GB1586491A - Apparatus for determining fatigue damage - Google Patents

Description

(54) IMPROVEMENTS IN OR RELATING TO APPARATUS FOR DETERMINING FATIGUE DAMAGE (71) We, ELLIOTT BROTHERS (LONDON) LIMITED, of Marconi House, New Street, Chelmsford, Essex CMl lPL, a British Company, do hereby declare the invention, for which we pray that a patent may be granted to us, and the method by which it is to be performed, to be particularly described in and by the following statement:- This invention relates to apparatus for determining fatigue damage.

The invention relates particularly to apparatus for determining fatigue damage of rotating parts of machines, more especially though not exclusively, aircraft engines.

The rotating parts of an aircraft engine vary in revolutions per minute (RPM) during a flight in response to pilots throttle of fuel control variations. After a large number of flights the resultant variation in stress can lead to a fatigue failure in a rotating part and thus produce a serious hazard to the aircraft.

To minimise such an occurrence, a life limitation, expressed in terms of reference fatigue cycles, is imposed on certain rotating parts of the engine. The life limitation is based on the results of rig tests of such parts under simulated operating conditions or on the evidence of satisfactory operation in service of similar parts.

These reference fatigue cycles, also referred to as major cycles, are defined as varying from zero stress to stress at 1000/, RPM at a specified operating condition.

During a typical sortie the RPM starts at zero, increases to a high value, with intermediate variations, and continues to vary throughout the flight before finally returning to zero when the flight is over.

The intermediate RPM variations are termed minor cycles and, although they individually produce less fatigue damage than the major cycles, their contribution cannot necessarily be ignored.

On post flight examination of a typical sortie pattern the minor cycles are identified by selecting first the cycle with the smallest difference between RPM turning points and continuing until all cycles up to the largest one have been accounted for.

The minor cycles are then summated into equivalent major cycles by using the following linear damage equation for the total sortie; Total sortie damage =

Where:-- Nmax is the highest speed achieved, thus representing the major cycle; NH and NL are the highest and lowest speeds associated with any minor cycle n; C is the number of minor cycles and K and P are constants whose values depend on the machine part whose fatigue damage is being measured.

This method of determining fatigue damage is hereinafter referred to as the definitive method.

In this way the effect of the major and all minor cycles can be calculated.

The present invention provides an apparatus for determining fatigue damage of a rotating part of a machine by the definitive method wherein the minor cycles are identified during a period of operation of the machine and the computation of fatigue damage due to minor cycles takes place during said period of operation leaving only the contribution of the major cycle to be calculated at the end of the period of operation.

The invention will now be further explained, and two arrangements in accordance with the invention will be described, by way of example, with reference to the accompanying drawings in which: Figures 1 and 5 show speed change patterns for an aircraft engine during a sortie; Figure 2 illustrates the operation of a store incorporated in an apparatus in accordance with the invention; Figures 3 and 4 are flow diagrams illustrating certain aspects of the operation of the apparatus; Figure 6 is a block schematic diagram of one of the arrangements to be described by way of example; Figure 7 is a diagram illustrating the operation of part of the arrangement of Figure 6; Figure 8 illustrates an RPM banding arrangement used in the arrangement of Figure 6; Figure 9 is a block schematic diagram of the other arrangement to be described by way of example; and Figure 10 is a diagram of circuitry used in interfacing either of the arrangements with an engine under test.

In an apparatus in accordance with the invention the turning points of the speed change pattern of the machine part under test (i.e. for an aircraft engine, the sortie pattern) are examined as they are generated to determine sufficient information to identify and thus evaluate most minor cycles at or soon after their completion.

If on arrival at a peak value (an NH) it is found to be higher than a previous NH then one or more minor cycles can be identified and evaluated.

Similarly, if on arrival at a low value (an NL) it is found to be lower than a previous NL then one or more minor cycles can be identified and evaluated. The algorithm applied to identify the cycles is as follows: (a) For Confirmed Peak Values (NH) Determine Whether the Current NH is Higher than or Equal to the Last Unresolved (Not Yet Identified with a Specific Cycle) NH? If, Yes, Pair the Last Unresolved NH with the Last Unresolved NL.

Keep Repeating the First Step Until the Answer is No. then Store Current NH as the Latest Unresolved NH.

(b) For Confirmed Low Values (NL) Determine Whether the Current NL is Lower than or Equal to the Last Unresolved NL? If Yes, Pair the Last Unresolved NH with the Last Unresolved NL.

Keep Repeating the First Step Until the Answer is No, then Store the Current NL as the Latest Unresolved NL.

Before considering an example one must image the existence of two small stores A and B, one for "unresolved N highs" (store B) and one for "unresolved N lows" (store A). When stored numbers are paired according to the rules above, they are removed from the stores in alast-in, first-out manner i.e. the stores operate as push down stacks.

Figure I shows a simple example of a sortie pattern. The definitive method of identification of cycles is shown by the horizontal dotted lines, i.e.

NH I with NL2 (minor cycle) NH2 with NL4 (minor cycle) NH3 with NL3 (minor cycle) NH5 with NL5 (minor cycle) NH4 with NLI (major cycle) In an apparatus in accordance with the invention these cycles are identified and fatigue damage. is computed by the following procedure: (i) Enter NLI into the unresolved low store A.

(ii) Is NH I higher than or equal to the last unresolved NH? Answer NO since the unresolved high store B is empty. Enter NHI into the unresolved high store B.

(iii) Is NL2 lower than or equal to the last unresolved N low? NO since NL2 is greater than NLI (zero). Enter NL2 into store A as the 2nd entry.

(iv) Is NH2 greater than or equal to the last unresolved N high? YES since NH2 is greater than NHl. Therefore the least unresolved high, NHl, and the last unresolved low NL2, can be paired as a minor cycle, the effective damage calculated and logged and NH 1 and NL2 removed from their respective unresolved stores.

Asking the question again this time gives a NO answer. NH2 is then loaded into store B as the only entry. Hence NHl and NL2 have been identified as a minor cycle and their damage effect logged at the point in time associated with NH2.

(v) Is NL3 lower than or equal to the last unresolved N low? NO since NL3 is not less than NLl. Enter NL3 into the unresolved low store as the last entry (now 2 values in store A).

(vi) is NH3 greater than or equal to the last unresolved high? NO since NH3 is not greater than NH2. Enter NH3 into the unresolved high store.

(vii) Is NL4 lower than or equal to the last unresolved low? YES since NL4 is less than NL3. Therefore the last unresolved high NL3 and the last unresolved low NL3 can be resolved as a minor cycle, the effective damage calculated and accumulated and NH3 and NL3 can be removed from their respective unresolved stores.

Asking the question again gives a NO answer. NL4 is therefore loaded into the unresolved low store. Hence NH3 and NL3 have been identified as a minor cycle and the damage effect logged at a point in time associated with NL4.

(viii) Is NH4 greater than or equal to the last unresolved high? YES NH2 can be resolved with NL4, damage effect calculated and logged and NH2 and NL4 removed from their respective unresolved stores.

Asking the question again gives a NO answer. NH4 can therefore be loaded into the unresolved high store. Thus at this point in real time NH2 and NL4 are paired as a minor cycle.

(ix) is NL5 lower than or equal to the last unresolved low? NO since the only stored NL is the original entry of NLl which is zero.

Therefore NL5 is entered into the unresolved low store as the last entry (i.e. top of the stack).

(x) Is NH5 greater than or equal to the last unresolved high? NO since NH5 is not greater than NH4. Therefore NH5 is entered into the unresolved high store as the latest entry.

(xi) Is NL6 lower than or equal to the last unresolved low? YES since NL6 is less than NL5. Therefore the last unresolved high and the last unresolved low (top of the stack in both cases) can be resolved.

Hence NH5 and NL5 are paired as a minor cycle, their damage effect calculated and logged and NH5 and NL5 can be remdved from their respective stores.

Asking the question again, is NL6 less than or equal to the last unresolved low? YES NL6 is equal to NLl. Therefore the last unresolved high can be resolved with the last unresolved low. Hence NH4 and NLI are paired and resolved, thus adding in the major cycle contribution.

This simple example illustrates the effective way that the application of this procedure identifies, on-line, the exact pairing achieved by the definitive method technique.

Figure 2 represents stores A and B during the working of this example and it will be seen that at no time does the content of either store exceed two speed values, i.e. two words.

The procedure described above following identification of a confirmed NL or NH value can be summarised by the flow diagrams shown in Figures 3 and 4, Figure 3 illustrating the procedure on identification of a confirmed NL value, and Figure 4 illustrating the procedure on identification of a confirmed NH value.

In Figures 3 and 4: NHi represents current computed maximum NHs represents maximum on top of stack, NLs represents minimum on top of stack, NLi represents current computed minimum, Confirmed means identified after filtering, Evaluate means solve with the definitive fatigue damage algorithm.

The reason for the provision of a "stack-full" box on Figure 3 will now be explained.

In the case of a typical sortie pattern with 160 confirmed turning points the required maximum capacity of the stores reaches five words.

Whilst this can be considered as typical of the actual requirement, sorties can be envisaged where the store size requirement may be greater.

The situation which causes the store size ever to increase above two words is where the speed profile with time shows consecutive minor cycles decaying in NH and increasing in NL (similar to a decaying oscillation). The sequence only being broken when the current value of NH exceeds one or more of the unresolved peaks or NL is lower than one or more of the unresolved lows.

For a practical apparatus sixteen-word stores are suitable.

A sixteen-word store provides for the case illustrated in Figure 5, where a sample in time for a sortie pattern shows a sequence of 16 decaying minor cycles as described. This is a situation which is so unlikely that it is probable that it will never occur in practice.

Since there is an equal chance during any sortie of any NH being smaller than the previous NH and an equal change of any NL being higher than the previous NL then the probability of 15 consecutive decaying minor cycles following any initial cycle is: 2-30 or 9.32x10-10 Samples of sortie patterns show that a typical number of confirmed turning points for a sortie is of the order of 200. This gives the possibility of a sixteen-word store filling as 200x9.32x 10-'0 or alternatively one chance in 5x 106 sorties.

Pursuing this probability argument a little further we can obtain estimates of the likely maximum store sizes for varying numbers of confirmed turning points in a sortie, as shown in the Table below.

TABLE Number of Confirmed Turning Points Per Sortie Stack Size (After Filtering) (Words) 4 2 16 3 64 4 256 5 1024 6 4096 7 16384 8 This supports the findings for the typical sortie pattern mentioned above. The results shown in the Table indicate a stack size requirement of 5 for a sortie having 160 turning points.

The likelihood of sixteen words of storage ever being fully used is so remote that the possibility can be theoretically ignored.

However should this situation ever occur in practice the values of NH and NL in the last store locations are paired as a minor cycle thus keeping available one pair of store locations for when the sequence is broken. The "stack-full" box in Figure 3 initiates this procedure, the NL store always being the first to be filled.

This approach is justified for the following reasons: (a) It has a 50',/, chance of being the correct pairing anyway.

(b) If the situation ever occurs it is likely to be a very minor cycle producing such a small damage contribution that it will be insignificant in the final result for the mission. Indeed it is most likely to be below the cut off value.

(c) The occurrence is so unlikely that over a series of sorties any error will be even more insignificant, and (d) It is necessary to optimise the hardware content in order to produce a cost.

effective solution.

The only purpose of describing this problem at all is that it does represent the only limitation of the apparatus and it is therefore necessary lo show how the apparatus will cope, even though it is considered unlikely that this feature of the programme will ever be used in the service life of the apparatus.

Two embodiments of apparatus implementing the above described procedure for pairing values of NH and NL and subsequently calculating minor cycle fatigue damage by the definitive method algorithm will now be described.

The first embodiment, which involves the use of logic circuitry, is shown in Figure 6.

In this embodiment an eight bit word representing engine shaft RPM is produced at intervals by interface circuitry 1 (described more fully later with reference to Figure 10). This value of RPM (N) is then steered to a maximum or a minimum RPM register 3 or 5 by comparing it in a comparator 7 with the values (NH and NL) already in these registers. The direction of the slope of these changing values is stored as a one bit number to enable the turning points to be determined, again by comparison. For example if the speed is increasing, the latched slope value switches a multiplexer 9 to permit NH to be compared with the input speed.

All the time this comparison gives N > NH, the value of N is written into register 5. When the answer is N < NH, this implies a turning point has been reached, i.e. the current NH value is confirmed as a peak value. A calculate routine is then triggered and the multiplexer 9 is switched over to compare NL to N, searching for and eventually confirming a subsequent low value NL in the same manner.

Each confirmed peak or low value NH or NL is compared with the value on the top of a push down stack 11 or 13 and the algorithm is implemented. The push down stacks are each a random access memory consisting of sixteen words of eight bits. There are two operations associated with each stack, termed "push" ard "pop". When a number is written into the stack, the stack is "pushed" and when a number is read, the stack is "popped". An alternative name for such a stacks a First-In Last-Out memory.

A stack can be compared to the analogy in Figure 7, a narrow tube containing balls on a sprung platform. Data can be removed from or added to the stack but only in a set order, i.e. the last ball in is the first out.

Paired values of NH and NL from stacks 11 and- 13 identified by implementation of the algorithm are used to calculate fatigue damage, as follows.

In order to evaluate the damage in this implementation, it is optimal to divide the RPM range into sixteen bands, as illustrated in Figure 8. As the bands require to be closer together as RPM (and hence damage/RPM difference) increases, an (RPM)2 distribution is chosen. The band limits are given by:

where n is an integer between 0 and 16.

In band select circuits 15 and 17 a read only memory of 256 words of 4 bits is used to convert each 8 bit speed derived from stack 11 or 13 to a 4 bit band number (4 bits gives values 0--15). The two band numbers corresponding to paired NH and NL values form an 8 bit word which is used to address a read only memory 19 which stores pre-programmed values of damage for all 128 allowable combinations of bands.

ROM 19 provides the damage associated with the paired NH and NL values using the algorithm:

This is achieved by using it as a look-up table. The values of damage associated with two band identities is calculated from the average value of N for the associated bands.

The value of damage obtained for a pair of values NH and NL is added into a staticising counter 21. This counter is loaded by this addition at the speed of the computing element, and decremented at the same rate as an output fatigue counter 23 which drives a display 25 via a module 27 is incremented, thus providing a buffering facility between the high speed computing circuitry and the lower speed display counter 23.

When the staticising counter 21 is empty, the fatigue counter 23 is disenabled.

When it is not empty the staticising counter 21 is decremented at the same time as the fatigue counter 23 is incremented. This enables an output counter of any operating speed to be used with the system.

The output fatigue counter 23 is suitably a high temperature electromechanical counter.

While the use of a banding system gives a short term accuracy of about lOY, in particular from sortie to sortie, long term accuracy is better than 1'S,. This is because the pre-calculated and stored values of minor cycle damage are average for the band pairings and so will become more representative as the number of turning points evaluated increases.

As described earlier the concept of dividing the speed range into a defined number of discrete bands is introduced as a simplification in order to reduce the quantity of hardware required to solve the specified algorithm, 16 bands being chosen which are spaced according to an n2 law.

Since the pre-calculated stress associated with pairs of bands is based upon average speed values selected within each band, the accuracy of the system must inevitably depend upon: (a) The number of bands.

(b) The spacing of the bands.

(c) The selected values chosen to represent each bands in the calculation of the pre-stored fatigue counts.

Theoretically the spacing of bands should depend upon equal fatigue weightings, i.e. an n2P law.

In this way an equal error, due to the banding concept, across the whole speed range results. However, this assumes an equal probability of turning points across the speed range.

In practice, by analysing a large number of sortie patterns it may be possible to produce a more representative probability distribution for confirmed turningpoints across the speed range; a distribution, peaking may be at 8090 /" Nmax.

The optimum positioning of bands therefore is a function of both equal fatigue considerations and the most probable minor cycle position distribution.

Referring to Figure 9, the second embodiment to be described involves the use of an Intel 8080 type eight bit microprocessor 27, associated with a RAM data store 29, a PROM program store 31, a status latch 33, a timing and control unit 35, output counters 37 and displays 39, and interface circuits 41. Up to 4 inputs can be software controlled by the microprocessor 27 using 2048 words of program store and 256 words of scratch pad.

The timing and control unit 35 generates a real time interrupt each time an interface circuit 41 produces an 8 bit word representing engine speed, and the 8 bit word representing engine speed is read into the processor. The software sorts this speed by comparison to previous maxima and minima, up-dating and filtering these values where necessary. When a turning point is detected, a software push down stack, using part of the scratchpad facility, is accessed.

This process produces the required paired RPM values. The depth of the push down stack can be increased from 16 if it is felt to be essential, although probability dictates that the likelihood of 16 being filled in a sortie is less than 10-'.

The damage algorithm is interrupted by the software as a binomial expansion, i.e. the power series of (x)P is calculated by P(P-I) (I+x)P=l+Px+ x2+..

2! A further real time interrupt occurs at the update rate of the output fatigue counters 37, and a software staticising counter determines whether a fatigue counter is incremented or not. This staticising counter is updated by the results oF the damage calculation.

The separate inputs and outputs to the processor are dealt with as peripheral devices under software control. Further expansion of the tasks governed by the processor can be readily achieved. The incorporation of pressure and temperature as inputs would require further input hardware for signal conditioning, and might require incrcased program store size. depending on the complexity of the algorithms chosen.

The program store is broken down into blocks of 256 words. Hence the algorithm and any other relevant constants can be grouped into one block and replacement of this block would alter the values of P, K etc.

The accuracy of this system is better than 1%. The only errors introduced compared with the definitive method of fatigue damage analysis are due to the 8 bit-word length of the processor proposed.

This arrangement exhibits considerable flexibility. By replacing the program blocks, the system constants, system algorithms and even the introduction of new algorithms can be readily achieved.

Circuitry for interfacing either of the above arrangements with typical engine transducers is illustrated in Figure 10. The circuit is designed to measure the frequency of the incoming signals by counting the number of pulses occurring in a fixed period of time. 100% RPM is represented by 70 Hz from a tachogenerator and 4200 Hz from a speed probe, both varying frequency linearly with engine speed from zero. Isolation is achieved by a signal transformer 43 for the speed probe, and by an optical isolator 45 for the tachogenerator. Loading of the transducers is negligible for the speed probe and less than 1 /" of the voltage output from the tachogenerator.

A link 47 is used to determine which input is routed via a comparator 49 to an eight bit binary counter 51 and to pass this information to the timing network of the apparatus. The latter enables the counting period to be modified by a factor of 60 for the two inputs, i.e. it takes 60 times longer to count the same number of pulses at 70 Hz than it does at 4200 Hz.

The comparator 49 has a high level of hysteresis to enable noise rejection from either input. A small number of high level noise spikes which break through this "filter" will not produce significant error in the eight bit binary count of frequency.

Self test is achieved by depressing a button 53 TEST to apply a test input to counter 51. If the apparatus is operating correctly a test lamp 55 lights after a time delay. The exact time delay is dependent on the type of input transducer used. The mechanics of this built in test equipment are as follows. Depressing the button 53 depowers the counter 51, avoiding the possibility of falsifying the existing fatigue count, and applies a digital waveform to the counter 51. This waveform alternates in frequency, thus simulating high and low RPM excursions of known value. The subsequent value of damage calculated by the apparatus is then compared to a stored value, and if the two are in agreement the test as determined by a comparator indicating lamp 55 lights.

WHAT WE CLAIM IS: 1. Apparatus for determining fatigue damage of a rotating part of a machine by the definitive method wherein the minor cycles are identified during a period of operation of the machine and the computation of fatigue damage due to minor cycles takes place during said period of operation leaving only the contribution of the major cycle to be calculated at the end of the period of operation.

2. Apparatus according to Claim 1 comprising means for monitoring the speed of rotation of the part under test thereby to identify as it occurs each high and low turning point of the speed change pattern of said part; means for pairing the speed values at the turning points to identify fatigue cycles by application of the following algorithm:: (a) for each high turning point speed value (i) determine whether the speed value is higher than or equal to the last unresolved (not yet identified with a specific cycle) high turning point speed value; (ii) if yes, pair the last unresolved high turning point speed value with the last unresolved low turning point speed value; (iii) keep repeating step (i) until the answer is no, then store the speed value as the last unresolved high turning point speed value; (b) for each low turning point speed value; (iv) determine whether the speed value is lower than or equal to the last unresolved low turning point speed value, (v) if yes, pair the last unresolved high turning point speed value with the last unresolved low turning point speed value, (vi) keep repeating step (iv) until the answer is no, then store the speed value as the last unresolved low turning point speed value; ; and means for evaluating and summing the fatigue damage for each identified cycle.

**WARNING** end of DESC field may overlap start of CLMS **.

Claims (6)

1. **WARNING** start of CLMS field may overlap end of DESC **.

   The program store is broken down into blocks of 256 words. Hence the algorithm and any other relevant constants can be grouped into one block and replacement of this block would alter the values of P, K etc.

   The accuracy of this system is better than 1%. The only errors introduced compared with the definitive method of fatigue damage analysis are due to the 8 bit-word length of the processor proposed.

   This arrangement exhibits considerable flexibility. By replacing the program blocks, the system constants, system algorithms and even the introduction of new algorithms can be readily achieved.

   Circuitry for interfacing either of the above arrangements with typical engine transducers is illustrated in Figure 10. The circuit is designed to measure the frequency of the incoming signals by counting the number of pulses occurring in a fixed period of time. 100% RPM is represented by 70 Hz from a tachogenerator and 4200 Hz from a speed probe, both varying frequency linearly with engine speed from zero. Isolation is achieved by a signal transformer 43 for the speed probe, and by an optical isolator 45 for the tachogenerator. Loading of the transducers is negligible for the speed probe and less than 1 /" of the voltage output from the tachogenerator.

   A link 47 is used to determine which input is routed via a comparator 49 to an eight bit binary counter 51 and to pass this information to the timing network of the apparatus. The latter enables the counting period to be modified by a factor of 60 for the two inputs, i.e. it takes 60 times longer to count the same number of pulses at 70 Hz than it does at 4200 Hz.

   The comparator 49 has a high level of hysteresis to enable noise rejection from either input. A small number of high level noise spikes which break through this "filter" will not produce significant error in the eight bit binary count of frequency.

   Self test is achieved by depressing a button 53 TEST to apply a test input to counter 51. If the apparatus is operating correctly a test lamp 55 lights after a time delay. The exact time delay is dependent on the type of input transducer used. The mechanics of this built in test equipment are as follows. Depressing the button 53 depowers the counter 51, avoiding the possibility of falsifying the existing fatigue count, and applies a digital waveform to the counter 51. This waveform alternates in frequency, thus simulating high and low RPM excursions of known value. The subsequent value of damage calculated by the apparatus is then compared to a stored value, and if the two are in agreement the test as determined by a comparator indicating lamp 55 lights.

   WHAT WE CLAIM IS: 1. Apparatus for determining fatigue damage of a rotating part of a machine by the definitive method wherein the minor cycles are identified during a period of operation of the machine and the computation of fatigue damage due to minor cycles takes place during said period of operation leaving only the contribution of the major cycle to be calculated at the end of the period of operation.
2. 2. Apparatus according to Claim 1 comprising means for monitoring the speed of rotation of the part under test thereby to identify as it occurs each high and low turning point of the speed change pattern of said part; means for pairing the speed values at the turning points to identify fatigue cycles by application of the following algorithm:: (a) for each high turning point speed value (i) determine whether the speed value is higher than or equal to the last unresolved (not yet identified with a specific cycle) high turning point speed value; (ii) if yes, pair the last unresolved high turning point speed value with the last unresolved low turning point speed value; (iii) keep repeating step (i) until the answer is no, then store the speed value as the last unresolved high turning point speed value; (b) for each low turning point speed value; (iv) determine whether the speed value is lower than or equal to the last unresolved low turning point speed value, (v) if yes, pair the last unresolved high turning point speed value with the last unresolved low turning point speed value, (vi) keep repeating step (iv) until the answer is no, then store the speed value as the last unresolved low turning point speed value;; and means for evaluating and summing the fatigue damage for each identified cycle.
3. 3. Apparatus according to Claim 2, wherein said means for evaluating fatigue

   damage comprises storage means for storing fatigue damage corresponding to any selected pair of S speed values, each of said S values lying in a respective discrete band of speed values, fatigue damage for a pair of identified speed values being taken to be the damage associated with the pair of said S speed values in whose bands said identified speed values lie.
4. 4. Apparatus according to Claim 3, wherein the limits of the speed value bands are given by the expression

   where n is any integer between 0 and S and N max is the highest possible speed value.
5. 5. An apparatus for determining fatigue damage of a part of a machine substantially as hereinbefore described with reference to Figure 6.
6. 6. An apparatus for determining fatigue damage of a part of a machine substantially as hereinbefore described with reference to Figure 9.