Detection of simultaneous mechanical faults in 6‐kV pumping induction motors using combined MCSA and stray flux methods

Konstantinos N. Gyftakis, School of Engineering, University of Edinburgh, The King’s Buildings, Edinburgh, EH93DW UK. Email. k.n.gyftakis@ieee.org Abstract The popularity of stray and air‐gap flux monitoring methods is increasing. This trend is justified by several advantages of such methods over the stator current monitoring that has been demonstrated for electrical fault detection in the induction and synchronous machines. However, the use of the magnetic flux for mechanical faults detections has not drawn this much attention, while in industry, the vibration analysis continues to be popular. This study comes to bridge this gap via the detection of mechanical faults of 6‐ kV induction motors in a pumping station. The diagnostic procedure mainly involves the stator current and stray flux monitoring, and harmonic index analysis. The localisation of the fault has been made possible via oscillometer readings. It will be shown that mechanical faults have a very different impact on the stator current and the flux signals, while the flux is not sensitive to the bearing fault mechanisms.


| INTRODUCTION
Electrical machines are the heart of the modern world, producing electric power or consuming it to produce mechanical work and consequently products and services. Due to their critical role in our sustainability, electrical machines, condition monitoring and fault diagnosis have undergone a significant advancement. This is because undetected faults will evolve into higher severity levels and lead to catastrophic machine failures with a series of negative persecutions such as high financial losses, production delays and compromisation of safety [1][2][3][4].
Surveys and reviews have shown that the failure modes are dependent on the size of the electrical machines [5][6][7]. Low voltage motors suffer mainly from mechanical faults, which consist of three-quarters of total faults. Medium voltage motors are quite interesting because bearing faults are the dominant fault but with a very low difference than stator faults. However, the use of sleeve bearings in large machines consists of stator faults, the most frequent faulty condition. The various faults frequency is shown in Figure 1 as a function of the size of the electric motor.
Various methods have been developed over the years aiming for the detection of different faults in induction motors. The favourite seems to be the motor current signature analysis (MCSA) [8][9][10][11]. The application of this method depends on the monitoring of the motor's current during operation and the analysis of its frequency spectrum via the fast Fourier transform (FFT). Despite its acceptance from industry, this method has been found unreliable in late years, leading to false alarms due to various conditions of induction machines [12][13][14].
Other diagnostic methods include the use of the motor's output torque [15,16], vibrations monitoring [17,18], stray flux [19,21] and input electric power [22,23]. Among those, the stray magnetic flux appears to be quite promising due to the associated low-cost and non-intrusive character, while it has proved to be immune to certain phenomena leading to false alarms with the MCSA [24]. The stray flux monitoring allows the capture and analysis of either the axial flux or the radial one or a combination of both depending on the positioning of the flux sensors on the machine body ( Figure 2) [25]. The electromotive force sensed with the flux sensors is machine geometry free, thus senses harmonics that are cancelled out or hidden in the stator current because it depends on the machine's number of poles.
This study demonstrates the application of a series of diagnostic methods to detect and locate mechanical faults, such as bearing faults and misalignment, in 6-kV induction motors used for water pumping applications. No prior knowledge of the mechanical system parameters existed. The study has been carried out with the simultaneous application of the MCSA and flux monitoring, while the localisation of the fault has been achieved by the supplementary use of an oscillometer. The results demonstrate that the stray flux is not sensitive to actual bearing faults, while the stator current is there superior. However, the stray flux monitoring seems sensitive to the misalignment fault and consequent mechanical oscillations. However, since in electrical faults monitoring, the stator current is not always reliable, for generalised monitoring, it is necessary for the two methods to work together focusing on different failures [26]. A detailed discussion will close this study.

| THE PUMPING SYSTEMS' OPERATING CHARACTERISTICS
The primary aim of the monitored system is to pump water from the torrent to the main lake of the hydroplant. The water is pumped up to 20 m. To serve this purpose, five vertically installed motors driving pumps are used. During the visit, one of the motors (Motor 1) was stopped due to pump repairs and thus has not been considered in this study. Four motors' nameplate data are presented in the following Tables 1 and 2.

| MONITORING OF ELECTROMAGNETIC SIGNALS
The four motors were operating under rated conditions when the condition monitoring took place. More specifically, current clamps were installed to each motor's supply to capture the stator current waveforms ( Figure 5). At the same time, flux sensors were placed on the motors' bodies. Aiming to detect possible faults with axial impact, such as a possible level of inclined eccentricity or a bent shaft, two flux sensors were installed on each motor perfectly aligned and in parallel with the motor shaft. Their surface is perpendicular to the radial direction; thus, they monitor a combination of axial and radial flux. However, due to their long distance from the ends of the machine, the main dominant contributor is the radial flux component. The installed flux sensors on one of the motors are illustrated in Figure 6.
The stator current measurements were carried out without an MCSA commercial equipment. Current clamps were used with a sensitivity of 10 mV/A and an accuracy of ±1% of reading at ±100/±500 mA. Providing a safety Bayonet Neill-Concelman connector, this measurement was at a later stage logged onto a digital high-resolution buffer memory unit used for acquisition of signal waveforms and data. All three stator currents were recorded for three-phase inspection and detection of possible asymmetries.
There are two options for monitoring the stray flux, namely the rigid coil sensors and Hall sensors. In this case, coil sensors have been applied. The sensors were built in the laboratory using a custom made winding machine and a 3D printer. The sensors geometrical features are shown in Figure 7.
Due to the high reluctance of the air, the stray flux is generally weak. This is why, the flux sensors were designed under two constraints: low coil length and high number of turns. The coil was built with a very fine copper wire of 0.1 mm diameter leading to a total of 3500 turns. The per-length resistance of the wire is 3.4 Ω/m. During operation, the rotating magnetic field will induce an electromotive force to the sensors, so in order to record that, a voltage probe was installed at every sensor recording the voltage across the coil turns.
The current and flux sensor signals are captured by a portable high resolution, deep memory, eight-channel oscilloscope. Offering a 12-bit resolution and serial bus decoding with 256 MS buffer memory and a 20 MHz bandwidth, each signal waveform was captured within 12 frames of 10 s each, providing the ability to gather extended waveforms over the steady-state of the motors for reliable signal representation in both the time and frequency domain. The sampling frequency has been 15 kHz.

| STATOR CURRENT MONITORING
Due to lack of history and healthy condition data, the stator current spectra are compared between identical motors of the same power. The MCSA results for Motors 3 and 5 are shown in Figure 8. The stator current spectra analysis reveals the existence of additional harmonics in Motor 5. The harmonics appear as sidebands of the fundamental harmonic at 50 Hz. The first local maximum is located at 42.38 Hz, that is: Hz. However, the right sidebands appear as well together with multiples of this frequency. The recorded amplitudes of the f s ± k · f o sidebands are shown in Table 3. Interestingly, extra sidebands appear next to each one of the above-mentioned harmonics. The right one has distance f or ¼1.5 Hz, while the left f ol ¼1.08 Hz from the f s ± k · f o ones. The amplitudes of all sidebands are summarised in Table 4.
The observed fault harmonics in Motor 5 do not obey to the well-known formulas revealing rotor eccentricity ( f ecc ) [4] or broken rotor bar/end ring failures ( f bb ) [27], as shown below: where k ¼ 1; 2; 3; …, R : the rotor slot number, s: the slip, p: the number of pole pairs, f s : the fundamental frequency, n d ¼ 0 ðf or staticÞ; 1; 2; 3; …ðf or dynamicÞ, n sa : the saturation effect (=1, 2, 3,…) and n ws : time-harmonic rank (=1, 2, 3,…). It is to be noted that the first part of Equation (1) is for mixed eccentricity, while the second for only-static, only-dynamic or a combination of both.
Since the observed signatures are not satisfied by either Equation (1) or (2), they should be caused by some mechanical failure. Bearing faults is the first that comes to mind. Rolling element bearing faults cause signatures at frequencies [4,28]: where m = 1, 2, 3,… and f i;o and f c given by: where N b ; f r ; D b ; D c and β are the number of bearing balls, the mechanical rotor speed frequency, the ball diameter, the bearing pitch diameter and the contact angle of the balls, respectively. The frequency subscripts i; o; c stand for the ball pass inner raceway, the ball pass outer raceway and the fundamental cage. As a rule of thumb and if the number of balls in between 6 and 12, the corresponding values can be calculated as follows:

| MAGNETIC FLUX MONITORING
As mentioned before, two sensors have been installed on every induction motor, and they monitor a combination of axial and radial flux. The electromotive force (EMF) waveforms have been analysed with the application of the FFT, and their frequency spectra are presented in Figures 9  and 10.
From Figures 9 and 10, no axial dissymmetry is detected in both motors as the spectra from both sensors are almost identical. This is further supported by the flux waveforms over time such as the one for Motor 5 shown in Figure 11 -647 5 is characterised by increased harmonic index at frequencies associated with the mechanical speed: f r ¼ ð1−sÞ p f s . More specifically, both sensors mounted on Motor 5 have captured harmonic components at frequencies: f s ± k · f r ; k ∈ ℕ: This signatures series exists very clearly for values: k ¼ ½−3; 3�. On the other hand, Motor 3 has only the main sidebands for k ¼ ½−1; 1�. These harmonics are well known to be associated with mixed eccentricity [29] and misalignment [30] conditions; however, it is worth mentioning that they have been lately associated with rotor electrical faults as well [31]. However, no signs of rotor electrical faults exist in the stator current, a fact that leads to the eccentricity or misalignment as the main suspect for producing the above harmonics in the flux. The frequency f r has been found to be 12.45 Hz, which is synonymous to a mechanical speed of 747 rpm. The amplitudes of all eccentricity harmonics have been extracted and presented in Tables 5 and 6. The increased harmonic index om Motor 5, associated with mixed eccentricity and load imbalances, supplements the harmonic index of the stator current that points towards bearing failures while these two conditions are always related and often appear together. Interestingly, the mechanical frequency related harmonics do not clearly appear in the stator current, possibly due to low oscillations level.

| ADDITIONAL INVESTIGATION AND DIAGNOSTIC ANALYSIS
The properties of the mechanical system's are unknown while involving many different bearing types, namely: motor bearings, shaft bearings and pump bearings. In any case, the multiple harmonics of the current spectra indicate a complex failure. Absent geometrical data, Equation (6) is applied while taking into account the mechanical frequency of 12.45 Hz. For 6 < N b < 12, Equation (6) gives: The harmonics described by Equation (7) need to be subtracted from the fundamental to point at possible fault signatures in the stator current spectrum (Equation [3]). Table 7 summarises all the possible signatures' locations. The signatures that look to be close to the ones monitored in the MCSA spectrum appear in bold. Of course, this is just an indication and not proof because the harmonics have been calculated by the approximated formulae. In any case, the inner raceway and cage faults may produce the observed stator current harmonics. As a result, and since broken rotor bars or other rotor electrical failures have been excluded as possible, the diagnostic conclusion is that the fault is of mechanical nature and most probably a combination of different faults.
Though the source of the fault is still unknown. In such systems with a long distance between motor and pump, it is of interest to locate the fault with accuracy as this would save a lot of money and time for the inspection and repair. To pinpoint the exact source of the main failure, a supplementary diagnostic step was taken. Extra measurements were taken using a portable oscillometer with the following characteristics: (parameters are specified at 25°C) ➢ Sensitivity: 20 mV/mm/s ±5% at 100 Hz. ➢ Sensitivity temperature coefficient: 0.2%/°C. ➢ Frequency response: 4.5-1000 Hz + 0, −3dB.
The use of the oscillometer at the induction motor's load side is presented in Figure 12. Detailed recordings were taken at the motor and the pump sides.
The oscillation measurements were as follows at the different points of interest: � Upper motor part (x-axis 35 μm p-p, y-axis at direction of the discharge tube 27 μm p-p). � Lower motor part (x-axis 18 μm p-p, y-axis at direction of discharge tube 20 μm p-p). � Middle pump part (x-axis 120 μm p-p, y-axis at direction of discharge tube 110 μm p-p). � Lower pump part (x-axis 72 μm p-p, y-axis at direction of discharge tube 70 μm p-p, z-axis 70 μm p-p).
The additional investigation using the oscillometer lead to the conclusion that increased oscillations which appeared in the middle and lower part of the pump. The diagnostic outcome from the motor's behaviour led to the conclusion that the observed electromagnetic index was due to shaft misalignment causing faster ageing of the bearings and possible impeller damage. Therefore, the shaft had to be dismantled ( Figure 13) and the impeller replaced. The inspection showed extensive degradation of all of them.
Degradation of the load part with such a degradation severity level was improbable to have left the motor completely intact. That logical conclusion, however, was not supported by the measurements. On the one hand, the MCSA analysis did not reveal any important eccentricity harmonics in the stator current, while on the other, the oscillometer readings were quite low and non-alarming. Despite that, since the system was not operational due to the load repair, a decision was made to inspect the induction motor as well. The inspection also revealed that the misalignment had actually caused significant damage at the induction motor's bearings, which also had to be replaced ( Figure 14). The damage was distributed in a uniform -649 manner along the circumference of the bearings. Interestingly though, the degradation had affected the bearings in the axial direction and not the radial, possibly due to axial oscillations. It has been concluded that the vertical alignment of the installation together with the uniform degradation of the motor's bearings was probably shielding the fault's existence from both diagnostic approaches applied in this study.

| CRITICAL DISCUSSION
In this session, an effort will be made to critically appraise the outcomes from this study with the literature and extract useful conclusions for future research efforts.
First, this study clearly shows that the MCSA is highly sensitive to mechanical failures. The signatures were very clear and with significant amplitudes clearly leading to a diagnostic alarm. The method relies on the adoption of the induction motor itself and, more specifically, its stator winding as the sensor to depict failures in the whole system, in this case, faults coming from the shaft bearings and the pump. There is superiority of the MCSA compared to vibration analysis associated with the low cost and remote monitoring capabilities.
Second, the stray flux is totally insensitive to the mechanical faults and consequent mechanical oscillations that originate from the load. None of the signatures that appeared in the stator current spectrum existed in either of the two installed flux sensors. Despite that, the flux sensor was more sensitive to the rotor eccentricity and misalignment than the MCSA, leading to a family of many sidebands. Overall, the diagnostic contribution of the stray flux monitoring is this particular case was of low value.
Furthermore, when a system is considered, the diagnosis of the fault is the first necessary step. However, localisation of the fault is then required and to that purpose, secondary methods are required (in this case, the oscillometer measuring displacements and mechanical oscillations).
In recent studies, the stray flux appears to be the solution to many misdiagnosis cases associated with the MCSA such as the rotor axial cooling air ducts, which are misinterpreted for broken rotor bars [24]. Past experience in the field strongly suggests that relying on the MCSA alone is not a reliable way to perform diagnostics. Furthermore, stator current-based techniques proved insensitive to early stator inter-turn faults recently [32]. On the other hand, past studies have shown that the stray flux monitoring is insensitive to load defects [33]. This study clearly indicates the incapability of bearing faults detection of the whole system (motor, shaft, and pump) via the stray flux. The mechanisms that lead to this insensitivity should be quantified and understood in the future by more research and testing in the field to cover a statistically important population of electrical machines. However, the stray flux seems sensitive to the misalignment fault.
To summarise and while thinking about the current literature on diagnostics, it seems that one method alone cannot provide a full screening of the electrical machines' health. The combination of at least two methods will guarantee the reliability of the diagnosis, while the stray flux methods should replace MCSA on the electrical faults monitoring with MCSA being a good and advantageous replacement to the vibration analysis. Other secondary techniques such as oscillations here F I G U R E 1 3 The dismantled shaft F I G U R E 1 4 The motor without protective side covers: (a) before and (b) after the replacement of the bearings or thermography could localise the fault and add more information at a post-diagnostic and pre-service stage.
Finally, vertically mounted induction motors have very different degradation mechanisms of their bearings compared to horizontally mounted ones and which are the majority. The geometrically uniform ageing of the bearing as well as the fact that the damages and cracks were formed in the axial and not radial direction led to this serious faulty case going completely undetected with conventional methods. Here, the oscillometer did not sense any important displacements or oscillations at the two motor ends. More research is required to fully understand how such failures can be reliably detected.

| CONCLUSION
This study has presented all the steps from the application of the MCSA and stray flux monitoring aiming to detect mechanical faults during a routine test in 6 kV induction motors driving pumps. At a time where the stray flux wins a significant ground over the MCSA for electric faults detection, it becomes clear that it may be less sensitive to faults of mechanical nature which are still detected via the stator current reliably. This study has demonstrated a procedure to detect such faults without any prior knowledge of the geometrical and other features of the monitored system. Furthermore, a new challenge has been identified and concerns the detection and understanding of the underlying degradation mechanism of bearing faults in vertically mounted induction motors.