Induction motor faults: basics, developments and laboratory-scale implementation (part b)

<<< cont. from part a

__ 5 Laboratory-sale implementation of induction motor faults

Any fault diagnosis procedure will be nonsense if experimental investigations and validations are not included in the procedure. In fact, after all the theoretical and simulation-based analysis, what really matters in industry is measurements reflecting the real motor behavior in faulty conditions. Moreover, although developing the basic ideas mathematically or even by means of simulations, whether accurate FE or analytical models, helps researchers and industries investigate practically the impossible number of situations, models still suffer from some sort of inaccuracy depending on the number of assumptions made to build models.

Therefore, experimental study of what happens in reality should be an inseparable part of any diagnosis procedure of electrical machines. More importantly, some of the real world influential factors such as the thermal stresses, the non-homogenous magnetic materials and also every single multi-physics-related problem cannot be analyzed accurately unless a real motor and drive system is incorporated. This discussion necessitates presence of a straightforward implementation of the motor- drive systems along with the corresponding faulty conditions including all three major types of fault discussed in this section.

There are some prerequisites which should be considered prior to any setup preparation for the experimental test rig. The prerequisites are as follows:

--The motor should not be so small that implementing the fault is difficult or practically impossible. This really makes sense as motors in which the faults take place are of a large power, for instance hundreds of kilowatts.

--The motor should not be very large because of two main reasons. First, as the focus is on preparing a logical laboratory-scale setup, dealing with a relatively large motor during the fault implementation will be a very tricky and over whelming task. Second, the larger the motor is, the higher the investment will be. Most of the time, academic research budgets are limited, and having a solid investment plan is essential.

--Peak a motor for which an industrial drive could be easily found. Do not go through specifically dedicated motors built for special applications such as ultrahigh speed ones. Catching a drive for this kind of applications is not easy; moreover, research output will not probably be general enough to be used in other applications.

--Sometimes, more than one rotor is required if the goal is to study the effect of various broken bar positions. Therefore, include the corresponding prices in the investment plan.

--Keep in mind to apply reversible faults such as eccentricity prior to applying an irreversible defect like bar breakage if more than one type of fault should be investigated.

--Take the safe side and start from the low fault levels. Even if a reversible fault such as eccentricity of the rotor is studied, the reassembled healthy motor might not be as the same as it was initially right after the factory production.

If a higher fault level is applied, the motor might be subjected to un-repairable damage and not be able to be used later.

--Again, choosing a medium or not very small power motor is usually preferred as it is also robust enough against the environment noise which might affect the faulty motor signals and consequently the diagnosis procedure.

--Try to prepare a general application drive such ABB ACS800 which can be applied to all types of motors regardless of their specifications. The only restricting factor in choosing general application drives is their power limits which should be matched with the motor power.

--Try to prepare a drive consisting of both the open- and closed-loop control strategies. Unfortunately, there is only a DTC or FOC strategy implemented in one single drive. If one looks for a wider range of useful strategies, s/he should go through providing more than one drive. It is not recommended to build the drive on your own. If you wish so, make sure to implement every influential factor which might be included in an industrial drive. Otherwise, the results might not be convincing.

--In the case of a line-start application, make sure to have balanced three-phase network. Otherwise, the diagnosis results achieved from one phase might be different from the one obtained from another phase. To deal with an unbalanced network, using an autotransformer is a good solution. In the case of inverter-fed applications, it is not a requirement as the drive itself outputs balanced three-phase signals unless a switch fault or problem occurs.

--Prepare safe data acquisition instruments.

--Prepare a processor such as a computer or digital signal processors (DSP) to process the measured data to extract the fault indicators. A very useful and comprehensive discussion on this topic will be shortly provided in this section.

Tbl. 3.2 Induction motor specifications Fig. 19 (a) Tested induction motor and (b) corresponding winding layout 3.5.1 Three-phase induction motor The key element of any test rig built for the diagnosis purposes targeted in this guide is an induction motor. The discussion is continued by considering a three-phase induction motor with the following specifications (see Tbl. 3.2):

The real motor is shown in Fig. 19. This is exactly the same motor as what was previously simulated using the FE approach. So it consists of 28 rotor bars and 36 stator slots. The stator and rotor are both made of silicon steel. The rotor bars are made of aluminum. The rotor consists of two cages facilitating the motor start-up by increasing the start-up resistance of the rotor. The number of slots per pole per phase is equal to 3, so the winding layout should be the same as Fig. 19(b) as it consists of a concentric single layer winding topology. In high-power applications, double layer winding topology is more popular. By ''layer,'' we mean the number of separate layers of phases placed inside one single stator slot. The available network to supply the motor contains three phases with an effective value of 380 V. There fore, the peak value of the supply voltage is equal to 538 V. The network should be connected to the motor terminals with a delta connection (see Fig. 19(a)). For a star connection, the motor operation will be degraded due to the lack of supply voltage. Coupled with the mentioned points, presence of a rigid bench to which the motor is hanged firmly is another crucial requirement which reduces the unwanted vibrations caused by improper or loose motor placement. The more the degree of freedom of the bench is, the more flexible the motor assembling and disassembling would be.

Fig. 20 Autotransformer

__ 5.2 Autotransformer

Fig. 20 shows a typical autotransformer used in this study. There is one input and one output set containing four terminals three of which is related to phases, and the remaining one is the ground terminal if required. The adjusting wheel is used to control the voltage level. An autotransformer is an electrical device with only one single coil per phase and one or more terminals at the secondary to provide various voltage levels for the user. The utilized autotransformer consists of the adjusting wheel instead of the output terminals to make it possible to obtain smoothly vari able voltage range. The network voltage is applied to the primary winding(s) and the secondary winding(s) returns the required voltage(s). The scale of the voltage change applied by the transformer is not considerably large in fault diagnosis procedures. The main responsibility of this device is to stabilize the voltage amplitude partly different from the rated one. This apparatus has nothing to do with the supply frequency and is simply used to control the voltage level.

SAFETY NOTE: Before turning on the setup, ascertain that the adjusting wheel is located in a fair position in terms of the ratio of turns. Autotransformers are normally capable of increasing the voltage up to three times the rated one and this might cause life-threatening risk if it is not taken into account.

__ 5.3 Drive

Another key point to be considered is how the motor is managed to be used in inverter-fed applications. To do so, a very famous drive, i.e., ABB ACS800 is used in this study (see Fig. 21).

ACS800 is an industrial drive including the following features:

--controlling the time-domain characteristics of the motor speed and torque

--regenerative Braking mode

--DC magnetization to obtain the maximum torque capabilities of the motor

--flux weakening mode

--parallel motor control mode.

One of the most incredible advantages of this inverter is its ability to control motor speed in sensor-less method. In fact, there is no resolver or encoder connected to the motor shaft to measure its mechanical speed, although it is also applicable using this inverter. Mechanical speed is estimated by measuring electrical variables with a very high precision. The operating mode can also be selected among SCALAR and DTC. The main circuit of the drive is shown in Fig. 22. The significant parts are listed as follows:

--the control panel

--the start-up prevention switch (X41)

--the I/O board

--the input, output, DC bus and external braking resistor connections

--the six-step inverter.

Fig. 21 ABB ACS800

Fig. 22 Main circuit of ACS800 drive1

Fig. 23 Drive-shielded frame

The starting point of any motor-drive operation is to define the motor parameters including the rated values as well as its electrical quantities, namely the resistance and inductance of the drive, using the control panel. This is a must-do step prior to any other progress. The entered values facile the torque and speed estimation per formed internally by the drive if a sensor-less control strategy is used. Otherwise, the speed and torque signals can be passed to the drive through the I/O board using sensors which are going to be discussed shortly. Moreover, the estimated torque and speed signals are likewise available in the I/O board. So the I/O board is really helpful in case if an advanced data acquisition hardware is not accessible. Different drives provide different numbers and types of I/Os. Therefore, it is proposed to refer to the user manual to be informed of details. Furthermore, input and output power cables should be connected to the embedded ports shown in Fig. 21.

There are also three other output ports allowing users to access the DC bus voltage and also connect the external braking resistor to the drive. The latter is used to consume large regenerated braking power with the goal avoiding higher thermal tensions applied to the drive.

It is very important to know that there are some safety rules to follow. For example, the drive should always be installed inside a grounded metal frame (see Fig. 23) to prevent the external noise which might affect the control-level measurements. In addition, the life-threatening risk is reduced due to the fact that the frame voltage is deliberately brought to zero by connecting a ground wire to the body. A very important practical aspect is that if you use a computer to analyze the sampled motor signals, computer supply plug must be different from that of drive.

Otherwise, you will observe a very undesirable signal to noise ratio in your sampled signals. Another alternative is to use three-phase chock.

Generally, the installation steps are as follows:

--Identify the frame size based on the user's manual.

--Select the required cables including the power and control cables depending on the motor-drive ratings the environmental conditions.

--Check for the availability of the entire necessary module in the drive box.

--Prepare a chock to prevent inter-circuital interference between the drive and the data acquisition power supplies.

--Install the drive inside the metal frame.

--Connect the grounding cable of the frame.

--Connect and shield power cables. It is essential to use shielded power cables specifically where a long cable should be placed between the motor and the drive. This is the major requirement to surpass the radio frequency emissions radiated from the drive signals.

--Follow the cabling instruction to avoid improper installing angle of the cables at joints.

--Connect an external braking resistor if you might have a regenerative braking operation. This mostly happens when another drive which acts as the motor load is connected in parallel to the main drive. In this case, two drives share the DC bus voltage.

Finally, the motor is connected to the drive as shown in Fig. 24. A set of fuses should be necessarily utilized as an electric safe guard in case if any undesirable over-currents take place (see Fig. 24). The positive terminal of the braking resistor is regularly connected to the positive terminal of the DC bus, and a built-in terminal is embedded into the drive for the negative terminal. Both the motor and the drive bodies must be grounded together.

Usually two types of reference values are implemented in every drive, speed and torque reference values. Depending on the application in which the motor- drive system is used, different reference values might be chosen. Then, the rest of the operation is handled by the drive for controlling the motor speed or torque. It should be noted that precise determination of the motor quantities such as the stator inductance and the resistance are the essential task for presetting the drive.

Otherwise, the drive might fail to estimate the motor flux, torque and speed accurately. As a result, all the control process should be questioned.

Another must-remember point is the acceleration profile adjustment which takes place at the very beginning step of parameter determination for the drive.

Usually two general choices are available, linear and user-defined curves. The state of the art is to use a linear acceleration or even de-acceleration profile. However, various control strategies certainly lead to a different profile of the motor quantities. This makes the diagnosis procedure a very complex task. Investigating the motor-drive behavior in the faulty modes is still an open area of research and readers are referred to the author's publications to find the potential research areas.

Fig. 24 Motor-drive connections

Fig. 25 Motor-generator mechanical connection using a coupler

__ 5.4 Motor load

This is one of the most demanding aspects required to be as precise as possible if an accurate diagnosis investigation is really needed. First, it is the motor load level which affects all the motor quantities, so the necessity of presence of an accurate equipment for applying loads is one of the concerns. In addition, there are several types of equipment functioning as a load for any type of motors including:

--DC machines

--AC machines controlled by a drive

--advanced programmable dynamometers.

These are the general types of loads functioning in different ways but fulfilling the same goal which is fixing the motor load level at a specific value or changing the load based on a specific requirement such as oscillating load. Now, we are going to discuss each of the available options one by one from the simplest to the most advanced one sorted above.

__ 5.4.1 DC machines

This type of loads includes DC generators which are mechanically coupled to the shaft of induction motors, using a coupling device as shown in Fig. 25. The more reliable the coupler is, the less the amount of unwanted oscillations caused by improper placement of the motor or the generator on the bench will be. It is highly recommended that the motor and the generator be aligned along their shaft.

Otherwise, not only do undesirable oscillations harm the diagnosis procedure, but they also cause future mechanical defects if the system is used for a long time. DC machines are probably the simplest apparatus by means of which the motor load can be controlled. The underlying idea is to make the machine operate as a generator while it supplies a resistive load like what is shown in Fig. 26.

Significantly, the field and armature windings of the unitized DC machine should have a shunt nature as a series-connected topology suffers from the lack of field strength during the full-load operation. As a result, increasing the load, which is in turn achievable by adding a parallel resistive branch illustrated in Fig. 26, leads to a dramatic reduction of the field of a series-connected topology; hence, the generator starts to be demagnetized. This means that the DC machine would not be able to compensate for the increased load level, and the total system load drops down the desired value. This only happens if the rated power of the DC machine is almost close to that of the motor. To come up with a solution, a considerably larger DC machine is recommended. Otherwise, a shunt topology should be used.

One of the drawbacks associated with the mentioned approach of load control is that the load level indicator is the induction motor current or speed while what really matters in fault diagnosis applications is the motor slip. Nevertheless, the proposed system seems to be very simple and straightforward while some precautions in terms of the appropriate load-level control should be always kept in mind.

The shaft speed can also be measured by a tachometer.

The proposed scheme is considered as of the unsafe approaches if an amateur practitioner uses the equipment. The reasons are the naked terminals of both the induction and DC motor along with the fact that the load should be changed manually by switching the resistors. Therefore, the user should be alert enough to avoid undesirable risks.

__ 5.4.2 AC machines controlled by a drive

This scheme of loading the induction motor is very straightforward, useful, safe enough, but a little costly (see Fig. 27). The required ingredients of the scheme are as follows:

--tested induction motor as the main machine

--another AC machine, preferably a synchronous one, as the generator acting as a load

--two drives

--fuses.

The whole system is supplied by only one three-phase network connected to the machines and the drives through a set of fuses preventing the electrical stresses in the network side if any over currents happen in the system. Note that the fuses are supposed to be there any time. The three-phase fuse set should be used; moreover, two other fuses should connect the positive and the negative terminals of the DC buses of two drives. A choke is also required to prevent high-frequency PWM signals from circulating through the network. Drive#1 controls the induction motor while Drive#2 is connected to the AC machine, acting as the load, whether synchronous or asynchronous. The second motor-drive system should be set at torque control mode while the first system can be used in either torque or speed control mode depending on the test requirement. Preferably, the power, speed and torque ratings of the second motor-drive system which is used to act as the load should be similar to that of the induction motor. Otherwise, the AC machine power should be necessarily larger than that of the induction motor. Moreover, if there is so much difference of speed between the machines, a gearbox whose main operation is to change the speed while ideally transferring the same power is used to couple the shafts mechanically.

Fig. 26 Resistive load

Fig. 27 Machine-drive systems with a common DC bus

Due to its unique capability in accurate control of the torque or speed quantities, the closed-loop strategy should be hired in the load side. Otherwise, it is not guaranteed that a constant load is applied to the shaft if an open-loop strategy is used. So make sure to prepare a drive which is capable of handling closed-loop strategies, no matter if it is of a DTC or FOC nature, both work well. To reduce the budget, look for a drive with a sensor-less control included. Otherwise, a resolver or encoder should be prepared as well.

Whenever a single drive is connected to the network and the load is provided by the first scheme, the DC machine, all the required currents including the transient one come from the network and if a large power motor is utilized, other network-connected utilities might be affected harmfully due to the loading effect of the motor test setup. However, the second scheme, two motor-drive systems, is as demanding as the first scheme as the induction motor current is partly supplied by the AC machine which operates as a generator. So the current is circulated among two machines, and the network is only responsible for the power supply of drives.

The situation is somewhat different in transient start-ups while once the system comes to a stable point, the network normally operates with a minimum current defined by the drives.

A programmable drive #2 which facilitates applying a non-constant load to the system is recommended. This kind of drive is useful when phenomena such as oscillating loads are studied. Otherwise, studies should be limited to a constant but altering load level which means a part of actual operations are neglected. So the generality of the investigations directly depends on the capability of drives. This is an important feature which we are aiming at in this guide. With the use of the DC machine scheme, it is almost impossible to apply an oscillating load. This is another shortcoming of the first scheme highlighting it as an inefficient way of tackling fault diagnosis matters. To put it differently, it is not mainly proposed to go through the first scheme of motor loading.

On the other hand, the second scheme does not essentially ask for an auto transformer connecting the network to drive systems. The reason is that the drive itself takes care of stabilization of its own terminal voltages and eliminates the need for an extra stabilizing tool. This is another advantage of the mentioned motor- drive systems in use. Furthermore, the braking resistor is not required anymore if the motor and AC machines powers are properly matched, or in the best case, the AC machine power exceeds that of the induction motor.

There is one important question ''Is it possible to test a line-start induction motor, using the second scheme?'' The answer is ''YES.'' By removing the drive#1 and separately supplying the induction motor and the drive#2 to the network, a very fantastic line-start setup by means of which a pure constant load level can be easily applied is achieved. Do not forget to set the drive#2 at the torque control mode.

__ 5.4.3 Advanced programmable dynamometers

This is certainly the most accurate and appealing way of applying various load levels to an induction motor under the test. Dynamometers generally operate on the basis of absorbing the energy of the motor shaft and acting as a load. There are able to operate at various torque-speed profiles providing the commanded mechanical load for motors. Sometimes, there are measurement devices implemented to mea sure speed and torque to provide a proper command while some of the types of dynamometers do not necessarily require a speed measure ( Fig. 28).

Regardless of what is connected to the dynamometer shaft, it operates as a load-producing equipment guaranteeing a specific torque on the shaft. On the basis of what structure or material produces the braking force acting as a load connected to the shaft, various dynamometers are available. Three important ones are as follows:

--Hysteresis brake dynamometers

Fig. 28 Dynamometer

This is a perfect choice for experimenting motors ranging from fractions of a kilowatt to medium-power applications. A full range of motor speed including the free-run to locked rotor can be easily controlled by means of a hysteresis dynamometer. This is possible just because hysteresis dynamometers do not generally need a speed measurement to precisely output a torque. They are indeed accurate, and the maximum error, depending on the size and accuracy of the configuration, does not go beyond 1%. Considering the nature of hysteresis dynamometers, they can be programmed to be even used in varying load applications.

--Eddy-current brake dynamometer

The main feature of this type is the dependency of its torque to speed of rotation. It is something like fans in which the developed load is proportional to square of speed. So not all range of torque-speed profiles might be covered by this means. However, if a high-speed application should be tested, eddy current dynamometers are the best.

--Powder brake dynamometers

This type is usually recommended for a considerably high torque application such as traction motors while the corresponding speed level should not be so much high. Again, a very well-designed dynamometer of this type might reach a maximum error of 1% in the worst-case scenario.

Dynamometers usually come with a comprehensive measurement and analysis package called power and spectrum analyzers by means of which all the ordinary motor signals including the voltage, the current, the active and reactive power, the inductance and also the corresponding frequency and time-domain variations are explored. This gives users the opportunity to access the essential mediums of fault diagnosis procedure straightforwardly. However, such a measurement and loading equipment is supposed to be very expensive. An example will be provided later in this section.

Fig. 29 (a) Squirrel-cage induction motor and (b) magnified stator-rotor teeth

Fig. 30 Cage rotor

__ 5.5 Implementation of broken bar fault

Initially, let us provide a complete 2D view of an induction motor cross-section shown in Fig. 29(a). The stator and the rotor consist of 36 slots and 28 bars, respectively. Every 9 slots and 7 bars form pole out of 4 poles of the motor. Due to the symmetry of the rotor, it is not important which bar is assigned as the first bar of the rotor. Then, the other bars are numbered clockwise or counter clockwise from 2 to 28. The air-gap length of the motor is equal to 0.45 mm ( Fig. 29(b)), and the bars possess one upper and one lower parts dealing with the motor start-up and steady-state operation, respectively.

The air gap is normally so small that even a very negligible amount of bearing defect or eccentricity fault leads to a considerable percentage of air-gap distortion causing an unbalanced motor operation. A very promising way of doing this is to fix the motor housing on the working bench and clamp the shaft from both ends and then try to loosen and open the plates. To perform it more smoothly, preheat the plates close to the bearing area to make sure the bearing nests are loose enough.

Prior to any fault implementation, the first step is to disassemble the motor by opening the two end plates. Preferably, start with the load-end plate. Make sure to clamp the rotor shaft and not to release it while opening two end plates. Even if you intend to release the shaft and put it on inner surface of the stator, do it safely to prevent the stator material from being damaged or rubbed. If not so, the stator would be apparently useless for future fault diagnosis attempts. Finally, take the entire rotor body out and put it somewhere safe. Make sure to do it gently with the goal of protecting the rotor material. By ''material,'' we mean the silicon steel material, not the bar material because the bars are often intact and inaccessible while the outer layers of the silicon steel materials are of course in contact. The bearings are mainly attached and fixed to the two shaft ends. So leave them the same unless an eccentric rotor should be investigated (see Fig. 30).

In this specific rotor, the bars are in the shape of a deep bar which consists of two parts, the upper and the lower cages, as shown in Fig. 31. This makes the fault implementation somewhat tricky as there is not a uniform distribution of the bar cross-section across the rotor bar depth. So while applying the partial broken bar, it is not a 100% accurate practice. Nevertheless, applying a full broken bar is easy and precise. The main aspect of implementing the breakage is to know the exact dimension and shape of bars, as it is visible from outside of the rotor.

Therefore, having the structural map of the motor in hand is essential.

Considering the double cage rotor bars shown in Fig. 31, the maximum diameters of the upper and lower cages are equal to 6.2 and 5.4 mm, respectively.

So, it is obvious that a full broken bar requires an elimination of a crosssection covered by a width 6.2 mm, while the partial broken bar only asks for a smaller portion of the mentioned value. The question is that how the removal of the crosssection is done. Actually, the goal is to partially or fully eliminate the bar current and prevent the corresponding bar from contributing to the magnetic field generation. How is it handled? To apply broken bars fault to the rotor, the rotor bar should be drilled to pre vent the current from passing through the bar from one end to the other end. In general, there are two main locations to make the desired hole ( Fig. 32):

--somewhere between two end-rings along the bar length

--right at the connection of bar and end-rings joints.

Fig. 31 Rotor bar slot and different degree of broken bar fault in one slot including partial and full broken bar

The latter is the most probable location of the fault in real motors ( Fig. 32(a)).

The joint between bar and end-ring is not always as strong as it should be due to some internal cracks or bad welding or casting process. When the motor is subject to undesirable thermal or mechanical stress, the joint is broken and it is literally said that the bar is broken. It is not proposed to drill the hole at the joint of bar and end-ring as it probably damages the end-ring which leads to a worse fault level.

First of all, it is better to make a very small hole on the rotor bar, somewhere between two end-rings, using a very thin drill. Do not put so much pressure on the drill because the bar is made of aluminum which is a really soft and easily drilled material. Keep in mind to use fixture to prevent the rotor from moving when it is drilled. Otherwise, the silicon steel material might be damaged. If so, not only the core losses but also the flux distribution will dramatically be increased. To make sure that you are just drilling the bar not the core, you have to keep eye on color of swarf coming out of the drill. If the color is bright, it is the aluminum which is drilled. Otherwise, if it is dark gray, you are drilling the core material. So you have to stop immediately. Be careful to drill the bar with a very high speed and low force level to guarantee that the drill would not brake and stick into the bar.

Depending on the diameter of drilling area, different levels of the breakage is applied. Fig. 31 clearly illustrates four different possibilities, namely the 25%, 50%, 75% and 100% breakage.

Fig. 32 Drilling the bar (a) joint of bar and end-ring and (b) between the two end-rings

The formal course of drilling action is to utilize a stand-up drill like what is shown in Fig. 33. Do not lose the sight of the fact that the rotor should be certainly fixed by means of a fixture. Act very gently and smoothly! If you are going to investigate different fault levels ranging from partial to several full broken bars, go ahead with the smallest fault level, for example 25% partial breakage. Then, move toward the higher levels to keep the tested rotor less damaged. Note that the broken bars fault is an irreversible kind of fault leading to a permanent damage.

Fig. 33 Stand-up drill

Fig. 34 Different types of broken bars

A very critical point is to get more than one rotor ready for experiments as you might also intend to study more than one breakage at different poles. So one single rotor is not enough to fulfill the requirements.

Partial and full breakages are two types of broken bars. If it is a partial breakage, it is called partial broken bar in which a fraction of healthy current passes through the bar. By increasing the level of partial breakage, the bar resistance is increased up to almost infinity. Sometimes, there is full broken bar. This is simply known as broken bar fault. It is mostly considered that the resistance of a broken bar is infinite and no current passes through it, although it is not infinite when the bar is partially broken. What really happens is a bit different due to the existence of interbar currents. Although the current might not be able to pass through one bar in full breakage case, it might enclose its pass through cracks existing in core between two or more bars. This phenomenon has not yet been studied deeply, but the effects of interbar currents have been observed in different tests. On the basis of this claim, the resistance of a broken bar is not infinite at all. If you need to simulate the broken bar fault, you might choose a larger value compared to that of the healthy motor. This value will be discussed in the next sections. Sometimes, more than one broken bar, for example 2, 3 or more, might exist or happen in a rotor. The broken bars could be in adjacent or nonadjacent locations. So the broken bars fault is categorized into three main groups as follows:

--partial

--adjacent

--nonadjacent.

Fig. 35 Broken bar (bb) locations (a) healthy, (b) 1 bb, (c) 2 adjacent bbs (case 1), (d) 2 bbs at half pole-pitch distance (case 2), (e) 2 bbs at one pole-pitch distance (case 3) and ( f) 2 bbs at two pole-pitch distance (case 4). (bb: Broken Bar)

Fig. 34 shows partial, 1, 2, 3, 4 adjacent broken bars. As the entire rotor circumference cannot be shown in 3D figures, the nonadjacent broken bars located at different poles are shown in Fig. 35 by means of a 2D representation. P1, P2, P3 and P4 stand for first, second, third and fourth poles, respectively. Each small gray circle is the symbol of one bar. Red color means that the bar is broken.

The number of nonadjacent broken bars might vary from 2 to larger numbers. (Instead of color, special signs must be used to have clear figures in the black and white hard copy of the guide.) As you notice, there could be various degrees, numbers and locations of the broken bars. So three totally different concepts, briefly addressed before, are introduced as follows:

--Detection: Monitoring incipient broken bars fault occurrence

--Determination: Monitoring number of broken bars

--Fault-location: Monitoring location of the broken bars.

Combination of ''detection,'' ''determination'' and ''fault-location'' procedures builds a very comprehensive advanced industrial concept which is called ''Diagnosis.'' The same is valid for the other types of motor faults. Diagnosis is not a specific process of monitoring for electric machines. It is a very complicated process which is required in all engineering and medical applications to provide a safe and reliable operating condition not only for devices but also for technicians.

Taking the possibility of having different levels and locations of the fault into account, one can easily understand the struggling of the diagnosis procedures based on which users, researchers and technicians are trying to detect, determine and even locate the fault. It is a really overwhelming task unless a clear interpretation of faults and their effects on the motor signals is prepared. That is why a very basic to advanced discussion of every influential factor of motors, drives and faults have been and will be discussed in this guide. Ignoring one important aspect, something like the location of the fault, would probably result in unsatisfactory outputs leading to an incorrect maintenance process.

Having provided the experimental details in terms of the setup preparation and also the broken bar implementations, now it is time to tackle the eccentricity fault and its experimental sides. To learn more, have a look at the next subsection.

__ 5.6 Implementation of eccentricity fault

Fig. 36 Experimental implementation of eccentricity fault

Fig. 37 Experimental implementation of bearing fault

So far, different ways of producing eccentricity fault in a laboratory scale have been proposed in the literature. For example, static eccentricity can be the output of the following approaches:

--to use a faulty bearing

--to design and build specific plates enabling us to change the vertical position of the bearings

--to replace the bearing cage by an eccentric one and then use concentric rings around the cage to fill the empty spaces

--to place the concentric bearing cage between the eccentric rings inside and outside the cage

--to remove the plates, fix the rotor on stands and the stator on adjustable stand In addition, the following approaches can be used to apply a dynamic eccentricity:

--to couple an unbalanced load to motor

--to chisel the bearings housing and fill the empty spaces by means of eccentric rings

--to chisel the bearings housing and make an eccentric space and fill the empty spaces by means of non-eccentric rings

--to replace the bearing by bearings with a larger inner race diameter and fill the empty space by means of eccentric rings.

Using one static and one dynamic approach simultaneously introduces a mixed eccentricity fault. Some of the mentioned approaches are destructive. This means that the motor structure is damaged permanently after applying the fault. In the case of destructive approaches, returning the motor to its healthy state will be almost impossible. Moreover, the fault level cannot even be changed if a destructive approach is used. Some of the approaches require a sophisticated measurement tool to measure the level of eccentricity applied to bearings.

We decide to stick to an approach which reduces the number of disadvantages mentioned above. Mainly, the approach should not be destructive and not require a specific measurement tool. So various types and levels of eccentricity can be applied and eventually removed from the motor to get it back to a healthy operation. It is an extension of the forth dynamic approach. For this purpose, instead of the main bearing with the code 6309, another bearing called 6011 whose inner race diameter is 10 mm larger and outer race diameter is 10 mm smaller than that of 6309 is utilized. Placing the new bearing helps us with two empty spaces inside and outside it and allows us to fill both spaces by means of rings leading to the possibility of both the static and dynamic eccentricities. So the steps are as follows:

--to install a concentric ring inside and an eccentric ring outside the bearing to produce a static eccentricity. The fault severity is controlled by means of the level the eccentricity of the outer ring.

--to install an eccentric ring inside and a concentric ring outside the bearing to produce a dynamic eccentricity. The fault level is controlled by means of the level of eccentricity of the inner ring.

--to combine the previous two approaches to apply a mixed eccentricity fault.

Finally, the initial healthy motor can be assembled again using the replacement of faulty bearings by the healthy one. A very clear illustration of the mentioned steps is provided in Fig. 36. The challenging part of the implementation is to appropriately prepare the eccentric rings with an exact level of eccentricity. As the air-gap length is very small, a very accurate tool such as a computerized numeric control machine might be required. This is probably an expensive but appreciable practice.

Fig. 38 (a) Schematic implementation of short-circuit fault and (b) motor end winding

Now, let us switch to bearing fault and a very simple but intuitive ways of applying defects to it (see Fig. 37). Simply drill the inner or outer race to produce the inner race or outer race fault. To apply a ball defect, it is better to apply a crack or breakage to one of the balls. So it requires disassembling the bearing.

__ 5.7 Implementation of interturn short-circuit fault

In a healthy condition, the only accessible parts of the turns or coils are the end terminals allowing us to connect them to the supply while a short-circuit fault stands in need of new electrical connections as shown in Fig. 38. These new connections are definitely different from the terminals and provide a current pass through somewhere in the middle of one of the windings, a, b or c to the same winding, another winding or even the ground (motor housing). Respectively, they are called interturn, phase-to-phase and phase-to-ground fault. Therefore, to apply a short-circuit fault, the insolation of at least two turns should be partly removed to access the turns. So it is certainly a destructive task and needs a careful attention.

Even after removing the insolation, an additional resistance shown by Rf should be used to limit the circulating fault current to a safe value. Actually, if the fault connections are attached to each other, using a low resistance wire, the fault current in the new phase might go beyond control, and the phase winding(s) burns completely. In this case, the motor will be useless.

Depending on the fault severity, location and type, different values of Rf are demanded. For example, when dealing with phase-to-ground short-circuit fault, probably the larger Rf should be used as it is often the electric path with the smallest resistance. Hence, a larger circulating current will flow for a given path voltage.

However, the interturn fault, which is usually the most possible case, is of the smallest current, so the corresponding Rf might also be equal to zero, depending on the insulation quality. This is actually the insulation and partly the copper quality which define the maximum allowable circulating current and subsequently the Rf value. In fact, the more the insulation and copper endurance is against the temperature rise caused by a large circulating current, the smaller the value of Rf needs to be. Therefore, higher levels of fault might also be studied. High-quality insulation also makes it possible to sample a larger time period giving the opportunity to have a better spectral resolution. Rf should be a high-power resistance which is capable of tolerating large power dissipation. Then, Rf is to be electrically connected to the turns which are planned to be short-circuited. As the turns are not accessible through the stator stack length, the only part facilitating the process is the end-winding areas coming out from the stack.

In practice, as investigating real large current faults is impossible due to the fact that the motor will be permanently destructed, the trend is to only analyze lower levels of fault to first find proper indicators and then check how the proposed indicator depends on the fault increase or the other influential factors. The other faults are treated the same while some of them such as eccentricity fault are of a reversible nature pointing it as a motor friendly type of fault.

The aim of this section is to deliver the real sense of fault diagnosis and their laboratory-scale implementations. On the other hand, as the very beginning step of the diagnosis procedure, signals should be prepared for processing step. This is why measuring, or in other words, sampling motor signals including thermal, magnetic, electrical and mechanical signals are of a great interest as they are the mediums reflecting the possible faulty behavior. Motor signals such as voltage and current and their corresponding temporal and spectral variations are kind of signatures for any motor and differ depending on the motor type, the fault type, the processed signal and also the utilized processor.

Basically, an accurate measurement of proper motor signals is the most important concern of any diagnosis procedure. Therefore, we decide to establish a useful and comprehensive analysis of various types of sensing and sampling motor signals. To this end, move toward the next section which discusses signals and the corresponding sensor implementation.

__ 5.8 Signals and sensors

__ 5.8.1 Voltage

Fig. 39 (a) VT circuit and (b) LEM VT

One of the most well-known and significant motor signals is probably the voltage which is responsible for providing the amount of torque required for rotation.

Although in line-start applications, there is no point of generally using the voltage as a signal for the diagnosis purposes, in inverter-fed applications it plays a vital role in defining some specific aspects of faults such as justifying the possible differences between various supply modes. However, the voltage might also be useful in a line start application if calculation of quantities such as winding inductances is the main goal. Moreover, the time-domain samples of the line voltage are simply available in the I/O port of drives or by means of built-in packages or interfaces developed for a drive. Therefore, in drive-fed motors, using additional instruments is not recommended, but in case a line-start application is under the test or there is no access to drive interfaces, one might use a ''Voltage Transducer (VT)'' specifically developed for this kind of applications (see Fig. 39). VTs are promising tools to measure and output a scale voltage exactly similar to their inputs which is the motor terminal voltage. A common structure of VTs is illustrated in Fig. 39(a).

Any VT consists of a transformer used for isolating the primary and the secondary windings to increase the safety and also possibility of the measurement, using a small output voltage. The input voltage might be very high. A restricting resistance connects the terminals to the VT and it should have a considerably large value to avoid the loading effect of the VT. Then, the output of transformer is connected to an op-amp for outputting a scaled voltage measured in the M (measurement) terminal. There is also a series-connected measurement resistance to prevent the output current from running above fractions of an ampere. If the output of voltages of transformer becomes larger than the supply voltage of the OP-AMP, the output is saturated and will represent a flat-topped signal. So, make sure the maximum terminal voltage does not go beyond the rated limits of the VT.

Using VTs is very simple and straightforward, and the only necessary additional thing is a DC voltage source to supply the OP-AMP. It is worth noting that the supply provides a positive-negative polarity voltage. Otherwise, one side of the signal, either the positive or negative side, is cut off in the output. The LEM company sells a wide variety of VTs as well as current transducers (CT) ranging from very small to medium ratings (see Fig. 39(b)). Every single product of this company has its own specific datasheet in which the proper usage is provided.

Fig. 40 (a) CT circuit and (b) LEM CT

__ 5.8.2 Current

As a principle ingredient of diagnosis procedure, the current should be considered as the most tackled and useful signal in any procedure. Both the time and the frequency analysis of this practically appreciated signal have been matter of various investigations, so an abbreviated term called MCSA which stands for ''motor current signature analysis'' is usually assigned to the process of investigating the fault, using the motor current. The widespread use of the motor current is under lined by the fact that it reflects almost all the essential fault behavior and also it is the easiest and safest signal in terms of sampling. Unlike the motor voltage which might be very dangerous if precautions are not taken into account, the current sampling is painless. The only required instrument is a CT, especially a Hall-effect sensor (see Fig. 40). The connections look similar to that of a VT except the input connections. Instead of a simple isolated transformer, a Hall-effect sensor, which does not require an electric connection in the primary, is implemented. The primary is the motor terminal wires passing through the middle open space of the CT shown in Fig. 40(b). Using the magnetic induction on the secondary winding of the sensor, the transformer output is passed to the OP-AMP and then scaled to the output which is the same as that of VT.

__ 5.8.3 Torque

In general, there are two types of torque-measurement sensors as the following:

--invasive

--noninvasive.

Invasive type of measurements is normally coupled to the motor shaft mechanically; hence, there is always a mechanical coupling which might affect the motor behavior in a bad way. In fact, there is the possibility of the signals used for the diagnosis being affected. Therefore, it is called ''invasive.'' The terminology of ''invasive measurement'' used in this guide is way broader than what is usually referred to in the literature. In fact, any device or even device placement revealing any kind of potentially harmful or disturbing effect might be included in this category. Otherwise, it is called a ''noninvasive'' technique. Definitely, a non-invasive method should be practically preferred to an invasive one. However, the corresponding noninvasive sensors (see Fig. 41(b)) which are principally based on the Hall-effect do not cover a wide range of applications. So it is recommended to prepare a mechanically coupled sensor (see Fig. 41(a)) if a larger range of torque and speed variations should be investigated.

One of the disadvantages of the mechanically coupled sensors is the presence of a mechanical connection between the motor and the sensor shafts. Thus, if eccentricity is investigated, there is a possibility of a bended sensor shaft as well.

This not only applies asymmetry in a long-term use, but also might reduce the accuracy of the sensor. However, both types are usually robust enough to withstand sever applications up a certain point. Regardless of the type, the associated error of torque sensors today does not usually surpass 1% which is an acceptable range for the diagnosis purposes.

Both types consist of a scaled electrical output to hand in the motor torque instances mostly in an analogue regime. Actually, as they mostly use analogue devices, the output torque has a continuous nature while digital torque sensors providing described signals are also available. Close attention should be paid to the selection of a digital torque sensor in terms of the sampling rate. The higher the sampling rate is, the higher the resolution and the number of observable harmonic orders will be in the torque spectrum.

Fig. 41 Torque sensors (a) invasive method and (b) noninvasive method

__ 5.8.4 Speed

Encoders and tachometers are two mediums of measuring the motor speed, each having its own advantages and disadvantages (see Fig. 42). Encoders are mechanically coupled devices while tachometer should not necessarily be connected mechanically. What makes a difference between these two types is the way they are used to measure the speed. The following are the main differences:

--Encoders are mechanically connected to the motor shaft; hence, are considered as an invasive method of measurement. On the other hand, tachometers work on the basis of laser light which sends and receives high-frequency waves reflected from the shaft. Normally, unlike encoders, tachometers do not pro vide us with an output port of speed measurement, and they only consist of a digital indicator indicating an average value of speed.

--Encoders are of course capable of sampling the speed variation while tachometers often return one single number in average. So if the goal is to merely define the steady-state speed of the motor shaft, tachometers are the best choices. Otherwise, in case of a need for an accurate time-dependent step-by step speed development of the shaft, encoders should be used undoubtedly.

Discussing the details of operating principles of sensors is not a part of this guide, and the only focus is on their application in accurate signal measurement. For more information about the principles, readers are referred to datasheets distributed by companies.

__ 5.8.5 Flux

Fig. 42 Speed sensors (a) encoder and (b) tachometer

Magnetic flux density is the key agent of transferring power from the stator to the rotor. Without a magnetic coupling through the motor air gap, electromagnetic energy conversion is stopped and the motor operation fails. This is why it is some times considered as the heart of any electrical machine. On the other hand, although the original signal of supplying the motor power is the terminal voltage, what handles the rotation is the magnetic force or torque developed by the magnetic flux density.

Moreover, previous mathematical and simulation-based developments proved that any fault behavior, either in the stator or the rotor, is somehow transferred to the rotor or the stator only by means of the air-gap magnetic flux. The mentioned point necessitates the utilization of an approach to access the magnetic flux at least at air gap level. To this end, two approaches, namely an FE-based simulation and an experimental sensor, are available. The first approach is the subject of the next sections while here we are going to introduce the second approach.

Observing an induction motor structure, whether wound or cage rotor, one will simply understand that in an ordinary motor, there is no way of accessing the air gap as motors are totally encapsulated by the housing and there is no access inside. So the trick to access the air-gap quantities is to use an invasive technique called ''search coil'' shown in Fig. 43. The approach is called invasive as the motor plates should be opened, and then the search coil is placed around one of the stator teeth.

Fig. 43 Search coil

The larger the number of coils is, the larger the output voltage will be. However, even one turn works very well. Be careful not to increase the number of turns so much that it impacts the flux distribution of the main coils. The search coil provides two open terminals whose differentiated voltages should be equal to its induced voltage on it.

As the search coil is installed as close as possible to the air gap, the most dominant neighboring magnetic field is that of the air gap. According to the Faraday's law, while the magnetic field rotates and passes through the coil, a specific level of EMF is induced and measured at the terminals. This is a highly invasive but very useful technique. The magnetic flux density of the other parts of the motor including the stator tooth, the stator yoke, the rotor tooth and also the rotor yoke is not measurable at least by means of ordinary approaches. It is not also possible to implement such a device in a laboratory scale. This is why researchers rely on 2D or 3D FE simulations to study behavior of faulty motors. Furthermore, in the case of a cage rotor, rotor quantities are actually beyond reach, so the mentioned sensors will be useless unless a pre-implemented sensor, manufactured by factories, is used.

A very important point is to get familiar with the issues associated with magnetic characteristics of faulty induction motors. This is not achievable unless a very comprehensive FE model of motor is investigated. In fact, any posterior analysis including thermal or loss characterization highly depends on an accurate magnetic flux distribution prediction and FE approach makes this possible. Therefore, one complete section will be devoted to basics, formulations, implementation and postprocessing the FE analysis of healthy and faulty induction motors. Along with the other sections, the section related to the FE analysis definitely discriminates this guide from similar guides in which a shallow study of the motor quantities is addressed without providing justifications.

__ 5.8.6 Vibration

Fig. 44 (a) Vibration sensor and (b) possible installation locations

Practically, vibration analysis is the core diagnosis approach of induction motors.

Vibration in vertical, horizontal and axial directions returns astounding information on the motor behavior. Most of the time, it is by mistake assumed that vibration analysis only deals with mechanical defects and has nothing to with electrical faults. However, any type of fault, as what was previously mentioned, produce a specific oscillating trend causing monotonic mechanical pulsation of the motor body. Even the short-circuit fault which is a kind of naturally electrical fault also produces a pulsating air-gap flux density leading to a mechanical vibration. This is exactly what is sensed by a vibration sensor and returned as an electrical signal.

A typical vibration sensor, along with the possible installation locations, is shown in Fig. 44. Accelerometer is the conventional name of vibration sensors. They are magnetically mounted devices which can be rotated by user around the motor to capture vibration patterns of different parts of the motor.

There are types of two implementations of accelerometers, temporary and permanently located sensors, which are used in easily accessible and totally inaccessible operating areas, respectively. In other words, if the motor should be installed once and not accessed easily later on, it is usually preferred to install a permanently mounted wired sensor instead of a rotating accelerometer. If more than one accelerometer is mounted or used to monitor the vibration, data from all sensors is passed to a switch which is connected to a monitor or any kind of indicator. Then, sensors data is monitored one by one by switching between the sensors. Depending on the location of sensors on the motor housing, shaft, plates or even bearings, different patterns each revealing an aspect of faulty motor are achieved. It should be noted that it is also possible to mount a sensor internally inside the motor housing. Nevertheless, this might be considered an invasive technique and is not recommended. Having said that, the placement and location of sensors itself is a cause of decision change of a condition monitoring procedure.

If so, vibration analysis should be included in the category of invasive approaches as the way it is implemented affects the diagnosis procedure. However, once the sensors are implemented and fixed, outputs should be the same for a given faulty condition. So in this sense, it is a noninvasive technique. Equally important, vibration sensors are always mechanically connected to the motor.

Vibration sensors are comparably more expensive than VTs or CTs. Therefore, it might not be affordable by academic research centers whose budgets are limited.

In this case, it is proposed to use a search coil and measure the air-gap flux density.

Then, the radial force can be calculated and assigned as a vibration producing component, using (3.38). As the tangential component of the air-gap flux density is almost zero, this leads to an acceptable approximation of the radial force which is the main reason of the vibration. The other option is to use an accurate FE approach to simulate the vibration signals. However, vibration analysis has been used for several decades and is still of a great interest in industries.

Sound (noise) is a direct consequence of any kind of vibration. So in an environment that there is no additional noise or sound-producing factor except the tested motor, it is a premium approach to the analysis of the fault as it certainly has a noninvasive nature. The corresponding sound or noise sensors are in the form of antenna absorbing the noise or any sound disturbance produced by the fault. The significant requirement of any sound sensing is the presence of a noise-free room (antenna room) in which the motor should be tested. Although the required room should not be very large, it is a very dedicated and expensive facility not found easily everywhere (see Fig. 45). Sound analysis technique is a very promising way of diagnosing motor faults.

__ 5.8.7 Temperature

Fig. 45 Antenna room

Fig. 46 Thermocouple (a) circuit and (b) real one

Temperature sensing is one of the common monitoring techniques of electrical machines including motors and generator. Depending on the machine and fault type, different parts of the machine are subjected to a thermal tension requesting for a precaution in terms of possible future defects. In a healthy induction motor, end windings are the highly heated parts due to the higher leakage inductance while in a faulty motor, every part experiencing a dramatically increasing saturation level is the target of thermal tensions. For example, when a broken bar or short-circuit fault occurs, the area neighboring the fault reveals a considerably high temperature leading to an unbalanced temperature distribution of the machine. An eccentricity fault introducing a rotating saturated region called UMP point applies a total temperature rise of the motor. However, the other types of fault are also factors for generally increasing the motor temperature compared to a healthy operation.

Temperature monitoring is not usually performed as a sole-task, and it is combined with some other techniques such as the vibration analysis to return helpful information on motor operation. Moreover, the operating mode and the motor load level should be first defined in any thermal analysis as these are the principle factors in determining the temperature.

There are two basic monitoring approaches dealing with the thermal analysis of an induction motor:

--measuring local temperatures

--measuring a bulk image or motor temperature.

The first approach, which is a kind of invasive technique, requires several thermocouples (see Fig. 46) connected or installed at target part of the motor body including end windings, stator core, rotor core, housing and bearings, etc. This approach is based on a well-known effect called Seebeck which is the direct con version of a difference in temperature between to materials to an electric signal. It literally means that an electrical equivalent value is assigned to heat existing at a joint of two wires with different materials ( Fig. 46(a)). Two metals face the same heat source while the rate of temperature increase is totally different as the materials are not the same. Considering that any temperature change leads to a flow of electrons and assuming two materials with a great difference in their heat transfer capability, an electric voltage difference between two terminals in the measurement side takes place. This is what should be equivalently measured instead of real temperature difference. As thermocouples are very handy and small devices, they can be easily bonded on every part of motors body. This makes them one of the interesting ways of exact temperature monitoring.

Although thermocouples are very easy to use, their lifetime is somehow short compared to the other types of sensors discussed so far. The accuracy of the device might be affected by thermal tensions existing in the test environment. So calibrating the device and making sure if the device works very well should be one of the ongoing steps of using them.

Unlike the local thermal analysis, a global value indicating a total temperature change is sometimes assigned to the motor. The best candidate to reflect the change is the coolant temperature which is usually the air flowing in the air gap or close to it. So one can use a thermometer, regardless of type, to measure the time-dependent change of the motor temperature. This technique has nothing to do with an exact diagnosis procedure as the global temperature variation might be the product of any other unknown reason. Therefore, it is not proposed to go this way. Instead, utilizing a thermography camera is highly recommended (see Fig. 47). It is a potential device to return a very discriminative temperature distribution of motors.

In the case of stationary faults including the stator short circuit and also static eccentricity, it is expected to have a very clear representation of the fault location, and the detection is performed very well. However, if a rotating fault exists, it will not be easily detectable. As the thermal time constant is always larger than the magnetic time constant, incipient faults might not be detected by thermal analysis.

Considering the comments in terms of the motor signals and the corresponding sensing devices, all the sensors return electrical values which are relevant to the measured quantity. The products are electrical signals which should analyzed to conduct a diagnostic task. Therefore, another significantly important step is the so called ''data acquisition'' which is introduced below. This step comes ahead of any signal processing and fault diagnosis step as the latter cases are impossible to be done unless a signal is in hand.

Fig. 47 (a) Thermography camera6 and (b) motor temperature distribution7

Fig. 48 AdvanTech PCI-1710 HG (a) main board and (b) terminal box

Fig. 49 MATLAB data acquisition routine

__ 5.9 Data acquisition

We prefer to introduce a very useful general application data acquisition (DAQ) device called ''AdvanTech PCI-1710 HG'' which enables us to sample several signals at the same time with difference qualities. A typical device is shown in Fig. 48.

The specifications of the mentioned DAQ device are as follows:

--16 single-ended or 8 differential or a combination of analog inputs

--12-bit A/D converter, with up to 100-kHz sampling rate

--programmable gain

--automatic channel/gain scanning

--onboard FIFO memory (4,096 samples)

--two 12-bit analog output channels (PCI-1710/1710HG only)

--16 digital inputs and 16 digital outputs

--onboard programmable counter

--board IDTM switch.

A very compelling aspect of the device is the number of input channels which is equal to 16 if a single-ended topology is used. On the other hand, if a noisy environment exists, the differential topology consisting of 8 channels is recommended. The analogue to digital convertor implemented in this device is capable of sampling at a maximum rate of 100 kHz indicating that a maximum sampling frequency of 6,250 Hz can be assigned to every channel in a single-ended topology.

In the case of differential inputs, 12,500 Hz is the target. The corresponding sampling frequencies cannot be increased over the mentioned values unless some of the channels are free and not used. Then, the sampling frequency of used channels may increase if needed. The more the sampling frequency is, the more information preserved in a sampled signal will be. According to the Nyquist's law, the sampling frequency should be at least equal to twice the maximum frequency which should be included in the spectrum. Considering this fact and also taking the spectral resolution into account, it is suggested to take the safe side and go beyond the Nyquist's law and increase the sampling frequency as much as possible. The upper limit is usually forced by the data storage and also real-time analysis capability. Most of the time, it is favorable to have an online diagnosis technique which is able to detect incipient fault. In this case, there should be a trade-off between the accuracy and the required sampled data. In an offline case, users might even sample tons of data to increase the accuracy of the future investigations.

__ 5.9.1 MATLAB_ code for an AdvanTech device

AdvanTech boards always come with a terminal box allowing us to connect the sensors outputs to the AdvanTech board installed inside a personal computer. The interface between the terminal box and the board is implemented by a 68-pin parallel connector shown in Fig. 48(b). The whole system is controlled by the interfaces developed by the company or simply using MATLAB. Fortunately, MATLAB provides a fantastic DAQ interface not only for the AdvanTech products but also for a variety of other DAQ devices. Below is the routine which is used by MATLAB to input or output signals (see Fig. 49). The interfaces taking care of MATLAB/AdvanTech interactions include m-files, data acquisition engines and hardware-driver adaptors. The last two aspects are themselves controlled by m-files or even SIMULINK_ files.

This routine represents the possibility of inputting and outputting data while the focus of our work is on the inputting process. Assuming an inputting process, the steps are as follows:

--connect the sensors outputs to the terminal box

--connect the 68-pin parallel connector to the main AdvanTech board

--before going through defining a DAQ hardware, check if the hardware is available or not using the following command in the command line:

% show and save list of available and installed vendors vendor=daq.getVendors

--open an m-file in a MATLAB environment and start typing the following code:

% Create a session for the required vendor session = daq.createSession('advnatech')

% Add an input digital channel ch1=addDigitalChannel(session,'DeviceID',

'ChannelID', 'InputOnly');

% Specify channel specifications ch1.TerminalConfig = 'Differential';

ch1.Range = [-10.0 10.0 ];

% Specify duration of acquisition session.DurationInSeconds = 2;

% Specify the sampling frequency or rate per second session.Rate = 10000;

% Inquire about the limit of sampling rate determined by hardware session.RateLimit

% start session data = startForeground(s);

The steps have been explained as comments shown by percent symbols. Simply, one can call an AdvanTech Package and set the options according to the requirements and then start sampling input signals, using MATLAB. A real-time practice of data acquisition toolbox is also implementable by means of MATLAB/SIMULINK blocks. Moreover, the above-mentioned process shows a digital input scheme while the analogue one is also applicable if needed. For more details, readers are referred to the Mathwork website mathworks.com/help/daq/functionlist.html .

Some technical points while using any acquisition hardware or device should be taken into account:

--Do not touch the input terminals with a bare hand. Static electricity might de-calibrate the device.

--Calibrate the device before using it.

--Scale the input signals applied to the terminal box to a measurable range of the DAQ device, i.e., [ 10, 10 ] V for AdvanTech PCI-1710 HG. In the situation that inputs run over the mentioned range, the sampled signal's top will be cut off and look like a flat-topped curve. Furthermore, relatively large inputs, those which are higher than the tolerable range of the device, might lead to a complete device failure by introducing insulation failure.

--When connecting more than one input, if the number of inputs does not surpass the number of channels, assign distanced channels to inputs to reduce the possibility of loading effects of channels onto each other's signals. However, all the channels should be isolated in the manufacturing process.

What has been already discussed in terms of the DAQ devices is focused on a multitask hardware built by the AdvanTech Corporation. This device does not come with a set of sensors such as VTs or CTs, and one should first prepare the required sensors and then integrate them with a DAQ board. This means that although the provided measurement and acquisition setup components are very handy and useful, they might not perfectly match each other in terms of impedances, calibration, input/ output ranges and so forth. So another comparable option is to use an integrated digital oscilloscope accompanied by a matched set of current and voltage probes.

Tektronix is the best example of manufacturer of this kind. It provides high resolution multichannel digital oscilloscopes; moreover, very accurate current and voltage probes at different ratings are also available. The only drawback is related to the variety of type of probes (sensors) which does not cover the mechanical, magnetic and thermal sensors. So if one looks for a full control of the DAQ devices, the first approach is recommended while the second one is probably easier to use.

After being sampled, all the required motor signals are passed to a processor which can be a simple personal computer or a separate DSP such as those developed by the Texas Instruments company. The diagnosis procedure normally ends with the signal processing and fault indicator extraction step. This is a major step through the final goal of the fault diagnosis. Thus, it will be deeply discussed in Section 7. For now, it is assumed that the corresponding prerequisites are ready to go further through introducing an integrated fault diagnosis scheme below in the next subsection.

Fig. 50 Scheme diagram of a conventional cabled system implementation

__ 5.10 Overall scheme of the conventional cabled diagnosis system implementation

Taking all the mentioned aspects of a practical implementation of fault diagnosis procedure into account, a general scheme shown in Fig. 1 is introduced as the final solution for the experimental setup. In a few words, the following step-by-step procedure should be followed up to the end point which is extracting fault information, using a time, frequency or time-frequency analysis ( Fig. 50).

--If the motor is a wound rotor topology, short circuit the rotor windings.

--Connect the tested induction motor to the three-phase network/Drive, using a switch to take care of both the line-start and inverter-fed modes.

--The switch is used to switch between the line-start and inverter-fed modes.

--Connect the drive input to the three-phase network. One might also connect a three-phase choke to reduce the noise applied to the network.

--Connect the drives output terminals to the switch input terminals.

--Use shielded cables.

--Connect the switch output terminals to the motor terminals.

--Determine one of the phases as the diagnosis phase and then pass it through the CT. It is proposed to roll it several times around one leg of the CT to make sure the best primary coupling is made between the phase cable and the CT.

--Connect the test phase to the VT.

--Prior to connecting the electrical inputs of the motor, mechanical parts should be handled very well. It means that the encoder, the torque meter, the coupling and the load should all be connected and checked for being safe first.

--Prior to connecting the drive to the circuit, make sure to adjust the required initial parameters including the rise-time, the motor resistance, the motor inductance and most importantly the operating mode.

--Apply the temperature, flux and also vibration sensors as desired and then route all the sensing signals to the DAQ devices.

--Connect the DAQ devices outputs to a processing hardware which includes an ordinary CPU or a DSP.

--Obviously, the fault should be already applied to the motor. Furthermore, the speed and torque adjustments are supposed to be done.

--Finally, run the system and start sampling and analyzing the motor signals with the goal of extracting proper fault indicators. This step is the main focus of this guide and will be further analyzed, discussed and explained in the next sections.

One of the major aspects of the discussed scheme is the presence of a bunch of probably cumbersome wiring as all the signaling route is handled by means of cables. So if a long-distance analyzing center should be incorporated into the diagnosis procedure, this scheme, although provides a promising infrastructure, it will be somehow challenging in terms of signal routing. Instead, a wireless implementation of a condition monitoring system proposes a very tangible substitute for the conventional approach. The wireless condition monitoring is an almost new topic in the field, and not all researchers or even manufacturers have tried to address this approach. However, the existing literature clearly explains the idea.

__ 5.11 Wireless condition monitoring setup

Due to the recent developments achieved in the field of low-cost wireless sensors revealing a great ability in processing and communicating processed data, a couple of research attempts have been devoted to wireless implementation of a diagnosis setup. Consequently, online wired-monitoring systems, which indeed work well, might be totally replaced by wireless network alternatives. The main reason is probably the higher cost of shielded and isolated cable implementation while a wireless system does not require cabling at all. The only cost normally corresponds to the transmitters, receivers and likely the on-sensor processors if utilized.

Another disadvantage associated with the conventional monitoring system is its inability to provide an opportunity for temporary or specialized condition monitoring practices.

Wireless sensor networks (WSN) commonly provide a cheaper system implementation along with a more flexible and re-locatable monitoring system which does not restrict the location of system-utilizing technician. Nevertheless, no one can deny the existing shortcomings such as the potentially short range of operation and also possible electromagnetic interferences. On the other hand, WSNs have considerably constrained resources the significant of which is the associated battery life time. Regardless of its drawbacks, WSNs could definitely play a vital role in simplifying the diagnosis procedure.

In general, there are two kinds of topology for benefiting from a WSN:

--on-sensor processing

--off-sensor processing.

In the first topology, the signal processing, feature extraction and likely final fault diagnosis step are performed in the sensor itself. Therefore, sensors should provide an internally implemented microprocessor/microcontroller allowing processing the signals and then transmitting them to the monitoring unit. In the second topology, the sensor is just sensing and transmitting unprocessed signals to the receivers.

Finally, the processing and feature extraction are handled by the monitoring unit which consists of a processing device such as DSPs. What is usually preferred is the first topology due to the fact that the processing unit is fixed and located close to the motor while the monitoring unit can be easily moving around the operating environment. If an off-sensor processing topology is used, the monitoring unit would probably be larger and more difficult to be flexibly moved. Besides, electromagnetic interference, if not dealt with properly, might harm the unprocessed signals and lead to a loss of valuable information during transmission of the signal.

So the first topology is selected and shown in Fig. 51.

For each sensor, one processing and transmitting unit is assigned while there can be only one receiver receiving all the data simultaneously by multiplexing among the inputs. The DAQ, signal processing and feature extraction is handled by a microcontroller or microprocessor, and the extracted features are sent to the transmitter. There are some types of microcontrollers such as Jennic JN5139 which is capable of multitasking. This means that they conduct all the tasks ranging from receiving sensor data to transmitting extracted features at the same time. So not only do they lead to a considerable save in the implementation space, but they also reduce the price of integrating devices. The proposed scheme in Fig. 51 is applied to a line-start motor, but an inverter-fed application is also possible. To understand perfectly how a WSN works and is managed, a knowledge of micro controller programming is also required. At this point, we do not aim at going through the details as it is not the main topic of this guide, and it is assumed that readers have the required knowledge.

Fig. 51 Wireless sensor network-based condition monitoring setup

A brief overview of the current section reveals the attempts to understand the fundamentals of fault occurrence and the way how fault affects the motor time domain behavior. The attempts were totally devoted to explain the theoretical fault basis and consequently address the most significant time-domain variations by means of which a fault might probably be detected. Nevertheless, not all faults introduce a specifically detectable behavior, at least in a time domain. A step-by step laboratory-scale implementation of various types of faults, along with the required details, was provided. By means of this knowledge, a simple but very useful setup of fault diagnosis based on which one can study different fault aspects is indeed achieved.

Bearing this in mind, in the next three sections, we are going to target the best existing simulation approaches used for accurately modeling faults in induction machines. Not only the fundamental time harmonic components, but also all the possible spatial harmonic components are included and incorporated into the modeling process. The goal is actually to help researchers understand better how internal motor behaves, and it is not easily accessible by means of the existing experimental setups. For example, magnetic flux density at different parts of the motor or even the UMP is not the thing available at a cheap price in any laboratory or industry. Sometimes, it is even impossible unless an accurate modeling and simulation approach is hired. We believe no one's knowledge will get improved unless the upcoming information is studied and kept in mind. Therefore, it is highly proposed to study the following sections first before attempting to read and understand the sections coming after.