Volume 18 - Friction, Lubrication, and Wear Technology Part 8 potx

Fig. 6 Effect of windowing on time-domain signal Fig. 7 Effect of windowing on spectrum Averaging In many practical situations, data collection has to be carried out in a noisy environment, which means that the signal to be analyzed is contaminated by unwanted signals from other sources. A commonly used method of attenuating unwanted signals is to limit the bandwidth by conventional filtering. Because the noise is usually broadband, this cuts off the extra bandwidth beyond the highest frequency of interest, which eliminates much of the noise. Of course, this approach is of little use if the noise and the signal occupy the same bandwidth. If the unwanted noise signal has a near-zero mean value, then the signal-to-noise ratio (S/N) can be improved by averaging several blocks of data that come from the sensor. Usually, between 4 and 64 data blocks are taken. Signal averaging is most effective when a deterministic signal is buried in a random signal. It can be applied to the raw time data or to the resultant spectrum of each block. It is a very effective way to clean up (smooth) plots. The improvement in S/N ratio is equal to the square root of the number of blocks of data used. Random signals cannot be analyzed exactly (unless the sampling time stretches, from minus to plus infinity), unlike deterministic signals, where the sampling time can be confined to one cycle, because of the repeatability of the cycle. Finite sampling times applied to random signalswill introduce errors. If noise is present in the signal, then the quality of the result will depend on the averaging time, T av , and the ideal filter bandwidth, f. It has been shown that if the noise in the signal has a Gaussian distribution, then the measurement uncertainty, , is (Ref 7): (Eq 1) For example, if a random vibration is to be analyzed in 20 Hz bands, then the averaging time needed, if the measurement uncertainty is to be less than 5% of the mean squared signal, will be 20 s. Exponential averaging is a continuous summation average in which most of the recent data are added while earlier data are removed. For example, ten exponential averages means starting with ten summation averages, removing the first, and adding the eleventh, and so on. Therefore, each new piece of data counts for th of the average. Peak averaging considers as many instantaneous spectra as are selected and saves only the highest amplitude at each frequency seen during the data gathering period. Statistical Analysis Approach Failure in moving surfaces that are in contact can be due to a number of causes related, primarily, to excessive wear. Pitting occurs at points of maximum Hertzian contact stress. A failure of this type depends on the number of stress cycles, and results in small surface fatigue cracks. Small pits are formed, usually 0.5 to 1.0 mm (20 to 40 mils) in diameter. Large areas of material removal are known as spalling. Scuffing is related of the lubricating film and is caused by overheating, which could be due to friction and the sliding velocity between surfaces. This type of contact produces alternating welding and tearing, which removes metal rapidly. Once the condition begins, it is very difficult to reestablish a proper oil film. Severe surface welding is called scoring. Plastic flow is due to cold working of the surfaces caused by high contact stresses, together with rolling and sliding actions. The surface deformation results from the yielding of the surface and subsurface material. Abrasive wear, unlike the other types of surface damage, is not a local failure, but is likely to spread over large areas. The amplitude distribution of a vibration signal picked up from surfaces moving relative to each other can be expressed in terms of a probability density function (PDF). This represents an estimate (probability) of the time the signal remains in a particular amplitude window. It has been well established in the literature that machined surfaces are neither perfectly flat nor smooth. Experiments show (Fig. 2b) that the PDF closely approximates to a Gaussian distribution. As damage (wear) starts to occur, the PDF begins to deviate from this classic bell-shaped curve. Rather than examining the PDF in detail, Dyer and Stewart (Ref 8) decided to track the progress of this damage by examining the statistical moments of the data. Any signal, particularly a random signal, can be described by its statistical moments. The first three moments are: • Mean value, or average amplitude size • Mean squared deviation, or average power in the signal • Standard deviation, or the measure of how closely the data are clustered around the mean value A general integral defining all the statistical moments (Ref 9, 10) can be expressed as: M n = g(y) n p(y)dy (Eq 2) where M n is the nth moment, nequals 1, 2, 3, 4, . . ., g(y) is the amplitude function, p(y) is the probability density function, and y is amplitude in millimeters. For a surface that is in good condition: (Eq 3) Odd moments are related to information about the position of the peak density distribution in relation to the mean value, whereas even moments indicate the characteristics of the spread of the distribution. It has been well established mathematically (Ref 11) that if the fourth moment is normalized using the square of the second moment, then the Kurtosis coefficient takes on the unique value of 3.0, if the surfaces under consideration exhibit a Gaussian distribution of asperities. As the surfaces become damaged, the Gaussian indicator changes value, because the shape of the curve moves from the classic bell shape. Volker and Martin (Ref 12) showed that for roller bearings, a damage map could be created by plotting the Kurtosis coefficient against acceleration level. It can be seen from Fig. 8 that the types of damage in progress can be identified. The damage map represents a plot of the energy present because of the damage (x axis) versus a measure of the impulsive nature of the damage and the number of defects (y axis). Fig. 8 Damage classification for roller bearings Figure 9 shows the difference between dry and lubricated bearings using this approach. The accelerometer is placed on the bearing housing and the data is then filtered into five frequency bands on the graph. Fig. 9 Difference between dry and lubricating bearing. Points labeled 1 indicate frequency range from 2.5 to 5 kHz; points 2, 5 to 10 kHz; points 3, 10 to 20 kHz; points 4, 20 to 40 kHz; points 5, 40 to 80 kHz Figure 10 shows the effect of adding silicon powder to the bearing lubrication in order to simulate abrasive damage. The particles used were size #230 SiC, applied in 0.2 mg doses over a period of 8 min. Fig. 10 Effect of adding abrasive powder Relationship between Friction and Vibration As mentioned previously, the amount of contact between any two surfaces, in the process of cold welding of the asperities, is a function of the load. In the case of two flat plates, Bowden and Tabor (Ref 13) generated the data shown in Table 1. Table 1 Amount of contact between two flat plates Load True area of contact N kgf cm 2 10 -5 in. 2 Percentage of apparent area of contact (a) 19.6 2 0.0002 3.1 0.0010 49 5 0.0005 7.8 0.0025 980 100 0.01 155 0.05 4900 500 0.05 775 0.25 (a) Nominal contact area, 20 cm 2 (3 in. 2 ) The general relationship between load and contact area can be expressed as: (Eq 4) where A is the true area of contact (cm 2 ), F is the applied load (N), and H is the indentation hardness (N/cm 2 ). The volume of material worn away when one of the plates slides over the other is proportional to the true area of contact, the total distance of sliding, and the nature of the materials. (Eq 5) where V is the volume of material removed (cm 3 ), l is the total distance traveled (cm), and k is the wear coefficient. It has been found, in practice, that the wear coefficient, k, remains constant until the pressure between the plates exceeds a value greater than one-third of the hardness, H. Hence, the wear rate is reasonably linear. At pressures in excess of this, k begins to increase, and the wear rate rises rapidly and nonlinearly. It is suggested that running parts, which are in motion relative to each other, need to be carried out with relatively light loads, until an indication of constant wear rate shows that the surfaces have settled in to each other. This happens when the asperities have been ground down and the effective pressure falls, allowing higher pressures to be applied. A typical value for k would be 26 × 10 4 . This "running in" effect illustrated in Fig. 11, which shows the results for a new bearing, just out of the box, and a repeat set of results after 60 minutes running. Although the vibration level does not give any significantly information, the Kurtosis values indicate that after 1 h, the results are closer to 3.0. This indicates a reduction in peakness in the data, that is, the asperity distribution is closer to Gaussian, after the honing period. Fig. 11 Detecting the progress of "running in" The interaction of surface asperities during sliding causes complex vibration patterns in both the normal and tangential directions. It has been suggested (Ref 14) that the asperities act as a system of microsprings, with appropriate stiffnesses in two directions. These normal and tangential contact springs are, of course, nonlinear with loading, because the geometry of the peaks changes, and more of them come into contact. For dry friction conditions, the asperities can be modeled as a random distribution of microaxially loaded bars that are set in free vibration by the sliding motion between the plates. Calculations based on the above theory and experimental data indicate that the frequency of free vibration for metallic pairs experiencing asymmetrical deformation of asperities was found to be approximately in the range from 500 to 3000 Hz. The higher the relative velocity between the two surfaces, the more prominent is the normal component of impulses, because of collisions between the asperities, and the higher the amplitude of contact vibrations. References 1. F.T. Barwell et al., The Interaction and Lubrication of Rough Surfaces, Proceedings of the Symposium of the International Union of Theoretical and Applied Mechanics, IUTAM (Enschede, Holland), 1974, p 304- 329 2. M. Brock, Fourier Analysis of Surface Roughness, Bruel and Kjaer Technical Review, No. 3, 1983, p 3-45 3. S. Braun, Ed., Mechanical Signature Analysis, Academic Press, 1986, p 321-342 4. J. Chatigny, Piezo Film Yields Novel Transducers, Electron. Week, Aug 1984 5. R.V. Williams, Acoustic Emission, Adam Hilger, Bristol, 1980 6. J.W. Cooley and J. W. Tukey, An Algorithm for the Machine Calculation of Complex Fourier Series, Maths of Computation, Vol 19 (No. 90), 1965, p 297-301 7. J. Bendat and A. Piersol, Chapter 6, Random Data: Analysis and Measurement Procedures, Wiley- Interscience, 1971 8. D. Dyer and R. Stewart, Detection of Rolling Element Bearing Damage by Statistical Vibration Analysis, J. Mech Design (ASME), Vol 100, 1978 9. J.S. Bendat, Principles and Applications of Random Noise Theory, John Wiley & Sons, 1958 10. A. Papoulis, Probability, Random Variables and Stochastic Processes, McGraw-Hill, 1965 11. H.R. Martin, Review of Gear Damage Monitoring Techniques, Proceedings of th e First International Machinery Monitoring and Diagnostic Conference (Los Angeles), Society for Experimental Mechanics, 1989, p 183-189 12. E. Volker and H.R. Martin, Application of Kurtosis to Damage Mapping, Proceedings of the Fourth International Modal Analysis Conference (Los Angeles), Society for Experimental Mechanics, 1986, p 629- 633 13. F.P. Bowden and D. Tabor, "Friction and Lubrication," Methuen Co., 1956 14. B.V. Budanov, Mutual Relation between Friction and Vibration, EuroTrib 81, Vol 1A, 1981, p 240-246 Selected References • K.G. Beauchamp, Signal Processing, Allen and Unwin, 1973 • E.O. Brigham, The Fast Fourier Transform, Prentice-Hall, 1974 • I.P. Castro, An Introduction to the Digital Analysis of Stationary Signals, Adam Hilger/ESM, 1989 • N.H. Cook, Tool Wear Sensors, Wear, Vol 62, 1980, p 49-57 • G. Kivenson, Durability and Reliability in Engineering Design, Pitman, 1971 • W. Lenkiewicz, The Sliding Friction Process Effect of External Vibration, Wear, Vol 13 (No. 2), 1969, p 99-108 • P.A. Lynn, Electronic Signals and Systems, MacMillan, 1986 • J.S. Mitchell, Machinery Analysis and Monitoring, PennWell Books, 1981 • D.E. Newland, An Introduction to Random Vibrations and Spectral Analysis, 2nd ed., Wiley, 1984 • A. Papoulis, Probability, Random Variables and Stochastic Processes, McGraw-Hill, 1965 • R.T. Spurr, Frictional Oscillations, Nature, Vol 169, 1961 • T. Vinh and J. Blouet, Non Stationary Signal Processing Applications to the Study of Time Dependent Sliding Friction, Annals CIRP, Vol 30 (No. 1), 1981 Lubricant Analysis F.E. Lockwood, Pennzoil Products Company; R. Dalley, Predict Technologies Introduction LUBRICANT ANALYSIS PROGRAMS for predictive maintenance are now routinely practiced by a number of industries, including railway, aircraft, automotive and truck, chemical, refinery, and various others (Ref 1 2, 3, 4, 5, 6, 7, 8, 9, 10, 11). The physical appearance, viscosity, metals content, and other properties of oils sampled periodically from operating equipment are interpreted to provide an early warning of impending failure or to signal the need for oil changes and routine maintenance. Best known of such programs are the spectrometric oil analysis program (SOAP) and related programs practiced by the U.S. military services (Ref 12). Similar programs have also been developed for commercial aircraft, commercial truck and auto fleets, and industrial equipment such as compressors, gearboxes, and pumps. Continuous on-line oil monitoring may one day be standard practice. Oil/wear particle analysis also is often a valuable failure analysis tool, although less has been published on this subject (Ref 13, 14). Failure analysis generally is practiced by mechanical engineers who primarily examine hardware. However, in the area of automotive lubricants, extensive use of oil/wear particle analysis is made in "post-mortem" failure investigations. This expertise is similarly useful in many other areas. These two subjects, covering the details and references of applications of oil/wear particle analysis, are the focus of this article. The important oil analysis methods will be reviewed, and appropriate test programs for predictive maintenance and a methodology for failure analysis will be discussed. Acknowledgements Oil analysis cases and a number of helpful comments were provided by C.M. Comer, J.W. Fu, T.E. Rushing, and J. Torres of Pennzoil Products Company. Extensive information and figures were extracted from the Wear Particle Atlas, currently a publication of Predict Technologies. Oil/Wear Particle Analysis Methods Oil/wear particle analysis to determine lubricant condition includes both physical inspection and chemical examination of wear debris, contaminants, and reaction products from lubricants such as engine oils, hydraulic fluids, cutting fluids, greases, and synovial fluids from humans and animals (Ref 15). Because wear is an inevitable and anticipated consequence of surface contact between interacting machine parts such as shafts, bearings, gears, and bushings even in properly lubricated systems, oil/wear particle analysis can potentially be applied to all lubricated equipment. Equipment life expectancies, safety factors, performance ratings, and maintenance recommendations are predicated on normally occurring wear and lubricant service. Such factors as design complexity, unit size, intricacy of assembly configurations, and variations in operating conditions and environments can make maintenance or repair needs difficult to evaluate or detect without taking equipment out of service. Oil/wear particle analysis allows noninterruptive diagnostic determination of lubricant condition by determining the amount of wear and the lubricant reaction products. Based on this determination, equipment condition or impending failure can be predicted. Sampling of Service Lubricants. Lubricant analysis begins with the sampling procedure, and the validity of a particular analysis depends on how well this procedure is carried out. The ideal sample is taken immediately downstream from the lubricated surfaces for example, from a drain line off an individual bearing, prior to filtration, while the equipment is operating under usual conditions and temperatures. Care is taken to obtain a representative sample by discarding any volume that may have been stagnant in the drain line. The sample is captured in a clean nonmetallic container, sealed, and carefully labeled, including information about lubricant and equipment history. In practice, it is difficult to achieve this ideal. Samples often must be taken from sumps and recycle lines or large reservoirs. Deficiencies in sampling points, however, are to some extent compensated for by consistent sampling at the same point under the same operating conditions (Ref 5). Once in the laboratory, all samples should be brought to a uniform temperature and stirred condition before testing. This is particularly important when studying lubricant additive and wear metals, which may stratify in the lubricant under some conditions. Sampling frequency is another key concern. This depends on the type of equipment, service conditions, and critical nature of service. Equipment maintenance records should suggest a proper sampling frequency. Otherwise, it is a good idea to sample frequently (for example, weekly) until a track record is built; then, if desirable, the rate of sampling can be lessened. Once a possible problem is detected, the sampling frequency must be increased until a positive determination is made on equipment condition and the action to be taken. As general guidelines, aircraft should be inspected after every flight, while large industrial equipment with good filtration systems may require only monthly inspection. One rule of thumb is 25 to 40 h sampling frequency for gas turbines and 100 to 500 h for diesel engines. Recommended inspection guidelines for checking the condition of lubricants used in large industrial equipment are given in Table 1. Table 1 Recommended lubricant inspection intervals for selected engines, drive systems, and power generating units System Recommended operating intervals between inspections, h Aircraft gas turbine 50 Airborne hydraulic system 50 Aircraft derivative gas turbines 50 Diesel engine 200 Heavy transmission/gears 200 Surface hydraulic system 200 Heavy-duty gas turbine 200-500 Steam turbine 250-500 Large-bore reciprocating engine 250-500 Source: Ref 6 For each lubricant parameter that is measured, a control record is built that after a period of time will reveal normal operating ranges for a given type of equipment/lubricant. For parameters such as viscosity, direct comparison with lubricant and equipment manufacturer specifications also provides information on the acceptable operating range. In setting up a sampling and analysis program, it should be kept in mind that unless parameter operating guidelines are known beforehand, the program must provide enough information to set statistical guidelines for acceptable versus abnormal parameter limits. Physical Inspection. A simple physical inspection, particularly in the case of failure analysis, can speed diagnosis of poor equipment operation and wear. Typical lubricant indicators that are often observed are listed in Table 2. [...]... composition, shape, and texture of both metallic and nonmetallic wear particles The wear particles are classified to determine the type of wear and its source Types of Wear Particles Ferrography is used to differentiate abnormal wear conditions from the normal rubbing wear that occurs during stable machine operation and from break-in wear that occurs during start-up of equipment During break-in of a wear surface,... amounts of wear particles are detected by ferrography The various types of wear particles are illustrated in Fig 4 These particles are described in considerable detail in Ref 39, 40, and 41 Fig 4 Various types of wear particles (a) Normal rubbing wear particles (b) Sliding wear particles (c) Cutting wear particles (d) Fatigue particles (e) Laminar particles (f) Spheres (g) Red oxide sliding wear particles... 16 -1 8 Aug 1976 37 D.P Anderson and R.D Driver, Equilibrium Particle Concentration in Engine Oil, Wear, Vol 56, 1979, p 41 5-4 19 38 W.W Seifert and V.C Westcott, A Method for the Study of Wear Particles in Lubricating Oil, Wear, Vol 21, 1972, p 2 7-4 2 39 A.A Reda, R Bowen, and V.C Westcott, Characteristics of Particles Generated at the Interface Between Sliding Steel Surfaces, Wear, Vol 34, 1975, p 26 1-2 73... Publication 584 , National Bureau of Standards, 1 980 16 "Standard Test Method for Kinematic Viscosity of Transparent and Opaque Liquids (and the Calculation of Dynamic Viscosity)," D 455, Annual Book of ASTM Standards, Vol 5.01, ASTM, 1990, p 17 0-1 75 17 "Standard Test Method for Acid and Base Number by Color Indicator Titration," D 97 4 -8 7, Annual Book of ASTM Standards, Vol 5.01, ASTM, 1990, p 33 0-3 35 18 "Standard... Potentiometric Titration," D 66 4 -8 9, Annual Book of ASTM Standards, Vol 5.01, ASTM, 1990, p 23 6-2 42 19 "Standard Test Method for Water in Liquid Petroleum Products by Karl Fischer Reagent," D 174 4 -8 3, Annual Book of ASTM Standards, Vol 5.01, ASTM, 1990, p 67 7-6 81 20 "Standard Test Method for Detecting Glycol-Base Antifreeze in Used Lubricating Oils," D 2 98 2 -8 5, Annual Book of ASTM Standards, ASTM 21 M.P Granchi,... Will Optimize Both Lubricant and Equipment Life, Lubr Eng., Vol 45 (No 10), 1 989 , p 61 8- 624 33 R Cooper, Wear Debris Monitoring of Rolling Bearings, Br J Nondestr Test., Mar 1 983 , p 7 5 -8 3 34 P.S Baur, Ferrography: Machinery -Wear Analysis With a Predictable Future, Power Mag., 1 982 35 U.S Patent No 404 781 4 36 E.R Bowen and W Seifert, "Ferrography A New Tool for Analyzing Wear Conditions," paper presented... severe wear on the ball bearings and damage to the race and cage Subsequently, the damage was repaired, and the pump was cleaned and returned to service The 12/22 /86 sample showed a substantial decrease in the WPC and SI SEM analysis showed several strings of wear particles, which were identified as steel and leftover wear debris from the earlier failure The strings of wear particles were probably break-in... analytical methods Both particle counting and direct-reading ferrography detect the onset of severe wear as a rapid increase in the amount and size of particles Particle counting detects all particles, whereas ferrography screens primarily for ferrous wear particles Typical reporting formats require that the number of particles per milliliter of fluid volume be broken down into the following particle size categories... Ferrographic Condition-Monitoring Program in a Petrochemical Plant," Preprint No 82 -AM-4C-3, American Society of Lubrication Engineers, 7th Annual Meeting (Cincinnati), 1 0-1 3 May 1 982 5 F.E Lockwood, C.M Comer, J.W Fu, and F.D Davidson, Monitoring Wear Particles in Industrial Fluids by Ferrography, Proceedings Technische, Akademie Esslingen, 6th International Colloquium, 1 2-1 4 Jan 1 988 6 J.S Laskowski,... Optical and scanning electron microscopy (SEM) examination of the particles showed dirt particles and a small amount of normal wear debris At this point, the ferrographic analysis of this machine was no different from that of other unfiltered machines Fig 10 Wear particle concentration changes in a P-975 moist solvent pump tested over an 11-month period The fourth sample (11 / 18/ 86 ) gave a 10-fold increase . metallic and nonmetallic wear particles. The wear particles are classified to determine the type of wear and its source. Types of Wear Particles Ferrography is used to differentiate abnormal wear. wear particles. (a) Normal rubbing wear pa rticles. (b) Sliding wear particles. (c) Cutting wear particles. (d) Fatigue particles. (e) Laminar particles. (f) Spheres. (g) Red oxide sliding wear. of wear particles is within the detectable (<10 m) range. However, in abnormal wear situations, such as severe sliding, rolling fatigue, cutting and abrasive wear, and scuffing wear, particles