below. It should be noted that the energy within a frequency band Δω is equal to that of a sine wave with amplitude ai as indicated on Figure 7. x (t ) t S (ω ) 1 2 ai2 = S (ωi )Δω ai Ti = Δω ω 2π ωi ωi Figure 7 A stochastic process x(t ) and its energy spectrum S (ω ) Rain flow counting Cycle counting With narrow band time series as shown in Figure 8a, individual stress cycles can easily be identified and counted, e.g. by counting one stress cycle for each zero crossing (with positive slope) and take the stress range as the difference between the peak and the valley values. In broad band time series, where large cycles are interspersed with small cycles with varying mean level, Figure 8b, the question of what is meant by a cycle and the corresponding stress range becomes less evident. It is therefore necessary to use a cycle counting method that in an 9 unambiguous way breaks the stress history down into individual cycles which can be summed up into a stress range distribution. σ (t ) σ (t ) t t a) Narrow band process Figure 8 b) Broad band proces Time series from stochastic processes Several methods have been proposed for general cycle counting of broad band processes, generally leading to different results in terms of accumulated fatigue damage. Cycle counting procedures that give the most correct physical representation of the fatigue process are therefore often preferred. For narrow band time series the choice of counting method seems less critical, and most generally accepted methods lead to quite similar results. For high cycle fatigue the most frequently used cycle counting methods are the Reservoir Counting Method and the Rain Flow Counting Method. As the Rain Flow Counting Method is the most frequently used cycle counting method in the industry, the following will focus on the details of this method. Basic principle in Rain Flow Counting The method is designed to count reversals in accordance with the material’s stress/strain relationship including hysteresis loops. Thus, complex loading situations with local stress reversals can be accounted for in a consistent way. The principle may be illustrated by the strain history given in Figure 9 and the corresponding stress-strain path. ε σ 4 4 2 2 2′ 3 1 Figure 9 3 t 1 2′ 2-3-2´ closed hysteresis loop ε Strain history and corresponding stress 10 The strain history in Figure 9 consists of a large stress range 1-4, with a small closed cycle 2-3-2´ embedded. A basic principle in rain flow counting is to count a cycle each time a hysteresis loop is closed. Thus, the total count in Figure 9 is a half cycle with stress range 1-4 plus a full cycle with stress range 2-3. To illustrate the principle further a more complicated stress history is considered in Figure 10. This example is an extension of the strain history given in Figure 9. Figure 10 Rain flow counting based on a more complicated strain history The rain flow method has originally obtained its name from an analogy of rain falling down a pagoda roof. The time series considered X (t ) is converted to a point process of peaks and troughs as shown in Figure 11, and the process is considered a sequence of roofs with rain falling on them. Rules for rain flow counting: 1. A rain flow is started at each peak and trough. 2. When a rain flow path started at a trough comes to a tip of the roof, the flow stops if the opposite trough is more negative than the one the flow started from. 3. For a flow started at a peak it is stopped by a peak which is more positive than the one the flow started from. 4. If the rain flowing down a roof intercepts flow from an earlier path, the present path is stopped. Figure 11 Rain flow counting of process X (t ) 11 The rain flow count applied to the process in Figure 11 results in • • half cycles of trough generated stress ranges half cycles of peak generated stress ranges In both cases the stress ranges are found as the projected distances on the stress axis. Figure 12 shows the details of the rain flow count of the process from Figure 11. Half cycles of trough generated stress ranges 1-8, 3-3a, 5-5a, 7-7a, 9-10, 11-12 and 13-14 Half cycles of peak generated stress ranges 2-3, 4-5, 6-7 10-12b, 12-12a and 8-(9)-13 Total count Full cycles: 2-3-3a, 4-5-5a, 6-7-7a, 9-10-12b(9) and 11-12-12a(11) Half cycles: 1-8, 8-13 and 13-14 Figure 12 Rain flow count details It should be noted that for X (t ) sufficiently long, it can be shown, Ref. /3/, that any trough originated half cycle will be followed by another peak originated half cycle for the same range. This is also the case for short stress histories if the stress history starts and ends at the same stress value. Although the rain flow counting method widely is considered superior to other counting methods for fatigue calculation, a basic criticism of the method as used above has been that the fatigue damage procedure cannot account for the sequence of the stress ranges. This criticism is justified especially since fatigue tests have revealed that the sequence of the stress ranges in some cases is of importance. However, the effect of sequence has been found to even out in fatigue calculations where many time histories are considered, and currently no better procedures are available. References Ref. /1/ Wægter, J.: Fatigue design based on S-N data. Ramboll Oil & Gas, Denmark. 2009. Ref. /2/ Almar-Næss et. al.: “Fatigue Handbook – Offshore Steel Structures”. 1985. Tapir, Norway. 12 Ref. /3/ Wirsching, P. H., Mohsen Shehata, A.: “Fatigue Under Wide Band Random Stresses Using the Rain-Flow Method”. Journal of Engineering Materials and Technology. July 1977. Ref. /4/ ESDEP WG 12 Fatigue. Lecture 12.1: Basic Introduction to Fatigue. http://www.kuleuven.ac.be/bwk/materials/Teaching/master/toc.htm. 13