INTRODUCTION

The modern automobile consists of many mechanical systems such as

power seats, windshield wipers, mirrors, trunks, and windows, which are all

susceptible to breakdown. Without any condition monitoring system, the

breakdown is usually catastrophic, and requires an expensive part replacement.

Real-time condition monitoring allows for early detection of faults, which could

require a simple solution such as the application of a lubricant, to fix. This

prolongs the useful life of the component, and prevents sudden and unexpected

failure. Real-time condition monitoring can be accomplished by examining the

vibration signature of a mechanical system. For example, an automobile power

window consists of a DC motor and its associated bearings and couplings, a

gear reduction system consisting of worm and spur gears, and kinematic links.

Faults resulting in excessive vibrations may be caused by coupling

misalignment, bearing failure or gear train failure. Coupling misalignments

occur at the connection between the drive shaft and the driven shaft, and are

typically due to imperfect manufacturing. Bearing failure is caused by lack of

lubrication or moisture contamination causing rusting, while gear train failure is

caused by misaligned, broken, cracked or chipped gear teeth [1]. Each fault

occurs at a characteristic frequency, and so the state of the mechanical system

can be determined by monitoring the amplitudes of the relevant frequencies.

Vibrations due to coupling misalignments occur at harmonics of the shaft

rotational speed. Gear vibrations occur at the gear turn speed or at sidebands of

the gear mesh frequency. Ball bearing vibrations may be caused by outer

bearing race defects By monitoring the real time vibration signature of the

mechanical system, anomalies can be quickly identified and fixed. In the

proposed method, vibration signals are obtained using piezo-electric sensor

and motor current signature analysis is performed using Hall Effect sensor. The

features of the signal are analyzed using wavelet packet transform. Besides

other signal processing techniques, wavelet packet transform is preferred

because it has certain advantages. Traditional signal processing techniques like

Fourier transform can perform only on stationary signals. Since it is not well

suited for non-stationary signals short time Fourier transform (STFT) is used.

STFT uses a constant window function as a base to obtain the frequency

spectrum coefficients. The size of the window function cannot be changed

which led to the need for wavelet transform. Wavelet transform uses a varying

size window function as its base.

In wavelet transform low frequency signals are decomposed repeatedly to

obtain low frequency information. In wavelet transform the information about

high frequency signals are limited. In the proposed method, wavelet packet

transform decomposes both low frequency and high frequency information. It

can analyze both stationary and non-stationary signals.

There are many classifier models to effectively classify the faulty data from the

healthy one. They are: Analytical model-based methods, Artificial Intelligence-

based methods.

Analytical model based methods are efficient monitoring systems for

providing warning and predicting certain faults in their early stages. Artificial

Intelligence based methods are of two categories: Knowledge based models and

Data based models. When considering fault diagnostics of gear system it is

difficult to develop an analytical model that describes the performance of a gear

under all its operation points. It is difficult for a human expert to distinguish

faults from the healthy operation. Though analytical based methods and

knowledge based methods are effective classification methods, their

performance in induction motors is not good. Moreover conventional methods

cannot be applied effectively for vibration signal diagnosis due to their lack of

adaptability and the random nature of vibration signal. In such a situation, data

based models are used to classify faults in gear system.

Data based models are applied when:

— the process model is not known in the analytical form

— expert knowledge of the process performance under faults is not available

Some of the popular data based models are neural networks, fuzzy systems and

Support vector machine. Neural networks and fuzzy logic are widely used in the

field of fault diagnostics. Fuzzy logic provides a systematic framework to

process vague and qualitative knowledge. Using fuzzy logic it is possible to

classify a fault in terms of its degree of severity. Artificial neural network are

modeled with artificial neurons. Each artificial neuron accepts several inputs,

applies preset weights to each input and generates a non-linear output based on

the result. The neurons are connected in layers between the inputs and outputs.

Support Vector Machine, a novel machine learning technique is used in this

paper. It is based on statistical learning theory, and is introduced during the

early 90’s. SVM is opted in this paper since it is shown to have better

generalization properties than traditional classifiers.

Efficiency of SVM does not depend on the number of features of

classified entities. property is very useful in fault diagnostics, because the

number of features to be chosen to be the base of fault classification is thus not

limited. Gear system’s condition monitoring systems collect data from the main

components such as the generator, the gearbox, the main bearing, and the shaft.

The purpose of this data-gathering is to minimize downtime and maintenance

costs while increasing energy availability and the lifetime service of wind

turbine components. An ideal condition monitoring system would monitor all

the components using a minimum number of sensors.

There have been a few literature reviews on Industrial Motor’s condition

monitoring. This chapter aims to review the most recent advances in condition

monitoring and fault diagnostic techniques with a focus on wind turbines and

their subsystems related to mechanical fault. This section summarizes the

monitoring and diagnostic methods for the major subsystems in Industrial

Motor’s such as gearbox, bearing, and generator which are the primary focus of

this study.

2.1.1 Gearbox and Bearing

Gearbox fault is widely acknowledged as the leading issue for wind

turbine drive train condition monitoring among all subsystems [11-19]. Gear

tooth damage and bearing faults are both common in the Industrial Motor’s.

According to McNiff [27], bearing failure is the leading factor in turbine

gearbox problems. In particular, it was pointed out that the gearbox bearings

tend to fail in different rates. Among all bearings in a planetary gearbox, the

planet bearings, the intermediate shaft-locating bearings, and the high-speed

locating bearings tend to fail at the fastest rate, while the planet carrier bearings,

hollow shaft bearings, and non-locating bearings are least likely to fail. This

study indicates that more detailed stress analysis of the gearbox is needed in

order to achieve a better understanding of the failure mechanism and load

distribution which would lead to improvement of drive train design and sensor

allocation. Vibration measurement and spectrum analysis are typical choices for

gearbox monitoring and diagnostics. For instance, Yang et al. developed a

neural network based diagnostic framework for gearboxes in [28]. The

relatively slow speed of the wind turbine sets a limitation in early fault

diagnosis using the vibration monitoring method. Therefore, acoustic emission

(AE) sensing, which detects the surface stress waves generated by the rubbing

action of failed components, has recently been considered a suitable

enhancement to the classic vibration based methods for multisensory

monitoring scheme for gearbox diagnosis, especially for early detection of

pitting, cracking, or other potential faults.

Lekou et al. presented their study using AE in parallel with vibration,

temperature, and rotating speed data for health monitoring [29]. It was shown

that monitored periodic statistics of AE data can be used as an indicator of

damage presence and damage severity in Industrial Motor’s.

Chen et al. set up a finite-element (FE) simulation study of stress wave

based diagnosis for the rolling element bearing of the wind turbine gearbox. It is

noteworthy that FE analysis is a good complementary tool to the experimental

based study, with which the physical insight of various levels of faults can be

investigated. Notice that AE measurement features very high frequencies

compared to other methods, so the cost of data acquisition systems with high

sampling rates needs to be considered. Besides, it is noise-rich information from

AE measurement. Advanced algorithms are needed to extract useful

information. For mechanical faults of the drive train, the electrical analysis was

investigated. Diagnosis of gear eccentricity was studied using current and power

signals. It is noteworthy that the data were obtained from a wind turbine

emulator, incorporating the properties of both natural wind and the turbine rotor

aerodynamic behavior. Although the level of turbulence simulated was not

described, the demonstrated performance was still promising for practical

applications. Torque measurement has also been utilized for drive train fault

detection. The rotor faults may cause either a torsional oscillation or a shift in

the torque-speed ratio. Also, shaft torque has a potential to be used as an

indicator for decoupling the fault-like perturbations due to higher load.

However, inline torque sensors are usually expensive and difficult to install.

Therefore, using torque measurement for drive train fault diagnosis and

condition monitoring is still not practically feasible.

2.1.2 Generators

The Industrial generators are also subject to failures in bearing, stator, and

rotor among others components. For induction machines, about 40% of failures

are related to bearings, 38% to the stator, and 10% to the rotor. The major faults

in induction machine stators and rotors include inter-turn faults in the opening

or shorting of one or more circuits of a stator or rotor winding, abnormal

connection of the stator winding, dynamic eccentricity, broken rotor bars of

cracked end-rings for cage rotor, static and/or dynamic air-gap eccentricities,

among others. Faults in induction machines may produce some of the following

phenomena: unbalances and harmonics in the air-gap flux and phase currents,

increased torque oscillation, decreased average torque, increased losses and

reduction in efficiency, and excessive heating in the winding.

2.1.3 Machine Vibration Analysis

Vibration analysis is a proven and effective technology being used in

condition monitoring. For the measurement of vibration, different vibration

transducers are applied, according to the frequency range. Vibration

measurement is commonly done in the gearbox, turbines, bearings, and shaft.

For wind turbine application, the measurement is usually done at critical

locations where the load condition is at maximum, for example, wheels and

bearings of the gearbox, the main shaft of turbine, and bearings of the generator.

Different types of sensors are employed for the measurement of vibration:

acceleration sensors, velocity sensors, and displacement sensors. Different

vibration frequencies in a rotation machine are directly correlated to the

structure, geometry, and speed of the machine. By determining the relation

between types of defects and their characteristic frequencies, the causality of

problems can be determined, and the remaining useful life of components can

be estimated. The history of the equipment, its failure statistic, vibration trend,

and degradation pattern are of vital importance in determining the health of the

system and its future operating condition. Using vibration analysis, the presence

of a failure, or even an upcoming failure, can be detected because of the

increase or modification in vibrations of industrial equipment. Since an analysis

of vibrations is a powerful tool for the diagnosis of equipment, a number of

different techniques have been developed. There are methods that only

distinguish failures at a final state of evolution and there are others, more

complex, that identify defects at an early phase of development.

2.2 Review Conclusions

To achieve an accurate and reliable condition monitoring system for

wind turbines, it is necessary to select measurable parameters as well as to

choose suitable signal processing methods. In some examples, electrical sensors

installed around the generator are highly recommended as they are non-invasive

and easy to implement compared to the mechanical ones. In wind turbines,

because of the noisy environment due to the presence of power electronics

converters, signal to noise ratio of measured signals is low and the usage of

electrical parameters are often more problematic than in a lab environment.

Inaccurate signal analysis leads to various false alarms which makes fault

detection unreliable. To overcome this drawback, several approaches have been

proposed by introducing the vibration measurement and using vibrations as an

index for detecting mechanical fault in the system. However, those methods

have been applied mostly for drive train failure, bearing faults, and gear tooth

damage by using acoustic emission (AE) techniques for detection. Therefore, to

enhance the effectiveness and thorough of condition-based predictive

maintenance, dissertation proposes a vibration based monitoring system for

rotor imbalance conditions.

Literature review

Gearbox Speed Sensor Design and Performance Optimization

The speed sensing applications of a gearbox demand high operational air-

gaps, tight switching and timing accuracy, and excellent overall reliability. High

teeth density and smalldiameter helical (nonsymmetrical) teeth profiles with

fine teeth pitch pose significant challenges to rotational speed sensing. For

many embedded control systems, such as an automated mechanical

transmission, it is critical that before the transmission control module decides

whether to open or close a clutch, the speed of a shaft and the corresponding

gear have to be synchronized. This requires high fidelity in the accuracy of the

computed shaft’s speed. This paper presents a low-cost rotational speed sensor

module with high output accuracy and resolution for gear speed sensing in a

commercial vehicle gearbox. The sensor performance has been modeled using

finite element analysis, and the model has been validated against experimental

data from the sensor prototype. The complete sensor module is designed

for ease of manufacturing and application in harsh transmission environments

such as high temperature profiles ranging from −40 °C to 125 °C.

Application of MEMS technology in automotive sensors and actuators

Sensors and actuators are the critical system components that collect and

act on information in the analog environment and link it to the world of digital

electronics. The functional groups of sensors, software, controller hardware, and

actuators from the backbone of present and future automotive systems. Unit

volumes for sensors and actuators in the automotive industry are measured in

millions per year and at a unit cost of a few dollars. The design of sensors and

actuators has increasingly made use of microelectromechanical systems

(MEMS) technology. This technology is well suited to producing a class of

micromachined sensors and actuators that combines signal processing and

communications on a single silicon chip or contained within the same package.

This paper contains a discussion of the issues in producing MEMS sensors and

actuators from the concept selection stage to the manufacturing platform.

Examples of commercial and emerging automotive sensors and actuators are

given, which illustrate the various aspects of device development. Future trends

in MEMS technology as applied to automotive components are also discussed

Sinusoidal behavior of a dipole magnetization for position sensing

applications

Position sensing is needed in a large variety of applications, where more

specifically Hall-effect position sensors, due to their simplicity and versatility,

are elegant solutions. This paper describes a reverse engineering approach to

investigate the mismatch in measured Hall-effect voltage ofa dipole permanent

magnet ring. Using severalmagnetization patterns, the position signal is

reproduced, and it is concluded that by substitution of these patterns, the

measured voltage can be recreated

Planetary Gearbox Fault Detection Using Vibration Separation Techniques

Studies were performed to demonstrate the capability to detect planetary

gear and bearing faults in helicopter rotor transmissions. The work supported

the Operations Support and Sustainment (OSST) program with the U.S. Army

Aviation Applied Technology Directorate (AATD) and Bell helicopter Textron.

Vibration data from the OH-58C planetary system were collected on a healthy

transmission as well as with various seeded-fault components. Planetary fault

detection algorithms were used with the collected data to evaluate fault

detection effectiveness. Planet gear tooth cracks and spalls were detectable

using the vibration separation techniques. Sun gear tooth cracks were not

discernibly detectable from the vibration separation process. Sun gear tooth

spall defects were detectable. Ring gear tooth cracks were only clearly

detectable by accelerometers located near the crack location or directly across

from the crack. Enveloping provided an effective method for planet bearing

inner- and outer-race spalling fault detection

3.1 HARDWARE IMPLEMETATION

3.1.1 ADXL330 MEMS Accelerometer

Figure 3.1 MEMS Accelerometer

The ADXL330 is a complete three-axis acceleration measurement

system on a single monolithic IC. The ADXL330 has a measurement range of

±6g.The block diagram is illustrated in Figure 3.2. It contains a micro-

machined sensor and signal conditioning circuit to implement the open loop

acceleration measurement architecture. The output signals are analog voltages

that are proportional to acceleration. The accelerometer can measure the static

acceleration of gravity in tilt sensing applications as well as dynamic

acceleration resulting from motion, shock, or vibration. Deflection of the

structure is measured using a differential capacitor that consists of

independent fixed plates and plates attached to the moving mass. The fixed

plates are driven by 180° out-of-phase square waves. Acceleration deflects the

moving mass and unbalances the differential capacitor resulting in a sensor

output whose amplitude is proportional to acceleration. Phase-sensitive

demodulation techniques are then used to determine the magnitude and

direction of the acceleration.

Figure 3.2 Block Diagram of ADXL330 MEMS Accelerometer

The demodulator output is amplified and brought off-chip

through a 32kΩ resistor. The user then sets the signal bandwidth of the

device by adding a capacitor. This filtering improves measurement

resolution and helps prevent aliasing. The user selects the bandwidth of

the accelerometer using the CX, CY, and CZ capacitors at the XOUT,

YOUT, and ZOUT pins. Bandwidths can be selected to suit the

application, with a range of 0.5 Hz to 1600 Hz for X and Y axes, and a

range of 0.5 Hz to 550 Hz for the Z axis.

3.1.1.1 Features

3-axis sensing

Small, low-profile package : 4 mm × 4 mm × 1.45 mm

LFCSP

Low power: 180μA at VS = 1.8 V (typical)

Single-supply operation: 1.8 V to 3.6 V

10,000 g shock survival

Excellent temperature stability

BW adjustment with a single capacitor per axis

RoHS/WEEE lead-free complian

3.1.1.2 Pin configuration

Figure 3.3 Pin Configuration

Table 3.1 Pin function Description of ADXL330 Tri-axial AccelerometerPin no. Mnemonic Description Pin no. Mnemonic Description

1. NC No connect 9. NC No Connect

2. ST Self test 10. YOUT Y Channel Output

3. COM Common 11. NC No Connect

4. NC No connect 12. XOUT X Channel Output

5. COM Common 13. NC No Connect

6. COM Common 14. VS Supply Voltage (1.8V to 3.6V)

7. COM Common 15. VS Supply Voltage (1.8V to 3.6V)

8. ZOUT Z Channel Out 16. NC No Connect

Figure 3.4 ADXL330 Tri-axial Accelerometer Mounted on the Motor

Housing

The wireless sensor network is implemented and an

accelerometer is also integrated in this monitoring system for detecting

the vibration signals. Vibration signals were collected using ADXL330

tri-axial accelerometer mounted on the motor housing in Figure 3.3

All three-axis vibration signals are acquired at a sampling rate of

2kHz by a 12-bit ADC conversion.

3.1.1.3 Applications

Cost-sensitive, low power, motion- and tilt-sensing

applications

Mobile devices

Gaming systems

Disk drive protection

Image stabilization

Sports and health devices

3.1.1.4 Advantages

Apart from the significant cost saving over traditional force-

balance accelerometers, due to the nature of their design micro-

electromechanical systems sensors have a much better high frequency

response. Where most earthquake accelerometers are specified as

having a frequency response of DC to 50Hz, 100Hz or in some cases

200Hz, the seismic-oriented MEMS sensors have a much higher

frequency range. For example, the Silicon Designs units used in the

ESS-1221 sensor have a frequency response of DC to 400Hz, and the

Colibrys SF3000L MEMS sensors extend to 1000Hz.

Frequency response is important when recording strong motion,

particularly for events at close range where high frequencies have not

been attenuated with distance. In blast monitoring, where the source

can be only dozens of metres away from the sensor, frequencies of

5000Hz or more can be recorded, so it is possible that large, nearby

earthquakes could achieve their peak accelerations in frequencies above

200Hz. Earthquake recorders typically record data at 100sps and

200sps, meaning that frequencies above 50Hz or 100Hz are not

recorded. More can be learnt about earthquakes by using MEMS

accelerometers and recorders capable of sampling at up to 2000sps.

Condition monitoring and fault diagnosis of induction motors are

of great importance in production lines. It can significantly reduce the

cost of maintenance and the risk of unexpected failures by allowing the

early detection of potentially catastrophic faults. In condition based

maintenance, one does not schedule maintenance or machine

replacement based on previous records or statistical estimates of

machine failure. Rather, one relies on the information provided by

condition monitoring systems assessing the machine's condition.

Thus the key for the success of condition based maintenance is

having an accurate means of condition assessment and fault diagnosis.

On-line condition monitoring uses measurements taken while a

machine is in operating condition.

There are around 1.2 billion of electric motors used in the United

States, which consume about 57% of the generated electric power.

Over 70% of the electrical energy used by manufacturing and 90% in

process industries are consumed by motor driven systems.

Among these motor systems, squirrel-cage induction motors

(SCIM) have a dominant percentage because they are robust, easily

installed, controlled, and adaptable for many industrial applications.

SCIM find applications in pumps, fans, air compressors, machine

tools, mixers, and conveyor belts, as well as many other industrial

applications. Moreover, induction motors may be supplied directly

from a constant frequency sinusoidal power supply or by an a.c.

variable frequency drive. Thus condition based maintenance is essential

for an induction motor.

It is estimated that about 38% of the induction motor failures are

caused by stator winding faults, 40% by bearing failures, 10% by rotor

faults, and 12% by miscellaneous faults. Bearing faults and stator

winding faults contribute a major portion to the induction motor

failures. Though rotor faults appear less significant than bearing faults,

most of the bearing failures are caused by shaft misalignment, rotor

eccentricity, and other rotor related faults. Besides, rotor faults can also

result in excess heat, decreased efficiency, reduced insulation life, and

iron core damage. So detection of mechanical and electrical faults are

equally important in any electrical motor.

In the proposed method, vibration signals are obtained using

piezo-electric sensor and motor current signature analysis is performed

using Hall Effect sensor. The features of the signal are analyzed using

wavelet packet transform. Besides other signal processing techniques,

wavelet packet transform is preferred because it has certain advantages.

Traditional signal processing techniques like Fourier transform

can perform only on stationary signals. Since it is not well suited for

non-stationary signals short time Fourier transform (STFT) is used.

STFT uses a constant window function as a base to obtain the

frequency spectrum coefficients. The size of the window function

cannot be changed which led to the need for wavelet transform.

Wavelet transform uses a varying size window function as its base. In

wavelet transform low frequency signals are decomposed repeatedly to

obtain low frequency information. In wavelet transform the information

about high frequency signals are limited. In the proposed method,

wavelet packet transform decomposes both low frequency and high

frequency information. It can analyze both stationary and non-

stationary signals. There are many classifier models to effectively

classify the faulty data from the healthy one. They are: Analytical

model-based methods, Artificial Intelligence-based methods. analytical

model based methods are efficient monitoring systems for providing

warning and predicting certain faults in their early stages. Artificial

Intelligence based methods are of two categories: Knowledge based

models and Data based models. When considering fault diagnostics of

induction motor it is difficult to develop an analytical model that

describes the performance of a motor under all its operation points. It is

difficult for a human expert to distinguish faults from the healthy

operation. Though analytical based methods and knowledge based

methods are effective classification methods, their performance in

induction motors is not good. Moreover conventional methods cannot

be applied effectively for vibration signal diagnosis due to their lack of

adaptability and the random nature of vibration signal. In such a

situation, data based models are used to classify faults in induction

motors.

Data based models are applied when:

— the process model is not known in the analytical form

— expert knowledge of the process performance under faults is not

available

Some of the popular data based models are neural networks,

fuzzy systems and Support vector machine. Neural networks and

fuzzy logic are widely used in the field of fault diagnostics. Fuzzy

logic provides a systematic framework to process vague and

qualitative knowledge. Using fuzzy logic it is possible to classify a

fault in terms of its degree of severity. Artificial neural network are

modeled with artificial neurons. Each artificial neuron accepts several

inputs, applies preset weights to each input and generates a non-linear

output based on the result. The neurons are connected in layers

between the inputs and outputs.

Support Vector Machine, a novel machine learning technique is used in this

paper. It is based on statistical learning theory, and is introduced during the

early 90’s. SVM is opted in this paper since it is shown to have better

generalization properties than traditional classifiers. Efficiency of SVM does

not depend on the number of features of classified entities. property is very

useful in fault diagnostics, because the number of features to be chosen to be

the base of fault classification is thus not limited.

1.MICROELECTRO MECHANICAL SENSORS

Microelectro mechanical systems (MEMS) inertial sensors provide a small

footprint with sensitivities that are either comparable or exceed any macro

sensor, along with the capability of mass production and low unit cost. These

sensors utilize compliant micro-flexures attached to a proof mass that

displaces in response to an environmental acceleration. Many transduction

mechanisms have been developed that convert the displacement into a

measurable electric signal and include thermal, piezoresistive, piezoelectric,

optical, and capacitive methods. Most MEMS sensors are silicon based, and

are fabricated using surface or bulk micromachining. Surface micromachining

creates free standing movable structures on top of a substrate using a

combination of sacrificial layers, and structural layers which are commonly

polysilicon [3]. In bulk micromachining, the mechanical structures are

defined using a removal process where bulk material, typically silicon, is

etched away. MEMS sensors have been used for vibration and shock

monitoring on industrial systems and robotics, guidance and navigation in

global positioning systems (GPS), seismometry in earthquake prediction,

image stabilization in digital cameras, and automobile safety and stability [4].

2.AUTOMOTIVE SENSORS

Sensors cover every major aspect of a modern automobile: power-train

sensors monitor fuel combustion and emissions, chassis sensors monitor road

traction and tire condition, and body sensors facilitate air-bag deployment and

vehicle proximity for radar guided cruise control [5]. Pressure sensors,

whichtypically consist of a piezoresistive strain sensing element attached to a

silicon diaphragm that deflects when exposed to an applied pressure, are one

of the first micro-machined sensors used in an automobile. Implemented as

manifold absolute pressure (MAP) sensors, they allow precise control of the

air fuel ratio, which allows the catalytic converter to efficiently reduce

tailpipe emissions [6]. Variable reluctance sensors based on electromagnetics

are used for automobile traction control, and produce a voltage output that is

dependant on the magnetic flux variations between a rotating component and

the sensor bias magnet [5,7]. MEMS based linear accelerometers are utilized

for airbag deployment upon impact, and provide a lower cost and small

package over the older ball-in-tube accelerometers [6].

3.EXISTING METHODS FOR FAULT DETECTION

As this thesis develops a sensor for fault detection using vibration

monitoring, this section overviews three existing methods for fault detection in

mechanical systems. First, motor current monitoring is discussed as it has the

ability to measure significant bearing faults in motor systems. Next, temperature

sensors specifically thin film thermocouples (TFTC) and MEMS temperature

sensors, are discussed. These sensors are mounted in the chamber of a

lubricating fluid or in close proximity to a moving component, and monitor

temperature fluctuations caused by increased wear. Finally, acoustic emission

(AE) sensors are discussed which have the ability to detect ultrasonic emissions

generated by plastic deformation in the crystalline lattice of a mechanical

component.

3.1TEMPERATURE SENSORS

Temperature sensors are largely used in high temperature, high load and

high speed environments, typically turbine engines and industrial tools.

Thermocouple sensors consist of two dissimilar metals joined together at two

junctions, and when exposed to a temperature difference generate a

thermoelectric voltage dependant only on the material properties and junction

temperature difference [11]. In turbine engines these sensors can be placed on

bearing casings or in the lubricant to monitor temperature fluctuations due to

increased friction. The thermocouples are fabricated using bulk materials or

micro-machined thin films (TFTC), with the latter providing a faster response,

small package and substantially lower production costs [12]. The TFTC can be

embedded for in situ monitoring of processes in hostile environments: One

study successfully embedded TFTC into electroplated nickel and attained

device characteristics on par with surface mounted devices [13]. More recently

MEMS temperature sensors have been developed that consist of micro-

machined semiconductor material that undergoes structural deformation with

temperature changes. A common implementation is a bent beam structure

whose temperature induced deflection, δ, is expressed as [14]: Temperature

sensors are well suited for localized fault monitoring and detection, but can only

characterize a mechanical system consisting of multiple components through

the use of multiple sensors. 9

3.2 ACOUSTIC EMISSION SENSORS

Acoustic emission (AE) sensors have been used to characterize wear in

machine tools, and monitor bearing and gear problems in centrifugal pumps

[15,16]. First developed as a Non-Destructive Testing (NDT) technique to

detect cracks in civil structures, these sensors detect acoustic emissions

generated by the release of vibration waves in a crystalline lattice due to plastic

deformation or crack growth [16]. Measurements are made using piezoelectric

transducers with high natural frequencies, 100 kHz to 1 MHz, to capture the

ultrasonic AE emissions. An AE sensor is useful as it has the ability to detect

subsurface cracks in gear teeth or bearings before appearing on the surface

causing further damage [17]. More recently MEMS acoustic sensors have been

developed, and one design includes multiple transducers on a single substrate,

which each detect acoustic emission energy at different frequencies [18]. This

helps distinguish spurious acoustic emissions arising from impact and friction,

from those arising from plastic deformation. When compared to typical

piezoelectric sensors, the MEMS devices have lower sensitivities and fail to

detect some acoustic emissions [18,19]. In addition, the acoustic emission signal

suffers severe attenuation as is crosses various interfaces, such as a gearbox or

bearing casings. In one experiment consisting of a pinion gear and an associated

bearing, a 44dB attenuation was seen between an AE sensor placed directly on

the pinion to one placed on the bearing casing, and in some cases, intermediate

loss of the signal was observed [20].

4. MEMS VIBRATION SENSORS

This thesis focuses on the development of a MEMS vibration sensor for fault

detection, as the vibration signature of a mechanical system has the potential to

give an overview of the entire system. Many types of MEMS vibration sensors

exist with different operating principals based on resonance, the piezoelectric

effect, and displacement variation.

4.1 RESONANT SENSORS

Resonant sensors are based on the resonant frequency shift of a fixed-

fixed micro-beam that is excited into resonance using electrostatic, thermal or

piezoelectric methods. The performance of resonant sensors are dependant on

the quality factor, Q, which is the ratio of total energy stored in the system to

the energy lost per cycle, and determines the sharpness and amplitude of the

resonance peak. A high Q resonator develops a peak that is easily

distinguishable in a phase locked loop control circuit, and allows for improved

performance and resolution [3]. A resonator Q value is lowered by energy

losses to the surrounding fluid, attached supports, and material. Losses to the

fluid surrounding the MEMS structure is the dominant loss mechanism for

MEMS sensors operated in air, and can be greatly reduced by vacuum

packaging, which can be accomplished using glass-silicon anodic bonding [22].

In addition to increasing fabrication cost, the long term stability of the vacuum

degrades with ultimate device failure stemming from leakage through micro

cracks and defects, and out-gassing [22]. Support losses arise from the restoring

forces they generate, and can be reduce by balanced designs such as a double

ended tuning fork (DETF) or operation in higher modes [3]. In addition, the

sensitivity of a resonant sensor can be increased by the addition of micro-levers

to increase the axial force. In one such design, a two stage micro-lever increased

the sensitivity by an order of magnitude over a more conventional single lever

arm [23,24].

4.2 PIEZOELECTRIC SENSORS

Piezoelectric materials produce a charge in response to an applied force.

This is an intrinsic effect in materials such as Zinc Oxide (ZnO), whereas

materials such as Barium Titanate or Lead Zirconate Titanate (PZT) need to be

poled by placing the material in a strong electric field at an elevated temperature

[3]. The subscripts i=1,2,3, j=1,2,3,4,5,6, and k=1,2,3 represent the direction to

which the physical property is related. For micro-sensors poled in the out of

plane direction, only the D3 term is relevant, and if fabricated using ZnO or

PZT only the charge coefficients d31, d32, and d33 are relevant [25]. The

piezoelectric material is commonly deposited as a thin film on the surface of

compliant structures in bulk micro-machined sensors or incorporated as a layer

in surface micro-machining on a similar structure [26]. However, the inclusion

of multiple materials introduces large temperature variations, causing out of

plane bending especially in surface micro-machined devices [26,27]. Poled

piezoelectric

materials could be depolarized, which serves to reduce the piezoelectric affect

in the material. This can occur if the sensor is exposed to a strong electric field

of the opposite polarity, high temperatures in excess of the Curie point, or high

mechanical stress. Also, the piezoelectric effect reduces over time, an effect that

can be reduced with the addition of composite elements [28]. Foundry processes

such as PolyMUMPS and SoiMUMPS do not incorporate piezoelectric layers,

thus sensors using this method of transduction must use a customized process.

The piezoelectric sensor outlined in [26] uses polysilicon surface micro-

machining similar to PolyMUMPS, however it incorporates a ZnO layer that is

not available in the foundry process.

4.3 DISPLACEMENT VARIATION SENSORS

Displacement variation sensors consist of a proof mass connected to a

compliant micro-flexure. The movement of the proof mass is sensed using

optical, electron tunneling, or more commonly, capacitive methods. A recent

optical accelerometer utilized a novel multilayer nano-grating for in-plane

displacement sensing of the proof mass [29]. For small space variations between

the nano-gratings, a large change in the optical reflection amplitude of the

grating was observed [30]. Even though optical sensors offer resolutions

approaching the Brownian noise limit, they are not available in small packages

that are easily placed in space constricted areas for a low cost [29].

The electron tunneling effect is observed when the proximity between

two electrodes, one flexible and the other fixed, is on the order of 10Ǻ [31].

When the flexible electrode moves in response to acceleration, a feedback

controller maintains the original distance while determining the applied

acceleration. Electron tunneling sensors are able to sense accelerations in the

μg-ng range due to the exponential relation between tunneling current and gap.

However, these highly sensitive devices are susceptible to thermo-mechanical

noise, which could be reduced by operation at low pressure [32]. It is also

important for the two electrodes to be coated with metal, which is difficult to

achieve with foundry fabrication.

Capacitive sensors are based on the parallel plate capacitor, and are

implemented using inter-digitated comb fingers in an in-plane lateral or

transverse drive, or in an out of plane parallel plate drive [33]. These sensors

have the advantage of high sensitivity, low power consumption, small package,

low temperature dependence and are easily integrated with existing foundry

fabrication processes [34]. The lateral drive offers a fully linear response but

low sensitivity, whereas the transverse and parallel plate drives achieve high

sensitivity if gap distances are of a few microns and large capacitance areas are

utilized. However, both drives have a non-linear response and are prone to pull-

in. Also, if these sensors are fabricated using a surface micromachining process

such as PolyMUMPS, where the ground plane is a few microns above the

structural layer, they are prone to electrical failure due to shorting. This was

observed in an experiment where similar sensors were placed in a vibrating

environment with a peak of 120g, and shorting between the structural layer and

ground plane was found to be the predominate mode of failure [35]. However,

with certain fabrication processes, such as SoiMUMPS, this issue can be

neglected as the substrate material below any moving component is removed.

Most capacitance sensors are fabricated using an integrated CMOS

(Complementary Metal Oxide Semiconductor)-MEMS technique, where the

MEMS sensor is fabricated before, during or after the CMOS circuit fabrication

[36]. Although these sensors offer high-sensitivity and small parasitic

capacitance, they suffer from in-plane and out of plane curling of the beam

which reduces the capacitance between adjacent comb fingers [37]. In addition,

multiple processing steps are required for device fabrication.

More than the other transduction methods, MEMS capacitive

displacement sensors are attractive for practical implementation in an

automobile, and therefore is the selected method for the vibration sensor

presented in this work. Their low temperature dependence is an ideal

characteristic for an automobile environment that faces varying temperature. If

implemented with off-chip readout circuitry, they require a standard foundry

fabrication process, which provides a strong platform for mass production and

low unit cost.

Damaged Gearbox Investigation

Experiments were conducted to investigate the progression of a tooth

defect on a high speed shaft (HSS) pinion, which was introduced into the

WTCMTR at variable-speed and generator load. The behaviour of a healthy

pinion and of four faults of increasing severity were investigated by introducing

progressive damage to the leading contact edge of one tooth of the gearbox

pinion. shows the healthy pinion, Figures 3(b), (c), and (d) show early stages of

tooth wear, while Figure 3(e) depicts the entire tooth missing. The most

important components in gear vibration spectra are the tooth meshing frequency

and its harmonics, together with sidebands caused by modulation phenomena.

The increment in the number and amplitude of such sidebands may indicate a

fault condition. Moreover, the spacing of the sidebands is related to their source

[12]. Damaged Gearbox Investigation Experiments were conducted to

investigate the progression of a tooth defect on a high speed shaft (HSS)

pinion, which was introduced into the WTCMTR at variable-speed and

generator load. The behaviour of a healthy pinion and of four faults of

increasing severity were investigated by introducing progressive damage to the

leading contact edge of one tooth of the gearbox pinion. Figure 3(a) shows the

healthy pinion, Figures 3(b), (c), and (d) show early stages of tooth wear, while

Figure 3(e) depicts the entire tooth missing. The most important components in

gear vibration spectra are the tooth meshing frequency and its harmonics,

together with sidebands caused by modulation phenomena. The increment in

the number and amplitude of such sidebands may indicate a fault condition.

Moreover, the spacing of the sidebands is related to their source [12].