# 10th International Conference on Vibrations in Rotating Machinery: 11-13 September 2012, Imeche London, UK

### Summary

This book presents the papers from the 10th International Conference on Vibrations in Rotating Machinery. This conference, first held in 1976, has defined and redefined the state-of-the-art in the many aspects of vibration encountered in rotating machinery. Distinguished by an excellent mix of industrial and academic participation achieved, these papers present the latest methods of theoretical, experimental and computational rotordynamics, alongside the current issues of concern in the further development of rotating machines. Topics are aimed at propelling forward the standards of excellence in the design and operation of rotating machines. Presents latest methods of theoretical, experimental and computational rotordynamics Covers current issues of concern in the further development of rotating machines

## Book Preview

### 10th International Conference on Vibrations in Rotating Machinery - Institution of Mechanical Engineers

Addresses

**Modelling, dynamic behaviour and diagnostics of cracked rotors **

**P. Pennacchi *, Politecnico di Milano, Dept. of Mechanical Engineering, Italy **

**ABSTRACT **

One of the most common incipient losses of structural integrity in mechanical structures is the development and propagation of a crack. This happens also in rotating machinery and a very rich, but also in some way confusing, literature about cracked rotors has appeared in the last 30 years.In the paper, a general and wide overview about the behaviour of cracked rotors will be presented, covering several aspects of this topic. In particular modelling, dynamic behaviour and diagnostics of cracks will be analyzed in detail, introducing also practical examples and cases of industrial machinery, including some topics barely documented in literature such as helical and annular crack development.

**1 INTRODUCTION **

The propagation of shaft cracks is one of the most dangerous faults that can occur in rotating machinery [**1]. If this fault is not early detected, it can cause serious damage as well as long outages and expensive maintenance actions. **

Transverse cracks that propagate in the shafts of rotating machines cause a change of the local flexural stiffness of the rotor. Moreover, in horizontal machine-trains, the shaft weight causes rotor bending that gives rise to tensile axial stresses, acting in the lower area of the faulty cross-section, which tend to open the crack. Conversely, the axial compressive stresses generated in upper area of the cracked cross-section tend to cause the contact between the crack surfaces. Therefore, during a complete revolution of the shaft, a transverse crack can be subjected to a *breathing phenomenon*, that is a periodic closure and opening of the crack.

Transverse cracks can propagate in the cross-section of the shafts of rotating machines as a consequence of very high fatigue stresses. In general, the stress distribution in a rotor is rather complex, being the consequence of many concomitant static and dynamic loads. Owing to some of the most important of them, the cracked section is subjected to time-varying axial stresses generated by different causes, in operating condition, both at low and high rotational speeds. Some of these stresses are *periodic *as they depend on mechanical phenomena that are influenced by the shaft angular position. Other important axial stresses are *aperiodic *as they are caused by quasi-static loads. With regard to this, the stresses caused, for instance, by changes of the machine thermal condition can be very important. In general, these thermal transients occur within time intervals that are considerably longer than the revolution period of the shaft. Therefore, the corresponding axial stresses gradually increase, or decrease, during sufficiently long time intervals. Detailed analyses of thermal effects on breathing mechanism are presented in [**1][2]. **

In the case of very thin cracks, having planar surfaces, the time-varying axial stresses acting on the cracked section of the shaft can cause a temporary opening, or closure, of the crack. Negative compressive stresses can cause contact between the fracture surfaces, whereas positive tensile stresses tend to open the crack. This phenomenon is often called crack breathing.

**2 CRACK BREATHING MECHANISM **

The breathing mechanism has been analysed in literature by several authors: Mayes and Davies [**3] proposed the so called switching crack, in which the crack is either completely closed or open and passes abruptly from one state to the other. Nelson and Nataraj [4] proposed a switching criterion based on the curvature of the shaft, while Wauer [5] based his criterion on the sign of the total axial strain at extreme fiber of the cross section of the shaft. Mayes and Davies [6] improved their model by introducing a smoothing trigonometric function of the rotation angle of the section. However, all these approaches, along with similar ones [7][8], suppose that the area of the cracks that is in contact for various rotation angles is a priori known as observed by Andrieux and Varé [9]. Functions that rule the breathing of the crack are still used in the literature, as in Sinou and Lees [10], Patel and Darpe [11], Sawicki et al. [12], Ishida et al. [13][14] and Sinou and Faverjon [15]. **

A different approach was proposed by Dimarogonas and Papadopoulos [**16] by considering shaft elasticity and stress intensity factors, with a typical approach of Solid Mechanics. Starting from that and by the introduction of the concept of crack closure line by Darpe et al. [17][18] and of other improvements by Papadopoulos [19]. This method has been widely used (see for instance Papadopoulos [20] and Sekhar [21][22][23]). However, some limitations of this approach have been highlighted, such as the limitation to consider cracks deeper than 50% of the diameter [19] or thermal stresses or crack closure effects [1]. **

Other approaches, able to manage the previously mentioned limitations have been proposed in [**9][24] and [25]. **

However, owing to the interest in defining the true breathing behaviour of cracked shafts, some tests have been performed in the laboratory of the Dept. of Mechanical Engineering of Politecnico di Milano, as described in [**1], to measure the strains in different points of a cracked shaft, with 70 mm diameter, under different load conditions. To this aim, a series of strain-gauges have been applied close to the crack and also directly across the crack lips (see figure 1). **

**Figure 1 **Detail of strain-gauge positions close and across the crack, from [**1]. **

The horizontal cracked specimen was subjected to different stationary loads and rotated in different angular positions in order to excite the breathing of the crack.

Two different non trivial effects have been observed: the *crack closure effect *and the *local contact conditions of the crack lips *in closed crack configuration, which are now discussed:

a) Crack closure effect: Small loads, generating small bending moments in the locality of the crack, were not able to open the crack. The crack closure effect generates an internal bending moment that holds the crack closed. Only when the external bending moment overcomes the internal bending moment, then the lips of the crack start to open.

b) Local contact conditions of the crack lips: When the crack is closed, with an external bending moment that sums up to the internal bending moment, the measured compressive strain is much higher than the theoretical strain calculated assuming a linear compressive stress distribution over the cracked section. This can be explained by assuming that the contact is not spread over all the cracked area when the crack is closed, but it occurs only on a smaller portion of the cracked surface, or on the crack lips only, determining higher strains associated also to stress intensity factors. This aspect is also related to the crack closure effect.

Despite the fact that crack closure effects have been studied by several researchers (see for instance [**26]), their influence, on the breathing behaviour of rotating shafts, has never been modelled suitably, to the author’s knowledge. **

Despite the highly non-linear stress and strain distribution in the cracked area and the non-linear breathing behaviour, the overall load-strain behaviour results are quite linear. The overall load vs. deflection law can therefore easily be represented by a linear model like the FLEX model presented in [**1]. **

Cracked shaft diameter was larger than 70 mm when the crack had been initiated by means of a small notch generated by electro-erosion and had propagated roughly up to half way the shaft cross section by applying a constant bending load to the rotating shaft. The specimen was then machined and the diameter was reduced to the final one by turning, so that the initial slot was removed.

The final cracked section has the shape shown in **figure 2, as determined from ultrasonic test measurements. The cracked shaft has been clamped at one end but can be rotated around his axis by steps of 15° each. A vertical load has been applied at the other end and has been increased by steps. Theoretical stresses and strains are calculated assuming no crack. **

**Figure 2 **Shape of the crack and strain-gauge positions.

Strain-gauges from A1 to A11 were applied each 15°, as close as possible to the crack. Strain-gauges from A12 to A15 were applied to an integer section, which was sufficiently far away from the crack to be not influenced by its breathing behaviour and their measurements are used as reference signals. Strain-gauges from A16 to A28 were applied at 15° in correspondence of the crack, but on the integer part opposite to the crack. Strain-gauges from B31 to B39 were applied across the lips, as it is shown in **figure 2. **

Typical results are shown in **figure 3 for measuring point A6, in which the maximum effect of the crack is expected. Angle 0° indicates that crack axis (which passes close to measuring point A6) is directed vertically downwards (crack open), angle 180° indicates that crack axis is directed vertically upwards (crack closed). A flat zone in strain level (about 140 με) is clearly recognisable between 0° and 60° and between 300° and 360°, for a total range of angular positions of 120°. In this range of angular positions where the strain does not change, provided that sufficient load is applied, the crack is supposed to be definitely open. The strain for the open crack is not zero but a tensile strain of 140 με is measured on the open crack lips. **

**Figure 3 **Strain vs. rotation measured by strain-gauge A6.

This can be explained by the presence of an internal pre-stress on the crack lips due to the crack closure effect, generating a compressive strain of 140 με, when the strain-gauges have been applied to the shaft with no external loads. Only the loads that generate a similar value of tensile stresses are they able to open the crack. Similar results, but generally with higher values (up to 240 με) of compressive strains due to the crack closure effect, have been found in the other measuring points A3, A4, A5 and A7. The behaviour is different in positions that are closer to the crack ends, i.e. also closer to the crack tip, as in measuring point A10.

When the crack is closed (at the angular position of 180°), the compressive stress of 350 με in correspondence of maximum load, is much higher with respect to the theoretical value of 190 us, calculated considering the load applied on a un-cracked specimen. This can be attributed to the fact that only a smaller part of the cracked surface is in contact, generating locally higher stresses and also stress intensity factors which would be absent if the crack faces were completely in contact each other over the complete crack area. This strain magnification is present for all loads in proportion with the same value, but its value is different for different measurement points and tends to unity (no strain magnification) close to the crack ends (in measuring points A1 and A10), as expected.

At the measuring points close to the crack ends, when the loads generate tensile stresses, during the gradually opening of the crack, rather high stress intensity factors have been measured that are due to the closeness of the crack tip to the measuring points. At the measuring point A16 (opposite to A6), the maximum tensile strain is 240 με (opposite to the closed crack), which is somewhat higher than the theoretical value of 190 με: this might be caused by the fact that not all the cracked area is in contact with the opposite face.

The measured behaviour has been also simulated, with the aid of a model of the crack. The FLEX model used for calculating the breathing behaviour and the reduced stiffness of the cracked shaft is described in [**1] and the details are not reported here for the sake of brevity. The crack closure effect has been simulated by an external bending moment that tends to hold the crack closed, generating a maximum of 140 με of compressive strain. Figure 4 shows the comparison of calculated results obtained with the above specified model, with the experimental results in measuring point A6 for the maximum load. It is surprising how good the simplified model is able to reproduce the experimental behaviour. This occurs at almost all measuring points. The quasi-linear breathing behaviour model can be considered completely validated with these experimental results. **

**Figure 4 **Comparison of calculated and experimental results at point A6.

**3 CRACK PROPAGATION AND SHAPE **

A crack may propagate from some small imperfections on the surface of the body or inside of the material and it is most likely to develop in regions of high stress concentration. Cracks in rotating shafts are most likely to appear due to sharp changes of the diameter or of the geometry of the shaft (for instance due to the presence of holes, slots for keys, threads and so on) in regions of high stress concentration.

Also thermal stresses that develop as a consequence of the working fluid in thermal machines, such as steam and gas turbines, and thermal shocks are responsible for generating high local stress intensity factors, which can cause the initiation of a crack and its propagation.

In rotating shafts, the cracks propagate generally in a plane perpendicular to the shaft axis, when axial bending stresses are prevailing, generating a *transverse crack*. A typical example is shown in **figure 5, where the actual shape of the crack in a generator was found definitively after dismantling the rotor and breaking it. The crack had propagated to nearly 50% of the cross section area. The starting point was in correspondence to a defect in a slot for the windings (see figure 5). **

**Figure 5 **Transverse crack in a generator (left); starting point (right), from [**1]. **

However, conical crack surfaces and helical crack surfaces, as well as annular and longitudinal cracks, have also been reported in the literature.

The crack may have any orientation at its starting point, depending on local conditions, but when the crack propagates more deeply, its direction is mainly radial and the cracked surface, although not exactly, is roughly planar, with vanishingly small curvatures.

The propagation velocity may change from case to case in rotating shafts. Propagation times of 74,000 up to 101,000 hours of operation are reported from detection, or suspected presence, to dangerous crack depth, but only 2,500 hours of operation has been enough to deepen consistently the crack in other cases. Very frequently the crack moves in steps, alternating in growth to stop: both can be seen on the cracked surface pattern where the rest lines called beach marks are recognizable. These are evident in **figure 5. **

Generally, when the crack approaches a dangerous depth, it propagates more quickly, with a propagation velocity that increases almost exponentially. The final growth, up to a critical dangerous depth, takes sometimes only few days of operation.

**3.1 Helical cracks **

Generally, cracks propagate in surfaces which are roughly planar, even if not exactly as analyzed in detail in [**27][28], and perpendicular to the rotation axis of the shaft. However, if a high torque combines with high bending loads, the crack may also propagate along a helical path, therefore these cracks are called helical or slant cracks. **

Cracks caused by torsional stresses develop mostly along helical surfaces rather than along planar inclined surfaces. Slant cracks (with helicoid angles up to 45°) can develop only when torque is alternating, which is usually not the case in turbomachinery.

The crack studied in [**1][25], see figure 6, developed along a helical path with an angle of 6° only on the outer surface of the shaft, owing to the combined action of bending and torsion. Such a crack could develop at mid-span of double flow steam turbines in high power turbogenerators, where the maximum bending moment due to weight combines to the transmitted torque. **

**Figure 6 **Slightly helical crack highlighted by dye penetrant, from [**1]. **

The results of the theoretical and experimental analyses showed that at full load the differences between helical and flat crack are so small that they could be neglected.

When torsional load is removed, then higher differences arise, but mainly for the torsional degree of freedom, which is excited by the bending load by means of a coupling effect.

In this condition torsional deflections are generated by bending moments due to coupling effects and torsional vibrations are excited in rotating shafts. This constitutes a clear symptom of the presence of a helical crack in a shaft line: when during a run-down transient (at no torsional load) the torsional natural frequencies are strongly excited at the corresponding rotating speed, this could be due to the relevant coupling effect of the helical crack.

**3.2 Annular cracks **

Occasionally, multiple crack initiations can occur along the circumference of the external surface of the shaft. This phenomenon can give rise to the propagation of a full-annular transverse crack [**2][29]. In general, the contour of the corresponding residual section is nearly a circle, the centre of which can be affected by an eccentricity with respect to the shaft axis (see figure 7). **

**Figure 7 **Cross-section of shaft showing an annular crack, from [**2]. **

The analysis of the breathing mechanism of annular cracks shows that the highest values of the eccentricity, as well as the lowest values of the radius, cause the highest values of the crack depth. The highest changes of the angle of the neutral axis, with respect to the reference, occur during a complete revolution. However, owing to the crack shape, even the highest values of the inclination angle of the neutral axis are limited. This reduces the amount of the complex flexion of the shaft, caused by its weight, and the levels of the super-synchronous vibrations.

Therefore, this type of crack, the propagation of which has been occasionally detected near high-pressure stages of steam turbines, can cause only small levels of the twice per revolution vibrations of the shaft that, in general, are considered the most common and characteristic symptom of the presence of a shaft crack.

**4 IDENTIFICATION OF CRACKS IN ROTORS **

Model based techniques are able to identify the presence, the position and even the depth of cracks in rotors. Crack diagnosis is performed by means of parity equations, by introducing *external forces *that are *equivalent *to the effects of the crack.

The starting point for the definition of equivalent external forces to cracks is the system of dynamic equations of the machine, obtained by considering the rotor modelled by means of finite beam elements, the bearings by equivalent damping and stiffness and the foundation by a suitable representation (pedestals, modal, transfer matrix or rigid).

The system of dynamic equations of a large rotating machine, with several d.o.f.s, is typically:

**(1) **

where the mass matrix [**M**] takes also into account the secondary effect of the rotatory inertia, the damping matrix [**C**] includes also the speed depending gyroscopic matrix and the stiffness matrix [**K**] takes also into account the shear effect.

Considering **eq. (1), it is difficult to identify the changes due to the developing fault in the matrices [ M], [C] and [K], which are of high order, from measurement of the absolute vibration xtotal in only few measuring planes along the shaft. **

In real rotating machines, the measurement points of the vibration along the shaft during operation are few and the transducers are normally in correspondence with the bearings. Only lateral vibrations are normally monitored, hence in the following only lateral behaviour of the rotor will be considered.

The consideration of further measuring planes along the rotor span, due to the presence of the casings and of the possible working fluids, is practically impossible, therefore methods that reconstruct modal shapes of the rotor cannot be used.

The right hand side (r.h.s.) external forces **F**(*t*) of **eq. (1) are composed of the weight (which is known) and by the original unbalance and bow (which are unknown). The system parameter changes due to the fault are indicated as[ dM], [dC] and [dK] and eq. (1) becomes: **

**(2) **

If the system behaviour is considered to be linear, which is acceptable for a wide class of faults [**30][31], then the total vibration xtotal can be considered as due to two superposed effects: **

**(3) **

It can be shown that the overall behaviour of a horizontal axis heavy cracked shaft is linear, under the conditions shown in [**1]. Only in extreme operating conditions, the non-linear effect of the breathing crack, which is weak in normal conditions, will influence its behaviour. **

The first vibration vector **x***ref *is the pre-fault vibration, which is due to the weight **W **. The second vibration **x***a *is due to the developing fault. The last is also called *additional vibration*. The vibration component **x***a *may be obtained by calculating the vector differences of the actual vibrations (due to weight, original unbalance, bow and fault) and the original vibrations measured, in the same operating conditions in a reference case (rotation speed, flow rate, power, temperature, etc.) before the fault was developing. A discussion about the possible errors introduced and their tracking is presented in [**32]. Recalling the definition of the pre-fault vibration xref., the following equation holds: **

**(4) **

which substituted in **eq. (2) with eq. (3) gives: **

**(5) **

The r.h.s. of **eq. (5) can be considered as a system of equivalent external forces which force the fault-free system to have the change in the additional vibration xa that is due to the developing fault only: **

**(6) **

A rather complete overview of the equivalent forcing systems to the most common faults in rotating machinery is presented in [**30][31]. **

Note that in **eq. (6) system parameters [ M], [C] and [K] are time invariant and known, but normally [C] and [K] are functions of the operating speed with regard to the gyroscopic effect and the bearing coefficients. Using this approach, the problem of fault identification is reduced to an external force identification by means of parity equations. The effect of a crack on the statical and dynamical behaviour of the rotor can be simulated in the frequency domain, by applying different sets of equivalent forces to the rotor in correspondence of the cracked beam element, one set for each of the harmonic components considered. **

The stiffness of a cracked shaft is periodic due to the breathing and the rotation of the crack, see [**1] where it is shown that the equivalent crack forces are given by the following expressions: **

**(7) **

**(8) **

**(9) **

**(10) **

of which **eqs. (8), (9) and (10) are respectively the first, the second and the third harmonic components of the crack force system. Usually higher order harmonic components are not considered for a practical reason: actual condition monitoring systems installed on real machines that performs order analysis normally store only 1X, 2X and sometimes 3X components. Note that the projections along reference axes of the harmonic component of the force system are not necessarily equal, thus the equivalent force system is not in general represented by rotating forces. **

From the experimental point of view and under the previously exposed hypothesis of linearity, the difference **x***a* = **x***tatal *– **x***ref*, between the measured vibration **x***totd *of a rotor system that has a fault and the reference case **x***ref*., represents the vibrational behaviour due to the fault, i.e. the *additional vibrations*. These vibrations are used in the identification procedure, since they are due to the impending fault only. In fact, the reference case vibrations **x***ref*. are given by **eq. (11): **

**(11) **

while those caused by the developing crack are given by:

**(12) **

If **eq. (12) is considered for an unknown crack, also [ Km] is unknown. Anyhow it can be approximated by [K] of the un-cracked shaft, from which it differs only very little: the crack affects the stiffness of one element only. Therefore the additional vibrations are given by: **

**(13) **

By applying the harmonic balance criteria in the frequency domain to **eq. (13) and considering the harmonic components of the additional vibrations Xn, the following equations are obtained for each one of the considered harmonic components: **

**(14) **

in **eq. (14) are those to be identified. For the sake of clarity and simplicity, a single fault is considered, while multiple faults can be handled using the method fully described in [30][31]. **

In the r.h.s. of **eq. (14), the equivalent force system is applied to the two nodes of the element that contains the crack. Since lateral vibrations only are considered, the model has 4 d.o.f.s per node (see figure 8) and the equivalent force is a vector of eight generalized forces. **

**Figure 8 **Degrees of freedom of an element and bending moments equivalent to crack.

Owing to energy considerations, the most important among these forces are the bending moments that are roughly equal and opposite on the two nodes of the cracked element (see **figure 8). **

In fact, cracks develop and propagate mainly due to axial stresses generated by bending moments in rotating shafts; therefore generally very high bending moments are present in correspondence of cracked elements. It can be shown that in a beam element loaded by shear forces and bending moments, the elastic deflection energy associated to the shear forces is much smaller than the energy associated to the bending moments. Therefore the first one is often neglected in comparison to the second one, as it has been done also in the Bernoulli’s beam in comparison to the Timoshenko’s beam. For this reason, the additional deflections due to the crack can be attributed to equivalent bending moments only.

Therefore the set of equivalent forces in the case of a crack is reduced to a couple of bending moments (with their relative phase) along two orthogonal directions for each *n*-th harmonic component.

, respectively each one on the reference system directions, are not full-element vectors that are convenient to be represented by means of:

**(15) **

where {**L**(*x*,*y*is a complex number representing the amplitude and the phase of the fault.

For a crack located in the *j*-th element of the shaft line, the corresponding equivalent force systems are:

**(16) **

where the only terms different from zero are those relative respectively to the rotational horizontal and vertical d.o.f.s of the extreme nodes of element *j*.

**Equation (14) can be now rewritten considering that, if the machine performs a speed transient, the condition monitoring systems collect data for many rotating speeds, so the additional vibrations could be available for several rotating speeds and a set of np rotating speeds is considered: **

**(17) **

Then, introducing the admittance matrix [**E **(*n***Ω**)] of the system:

**(18) **

**eq. (14) becomes: **

**(19) **

However, the equivalent force fault identification problem in **eq. (19) is overdetermined since the number of the observations (the measured vibrations at different rotating speeds) is greater than the number of the equivalent crack forces that have to be identified. The procedure used to solve the problem is determined, by recalling that the vibration measuring planes in real machines are normally corresponding to bearing location. **

Let the machine model have *nr *nodes and *nb *are measured at *np *rotating speeds in two orthogonal directions, and the supporting structure is represented by means of a foundation with *nf *d.o.f.s or modes [**31]. In case of rigid foundation, obviously nf = 0. The fully assembled model d.o.f.s are (4nr + nf) while per each rotating speed are measured only nb d.o.f.s. Note that ne, number of elements, is nr – 1 and nr >> nb. **

Since the fault has to be identified not only in its severity but also in its position, in general the procedure has to be repeated per each node of the rotor, unless the research of the fault is limited in a specified interval of the nodes.

**Equation (19) represents the general system of equations for all the d.o.f.s of the considered fully assembled machine. The admittance matrix [ E (nΩ)] has order ((4nr + nf)np × (4nr + nf)np). **

Now, least square identification is used here in order to evaluate the module, the phase and a residual of the fault, starting from the first element and moving along the rotor from element to element. Weighted least square technique, or more sophisticated techniques, can be employed in order to increase the robustness of the identification as described in [**33][34]. **

The equivalent force system, is applied in each element of the rotor model, so for all the rotating speeds the fault vector is of order ((4*nr* + *nf*)*np ×* 2). Therefore, in the first element, the localization matrix is obtained by combining the localization vectors of **eq. (16) for the i-th rotating speed: **

**(20) **

and for all the *np *rotating speeds:

**(21) **

, which vector is of order (2*nbnp* × 2), due to unitary force systems applied to the model is now calculated. This is done by first substituting **eq. (20) in the r.h.s. of eq. (19), and inverting matrix [ E(nΩ)], obtaining the matrix [H(nΩ)]. **

**(22) **

Then, the vibrations of the d.o.f.s, which are measured, are separated from the remaining d.o.f.s of the system, by considering only the rows of [**H**(*n***Ω**)] corresponding to the measured d.o.f.s. The partitioned matrix is of order (2*nbnp* × (4*nr* + *nf*)*np*) and following **eq. (23) is obtained: **

**(23) **

Now the complex vector:

**(24) **

, of order (2*nbnp* × 1), has to be estimated. The fitting is done in a least square sense, since the number of the unknowns (the module and the phase) is less than the equations (recalling that data are corresponding to several rotating speeds and each of the sets is composed by several measuring planes). The problem is equivalent to:

**(25) **

whose general solution is given by means of the pseudo-inverse calculation:

**(26) **

are the equivalent bending moments due to the crack, in the selected element, which is the first element as indicated by superscript (1) in **eq. (26). Finally the residual in the selected element is determined, by first obtaining the calculated response due to the identified fault in the selected element: **

**(27) **

and then by normalizing it:

**(28) **

The procedure is then iterated for all the *ne *elements of the rotor (or of the subset if the research is limited to a rotor part). If a fault only is considered, a set in ### of relative residuals given by **eq. (28), ordered by the element number, is obtained: **

**(29) **

indicates the most probable location of the fault, whose amplitude estimation is given by the corresponding values:

**(30) **

of **eq. (26). The closer to zero the minimum value of eq. (29), the better the estimation of the faults per each harmonic component. **

An example of model based identification of a crack, actually an annular crack, in a steam turbine from 1X vibrations is reported in [**2][29]. Figure 9 shows the identified position of the crack in correspondence of the 1st high pressure stage. This was confirmed after removing the blades and the annular crack was in the bottom of blade bucket. **

**Figure 9 **Identification of the crack position in a steam turbine (left); crack in the bottom of blade bucket (right).

It is worth noting that the 1X vibration components are due both to the breathing mechanism of the crack and to the local bow which generally has developed during the crack propagation. Therefore, the 1X component is useful for the localization of the crack, but not for the identification of its depth.

The 3X component is rather small and generally masked by some noise. Often this component can be recognized only when approaching the resonant condition at a rotating speed equal to ⅓ of the rotor 1st critical speed.

The 2X component is therefore the most suitable symptom for detecting position and depth of the crack; the highest values are obviously reached during a speed transient when approaching the resonant condition at ½ of the rotor 1st critical speed.

The final remark about the model based method regards the accuracy of the machine model. A full discussion would be far from the scope of this paper, but a reader interested in measures of accuracy in model based identification can refer to [**32]. **

**4.1 Crack depth identification **

Crack depth identification has been presented in [**35] for transverse cracks. The method, described in the previous section, identifies the crack position in a particular element of the rotor, whose length is known from the f.e.m. and equal to l (see figure 8). Figure 8 shows also the crack depth a, the d.o.f.s of the extremity nodes of the element and the bending caused by the equivalent moment components Mn (1X, 2X and 3X components) that are applied to this element. These equivalent bending moment components M1, M2 and M3 have been calculated from the corresponding 1X, 2X and 3X measured vibrations. **

Then the static bending moment **M **in correspondence of the same element, due to the weight and to bearing alignment conditions, is calculated from model data.

The moments **M**1, **M**2 and **M**3 represent the amplitude of the equivalent crack forces defined respectively by **eqs. (8), (9) and (10). If we neglect all terms of eqs. (8), (9) and (10) except the first one, we obtain: **

**(31) **

This assumption is reasonable if heavy, horizontal axis, industrial rotating machineries are considered, since the static deflection, in the shaft sections loaded by consistent bending moments, is much higher than the vibrations, especially when operating conditions far from resonances in critical speeds are analyzed. Moreover, it is possible to show that the projections along the reference axes are equal in case of 2X component, i.e. *M*2(*x*) = *M*2(*y*) = *M*2 and *φ*2(*x*) = *φ*2(*y*), and the equivalent force system is a rotating moment of amplitude *M*2.

have order (4*nr* + *nf*) × 1, the only elements different from zero are those corresponding to the d.o.f.s of the cracked element *j*-th. Therefore, it is more convenient to consider only the sub-matrix [*Δ***K***n*]|*j *of order 8 × 8 corresponding to those d.o.f.s and the displacement vector **x***s*|*j *of the nodes of the cracked element (see **figure 8): **

**(32) **

By recalling **eqs. (16) and (31), it follows that: **

**(33) **

and for the modulus:

**(34) **

The matrix [**ΔK***n*] is restricted to the nodes of the cracked element, therefore only the static deflections of the nodes of the cracked element are required in **eq. (34). Since in the crack, as in the Bernoulli’s beam, the shear force can be neglected with respect to bending moment flexibility, xs|j is composed mainly by rotational deflections of the extremity nodes. Therefore xs|j can be calculated by considering the statical bending moments due to static, fixed in space, force vector W acting on the cracked element. These bending moments do not change much from node to node of a element, for a suitably refined mesh of the rotor, and can therefore be considered constant in the cracked element and equal to M. The moment M is then proportional to the relative rotations of the extremity nodes of the cracked element, ψj + 1 – ψj (see figure 8). **

By considering the sub-matrix [**K***m*]|*j *of order 8 × 8 corresponding to those d.o.f.s extracted from the matrix [**K***m*], i.e. the mean value of the stiffness matrix of the cracked element, the statical deflections of the cracked element are given also by:

**(35) **

By replacing **eq. (35) in eq. (34), it finally results: **

**(36) **

**Equation (36) shows that ratio Mn / M depends only on the crack relative depth p = a/D, since [ΔKn]|j and [Km]|j of the cracked element are function of p only, if regular shapes with rectilinear tips are considered for the transverse crack. **

A similar ratio *Mn*′/*M *has been calculated for different cracks depths considering the equivalent length *lc *of a reference cracked beam element, instead of the length *l *of the cracked element in the model of the shaft. This equivalent length *lc *has been tuned by means of numerical 3D analyses as shown in [**1]. **

Moment *Mn*′ gives the same relative rotation of the nodes of the beam with length *lc *as *Mn *does with a beam with length *l*:

**(37) **

This is represented in **figure 10 for the 1X, 2X and 3 X components and expressed by the relationship in: **

**Figure 10 **Ratio of the bending moments on the equivalent cracked beam, as a function of crack relative depth *p* = *a*/*D *for the *n*X component.

**(38) **

In the same figure the curve *M*2*slot*′/*M *for an always open crack (a slot or notch) is shown: in this case the 2X component only is present and 1X and 3X component are absent.

**Equation (38) can then be used for determining the crack depth. However, as shown in [1], the length lc of the equivalent, reduced stiffness, beam element that simulates the behaviour of the cracked beam, is also depending on the relative crack depth p: **

**(39) **

The function *g*(*p*) is represented in **figure 11. Anyhow, coming from the identification procedure, the equivalent bending moments Mn are applied to an element with an actual length l instead of lc. It is worth noting that the nX measured displacements are the effect of the relative rotation of the cracked element extremity nodes, which is proportional to the product Mn · l of the identified nX bending moment component Mn applied to one element of the finite element model of the rotor, multiplied by its length l. **

**Figure 11 **Relationship between the crack relative depth *p*, the diameter *D *and the length *lc *of equivalent beam.

The equivalent bending moment component *Mn*′, applied to an equivalent cracked beam element of length *lc*, can therefore be calculated as:

**(40) **

By assuming that the static bending moment *M *applied to the original element of length *l *does not change along the element, the same *M *can be considered applied to the element with equivalent length *lc*. Recalling **eq. (38) we can derive: **

**(41) **

and using **eq. (39) we obtain: **

**(42) **

**Equation (42), shown in figure 12 for the nX components, can then be used for determining, from the known left hand side, the relative depth of the crack. **

**Figure 12 **Function for the calculation of the crack depth.

The proposed method has been validated by means of experimental tests performed on suitable test rigs, as described in detail in [**35][36]. An example is shown in figure 13. The crack position along the rotor corresponds again to element 20 and shaft diameter is 70 mm. Crack position is correctly identified by the minimum of residual curve. At the end of the tests the shaft was broken to check the actual shape and depth of the crack. From the analysis of figure 13, the crack average depth results of only about 14% in average, with a maximum depth of 11.3 mm equal to 16.1%. A relevant part of this depth is due to the notch used to let the crack starting the propagation, but form the consideration of figure 12, there is no remarkable difference between slot and 2X f(p) · g(p) curves for small crack depth. **

**Figure 13 **Crack depth identification in a test rig (left); crack shape and depth after shaft breaking.

**5 CONCLUSIONS **

The presence of cracks in rotors is one of the most analyzed topics in rotordynamics, due to the serious damages that may happen if this fault is not early detected. Several aspects of cracked rotors are covered in this paper, with special care to those which may affect real rotating machinery and with the support of experimental testing.

In the first part, the breathing mechanism is analyzed in detail and experimental results are presented in order to show the actual breathing of a cracked rotor. The second part presents some examples of different crack shapes and discusses crack propagation. Finally, the third part presents a model based method that, by means of inverse dynamics approach, allows the position and the depth of a crack to be identified using shaft vibrations measured by the condition monitoring system of the machine.

**6 REFERENCE LIST **

[1] Bachschmid, N., Pennacchi, P., Tanzi, E., Cracked Rotors, A Survey on Static and Dynamic Behaviour Including Modelling and Diagnosis. Springer, Berlin – Heidelberg, 2010.:1–401.

[2] Vania, A., Pennacchi, P., Chatterton, S. Effects of thermal transients on cracked shaft vibrations. In: *Proceedings of ASME Turbo Expo 2011 - GT2011*. British Columbia, Canada: Vancouver; 2011:1–8. [paper GT2011-46181, June 6-10].

[3] Mayes, I.W., Davies, W.G.R., The vibrational behaviour of a rotating shaft system containing a transverse crack. IMechE Conf. Vibration in Rotating Machinery, paper C168/76, Sept. 1976. 1976:53–64. [Cambridge, UK].

[4] Nelson, H.D., Nataraj, C. The dynamics of a rotor system with a cracked shaft. *Journal of Vibration and Acoustics*. 1986; 108:189–196.

[5] Wauer, J. Modeling and formulation of equations of motion for cracked rotating shafts. *International Journal of Solids and Structures*. 1990; 26(8):901–914.

[6] Mayes, I.W., Davies, W.G.R. The vibrational behavior of a multi-shaft, multi-bearing system in the presence of a propagating transverse crack. *Journal of Vibration and Acoustics*. 1984; 106:139–145.

[7] Grabowski, B., The vibrational behavior of a turbine rotor containing a transverse crack. ASME Design Engineering Technology Conf., Saint-Louis, Paper N◦79-DET-67. 1979.

[8] Ishida, Y., Ikeda, T., Yamamoto, T., Masuda, N. Vibrations of a rotating shaft containing a transverse crack. *JSME International Journal Series III*. 1988; 31(1):22–29.

[9] Andrieux, S., Varé, C. A 3D cracked beam model with unilateral contact. *Application to rotors. European Journal of Mechanics A/Solids*. 2002; 21:793–810.

[10] Sinou, J.-J., Lees, A.W. Influence of cracks in rotating shafts. *Journal of Sound and Vibration*. 2005; 285(4–5):1015–1037.

[11] Patel, T.H., Darpe, A.K. Influence of crack breathing model on nonlinear dynamics of a cracked rotor. *Journal of Sound and Vibration*. 2008; 311(3–5):953–972.

[12] Sawicki, J.T., Friswell, M.I., Kulesza, Z., Wroblewski, A., Lekki, J.D. Detecting cracked rotors using auxiliary harmonic excitation. *Journal of Sound and Vibration*. 2011; 330(7):1365–1381.

[13] Ishida, Y. Cracked rotors: Industrial machine case histories and nonlinear effects shown by simple Jeffcott rotor. *Mechanical Systems and Signal Processing*. 2008; 22:805–817.

[14] Inoue, T., Yamamichi, T., Kato, M., Ishida, Y. Vibration analysis and detection of rotor crack in a multi-disk rotor system by periodic excitation using AMB. *Proceedings of the ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, DETC2007*. 2007; v 1:1205–1214. [PART B].

[15] Sinou, J.-J., Faverjon, B. The vibration signature of chordal cracks in a rotor system including uncertainties. *Journal of Sound and Vibration*. 2012; 331(1):138–154.

[16] Dimarogonas, A.D., Papadopoulos, C.A. Vibration of cracked shafts in bending. *Journal of Sound Vibration*. 1983; 91(4):583–593.

[17] Darpe, A.K., Gupta, K., Chawla, A. Couple bending longitudinal and torsional vibrations of a cracked rotor. *Journal of Sound and Vibration*. 2004; 269(1–2):33–60.

[18] Darpe, A.K., Gupta, K., Chawla, A. Transient response and breathing behaviour of a cracked Jeffcott rotor. *Journal of Sound and Vibration*. 2004; 272(1–2):207–243.

[19] Papadopoulos, C.A. Some comments on the calculation of the local flexibility of cracked shafts. *Journal of Sound and Vibration*. 2004; 278(4–5):1205–1211.

[20] Papadopoulos, C.A. The strain energy release approach for modeling cracks in rotors: A state of the art review. *Mechanical Systems and Signal Processing*. 2008; 22:763–789.

[21] Sekhar, A.S. Multiple cracks effects and identification. *Mechanical Systems and Signal Processing*. 2008; 22:845–878.

[22] Sekhar, A.S. Vibration characteristics of a cracked rotor with two open cracks. *Journal of Sound and Vibration*. 1999; 223:497–512.

[23] Sekhar, A.S., Kumar, Dey J. Effects of cracks on rotor system instability. *Mechanism and Machine Theory*. 2000; 35:1657–1674.

[24] Stoisser, C.M., Audebert, S. A comprehensive theoretical, numerical and experimental approach for crack detection in power plant rotating machinery. *Mechanical Systems and Signal Processing*. 2008; 22:818–844.

[25] Bachschmid, N., Pennacchi, P., Tanzi, E. Some remarks on breathing mechanism, on non-linear effects and on slant and helicoidal cracks. *Mechanical Systems and Signal Processing*. 2008; 22:879–904.

[26] Borri, B., Carpinteri, M., Chiaia, B. Contact, Closure and Friction Behaviour of Rough Crack Concrete Surface. Fracture mechanics of concrete structures. *Freiburg*. 1998; 3:1635–1644.

[27] Andrier, B., Garbay, E., Hasnaoui, F., Massin, P., Verrier, P. Investigation of helix-shaped and transverse crack propagation in rotor shafts based on disk shrunk technology. *Nuclear Engineering and Design*. 2006; 236:333–349.

[28] Stoisser, C.M., Audebert, S. A comprehensive theoretical, numerical and experimental approach for crack detection in power plant rotating machinery. *Mechanical Systems and Signal Processing*. 2008; 22:818–844.

[29] Pennacchi, P., Vania, A., Shaft crack detection in a steam turbine: experimental evidences and model-based simulations. Proceedings of ISMA2010, International Conference on Noise and Vibration Engineering including USD2010, Leuven, Belgium, September 20-22. 2010:1055–1070.

[30] Bachschmid, N., Pennacchi, P., Vania, A. Identification of multiple faults in rotor systems. *Journal of Sound and Vibration*. 2002; 254(2):327–366.

[31] Pennacchi, P., Bachschmid, N., Vania, A., Zanetta, G.A., Gregori, L. Use of modal representation for the supporting structure in model based fault identification of large rotating machinery: part 1 – theoretical remarks. *Mechanical Systems and Signal Processing*. 2006; 20(3):662–681.

[32] Vania, A., Pennacchi, P. Experimental and theoretical application of fault identification measures of accuracy in rotating machine diagnostics. *Mechanical Systems and Signal Processing*. 2004; 18(2):329–352.

[33] Pennacchi, P., Vania, A., Bachschmid, N. Increasing the robustness of fault identification in rotor dynamics by means of M-estimators. *Mechanical Systems and Signal Processing*. 2007; 21(8):3003–3029.

[34] Pennacchi, P. Robust estimate of excitations in mechanical systems using M-estimators – Theoretical background. *Journal of Sound and Vibration*. 2008; 310(4–5):923–946.

[35] Pennacchi, P., Bachschmid, N., Vania, A. A model-based identification method of transverse cracks in rotating shafts suitable for industrial machines. *Mechanical Systems and Signal Processing*. 2006; 20(8):2112–2147.

[36] Bachschmid, N., Pennacchi, P., Tanzi, E., Vania, A. Identification of Transverse Crack Position and Depth in Rotor Systems. *Meccanica, International Journal of the Italian Association of Theoretical and Applied Mechanics*. 2000; 35(6):563–582.

***© Paolo Pennacchi, 2012 **

**Model based signal processing in Smart rotating machines **

**I. Bucher *, Mechanical Engineering, Technion, Israel **

**ABSTRACT **

Model-based signal-processing bridges the gap between numerical modelling and experimental testing of rotating machines. The present paper focuses on signal-processing methods exploiting multiple sensors and capable of handling transient dynamics under varying rotation speeds. These methods rely on special features of rotating elements, e.g. cyclic-symmetry, gyroscopic effects, directional whirling and circumferentially traveling deformations. The new methods could be one of the building blocks leading towards smarter and hence better and safer rotating machines.The ‘eyes’ of ‘Smart Rotating Machines’ are the sensors and the accompanied, real-time signal processing methods play the role of a ‘brain’ in the assessment of measured data. Indeed ‘smart’ also means combining advanced sensing capabilities with an electronic brain aware of the underlying physics laws that are captured in a model. At the moment it seems that the pendulum leans heavily towards numerical modelling. Finite Element models are the basis for analysis and design, while testing and measurements provide only limited validation means for some of the model parameters partly due to poor deployment of sensors and simplistic signal processing procedures.The presentation will highlight the some of the advantages model-based signal processing offers over past and presently used methods and will try to point towards a path leading from older methods and techniques towards present, state-of-the-art methods and further into the future where smart machines will have ‘eyes’ and ‘brains’.Specifically, the presentation will describe spatial, temporal and directional decomposition of rotating machine vibrations during rapid rotational accelerations. Real time signal processing methods that exploit Hilbert transform based decompositions; directional order-tracking and time-frequency maps will be demonstrated via simulations and experiments. The spatial and temporal decomposition method enables a Smart-Machine to assess the true stress and strain levels on parts rotating relative to an array of sensors and thus to enhance safety.The advance in sensors and electronics alongside their price and size reduction together with improved wireless capabilities will benefit the proposed model-based signal processing.

**1 INTRODUCTION **

Rotating machines exhibit complex vibration patterns combining several response components having different time and speed-dependent frequencies [**1–4]. Some of these components overlap in the frequency domain despite representing different kinematic and physical features. The rich dynamics of rotating machines makes it difficult to isolate particular vibratory modes from typical measured signals. It turns out that rotating structures have some unique features that can be exploited to isolate otherwise hidden signal-components when suitable model-based signal processing methods are employed. Two main features of rotating structures become useful for model-based signal processing and identification: (i) The inherent polar cyclic-symmetry of typical machines and (ii) Rotation which creates relative motion with respect to the foundation or sensors. **

The cyclically-symmetric structure of rotating machines creates speed dependent modulated responses [**5–8] where the rotating and vibrating structure exhibits wave-like propagating deformations. This fact has been often exploited to model rotating (bladed) disks [9–11] and to extract information about the nature of rotating disk and shaft vibration [12–16]. **

Rotation interacts with stationary forces and with the supporting structure in such a way that the spatial distribution of mass and stiffness come into play. Indeed, this fact complicates the analysis [**5,6,14] considerably, but with the aid of an externally controlled excitation, one can extract meaningful information about imperfections in the rotating parts [17,18]. The latter makes use of parametric excitation that takes place when special conditions coincide or when an active magnetic bearing can create them [19]. **

Rotation poses some unique difficulties alongside some special opportunities to obtain information. Any smart identification or fault detection schemes have usually limited accessibility to rotating parts and sensors must be deployed at the specific spatial locations in order to enable the appropriate extraction of information. Indeed, sensing is an important issue that finds new answers with the advent of technology [**20–23,25], i.e. scanning laser sensors, or by deploying circular arrays of sensors [8,12]. **

Signal processing plays an important role in the identification of rotating machine dynamics. Recent approaches include computed order tracking [**24,25] and various de-modulation techniques [3], Hilbert transform [26] and more traditional approaches [27]. **

The paper discusses 3 generic examples where model based signal processing helps in obtaining deeper insight into the dynamical behaviour of rotating machines. These examples include (a) decomposition of disk-like vibrations in real time (b) engine order decomposition (c) active detection of asymmetry in rotating parts.

**2 Detecting Traveling Wave Direction -Multiple Sensors **

Consider an elastic structure vibrating and rotating simultaneously. Time-varying deformations are often measured by sensors that are either fixed in space (e.g. eddy-current probes), rotating with the system (e.g. laser [**20,28]) or moving relative to the vibrating structure. **

Every sensor deployment can capture a part of the temporal-spatial vibration state while a full picture can be obtained only for some cases or when multiple sensors are deployed in special patterns along the structure. Cyclic-symmetry [**10,11] gives rise to groups of modes with similar spatial characteristics and close natural frequencies. The special properties of rotating machines and the forces caused by rotation give rise to deformation patterns moving rotationally relative to the structure, these special combinations are referred to as ‘traveling waves’. Fig. (1) describes a wave traveling on a bladed disk by capturing its state at several discrete time instances. **

**Figure 1 **Traveling 2-nodal diameter mode at 3 different time instances and a single sensor, fixed in space, measuring the out-of-plane vibrations

**Figure 2 **Deformed disc state as a combination of 2,3 and 8 nodal diameter modes and an array of 8 sensors

While **Fig. (1) shows a single mode of vibration, in reality several modes of vibration combine to create the deformed state, as shown below for a vibrating disc. **

Being cyclically symmetric, the response measured at a fixed radius *r*0 is a spatially periodic function of θ (the angle of rotation). Assuming that the rotating disk operates under time varying conditions, its response, as measured by a sensor located at the polar coordinates -(*r*0,θ), can be expressed as:

**(1) **

**Equation (1) does not assume constant speed of rotation and therefore an (t), bn (t) are general functions of time. Indeed, the sensors spread around the rotating disc ‘see’ the material point they illuminate, as illustrated in Fig. (3). The integer n appears in cos n θ, sin n θ with n = – ∞ denoting the number of nodal diameters (ND). **

**Figure 3 **

In practice, many nodal diameters participate in creating the deformation of these structures, and the time varying coefficients in **Eq. (1) can be expressed as [8]: **

**(2) **

These functions are sinusoids whose instantaneous amplitudes (α*r*(*t*), β*r*(*t*)) and phases (φ*r *(*t*), ϑ *r*(*t*)) change with time. **Equation (1) can also be expressed as: **

**(3) **

and *f*(*t*) ∈ {*s*,*a*,*b*}, *H*[⋅] denotes the Hilbert transform.

Substituting **Eq. (2) in Eq. (1), one obtains: **

**(4) **

**, represent forward and backward traveling waves respectively (see [29]). The ℜ operator projects the result onto the real axis. **

One can easily separate the time-varying coefficients, at least theoretically, by integration the response along the circumference (see [**29]): **

**(5) **

**Equation (5) can separate signal components having different wavelengths (number of ND), even under non-stationary e.g. rotationally accelerating structures. **

In order to illustrate the necessity for separating the various components, a time-frequency distribution (TFD) or spectral map, is shown for simulated and real measurement taken from a rotating disc.

**Figure 4 illustrates via simulated and experimental data the density and overlap of spectral lines for a simple rotating disc. The overlap of these lines makes it very difficult to decipher and separate the physical mechanism behind the vibration patterns. **

**Figure 4 **Evolution of spectral contents vs. time for a rotationally accelerating disc; measurement taken from a single sensor (see **Figs. (1-3)). Left: simulated as detailed in the appendix; Right: measured **

*Spatial separation with a discrete array of sensors *

*Spatial separation with a discrete array of sensors*

In **Eq. (5), the analysis was carried out assuming that the sensors are continuously distributed along the circumference. Clearly, sensors can be placed on the structure only at discrete locations. **

Consider the discrete Fourier transform of *N *signals coming from *N *. Uniform spacing in the entire range can guarantee exact results, completely identical to **Eq. (5) with no need to apply a spectral window [30]. It is important to stress that no numerical approximation is involved in this part. **

The *k*th coefficient (time function) can be computed exactly in a completely analogous manner to **Eq. (5), as long as N is large enough to rule out spatial aliasing ([29,30]): **

**(6) **

Now, computing the analytic signal of **Eq. (6) as before [28], one has: **

**(7) **

where:

**(8) **

The Hilbert transform of each measured sensor is denoted:

**(9) **

One can now obtain an explicit expression for the analytic signal of the *k*th wavelength:

**(10) **

Separating real and imaginary, one finally obtains:

**(11) **

It is worth mentioning that the negative on-going vibration waves are obtained for *k *, i.e.:

**(12) **

The wave components are inevitably complex functions of time thus they represent both amplitude and phase information at any moment in time:

**(13) **

**Equation (11) is the result of both a Hilbert and a subsequent spatial Fourier transform. The process produces time functions representing the instantaneous variation of the nth nodal diameter modes traveling in either the positive or negative direction. **

The number of sensors, *N*, should be chosen according the shortest wavelength (i.e. the largest number of ND) one expects to be excited within the frequency range of interest ([**31]). Assuming that nmax is the largest number of nodal diameters (ND) present in the response, one should place at least N = 2nradians apart at some constant radial distance in order to avoid spatial aliasing. **

The proposed algorithm can be described schematically in a diagram as shown in **Fig. (5). **

**Figure 5 **Schematic description of the proposed decomposition per ND and travel direction

The process requires the user to apply the Hilbert transform to the measured time data. A second stage combines the data from all the sensors to perform the spatial and directional separation so as the time data (in a complex form) related to each wavelength (no. of ND).

*Spatial separation of transient vibrations – simulated example *

*Spatial separation of transient vibrations – simulated example*

Consider a rotating disk whose dynamics are measured by *N* = 16 equispaced sensors. The simulated measurements for each one of the 16 sensors are computed via:

**(14) **

The instantaneous phase of each component, representing either run-up or run-down of the rotating system, is computed by:

**(15) **

and the instantaneous amplitude, as can be seen in **. The set of parameters that was used in the simulation is provided in Table 1 at the appendix. Applying the proposed decomposition, Fig. (6) is obtained. **

**Figure 6 **Left: Decomposed time vibrations *pnR*(*t*), Right: Raw data from sensor #1

**Figure (6) shows the effectiveness of the proposed separation as outlined in Fig. (5) and in Eq. (9)-(16). While the raw data (Fig.6-right) looks erratic, even when smaller time fragments are considered, the decomposed components have a clear structure and the original simulated components have been restored. Opposed to the complex looking time frequency distribution of the present case as shown in Fig. (4), the time-frequency distribution (TFD) of one of the components in Fig. (6) has nearly little to none cross-coupling effects. Examination of Fig. (7) shows two crossing spectral lines that belong to forward and backward travel directions respectively. **

**Figure 7 **Decomposed time vibrations of a single component - *p*3*R*(*t*)

Clearly, **Fig. (7) shows two components, both related to n = 3, which have been extracted from the complicated signal (see Fig. (4)). These two components can be further separated as they represent forward and backward traveling waves. The direction of travel is commonly represented in terms of positive and negative frequencies, respectively [2,8,12,32,33]. **

*Spatial and directional separation of transient vibrations – experiment *

*Spatial and directional separation of transient vibrations – experiment*

The experimental system, shown in **Fig. (8), consisted of a 20 mm diameter flexible shaft, which was 1.0 meter long. The shaft was mounted on two self-aligning ball-bearings and it was driven by a 20 Nm rated AC brushless motor capable of running up to 4500 RPM. A flexible disk was mounted on the other end of the shaft. A virtual cut-through is shown in the middle of the shaft through which an active magnetic bearing can be seen. In the present test, the magnetic bearing produced a small, band-limited random force to enrich the measurements and simulate the effect of turbulence in real jet-engines. Not shown in Fig (8) are the exciter driving the disk and the sensor-array measuring the disk vibration. The shaft vibrations were measured by two sensors connected at a right angle as can be seen in Fig (8). The electromagnet driving the flexible disk was fed by a DC biased random current to excite the otherwise unexcited disk modes with more than 1 nodal diameter. The motor speed was increased until it reached a predetermined upper limit, but the motor controller’s gain was set to a low value so as to increase the interaction between the shaft vibration and speed of rotation and thus to create some speed fluctuations, as can be seen in Fig. (4). The fluctuations in speed are manifested in the deviation of the engine orders (1X,2X,etc.) from a straight line during the acceleration. **

**Figure 8 **Left: Cad model of rig with disc and magnetic bearings: Right photograph

The run-up data time-frequency