Vibration Analysis on Bearings

Rolling Element Bearing Analysis and Condition Assessment

By Ron Frend

FRANK VOWLES ENGINEERING

Background

Bearing condition assessment is one of the most common subjects of training courses and technical papers in the PDM industry and is often the topic of conversation wherever maintenance maintenance decisions decisions are being made. When you you consider that the single most common failure of rotating machinery is the failure of these bearings and that the vast majority of machines are reliant on the health of their support bearings for correct operation this is perhaps no surprise. Due mostly to financial restraints and a lack of confidence in predictive techniques approximately two-thirds of rolling element bearing machine operators use corrective, run-to-failure maintenance. Preventive maintenance accounts for most of the remainder with only a small percentage using some form of predictive strategy.

Our objective

The primary objective of this course is to highlight technologies and methodologies that can improve your your ability to diagnose bearing health. To highlight the impact of predictive technologies on operational and maintenance costs and recommend approaches that will maximize your effectiveness with the prediction of rolling element bearing failure.

This document is protected under copyright, it may not be reproduced in whole or in part without written consent of Ron Frend Page 1

CONTENTS SECTION

PAGE #

1.0 UNDERSTANDING BEARINGS AND THEIR FAILURE MECHANISMS

7

1.1 TYPES OF ROLLING ELEMENT BEARINGS IN USE

7

1.2 LUBRICATION METHODS

8

1.3 BEARING FAILURE MECHANISMS

9

1.3.1 Fatigue Spalling 1.3.2 Lubricant Loss 1.3.3 Poor Assembly 1.4 BEARING LIFE ESTIMATION 2.0 TYPES OF MAINTENANCE USED IN INDUSTRY

2.1 COST OF MAINTENANCE

15 16 17

2.2 COMPARISON OF MAINTENANCE TECHNIQUES 2.2.1 Case history 1 : Comparison of maintenance techniques

3.0 RECOMMENDED PDM DIAGNOSTIC TECHNIQUES

3.1 VIBRATION SENSOR TYPES AND MOUNTING EFFECTS

21 21

3.1.1 Vibration sensor types Accelerometers Velocity Sensors Displacement Sensors

3.1.2 Sensor Mounting Effects 3.2 VIBRATION BASED ROLLING ELEMENT BEARING ANALYSIS

25

3.2.1 Bearing failure characteristics


CONTENTS SECTION

PAGE #

1.0 UNDERSTANDING BEARINGS AND THEIR FAILURE MECHANISMS

7

1.1 TYPES OF ROLLING ELEMENT BEARINGS IN USE

7

1.2 LUBRICATION METHODS

8

1.3 BEARING FAILURE MECHANISMS

9

1.3.1 Fatigue Spalling 1.3.2 Lubricant Loss 1.3.3 Poor Assembly 1.4 BEARING LIFE ESTIMATION 2.0 TYPES OF MAINTENANCE USED IN INDUSTRY

2.1 COST OF MAINTENANCE

15 16 17

2.2 COMPARISON OF MAINTENANCE TECHNIQUES 2.2.1 Case history 1 : Comparison of maintenance techniques

3.0 RECOMMENDED PDM DIAGNOSTIC TECHNIQUES


21 21

3.1.1 Vibration sensor types Accelerometers Velocity Sensors Displacement Sensors

3.1.2 Sensor Mounting Effects 3.2 VIBRATION BASED ROLLING ELEMENT BEARING ANALYSIS

25

3.2.1 Bearing failure characteristics


3.2.2 Recommended Vibration based analysis techniques Demodulation theory The demodulation process Resonance sources

Vibration case 1 AC motor bearing defect using demodulation Vibration case 2 DC motor bearing defect using demodulation Vibration case 3 Press bearing damage using demodulation

Using wave audio with demodulation for bearing analysis

The Application of Demodulation in a PDM approach to bearing analysis 3.2.3 Evaluation of Damage Severity Severity assessment using demodulation and velocity spectrum

3.3 DEMODULATION OF ULTRA SOUND DATA FOR BEARING ANALYSIS

49

3.4 OTHER TIME DOMAIN TECHNIQUES FOR BEARING ANALYSIS

51

3.4.1 Use of Kurtosis to identify bearing damage 3.4.2 Use of shock pulse monitoring to identify bearing bearing damage

3.5 OTHER FREQUENCY DOMAIN TECHNIQUES FOR BEARING ANALYSIS

52

3.5.1 Use of Broad band Frequency domain bearing analysis 3.5.2 High-Frequency Resonance Techniques

3.6 THE USE OF OIL ANALYSIS FOR BEARING DIAGNOSIS

53

3.7 THE USE OF TEMPERATURE MEASUREMENTS FOR BEARING DIAGNOSIS 55 3.7.1 Factors that effect the collection o f quality thermographic data


3.7.2 Evaluation of Severity using temperature measurement Thermographic case 1: Spindle bearing over packed with grease Thermographic case 2: Over torqued bearing housing

4.0 WHY PDM ? A MAINTENANCE SELECTION STRATEGY

4.1 CONSIDERATIONS FOR SELECTING THE TYPE OF MAINTENANCE

61

61

4.1.1 Useful questions to answer in evaluating the feasibility of PDM 4.1.2 Useful questions to answer in understanding the costs of PDM

5.0 CONCLUSIONS AND RECOMMENDATIONS


63

ILLUSTRATIONS

Figure

Title

1.

Most Commonly Used Bearings

2.

Lubrication Methods Used in Power Plant Machinery

3.

Failure Modes in Different Types of Power Plants

4.

Early Fatigue - Ball Bearing Courtesy of the Barden Corporation

5a.

Developed Fatigue on Roller Bearing Courtesy of the Torrington Company

5b.

Loss of Lubricant - Ball Bearing Inner Race Courtesy of the Barden Corporation

5c

Loss of Lubricant - Roller Bearing Courtesy of the Torrington Company

6.

Grease Life Expectancy for the type 206 and the type 9109 bearings

7a.

Installation Damage Loose Fit - Ball Bearing Outer Ring

7b

Installation Damage Loose Fit - Ball Bearing Outer Ring

8.

Percent of Machines Subject to Different Types of Maintenance

9a.

Maintenance Costs using corrective maintenance

9b.

Maintenance Costs using preventive maintenance

9c.

Maintenance Costs using predictive maintenance

10.

Types of accelerometer construction

11.

Piezoelectric Accelerometer Mounting Methods and Examples of Typical Frequency Responses

12.

Summary of common mounting techniques

13

Ball Bearing Terminology

14

Time and frequency domain data showing modulation

15

Waterfall of frequency domain data showing modulation

16

Time domain data showing start of demodulation process

17

Time domain data showing rectification and smoothing

18

Illustration of Time and frequency domain demodulated data.

19

Illustration of data from motor bearing

20

Illustration of waterfall frequency domain data showing modulation

21

Waterfall Trend and Spectral plot showing bearing defect

22

Illustration of modulated and demodulated velocity spectrum

23

Illustration of time domain data from DC motor bearing showing faults

24

Illustration of demodulated spectrum showing bearing defect frequency

25

Illustration of velocity spectrum indicating bearing defect frequency

26

Time domain trace for press flywheel showing bearing damage

27

Demodulated acceleration spectrum showing bearing defect on press flywheel


28

Velocity spectrum showing bearing defect on press flywheel

29

Demodulated acceleration spectrum obtained from recorded wave audio

30

Velocity spectrum obtained from wave audio file

31

Time waveform display of wave audio file

32a

Bearing Failure Sequence diagram general technologies

32b

Bearing failure sequence diagram for use with demodulated vibration data

33

Demodulated, heterodyned, ultra sound measurement showing bearing fault

34

Good V’s bad bearing shown using demodulated ultra sound

35

Probability Density in Decibels of Normalized Acceleration of a good and bad bearing

36

Trend of high frequency noise for a damaged bearing

37.

Microprobe Spectrum of a Bearing Metal Particle

38.

Trend of oil sample wear elements

39

Example of hot V’s cold motor bearings

40

Example bearing image 10 minutes after deliberate overpacking with grease

41


42


43

Example spectrum from cool bearing

44

Thermographic image showing location of vibration measurement points

45

Example spectrum from hot bearing

46.

Simplified Maintenance Selection Diagram


1.0 Understanding bearings and their failure mechanisms In order to be effective with condition assessment for rolling element bearings it is essential to have a grasp of the primary bearing types and failure modes experienced in industry.. industry.. This section is intended to introduce the primary forms forms of failure. Once a course of diagnosis is embarked upon we we have found that it is important once bearing failure has been suspected, that the bearing is removed from service and is dissected in order to visually confirm the suspected failure and build confidence in the adopted program The majority of this handbook references specific failure modes and characteristics that identify the existence of a particular failure mode, for this reason it is vital that theses failure modes can be visualized.

1.1 TYPES OF ROLLING ELEMENT BEARINGS IN USE Three types of rolling element bearings are commonly used in industry, ball, roller, and tapered roller. Each of these types are more suited to different applications. For example, ball bearings bearings are used predominantly for high speed applications where axial and radial load is required, deep groove bearings are suitable for grease-packed lubrication, and angular contact bearings are predominantly used in oillubricated systems. Furthermore, for high radial load applications cylindrical roller bearings are preferred and for a combination of high radial load and thrust load tapered rollers are the best choice. Outside of these general parameters, specific conditions exist that mandate the selection of a bearing type such as the avoidance of critical speed operations, since roller bearings generally show higher stiffness than ball bearings. Theses examples are not intended intended to act as selection criteria you must reference reference your bearing manufacturer to match the correct type of bearing for a specific application. The following figure illustrates the mechanics of both ball and roller bearings


Figure 1 Most commonly used bearings

Considering the variety of applications that require different operational survivability and performance, it is surprising that the ball bearing is so common. Its all round performance has shown it to the best choice for so many applications, in fact a recent EPRI study based on the utility industry shows, overall, 87% of all rolling element bearings are ball bearings, with only 7% being tapered roller bearings, and 6% are cylindrical roller bearings.

1.2 LUBRICATION METHODS Since the most common type of bearing you will encounter is the ball bearing, the next most critical fundamental issue is the type type of lubrication. lubrication. As illustrated in Figure 2, a study performed performed recently in the utility industry found that most rolling element bearings are grease lubricated. The second most common being oil-ring lubrication , accounting on the average for 33% of all bearings. Forced-feed lubrication, although not the most common, is arguably the most effective. It, however requires the largest and most expensive support systems providing direct injection of oil into the bearing. In general it seems clear to state that: One, grease-packed ball bearings are by far the most common and appear to be the type of choice for most industries. Two, Forced feed or oil bath lubrication methods only seem to be found in larger, more expensive pieces of machinery.


Figure 2 Lubrication Methods Used in Power Plant Machinery

1.3 BEARING FAILURE MECHANISMS A great deal of bearing failure experience exists both in industry and with the bearing manufacturers, this experience indicates that there are six primary causes of bearing failure. Fatigue spalling Lubricant loss Poor assembly Contamination Brinelling Overheating. A detailed look at one industry, the utilities, indicates that fatigue spalling is the predominant failure mode constituting (19%) in nuclear utilities and (22%) in coal fired plants. The combination of loss of lubricant and overheating failures, common to grease-packed bearings, account for 21% of all failures. Specific operational conditions tends to effect theses percentages, a case in point being that 52% of all bearing failures in the oil and gas industries were caused by overheating and loss of lubrication, probably due to the average operational speed being higher. Likewise, unusually high contamination contamination accounts for 25% of failures in the Gas/Oil Gas/ Oil power plants.


Table 1 Failure Modes in Different Types of Power Plants FAILURE MODES (% of Machines) Plant Type Failure Mode

Lubricant Loss

Nuclear

12

Coal

11

Gas/Oil

Overall

34

15

Corrosion

2

17

4

Electro-pitting

2

--

1

Overheating

7

17

6

Contamination

5

11

25

10

--

17

Fatigue Spalling

19

22

Brinelling

12

4

Poor Assembly

12

11

--

10

Others (unidentified)

29

42

7

30

7

( Detailed information provided by EPRI bearing condition assessment survey ) The data in Table 1 above shows that fatigue spalling, lubricant loss, and poor assembly are by far the most common of all failure modes. Contamination, brinelling, overheating, and other failure modes such as corrosion and electro-pitting, are either extensions of the major failure modes or, when independent, produce outward signals that are similar to those generated by the listed modes. For example, overheating results commonly from ineffective lubrication caused by either lubricant loss or overpacking. In either case, the function of the lubricant becomes greatly impaired, and wear and/or fatigue will take place. Brinelling, contamination, or corrosion, damages bearing surfaces and produces vibration patterns similar to those of fatigue etc.

Fatigue Spalling Fatigue failure or spalling results from mechanical materialogical failure of the bearing. Literally a stress related failure of the material which results from cyclic stresses due to operation at high loads. This fatigue may be initiated on the surface or beneath the surface. Surface fatigue is usually caused by scratches on races, balls, or rollers, abrasive contamination, or brinelling. These marks produce “stress raisers”, a point on the bearing surface that experiences abnormally high stress due to the physical conditions at that spot. Simply, a given load over a given area produces stress. If a crack or contaminant is found at that location, the load is distributed over a different (often smaller) area and therefore greatly increases stress at that point. This phenomenon limits the number of cycles a bearing can survive. These raised stress areas provide a start point for micro-crack formation that leads eventually to pitting, spalling, and wear.


Subsurface fatigue is usually caused by voids, foreign matter or coarse carbides introduced into the material at the time of formation. These material anomalies again provide for a point of crack formation if they fall within a high stress area. and once a crack is formed beneath the surface, it works its way outward and eventually develops into a spall. With time, both surface and subsurface fatigue flaws spread over the active bearing surfaces, causing bearing wear, growth in spalls, and eventual machine failure. The metal contaminants or wear particles removed from the bearing during spalling either are washed out with the oil, in oil-lubricated bearings, or are trapped in the bearing, as is common in sealed and grease-packed bearings. In these latter bearings, continuous recirculation of the particles causes progressively higher wear, to the point where either the bearing becomes excessively loose and fails to support the load suitably or the induced damage leads to failure.

Figure 3 Fatigue - Ball Bearing Courtesy of the Barden Corporation

Lubrication Loss Loss of lubricant is arguably the most common cause of bearing failure and occurs most frequently in grease-packed bearings. However, inadequate oil lubrication or oil pump failure, does produce similar results in oil-lubricated systems. Loss of lubrication in the case of grease or oil systems is usually not necessarily the physical loss of lubricant but the loss of the oil or grease’s properties. Since the life of a grease lubricant is strongly temperature dependent and since grease loses half of its life for every 20°F rise in temperature, this can be seen to have a very significant effect on bearing health. Oil on the other hand is not as bad, since while oil oxidation rates double every 180°F, when oil temperatures exceed 2000°F there are no other significant temperature effects. It is important to realize that the temperatures referenced here are the lubricant temperatures experienced at any point in the system and not necessarily the bearing temperature nor the lubricant at the point of measurement. Figure 6 shows an example of the life expectancy of one kind of grease. Note the difference in life for two bearings of different sizes running on the same shaft. In order to avoid the temperature effect due to overpacking, bearing manufacturers usually suggest that bearings are packed with grease to between 15% to 20% of the bearing's free volume. We know that overpacking is the most common cause of raised bearing temperature, which leads to reduction in grease


life and eventual failure. Under-packed bearings, on the other hand, or bearings that have lost grease due to physical migration, may generate high bearing temperatures when running at high speeds. Low speed, starved bearings on the other hand, usually simply wear into a condition of excessive looseness and fail without appreciable temperature increases. This is the more normal failure we think of when we think of loss of lubrication and although it is common place, surprisingly it is not the most common.

Figure 4 Developed Fatigue on Roller Bearing Courtesy of the Torrington Company


Figure 5 Loss of Lubricant - Ball Bearing Inner Race Courtesy of the Barden Corporation

Figure 6 Loss of Lubricant - Roller Bearing Courtesy of the Torrington Company


12.5 years

15 Months

45 Days

Figure 7 Grease Life Expectancy for type 206 and the type 9109 Torrington bearings

Poor Assembly The most common assembly problems that effect bearing health are : Inadequate alignment Cocking of the bearing retainer Poor seating of bearing races Lack of preload (as specified) Brinelling due to pounding Overpacking Contamination Inadequate balancing. The effect of any of the above, results in bearing life that is significantly reduced or, if the effect is dramatic enough, then bearing failure mostly occurs very shortly after installation. Poor seating, inadequate balance, and bearing cocking produce premature failures similar to those discussed under fatigue, however rotating loads and misalignment may also cause the cage to fail. Figure 7a shows the results of a loose ball bearing fit on the outer race. Figure 7b depicts a chipped-off inner race shoulder on a double roller bearing.


Figure 8 . Installation Damage Loose Fit - Ball Bearing Outer Ring

Courtesy of the Barden Corporation

Figure 9 Installation Damage Loose Fit - Ball Bearing Outer Ring

Courtesy of the Barden Corporation


1.4 BEARING LIFE ESTIMATION Bearing life varies from application to application in accordance with some of the fundamental influences such as speed, load, temperature etc. A bearings life from new can be estimated however by using the following equation:

L = 16700 (C/P)^3 N where: N = rpm C = bearing life coefficient (obtained from the manufacturer) P = static load on bearing This equation as stated gives estimated life. The subject of this course is to discuss the effect of various influences on bearing life, and vibration is one of the primary technologies discussed. In the rest of this course we discus the use of vibration analysis to determine failure, and the measurement of the overall vibration level as an indicator of failure, however theses discussions are based on the measurement of the vibration created by the bearing. Here we will show that not only does a bearing create vibration that can be used as an indicator of its health, but the vibration caused by the bearings environment also has an effect on the bearings health, and we do not mean to confuse this issue with the environments impact on the complexity of vibration spectrum. We mean that a bearing subjected to vibration will last for less time than a bearing that is not. The relationship can in fact be calculated as below:

L = [C/(P + .00006773 x MVF)]^ 3 x (1 6667/RPM) where: L = bearing life P = bearing load F = frequency (cpm) C = load capacity M = mass of vibrating part V = velocity (in/sec) Using this equation examples of life can be shown that illustrate this relationship for a typical motor ball bearing. Vibration (in/sec pk) 0.0 0.2 0.4 0.6 1.0 1.5 2.0 3.0

Bearing Life (years) 8.60 3.70 1.94 1.15 0.47 0.21 0.12 0.04


2.0 Types of Maintenance Used in Industry The type of maintenance adopted is dependent on several factors,. It has been seen that these factors include :Plant history and culture Criticality of plant or process Operational costs and or cost of down time Cost of maintenance Accessibility of plant or process Economic constraints There are three fundamental types of maintenance procedures. ·

Corrective - the machinery runs until it breaks or wears out. Depending on the state of the machine, repairs are made or the entire machine is replaced.

·

Preventive - specific time intervals for maintenance are prearranged, based on past experience to minimize risk of on-line failure.

·

Predictive - machines are monitored regularly. The use of appropriate information leads to determinations of the machine's state of health. Based on the results, decisions are made for maintenance.

As shown in Figure 8, the types of maintenance practiced on rolling element bearings are, with minor exceptions, corrective or preventive. Predictive maintenance (PDM), although becoming more widely understood is still only seen in specific industries with any great frequency, in industries such as Pulp and Paper and Petrochem. In the nuclear power plants for example according to a recent EPRI study only about 8% of nuclear plants practice some form of PDM on capital or critical equipment with the other forms of power generation implementing far less PDM. When you consider that the majority of fossil plant fall within the twenty to forty year age band, it is not surprising that the cultural effects influence the type of maintenance performed. However, as the age of plants increase the benefits of PDM should be more plainly visible as plant failure rate increases. The predominance of PDM in other industries outside of power generation is no different, history has shown that practices in the most part follow experiences, and the practices considered are a matter of education. TYPES OF MAINTENANCE (% of Machines) Plant Type Maintenance

Nuclear

Coal

Gas/Oil

Overall

Preventive

22

33

43

27

Corrective

56

67

57

58

Predictive

8

Above Combined

14

5 10


Figure 10 Percent of Machines Subject to Different Types of Maintenance

2.1 COST OF MAINTENANCE The actual costs of maintenance are difficult to detail since each industry, even each plant, has its own mode of operation, its own unique costs of labor etc. However it can be seen that costs are a function of downtime, machine complexity, required labor hours, replacement parts, extent of damage etc. Within the utility industry for example, gas/oil-fired plants report the highest maintenance costs, averaging $297 per MW-yr. The average overhaul costs for nuclear plants are $90 per MW-yr, and for coal-fired plants, $82 per MW-yr. The actual spread between the least and most expensive maintenance is shown in the following table. The highest costs for gas/oil-fired plants are due in most part because they are based on complex and or large machines, however it is not so important what theses figures are, only that they have been prepared and therefore some kind of measure can be applied to the savings available through the implementation of specific practices. Maintenance Cost Summary

Plant Type

No.

Nuclear

Coal

Gas/Oil

Capacity (MW)

1

Annual Cost ($)

655

$/MW-Year Cost

108,000

165

2

2,300

93,000

40

1

1,275

227,000

179

2

1,678

52,600

31

1

316

108,000

342

2

850

232,800

274


It is important to realize when reviewing data such as this that the figures are highly dependent on the detail of the plant maintenance logs. For example, the nuclear plants seem to have the most detailed failure histories, but even in these cases, the costs reflected are based only on machinery overhaul, not unplanned maintenance. Furthermore, the cost indicated may be unfairly compared to other industries due to the improved maintenance techniques practiced by many nuclear plants. When considering the pros and cons of each of the maintenance techniques implemented today, PDM has the most promise. One of the primary, less obvious benefits of PDM, is that it is built upon diagnosis which leads to thinking in terms of cause and effect relationships, and this train of thought leads to better paper practices, records and therefore optics for your program. PDM has potential for timely identification of flaws inherent in the machine design, assembly and operation in addition to preventing expensive or even catastrophic failures. PDM surprisingly may lead to the reduction in overall maintenance costs not only in the cost of maintaining the plant, but also as a tool to streamline maintenance activity. Although PDM does require investment in equipment and dedicated personnel, it has been proven that even with limited resources, benefits can still be realized from prioritization of equipment and asset utilization. The following cases cited from actual experience demonstrate the extent and, to a certain degree, the way in which benefits can be realized when PDM techniques are used.

2.2 COMPARISON OF MAINTENANCE TECHNIQUES

Comparison of Maintenance Techniques

Contributing factors

Corrective

Preventive

Predictive

Initial Equipment Investment

Low

Low

Moderate

Prevention of failures

None

None

Good

After maintenance checkout

None

None

Good

Part requirements/Readiness

Poor

Good

Good

Repair Costs

High

Medium

Minimized

Maintenance Period

Time of failure

Preset

Maximized

Downtime

Unscheduled

Scheduled

Fits Scheduled

Differences between the various maintenance practices can be shown through the analysis of any one of many real world maintenance histories. In some cases the differences are more easily recognized such as in the case for Gulf Oil.


Case history 1 : Comparison of maintenance techniques A heat exchanger that is prone to plugging regularly causes the shutdown of an oil production line (Gulf Oil Canada) with maintenance costs of 16 hours labor for a full cleaning. The cost and frequency of repairs when the equipment was subject to corrective maintenance was $480,000 per year (Figure 9a.) Based on the data gathered over a period of one year, it was observed that the plugging pressure was gradually increasing and, within the last two weeks prior to failure, began to increase at a much faster pace. The shortest period prior to the last two-week rise was four weeks; hence, to be on the safe side, preventive maintenance at four-week intervals was initiated. This more than doubled the overhaul from 6 to 13 times per year but prevented severe plugging, saving 14 hours of cleaning time per overhaul (Figure 9b). PDM reduced the number of overhauls from 13 to 6, retaining the 2-hour cleaning time (Figure 9c). A 50-psi pressure rise was used as an indicator of removal for overhaul. This brought the maintenance cost down from the original $480,000 to $60,000 per year. PDM cost saving compared to corrective of $ 420 000 per year

Figure 11 Maintenance Costs using corrective maintenance

Source EPRI report TR - 100160s

Figure 12 Maintenance Costs using preventive maintenance

Source EPRI report TR - 100160s


Figure 13 Maintenance Costs using predictive maintenance

Source EPRI report TR - 100160s Beyond the obvious savings that PDM offers in the direct reduction of machinery maintenance costs due to the early detection of failures, the costs of maintenance can be reduced still further by the extension of maintenance time intervals. These savings while not being readily calculable can be substantial, depending upon the type of plant, its age, state of machinery, maintenance practice etc. To provide an example of what return on investment may be realized, by applying PDM techniques, EPRI commissioned a study of a 465 MW plant and showed that PDM saved $465,000/yr , assuming a fault detection probability of 75% . The estimated cost of PDM equipment at this plant was $250,000, and operational costs estimated at a half a man-year was required for data acquisition and processing with a cost of $75,000.


3.0 Recommended PDM Diagnostic Techniques As industries experiences with maintenance technologies grows, then emerging technologies gain greater and greater recognition. It seems no sooner then a definition of maintenance technologies is made , then a new technology is developed. However the main categories of applied technologies can today be discussed, and the primary indicators are:Vibration analysis, time/frequency domain and demodulation Temperature analysis, measurement and imaging Acoustic emission, low frequency audible and high frequency acoustic emission. Tribology, the analysis of oil contaminants Vibration analysis quite uniquely requires not only knowledge of the data analysis techniques, but also a degree of knowledge of the fundamental parameters required in order to acquire good quality data. Within this course we define not only the techniques, but also the data collection quality issues you have to understand in order to implement the approaches discussed. An initial start point is to understand the methods employed to collect quality vibration data and to understand some of the limitations with sensors and mounting techniques.


3.1.1 Vibration Sensor Types Most of the sensors employed today sense either acceleration, velocity, or displacement. Contact between the sensor and target is almost always required to obtain a reading except in the use of proximity or displacement sensors. The sensors most frequently used are accelerometers, velocity probes, proximeters, with the introduction of Lasers, and Fotonic sensors in recent years. It is important to have a brief understanding of theses sensors, since the incorrect application will adversely effect data quality and your ability to acquire appropriate data for analysis:-

Accelerometers The most common forms of accelerometer consists of a piezoelectric crystal, mounted and spring-loaded in a metallic retainer. The crystals develop voltage when strained under the load produced by the accelerometer's spring/mass system in reaction to externally imposed excitation. The voltage is proportional to acceleration. There are three fundamental different types designed for specific applications giving a variety of frequency responses, sensitivities and environmental compliance. Each type has its own attributes that allows for a go od fit for a specific application.

Connector

Connector

Mass

Retaining Ring

Preload Stud

Mass Piezoelectric Element

Base

Base

Compressive

Shear

Flex

Figure 14 Figure 14 Types of accelerometers


Accelerometers are by far the most widely used sensor for collection of vibration data and are generally good for most applications within a frequency range of 10’s Hz to 10’s kHz . Special applications outside of this usual range are addressed with special designs and can be used down to a fraction of a Hz and up to the high 10’s kHz. Voltage output ranges are also specific to the application and can vary from 10 mV/G up to several volts/G for seismic applications. For general machinery maintenance applications, 100 mV/G is common with a frequency range of 2Hz to 10 kHz.

Velocity Sensors The most common types of velocity sensors illustrated on the left (compared to the Peizo-Velocity type on the right) are constructed of a stationary coil wound around a guided rod. The rod responds to forced vibration inputs transmitted through the housing, creating changes in the magnetic flux rate that are proportional to velocity. Velocity transducers, or Velometers, as they are most commonly known, are applied when low frequency data quality is essential, since for other applications, even though velocity is the most common measurement, accelerometers are used, with velocity acquired by integrating the signal from the accelerometer. Low speed rollers in paper plants or conveying systems is a common application where frequency response is essential to 0.1 Hz.

Figure 15 Displacement Sensors

Displacement Sensors Most displacement sensors are of the non contacting type. These transducers sense changes in impedance or capacitance in the gap separating the sensor's tip from the target surface and convert these changes into voltage, which is proportional to the change of the gap (or displacement). The most common application for displacement is again for low speed machines were low frequency data qualify is essential. Again displacement may be derived from an accelerometer signal with double integration, the process of integration adversely effects the low frequency quality. Displacement sensors are almost exclusively found in the utility and petro-chem industries for the monitoring of turbine bearings.

3.1.2 Accelerometer Sensor Mounting Effects Mounting the accelerometer on a machine requires specific attention if a clear indication of machine health is to be acquired. The effects of common sensor mounting methods are shown below and their effects on acquired data quality. Note that in figure 11a , the stud mounted sensor has a natural frequency of about 30 kHz. The response is fairly flat up to 10 kHz and becomes amplified between 10 and 30 kHz. Steep attenuation of the signal takes place above the 30 kHz natural (resonant) frequency.


The stud-mounted sensor in Figure 11b produces similar response patterns. In Figure 11c, a stud is not used; the sensor is held onto the surface with beeswax. Although the response is still comparable with that of a stud-mounted sensor, the surface finish and flatness of the mounting surface must be very fine (16 rms. maximum and flat within 0.001 in.) to produce the desired effects. Contamination with dirt drastically alters the response characteristics. Cementing an accelerometer may produce reasonable results as long as the cement is hard. Soft glue, as shown in Figure 11d may reduce the flat response to about 5 kHz. Adhesive tape may produce radically different responses depending upon tape thickness and adhesive strength, as is apparent from Figure 11e. In most applications, accelerometers are frequently temporarily affixed onto the surface with magnets. Although this kind of adhesion results in reductions in the frequency response as compared to ideal case it is useful for most PDM activities .


Thin film of silicone grease

Steel stud Max temp 1000°C (1800°F)

Thin film of silicone grease

Mica washer Steel stud Max temp 250°C (482°F)

Thin layer of bees wax

Max temp 40°C (100°F)

Methyl cyanoacrylate

Figure 11. Piezoelectric Accelerometer Mounting Methods and Examples of Typicalcement Frequency (super glue) Responses Methyl cyanoacrylate cement (super glue) soft glue Steel stud Max temp 80°C (178F)


Thin double sided adhesive disk

Thick double sided adhesive disk Double sided adhesive disk



Hand held probe

H a n d P ro b e n o i t a i v e D y t i v i t i s n e S

z H 0 0 1 . f e R ~ ) B d (

D u a l R a il M a g n e t

F la t M a g n e t

M o u n tin g P a d

A d h e s iv e M o u n t

S tu d M o u n t

+ 4 0 + 3 0 + 2 0 + 1 0 0 -1 0 -2 0 1 .0

1 0

1 0 0

L o g

1 0 0 0

F r e q u e n c y

1 0

0 0 0

1 0 0

0 0 0

( H z )

Figure 16 Summary of common mounting techniques


Data quality has shown itself to be only one of the attributes to data collection that is considered at the time of purchase, when attributes such as speed of the process are sold as being more important. The above summary indicates that whilst the most common technique, the use of magnets, is acceptable for most applications, there are a large number of people using hand held probes. It can be seen that hand held probes are only suitable for slow speed machines, i.e. max. 900 rpm and are not suitable for most of the vibration analysis techniques indicated for rolling element bearing analysis covered in this course.

3.2 VIBRATION BASED ROLLING ELEMENT BEARING ANALYSIS The diagnostic techniques employed for vibration based monitoring and detection of rolling element bearing faults in the most part rely on the detection of characteristic frequencies that are generated by bearing failure mechanisms. We have discussed the failure mechanisms in some detail earlier and it can be seen that all modes of failure have one thing in common: the degradation of the bearing surfaces. We have seen images of badly damaged surfaces, yet it is important to realize that the process of failure is of course a gradual one. The bearing surfaces become visually damaged at a fairly early stage in the bearings life. Since the detection of rolling element bearing faults using vibration analysis, like any other technique, is based on identifying failure related characteristics in data, it is important to understand the mechanisms that creates theses characteristics. The mechanism is simple, and can be compared to the effect of driving along a road with a damaged surface. Imagine you are riding on a roller or ball inside the bearing. Then each time you hit a flaw in the race, you feel a jolt. This jolt and the regularity of it is at the fundamental basis of vibration based analysis of rolling element bearings. When we use vibration analysis we commonly refer to time and frequency domain. Regardless of the technique used, monitoring of failure characteristics relies on the systematic checking of the presence of flaws indicative characteristics. When indications of bearing deterioration appear, the affected signals are tracked until trends develop that are indicative of a substantial change in the health of the monitored rolling element bearings. As a rolling element bearing begins to fail, the discrete bearing characteristics grow in amplitude.

3.2.1 Bearing Failure Characteristics The fact that rolling element bearings emit noise when in distress has been known perhaps since the invention of the bearing.. Before the onset of the era of high technology, skilled craftsmen could determine the state of a bearing merely by "listening" through a screwdriver pressed against the machine housing; a method later augmented by the use of stethoscopes. Both methods are still, and should still, be used when unusual noises emanate from a machine, particularly if you are caught short without your high tech. As time progressed, vibration meters became available, but the major emphasis was placed on monitoring noise levels that were supposed to be indicative of the bearing condition. In the last two decades, advances in the state of art of microprocessor and computer technology have brought about substantial refinement in the art of bearing failure diagnoses, trending, and record keeping. It is important to understand that as a bearing with a damaged surface rotates, the regularity, or frequency with which the roller or ball impacts on the defect indicate potential failure and allow us to determine the type of damage that exists. A number of characteristic frequencies are generated by a damaged bearing , and are known as:Ball train (BT) - same as cage frequency Ball pass with respect to the stationary outer race (BP/0) Ball pass with respect to a rotating inner race (BP/1) Ball rolling about its own axes with rotating inner race (BR)


Shaft frequency of rotation (FR). The following equations are used to calculate these frequencies BT = FR/2 [1 - (BD/PD) cos θ]

approximated by

FR(O.5 - 1.2/n)

BP/0 = nFR/2 [1 - (BD/PD) cos θ]

approximated by

FR(O.5n - 1.2)

BP/I = nFR/2 [1 + (BD/PD) cos θ]

approximated by

FR(O.5n + 1.2)

BR = (FR PD)/2BD [1 - (BD/PD)2 COS2 θ]

approximated by

FR(O.2n - 1.2/n)

where

BD

= ball (roller) diameter

PD

= pitch diameter

θ

= contact angle

n

= number of balls (rollers).

When bearing geometry is not known but the number of balls or rollers can be counted, it is suggested that the approximate equations be used to establish the bearing frequencies of interest. All the equations listed above show a direct dependence of the calculated frequency on the frequency of rotation.. The following figure illustrates the bearing geometry used in the above equations

Figure 17 Ball Bearing Terminology


·

In most cases this method for obtaining bearing defect frequencies is rarely used, instead the bearing reference numbers are looked up in a bearing database such as shown in the following example. You will see that some of the terminology is different, however the basic bearing geometry information and characteristic frequencies are displayed. The example illustrated is for an SKF 32230.

Such bearing databases contain most common bearings, in the case of the illustrated SKF Atlas program some 16000 bearings are included

3.2.2 Recommended vibration based analysis techniques As described in the previous section, both time and frequency domain techniques are utilized for rolling element bearing analysis. Simply, the use of time domain for data analysis focuses on the raw data as seen if you were to connect an oscilloscope directly to the sensor. The analyst must focus on the timing of the events seen in the data in order to ascertain characteristics that indicate failure. Frequency domain on the other hand provides for the display of the frequency characteristics that make up the time domain signal as previously discussed, using a mathematical technique called Fast Fourier Transformation or FFT. Both techniques are characteristics related indicative features in bearing more so than interpreted.

complex methods to visualize the machines operation and both contain all to that machines health. Both of theses techniques can be used to identify failure data, however a more useful technique has been developed that focuses on the the whole machine or at least reduces the data to a point that can more easily be

We have discussed the use of FFT’s to reduce or simplify the time domain data to display the rates of change involved in the data set, or frequencies. This is performed since industry has discovered that frequency related information can help identify the root cause of a machines problems, for example, the meshing of a pair of gears generates what is called a gearmesh frequency and is simply the rotational speed of that gear times the number of teeth on that gear. It can therefore be seen that if the rate of change of these characteristic frequencies themselves indicated a mechanism of failure, then simplifying This document is protected under copyright, it may not be reproduced in whole or in part without written consent of Ron Frend Page 29

the data further prior to display would be useful. This is in fact the case, and this technique is called Demodulation.

The demodulation of acceleration amplitude vibration data is now gaining greater acceptance as a valid predictive maintenance tool for the vibration analyst. Predictive maintenance, however, is not limited to vibration analysis. Any parameter which gives an amplitude that varies with time can also be passed through a FFT to identify the frequencies of the amplitude variations. Such parameters may be: sound, ultra-sound or even electric current. If the time domain signal has one or more frequency components which amplitude varies depending on the interaction of another component, then we may say that the one component is amplitude modulating the other.

Demodulation theory Before we look at any case histories using demodulation we should be clear exactly what is modulation. A signal may be said to be amplitude modulated if the amplitude of that signal is changing over a period of time because of the influence of another signal. The example below was taken from a large steam turbine running at 3600 rpm. The run speed signal is being modulated by a signal at 4 Hz which is probably a foundation resonance. This type of modulation is commonly found in maintenance applications but consider the example below.

Figure 18 Time and frequency domain data showing modulation


Figure 19 Waterfall of frequency domain data showing mo dulation

In figure 19 we see a vibration at 2 kHz which has been modulated slightly more than three times within the time period (50 ms which equates to 1 revolution of the inner race for this example). The 2 kHz vibration is the resonance of the bearing which is being excited by the bearing outer race frequency (3.07 x run speed). The excitation of the 2 kHz frequency by the bearing defect on the outer race causes the 2 kHz amplitude to be changed as seen in Figure 15. In other words the bearing outer race frequency is modulating the bearing resonance frequency. The demodulation process is based on the extraction of the modulating frequency to produce a time waveform which can be handled by the F.F.T. process and displayed in a simplified form. When we demodulate the above reading we are not interested in the 2 kHz frequency but we are interested in the outer race defect frequency which is: (1000/50*3.07) Hz = 61.4 Hz. As can be seen from Figure 15, the modulation is at this frequency. In vibration terms, demodulation is a way of extracting the rate of occurrence of high frequency resonance’s.

The demodulation process The time waveform of a machine with a bearing in the early stages of deterioration will look like the top plot in figure 16. The bearing excitation resonance is shown as small, high frequency pulses sitting on top of the high amplitude, low frequency vibration.


Figure 20 Time domain data showing start of demodulation process

The first stage in demodulation is to pass the signal through a high pass filter. To give the waveform shown in the lower plot of figure 16.

Figure 21 Time domain data showing rectification and smoothing


With the time domain signal in this format the F.F.T. conversion would give a single spike in the frequency domain at the resonant frequency which we have earlier said is not what we want. To modify the signal so as to be suitable for F.F.T. we must “envelope” each parcel of energy by first rectifying and then passing the signal through a smoothing R-C (resistance-capacitive) circuit. The signal is NOW passed through the F.F.T. and we get a spike in the frequency domain at the bearing defect frequency.

Figure 22 Illustration of time and frequency domain demodulated data.

Resonance sources Since the purpose of demodulation is to extract health related variations in machine resonant frequencies, it is clear that when taking a demodulated reading we must first decide on were to measure, how the measure, which filter setting etc. Conventional thinking will tell you that the resonance frequency which we are using as the carrier wave is always the resonant frequency of the bearing; while this is often the case it is not always so. For vibration readings, the accelerometer which we will use to detect the signal will probably be sitting on top of a magnet which will give a structural resonance in the 1.5 to 4 kHz range (typically). The bearing housing will have its own resonance, the machine structure will have its own resonance. In short, the carrier wave signal resonance could be coming from any part of the mechanical structure. Vibration case study 1: AC Motor bearing defect using demodulation Figure 19 shows the signal from the inboard bearing of a 35 H.P. A.C. motor operating a belt driven fan. The 2 upper plots are the time domain signal in two planes over a period of 640 mS. The lower plots show the time domain (left) and frequency domain (right) over a 50 ms period of the lower 640 ms plot. Note that the frequency spectrum shows spikes at 2 kHz and 3 kHz while the time domain plots show an “angel fish” pattern which is classic of a bearing defect.


Figure 19 Illustration of data from motor bearing Note also that the lower left portion of the plot is a zoom of the windowed part of the long time record. This shows a detail of the one “angel fish” and the amplitude can be seen to be passing from positive to negative and back again many times during the life of the angel fish - i.e. a high frequency oscillation. This leads us to the conclusion that this is the frequency of 2 and/or 3 kHz seen in the spectrum and one or both of these frequencies are the result of impacts and subsequent ring down and they are occurring at the resonant frequency of part of the mechanical structure.


Figure 20 Illustration of waterfall frequency domain data showing modulation Figure 20 shows a time/frequency cascade of the same time interval cropped below 0.001G. This clearly shows the modulation of the 2 kHz frequency while the 3 kHz frequency is static. The modulation has been calculated to be equal to the bearing outer race defect frequency of the motor inboard bearing. Every time one of the bearing balls passes a defect on the outer race, the ball impacts on the defect causing the 2 kHz vibration to suddenly rise and then ring down. The 2 kHz is the resonant frequency and the bearing defect frequency (outer race) is the modulating frequency. Figure 21 shows the demodulated spectrum on the left with waterfall plot on the right above the trend of the defect frequency.

Note that the demodulated spectrum is clean and extremely easy to analyze. The spikes occur at the bearing defect frequency (outer race) with multiple harmonics but there is no sign of the resonant frequency because this high frequency has been removed during its use in the demodulation process. The frequency range of the spectrum is such that the frequency of the impacts is clearly visible but we do not need to see the resonant frequency. The last spectrum in the waterfall is lower than the previous spectrum due to greasing of the motor bearings which lowered the amplitude at which the impacts caused the bearing to vibrate at resonance.


Figure 21 Waterfall Trend and Spectral plot showing bearing defect

Figure 22 Illustration of modulated and demodulated velocity spectrum Figure 22 shows a similar defect on another machine but here the velocity spectrum (left) is displayed beside the demodulated spectrum. Note that the demodulated spectrum is much cleaner and easier to analyze.

Vibration case study 2 : DC Motor bearing defect using demodulation

The time waveform in Figure 23 was taken from the drive end bearing of a 100 HP DC motor running at 1009 rpm. The reading was taken horizontally in the line of force of the drive belts. The motor drives a 600 ton press via vee-belts through a flywheel. Time waveform data for the evaluation of bearing defects should always be collected as acceleration so a special time waveform was defined in the velocity reading for this point. Note that the overall swing is almost 20 G’s peak to peak which is excessive for a bearing of this type in this operation and the fault and alert levels have been exceeded. The time between zero seconds and the first vertical bar is the time for one rev of the shaft (59 MS). Examination of the waveform does not give any form of angel fish pattern but we can see distinct high amplitude spikes at fairly regular intervals with a broad very high frequency band, oscillating around zero G’s at a fairly low amplitude.


Figure 23 Illustration of time domain data from DC motor bearing showing faults The frequency spectrum in Figure 24 is demodulated acceleration using the default high pass filter for the CSI 2120 data collector of 600 Hz. Immediately we see the high amplitude spikes at 5250 cpm with multiple harmonics. The spikes exceed the fault limit significantly and coincide exactly with the generated fault frequencies for the outer race defect frequency of the installed F.A.G. NU220 bearing. This now tells us that we have a bearing defect on the outer race with significant impacting, but we do still do not know if the bearing is spalled. Do determine the extent of the damage of the bearing we now look to the velocity frequency spectrum.


Figure 24 Illustration of demodulated spectrum showing bearing defect frequency The velocity spectrum below is unusual in that the bearing defect frequency is quite distinct although the amplitude is not what would normally be considered high for a non bearing defect. For bearing damage evaluation, however, a spike of this amplitude is considered severe and indicates that the bearing is spalled. The bearing was changed with no production loss and no secondary damage to the motor.

Figure 25 Illustration of velocity spectrum indicating bearing defect frequency


Vibration case study 3 : Press bearing damage using demodulation In the automotive industry, stamping presses with a capacity of 5,000 tons or more are in general use. Presses such as this operate in the region of about 6 cycles per minute and require massive flywheels to store the rotational energy for the short period of the hit - 50 tons flywheel weight not being excessive. These flywheels are typically about 10 or 15 feet diameter and rotating at about 250 rpm and are supported by two taper roller bearings (usually Timken) on a quill shaft which has the main drive shaft mounted inside. The drive is via a clutch from the flywheel to the main drive shaft. The body of the press is also quite massive, being about 250 tons for a medium size press so the capacity for dampening the vibration is significant. The case history presented here is typical of a flywheel bearing failure.

Figure 26 Time domain trace for press flywheel showing bearing damage The illustration above shows the demodulated time waveform data from a damaged flywheel bearing on a Danly 600 ton press. Note that the duration of the waveform is 1000 MS which allows a view of about four revs of the flywheel. The waveform is recorded after the demodulation circuit so appears to be DC negative. There are large repeatable vibrations every rev of the flywheel with higher frequency perturbations in between. The demodulated spectrum shows the effect of the FFT on the time waveform.


Figure 27 Demodulated acceleration spectrum showing bearing defect on press flywheel The high frequency effect shows itself as being the ball (roller) spin frequency x 2. This indicates that the rollers are damaged and impacting on both races.

Figure 28 velocity spectrum showing bearing defect on press flywheel The velocity spectrum shows that the rollers are not only marked but have, relatively, deep pitting.


The value of using demodulated data as well as velocity data was illustrated at one facility which removed a flywheel because of BPOR frequencies in velocity only to find that the bearing had a large clearance. The clearance caused the overhanging flywheel bearing outer race and cone to be tilted on the inner race with the result that every time a roller came into contact with the outer race it “jogged” the flywheel causing a velocity vibration at the outer race frequency. Apart from the excess clearance there was nothing wrong with the bearing. Using wave audio with demodulation for bearing analysis

Most discussion about the use of vibration for bearing analysis immediately brings images of data collectors to mind. It is worth remembering that in the days before data collectors, tape recording of data was common. Although tape recording technology has changed a great deal in recent years, with the advent of the DAT recorders for example most people tend to have moved away from this technology. There is one fundamental difference between a recording and a data file, that is that you can play back a recording, yes you can play back a digitized file, however you are limited by the data acquisition parameters used at the time of its collection, frequency range for example. Recorded data has the benefit of being reanalyzed time and time again with different techniques applied, for example, the use of amplitude demodulation, yet of course it has its disadvantage, particularly the use of fragile media, tape. Not to suggest that lap tops will replace data collectors, but modern laptop PCs have the capability to record and playback time waveform data at very high sampling rates - so high in fact that for all intents and purposes, the resultant waveform may be considered analog, with out the need for tape. Sampling rates of up to 44 kHz (CD quality) with 16 bit resolution give outstanding clarity and a signal to noise ratio of 96 dB. The data shown below was collected from a 125 H.P. A.C. induction motor by connecting a “Tee” B.N.C. to the CSI 2120 accelerometer signal and recording directly to a sound card (Sound Blaster compatible 16 bit) on a lap top P.C. with an sound card input impedance of over 80 kohm so signal quality may be considered to be unaffected.

Figure 29 demodulated acceleration spectrum obtained from recorded wave audio


Figure 30 velocity spectrum obtained from wave audio file

Figure 31 Time waveform display of wave audio file

A comparison of data quality between the above case histories illustrates that the use of wave audio provides for the same diagnostic capabilities as does a data collector. There is of course one significant draw back, to collect wave audio in this way you need to use your laptop PC in the field. However, for


applications that require a PC in the field, this may be an alternative method to obtain the same information, yet with one significant benefit. You can record for extended periods of time, and reanalyze your data in different ways, plus by using wave audio, as with more conventional recording, we can ensure that the same period of time is used for all spectra and time waveform analysis, thereby removing the chance of transient effects on only some of the data. The Application of Demodulation in a PDM approach to bearing analysis

Whenever you are trying to detect rolling element bearing deterioration, demodulated acceleration should be used - in conjunction with velocity. It is not usually necessary to take demodulated acceleration readings in more than one plane at each bearing as the resonant vibration is omni-directional. It will not be necessary to go to high frequencies in velocity as the bearing resonant frequencies will now be detected by the demodulator. Set the Fmax in velocity and demodulated acceleration to slightly more than 4 x BPIR to ensure enough defect harmonics are captured for analysis and to give good definition of run speed (and harmonics) and sub-synchronous frequencies. Place the demodulation readings on your route and collect them with your normal route readings. When analyzing exceptions you will probably notice a correlation between the 1-20 kHz H.F.D. rise with a rise in demodulation spectrum amplitudes as they are both looking at the same frequencies. Use your generated fault frequencies just as you would in velocity, and you will find the analysis much easier. Do NOT use demodulated acceleration in cases where frequency analysis is inappropriate (for example, a machine which ramps the speed up and down during data collection) unless you also use a speed trigger. 3.2.3 Evaluation of Damage Severity Once a bearing or other defect is suspected, the single most important question is whether to pull that piece of equipment off line. The consequences of an incorrect decision are amongst other things, being the loss of production. For this decision to be made correctly, an accurate assessment of the severity of the fault must be made. Experience has shown that the only true way of making a severity assessment is to compare the current health of a machine with its own history. However, machine history is not always available, so what do we do?. The vibration institute, along with other organizations have developed general purpose severity charts for this reason. However, they are not perfect. The biggest single disadvantage of such charts is the general categories of machinery they are applied to. It is obvious that all machines are not alike, what is acceptable on a small electric motor is not what is OK for a steam turbine. However saying this, the Rathbone chart is the most successfully used data chart in the absence of more specific information.


The Rathbone chart above is the most commonly used in industry


Other charts exist that are less frequently used, applied to a different set of machine types


In the case illustrated above this chart is tailored for the comparison and severity assessment using acceleration amplitudes. Regardless of the source of the chart, nothing beats your own experiences, we have discussed the use of a machines own history, and in its absence the use of industry severity charts, well, there is middle ground. GM have developed there own experiences into severity charts that are tailored to the types of machinery they are use to , the following examples for electric motors from GM standard specification V1.0 - 1993 illustrate this data.



Standard motor Line amplitude acceptance limits Special motor Line amplitude acceptance limits

.075

.04

.04

.04 .03

.04

.03

.03

.04 .03

.03

Fmax = 120 kcpm

Fmax = 120 kcpm

Running speed orders Standard motor - ulility operations

.03

Running speed orders Special motor - semi-finish operations

Precision motor Line amplitude acceptance limits

Band limited overall amplitude acceptance limits

.02 .02

Band 1 0.5g Standard motor & special motor .01

.01 .005

0.25g Precision motor

.005

Fmax = 120 kcpm

(g’s)

Running speed orders Precision motor - finish operations

Frequency

Fmax = 120 kcpm

Although more specific to a category of machine, this data is still very limiting with no machine history. Other methods of severity assessment are available based on the measurement of PDM related data, however, using a knowledge of the failure cycle of a bearing as a guide. We have discussed the increase of bearing temperature close to the failure of a bearing (discounting overpacking etc.) well, it could then be said that if the bearing is not hot, then it is not close to failure. Let us take a look at the failure cycle in more detail and use the following illustration as a guide.


Figure 32a Bearing Failure Sequence diagram general techniques

Stage 1 Over loading, over heating, under lubrication causes Elasto-Hydrodynamic problems. Stage 2 The bearing races become marked and vibration spectrum analysis works well, along with spike energy and high frequency techniques. Stage 3 The races become spalled high frequency works well, and spectral data will show defect frequencies Stage 4 Close to bearing failure, the temperature will rise, giving good indication by the use of Thermography.

It can be seen then that the presence of a characteristic indicator such as the presence of spikes at the defect frequencies in a velocity spectrum, or the temperature starts to rise, then these can be used to time remedial maintenance.

Severity assessment using demodulation and velocity spectrum


Figure 32b Bearing Failure Sequence diagram for the use with demodulated vibration data The most significant aspect of bearing analysis is when do you decide to replace a bearing. The answer is not entirely straight forward as the amplitude will change depending on load and speed, however, a general rule of thumb is shown in Figure 32. It is important to note that this rule of thumb includes demodulated acceleration readings and velocity readings. Demodulated acceleration will tell when a bearing is marked or suffering from lube problems but is not a good indicator of bearing damage in its own right so you should use demodulation in conjunction with velocity whenever possible. When a rolling element bearing is in good condition, is adequately lubricated, is operating within its design load parameters and is not over heated then there will be no significant spikes in the demodulated acceleration spectrum. if the amplitude is displayed in terms of G dB (re 0.001G) then there will be a 5dB scatter in amplitude. Note that a 3dB rise in amplitude means that the linear rise is approximately double, and a l0dB rise equates to a ten fold increase in amplitude. If the bearing starts to suffer from under lubrication then the demodulated amplitude will rise by 10 to 15dB without any increase in velocity at the bearing defect frequencies or defect frequency harmonics or sidebands (BDF/H/S)


The poor lubrication will eventually cause microscopic fatigue failure of the races at particular spots. This will result in minor impacting of the roller onto the corner of the fatigued spot which will give demodulated acceleration spikes of 15 to 20dB at the bearing defect frequency (possibly with harmonics). At this stage there will still be no BDF/H/S in velocity. As the fatigued area develops into spalling then the spike will rise to a level of 20dB above the carpet level and will develop sidebands. The damage will spread to other parts of the race and rollers causing random frequencies which will cause the carpet level to rise. At the same time the BDF/H/S in velocity will rise with a typical amplitude of about 0.01 i.p.s. As the damage becomes more severe and approaches failure then the demodulated acceleration sidebands multiply and grow in width (often as the bearing speed becomes erratic with loss of bearing clearance). This will have the effect of merging the defect spikes into the sidebands so as to cause "shoulders" around the defect frequencies. The spikes in velocity, (BDF/H/S) now become pronounced, particularly at third or fourth harmonics (this may vary depending on application) with an amplitude of less than 0.05 i.p.s. If the bearing reaches this stage then plans should be made to change the bearing without delay.

3.3 DEMODULATION OF ULTRA SOUND DATA FOR BEARING ANALYSIS . It has been known for years that you can hear a bad bearing. However you may not realize it until you listen to a good and bad bearing with the intent on recognizing the differences. Most audible analysis is performed using microphones and standard analysis hardware such as a spectrum analyzer or data collector. The analyst in this case is looking for exactly the same information as when using vibration data, and is challenged by exactly the same problems. That is, that a none processed spectrum or time waveform contains all machine operational characteristics and not just the one you are looking for. The uniquely poor attribute of audio analysis is the nature of the measurement. Although audible effects are what seems to make most sense with us humans, we are, if we focus on audible data, limiting ourselves significantly. When you consider for a moment the transmission path of the effect to your instrument it becomes quite obvious. Machine related faults are, as we know by now, related to mechanical events in most part, such as the spalling of bearing races or the impact of gears etc. We also know that metal impacts create vibration and that this vibration is of the material undergoing the impact. Therefore the direct measurement of the surface vibration is effected only by the transmission path through the metal to your point of measurement and the quality of the sensor and mounting method you have adopted. Unfortunately this is not the case with noise, OK, the source is the same, yet the noise you and the microphone hears is not the source itself, but the effect the impact or other mechanism has on the surface of the metal that then in turn excites the air and then to the mic. With this kind of analogy it can be seen why acoustic analysis is not widely adopted as a primary form of analysis, however it is adopted as a supporting technique or even primary if the source is difficult to measure with a none contact device. Ultra sound analysis on the other hand is becoming more popular as an analysis tool, however most U/S units do not h ave any frequency analysis capability. The UE2100 ultra sound gun, which was used for the measurements in this example, “1heterodynes” the ultra sonic frequencies to the audible range so that we hear the impacts of a bearing defect as they are converted from 33 kHz to between 5 and 8 kHz. If we could demodulate this data using the audible range as a carrier frequency we should be able to see the bearing defect frequencies in the spectrum. To enable the data collector to collect the signal from the UE2100, a 1/8” audio connector to BNC cable was used to take the headphones output from the UE2100 to the voltage input of the collector. The spectrum was set up to demodulate the data with a 1 kHz high pass filter. Figure 33 shows the result of a

1

Heterodyne - Conversion of data at one frequency to a different frequency while maintaining its’ relationship to other frequencies This document is protected under copyright, it may not be reproduced in whole or in part without written consent of Ron Frend Page 51

demodulated spectrum of ultra sound taken on the top bearing of a vertical centrifugal pump with a known defect.

Figure 33 Demodulated, heterodyned, ultra sound measurement showing bearing fault The large spike is the BPOR frequency of the bearing. Figure 34 shows this spectrum and a the same spectrum from a similar pump with a good bearing.

Figure 34 Good V’s bad bearing shown using demodulated ultra sound With the ultra sound gun used in this way it becomes nothing more than a different type of vibration transducer, but dedicated to very high frequencies. We have seen many times that the combination of


techniques may improve the benefit of their use. Using Ultra sound as an input to a data collector that can demodulate, clearly provides good diagnostic data for the analysis of rolling element defects. 3.4 OTHER TIME DOMAIN TECHNIQUES FOR BEARING ANALYSIS We have focused on the use of frequency and time domain techniques used in conjunction in order to identify bearing failure, however there are other techniques adopted that focus entirely on the time domain. It is obvious that the time domain methods of identifying defect frequencies may be used quite separately to the use of frequency domain or demodulated analysis and may have its own application. Many organizations cannot afford complex instrumentation yet may use oscilloscopes or PC based data collection technologies. In these cases a purely time domain approach make good sense and keeps the program cost way down. 3.4.1 Use of Kurtosis to identify bearing damage In addition to the time waveform techniques discussed, other methods focused on the time domain are used. One such mathematical approach is refereed to as distribution analysis or the recognition of Kurtosis to indicate bearing defects. This method is based on the plotting of the probability density of a time signal, a good bearing normally displays a normal Gaussian distribution. The appearance of faults distorts the curve shape. This is shown by curve A in Figure 35. Symmetrical deviation about a vertical line drawn through the center of the normal distribution curve is known as Kurtosis. Curve B of Figure 35 shows a typical Kurtosis effect. The base usually broadens while the center of the curve peaks out beyond the rounded off peak of the normal distribution. Some investigators proposed to use the change in Kurtosis as an indication of impending bearing failure since it can be mathematically calculated and expressed easily.

Figure 35. Probability Density in Decibels of Normalized Acceleration of a good and bad bearing 3.4.2 Use of shock pulse monitoring to identify bearing damage In recent years as technology has become more advanced and less expensive, we have seen it applied to the diagnosis of machinery faults. However older techniques that continue to show results are still in use in industry. Shock Pulse or Spike Energy instrumentation is still common place and provides an inexpensive method of PDM. We have discussed the use of time domain analysis and even mathematical approaches, however there is another family of instruments built around the use of the resonance of the sensor, to amplify the signals


emanating from a faulty bearing. This approach similar in nature to demodulation is arguably more sensitive than direct signal processing and can identify flawed bearings in both early and advanced stages of flaw development. The major disadvantage of this technique lies in its inability to identify the source of spike or peak activity. This can lead to erroneous decisions regarding the state of the bearing's health in the presence of other sources effecting the signal. 3.5 OTHER FREQUENCY DOMAIN TECHNIQUES FOR BEARING ANALYSIS 3.5.1 Use of Broad band Frequency domain bearing analysis Prior to the advent of spectrum or FFT technologies, industry applied hardware Band pass filters, Octave, Third Octave technologies successfully to the analysis of machinery health.. The simplest analogy that can be drawn to days technologies that indicates the limitation of these techniques is to consider the spectrum you are now used to seeing displayed in chunks of significant width, creating a spectrum that is stepped, with the amplitude of each step being the average, rms, or peak amplitude in the frequency range of that step. Clearly a form of PDM is available base on theses techniques, and the hardware is far less expensive, however the information available for diagnosis is limited as in the case of shock pulse monitoring. Although limited in diagnosis, these techniques still find a place in today’s industry, and are particularly well suited to quality control applications, were the focus is accept or reject, and not what is wrong. In some respects the use of One Third Octave analysis for quality control of a manufactured machine is superior to time or frequency based analysis in the ease of application and simplicity of information, for these reasons many successful QC applications are in place. 3.5.2 High-Frequency Noise As a bearing deteriorates, we have discussed the generation and growth of pits into spalls. We have also discussed the regularity of ball or roller impacts with those pits generating characteristic frequencies known as bearing defect frequencies. Weather you are looking at demodulated or regular spectral data those indicators are at relatively low frequencies, generally in the tens or low hundreds of Hz. The same effect that creates the defect frequencies also creates high frequency noise and that noise or broad band carpet level is most noticeable above 5 kHz in the early stage of damage and progressively lower and lower in frequency as the damage gets worst, until eventually the entire spectrum is dominated by broad frequency noise and all discrete frequencies are lost. This phenomena, has been used to indicate bearing damage as shown in fig 36 by trending the overall high frequency noise level, either by using inexpensive band pass filters attached to sound level meters or more expensively by setting an appropriate band on a data collector or spectrum analyzer. The disadvantage of using this technique solely is that it only indicates failure close to failure and is very significantly effected by lubrication.


Figure 36 Trend of high frequency noise for a damaged bearing The above figure illustrates the sudden rise in overall level within a 7 day period, which if your measurement interval were longer would be ineffective. As with most techniques that do not lend themselves to recommended first line PDM application this is a good supportive test. There is a useful test that can be performed at the time of proposed removal of a grease lubricated bearing that has shown not be missed, that is monitor the high frequency noise level, add grease to the bearing and see what happens to the noise level. If the noise level was high and returns to a high value quickly after adding the grease then the bearing is definitely damaged, on the other hand is the noise level reduces and remains reduced, then the cause of the high frequency noise was insufficient lubrication. 3.6 THE USE OF OIL ANALYSIS FOR BEARING DIAGNOSIS In the case of oil-lubricated machines such as most gearboxes, metallic wear of any element, be it gear wear or bearing wear, will be carried in the oil, where it will either settle or be caught in a filter as it recirculates. A lot of today’s lubrication systems are equipped with magnets that catch the magnetic wear products. The magnets can be provided either with an external indicator light, which will give warning when particles are caught, or the magnet can be periodically pulled and inspected for wear debris. The debris can be analyzed later using ferrography, a technique that separates and provides a count of magnetic particles. A microprobe analysis employing an electron beam microscope is sometimes performed to identify the atomic content of the sample from which the source of the debris can be determined. An example of a spectrum obtained with microprobe analysis is shown in Figure 37.


Figure 37. Microprobe Spectrum of Bearing Metal Particulate More common than the analysis of particles directly from a magnet or other catching device is the deliberate sampling of the oil and the analysis of suspended particles. Prior to vibration or thermographic analysis, oil analysis was the main stay of PDM. The technique relied on not only the trending of particulate count, but through the use of microprobe analysis to provide diagnostic information, such as brass particles indicate bushing wear, chromium for bearing wear etc. A lot of industries perform PDM using oil analysis as the primary tool and will continue due to the indirect benefits as well as the direct ones. Oil sampling is performed on a regular basis just like route based data collection in vibration programs. The indirect benefits come from the fact that a person is on the spot able to pass judgment with minimal training on oil quality and certainly to the level of oil or even absence of oil. Where water contamination is common this test on the spot has saved more drive trains then vibration or thermography combined, since it can detect visual anomalies in the oil quality prior to any bearing or other damage having taken place.

Figure 38. Trend of oil sample wear elements


As with other PDM techniques a trend of particulate count or metallic content can provide advanced warning of failure. Rates of change may indicate an appropriate time to replacement or overhaul. 3.7 THE USE OF TEMPERATURE MEASUREMENTS FOR BEARING DIAGNOSIS Practical experience has taught us that temperature is a good indicator of friction, for example sliding down the banister of your staircase as a child is probably the best reminder. Temperature in the work place for the analysis of bearing health is also commonly used, the most common example is if a bearing sounds “funny”, put your hand on it and if it is hot, then there probably is something wrong. Like the screwdriver to the ear for vibration, this form of predictive maintenance is widely adopted and successful. We know that as the rolling elements and bearing races deteriorate, friction increases and more heat is generated. In grease-packed bearings, we know that overpacking causes temperature rise. These indicators are obvious and may support a vibration based approach very simply. As the screwdriver has migrated to a data collector for collecting vibration spectrum, then the hand has also been seen to migrate to the Thermographic camera. Feelings and subjective assessments are now removed and objective evaluation is straightforward and, what’s more, within the domain of a less experienced individual. We are all used it the expression “A picture speaks a thousand words”, well look at the two pictures that follow, which bearing do you think is the hot one?

Figure 39 Example of hot V’s cold motor bearings Lets get back to basics and understand temperature measurement a little further. We have used the expression Thermographic cameras as an analogy to the migration of screwdriver to data collector. We now need to discuss the technology and its application to PDM. The use of temperature as a diagnostic technique relies on the use of what is called Infrared Thermography. Infrared = The next "section" above visible light in electromagnetic spectrum. Thermography = Thermo - "Heat" + graphy - "The study of' temperature distribution via infrared image. We know that some objects do not have a lot of heat, e.g. ice. Some objects have a lot of heat, such as a hot bearing. What is not commonly known is that as an object gets hotter it emits more radiation in the infrared spectrum. With a device that could measure infrared radiation you could measure the amount of heat in an object. We call this device an infrared camera and we call the science of its use “Infrared Thermography”. Infrared Thermography has been seen to become a larger and larger part of PDM programs and perhaps the simplest to implement.


3.7.1 Factors that effect the collection of quality thermographic data In order to collect quality thermographic data it is essential that some of the thermodynamic principles are understood and their effect on data quality. Some of the terminology used by thermographers is as follows:: Emissivity - the rate at which bodies emit radiation relative to a perfect emitter of infrared energy. Reflectance - the amount of energy a body will reflect Transmittance - the amount of energy a body transmits. Ahsorbance - the amount of energy a body absorbs Temperature range - the range to which the camera is set Visual resolution - the visual optics selected for use

Kirchoff Laws provide for E (Emissivity) = A (absorbency)

and that

E+ R + T = 1

were E = Emissivity R = Reflectance T = Transmittance Although good data collection does not require for a school book approach to be taken to the principles, it does require for an understanding of them. For example, we assume that the thermographer has seen to it that there is no energy being reflected onto the object being scanned. That there is no energy being transmitted through the object touching another etc. If these assumption hold true, then, 100% of the energy being emitted from a body is due to the amount of energy, held or absorbed by the body. Emissivity or an objects ability to reflect and emit energy is probably the most important aspect to understand if accurate measurement of a bodies temperature is to be taken. Some surfaces are naturally reflective and therefore have a relatively low emissivity approaching 0.0 such as. polished aluminum. On the other hand some materials are more absorptive in nature and therefore have an emissivity approaching 1.0. It is important to understand and adjust for the relative emissivities of the different materials that you are looking at. If emissivity is not adjusted for then the temperature reported by the camera will be inaccurate.


Fortunately industry provides “look up” tables for the most common materials, and others can be estimated accurately enough, examples of the most common are as follows. Heavily weathered aluminum Aluminum foil Roughened aluminum disk Common brick Polished copper Rusted iron Electricians’ black tape

.68 .09 .28 .81 - .92 .03 .91 - .96 .93

In predictive maintenance applications you do not need to be concerned with the actual, accurate temperature. Industrial thermographers are usually only concerned with the delta temperature or how much hotter one object is with respect to another. This is not to say that emissivity is not an issue. The wrong emissivity can lead to incorrect conclusions as to a machines condition, since the important thing is comparison, a knowledge of what your emissivity was set to last time, or for the other object of the same type, is usually all you need to know (90% of the time, leaving your Emissivity at about 0.90 will generally be accurate enough). The most common cause of problems getting overlooked is due to improper selection of temperature ranges and sensitivities, it is important we understand why their selection will have a great impact on our outcome. Chosen temperature ranges and sensitivities employed in an infrared detector act in similar fashion to physical blinders we would place on one's eyes. Blinders placed too high or too low will obviously restrict the information we can gather at the upper and lower temperature ranges respectively. Blinders set too wide will gather more information than is readily comprehensible. Blinders set too narrow provide a "tunnel vision" that limits what we will see. Improper temperature ranges and sensitivities will prevent us from "Seeing the Forest for the Trees". Visual resolution effects how clearly we look at our puzzle, especially when looking at small or physically confined components. Often when diagnosing a problem we have to distinguish a problem source with two or three possible sources located within a small area.. The clearer we see the individual component the better we can distinguish the source of the problem. The importance of high resolution will increase, as the size of the physical components we inspect continue to decrease. The optics we choose are an important component of our visual resolution. They not only serve to make our target large enough in our image for our eyes to distinguish, but the minimum focusing distances are critical to afford us the opportunity to use the proper perspective. 3.7.2 Evaluation of Severity using temperature measurement There are industry standards that can be applied for the assessment of severity of temperature change, however they are conservative standards and must be used with a significant experience factor. One good rule of thumb is that for every 10°C rise over the maximum allowable temperature for an object , the life expectancy of that component is halved.

Experienced Based standard Priority I Priority 2 Priority 3 Priority 4

Over 40°C 20 - 40°C 10 - 20°C 1 - 10°C


Thermographic case 1: Spindle bearing over packed with grease The following example illustrates the temperature effects due to the deliberate overpacking of a grease lubricated spindle.

Figure 40 Example bearing image 10 minutes after deliberate overpacking with grease

Figure 41 Example bearing image 30 minutes after deliberate overpacking with grease


Figure 42 Example bearing image 90 minutes after deliberate overpacking with grease Thermographic case 2: Over-torqued bearing housing Vibration spectrum are given collected from the hot and cold bearing on the spindle shown in the thermographic image figure 44. As discussed previously bearing degradation shows itself with a broad high frequency increase in vibration amplitude as indicated in figures 43 and 45. It also shows itself by a temperature rise induced by increased friction due to breakdown of the bearing races. In this case the root cause for the bearing damage and heat generation was found to be an excessive torque applied to the mounting fixtures. .

Figure 43 Example spectrum from cool bearing


Figure 44 Thermographic image showing location of vibration measurement points

Figure 45 Example spectrum from hot bearing


4.0 WHY PDM? MAINTENANCE SELECTION STRATEGY The appropriate selection of a maintenance strategy is crucial, since the impact on both direct maintenance and operational costs may be huge. This is not a discussion that should be made in a vacuum. Input from as many different sources as possible should be welcomed. In general there are a number of rules of thumb that can be applied to guide the decision process, these are expressed diagramatically in figure 46. The maintenance search sequence starts with a selected piece of equipment. This may be a complete set of machinery, such as a pump, coupling, transmission, and drive, or individual components that operate on rolling element bearings. Once the machine is identified, its criticality to plant operation must be established. The decision making process can conceptually be simply put as: “If the consequences and cost of failure are acceptable and no other adverse effects result, the equipment can be scheduled for corrective maintenance only”. When the consequences of failure can be tolerated but the cost is unacceptable, the extent to which the failure is predictable will determine the type of maintenance. A machine with a reasonably well predictable failure span can be subjected to preventive maintenance. When the span between failures is known, the question of whether PDM is practical must be asked. If not practical, preventive maintenance should be considered. If practical, the economic aspects of PDM must be evaluated before a decision on the use of PDM is made. 4.1 CONSIDERATIONS FOR SELECTING THE TYPE OF MAINTENANCE The selection of maintenance strategy and approach is probably one of the largest decisions made by maintenance management. The approach to this decision should be full informed and well structured. The following aspects of the implementation of maintenance technologies should be useful in making a strategic decision. Feasibility - Can I implement PDM at my plant? Cost - Can I afford to implement PDM ? Equipment required - Do I understand the requirements for equipment and resources? Machine location and environment - Is there anything about my machines that impact effectiveness or cost ? Information - Do I have process in place to collect the information I need for a decision ? The following questions should help answer theses main issues. 4.1.1 Useful questions to answer in evaluating the feasibility of PDM In assessing the feasibility of PDM you are trying to answer the question, will it work for me ? To answer this the following questions may be helpful. What does your maintenance history tell you about the expected machinery failure modes? Do you have a good understanding of PDM technologies and the type of failures they can predict? Are the primary failure modes identifiable with current technology? What percentage of the machine expected failures will PDM satisfy? Can the machine be properly prepared for instrumentation? How much data needs to be collected and of what type? Do you understand the equipment and resource requirements for interpreting machine data? Can the machine be subjected to invasive mounting if required? Do machine design attributes make correct sensor mounting unachievable? What are the environmental conditions, do equipment and or operator hazards exist? This document is protected under copyright, it may not be reproduced in whole or in part without written consent of Ron Frend Page 63

A negative evaluation of any of the above questions would indicate that PDM is not appropriate.

Figure 46. Simplified Maintenance Selection Diagram

4.1.2 Useful questions to answer in understanding the costs of PDM


Having satisfied the feasibility requirements, an examination of the economic viability of PDM is needed. You need to answer the question: “Can I afford it?”. The following questions will be helpful in making this decision. What is the cost of machine failure including cost of parts, labor, transportation and lost production? What is the effect on the work load of a change in maintenance strategy? What new or additional costs will be accrued by manpower, equipment etc.? Are technologies in existence that can reduce new equipment costs, such as networks etc. ? Will any existing equipment become overloaded with the new additions? Is the manpower in place sufficient to carry the expected PDM work load? How many machines are to be monitored and does this effectively satisfy your need for PDM? Do you know what equipment is necessary and what it will cost? Is the acquisition time and information achievable realistic to achieve your program goals? Is trained manpower available, if not do you understand the cost of training? Is the expertise available in-house to develop a PDM program? The above set of questions guides you to assess not only pay back, but also the financial commitment that is necessary both for implementation and for maintenance of the new strategy.

5.0 CONCLUSIONS AND RECOMMENDATIONS The conclusion that can be drawn after evaluating PDM technologies for the analysis and condition assessment of rolling element bearings is very favorable, and suggests the recommendation for implementation in industry. In the industries contacted during the preparation of theses materials and in the reference material utilized , a sufficiently high failure rate and repair activity exists to justify the use of PDM monitoring techniques on machinery supported on rolling element bearings. Experience has shown that the degree to which this practice will be cost effective depends upon the particular plant setup, however, data that has been collected from some industries clearly indicate savings in excess of the initial investment. A number of key reminders are worthy of repetition in conclusion : · • Of the different rolling element bearing types, ball bearings are most widely used, followed by tapered roller and cylindrical roller bearings. •

The majority of the bearings are grease lubricated, oil lubrication is employed mostly in the larger and more important types of machinery such as turbo machinery in the utility and Petrochem industries.

•

Loss of lubricant and overheating account for approach 25% of all failures. These failure modes are usually common to grease-packed bearings. The combined modes of fatigue spalling, contamination, corrosion, and brinelling account for almost 50% of all failures. Poor assembly accounts for 10%, and the rest are unexplained.

•

Poor record keeping makes the cost of failures difficult to estimate. This significantly effects the implementation of expensive PDM, since financial justification is made more difficult.

•

PDM is practiced by only a small percentage of rolling element bearing operators.


•

The most common types of rolling element bearing flaws that developed during operation consist of fatigue spalling, lubricant starvation, brinelling.

•

Advanced signal processing techniques, such as the use of demodulation, are yielding early indications of failure and should be incorporated into PDM practices.

•

The use of Thermography in PDM is indicating very good results and should be incorporated in standard programs.

•

Inexpensive yet limited PDM techniques exists that should not be overlooked for the implementation of PDM on tight budget.

•

Inaccurate or poorly informed diagnoses caused by faulty data acquisition methods or procedures quickly result in a lack of confidence in any system. Standard approaches covering data acquisition and recording should be developed as soon as practical good practices and record keeping is the key to a successful PDM program.


Vibration Analysis on Bearings

Recommend Documents