CASE STUDY: Roller Mill Gearbox. James C. Robinson. CSI, an Emerson Process Management Co. Lal Perera Insight Engineering Services, LTD.

Transcription

1 CASE STUDY: Roller Mill Gearbox James C. Robinson CSI, an Emerson Process Management Co. Lal Perera Insight Engineering Services, LTD. ABSTRACT Stress Wave Analysis on a roller will gearbox employing the peak value (PeakVue ) methodology identified an inner race defect on the inlet shaft. The severity of the fault advanced rapidly relative to what normally is experienced. The bearing fault was detected and classified as serious based on rate of growth of the impacting g levels. The bearing was replaced and serious inner race fault confirmed. Post replacement, PeakVue data confirmed the absence of the bearing fault. The post analysis of the PeakVue data showed the reoccurence of a fault experienced earlier on this gear box. It was concluded (previously and currently) that the observed impacting was a result of torsional resonance at a frequency equal to four times the first intermediate shaft running speed. In addition to PeakVue analysis, both strain gauge data and current data analysis confirmed the torsional resonance postulate. 1. Introduction This paper is focused on vibration analysis of a Roller Mill Gearbox employing the peak value, PeakVue, analysis methodology employed in the CSI hardware (portable and permanent). A broad discussion of the PeakVue methodology including theory, applications, rules for implementation, and alert/fault settings are presented in a white paper 1 which can be found on the CSI web site ( under support/technical papers/peakvue. 1 James C. Robinson and James E. Berry Description of PeakVue and Illustration of its Wide Array of Applications in Fault Detection and Problem Severity Assessment, support/technical papers/peakvue. 1

2 A brief review of the PeakVue methodology is presented in the next section. Section 3 contains PeakVue and velocity vibration representative data and analysis pertaining to the inner race bearing fault for one of the bearings on the inlet shaft to the roller mill gearbox. Severity assessment using PeakVue and velocity vibration data are presented. The latter part of Section 3 will focus on the torsional induced impacting which appeared after bearing replacement. The final section, Section 4, will be a summary of major findings and lessons learned. 2. PeakVue Methodolgy The motivation behind the development of PeakVue was to provide the capability to capture transient events (stress wave activity) which accompany impacting, fatigue cracking, and friction in industrial rotating machinery. Each event introduces stress wave packets of short duration (fractional to a few milliseconds) which propagate away from the initiation site at the speed of sound in metal. The frequency band over which the majority of energy is deposited can range in the nominal 1 to 5 khz range. The specific frequency band excited is dependent on the mass of the impacting parts, speed of the machine, and class of event responsible for the event (impacting different than fatiguing or friction). The rate at which the events occur are dependent on the fault type and speed of the machine. Thus to digitally capture events directly, a high sampling rate would be required (1 k sample/sec) for a time sufficient to monitor 6 + revs of the slowest fault (generally cage which repeats approximately once per 2.5 revs of the shaft). As an example, to resolve cage on a 6 RPM machine would require data to be captured at a sampling rate of 1, samples/sec for 15 s or 1,5, data points. This clearly is not practical. The frequency content within each individual event is of less importance than the rate of occurrence and the amplitude (use as a measure of energy) of each event. The need to capture these two equally important parameters was the driving force behind the development of PeakVue. For the rate of occurrence, the procedure is to define an analysis bandwidth which covers the possible fault frequencies which may be encountered, e.g., bandwidth greater than a bearing inner race fault or twice the gear mesh frequency. Once the bandwidth is defined (F max in data collectors), then use the standard sampling rate of 2.56*F max for defining sequential time intervals (1/SR) to 2

3 construct a PeakVue time waveform. Each sequential time interval contains the peak value encountered over that time interval. This sequence continues until all time intervals are filled (typically 2 N data points). Once the time block of peak values is obtained, spectral or autocorrelations analysis can be applied to resolve periodicity. The spectral or autocorrelation analysis of the PeakVue time waveform identifies periodicity (specific fault related to turning speed) or lack thereof (random events are often associated with friction, cavitation, etc.). The amplitude of the impacting events provide a generic trend parameter which can be correlated with fault severity. For an extensive discussion on theory and methodology of PeakVue, the reader is referred to Ref Case Study: Roller Mill Gear Box 3.1 Introduction In the next subsection, a description of the gear box (with measurement point locations) chosen for the case study are presented. This will be followed with selected PeakVue data as well as discussion of the data identifying the faulted inner race on the input shaft of the gearbox. The fourth subsection will focus on normal velocity spectra data relative to the bearing fault. The fifth subsection will focus on severity assessment employing trended PeakVue data and normal acceleration trended data. The last subsection will focus on the post bearing replacement PeakVue data. The data post bearing replacement shows no further indication of a bearing fault; however, a different recurring fault is present and will be discussed. 3.2 Description of Gear Box and Measurement Locations A cross sectional view of the gear box under study is presented in Figure 1. This gearbox is driven by a 2 hp 8 pole induction motor. It is approximately in size. The outlet shaft, which turns at 23 RPM, has sleeve bearings and the remainder bearings are of the rolling element type. This gear box has been monitored using vibration analysis for about 1 years. PeakVue has been used since the 1997 (when PeakVue was introduced) time frame. A photograph of the inboard end of the gearbox with measurement points G1 through G6 is presented in Figure 2. The input shaft is turning at 893RPM and has three 3

4 Figure 1. Cross Sectional View of Roller Miller Gearbox. bearings (see Figure 1). The table in Figure 1 identifies the third bearing as an SKF234 which was known to be incorrect. It was believed a FAG NU 234E was the third bearing which is a 14 rolling element bearing. It was identified later (after much bewilderment) to be a KOY NU 234 which is a 15 rolling element bearing. Bearing 1 is also a 15 rolling element bearing (FAG NU 336) having fault frequencies very near those of the KOY NU PeakVue Data The PeakVue data acquired from measurement point G2 on February 2, 22 showed significant increase in the overall peak g level relative to that acquired the previous month on January 9, 22 (51 g s peak on February 6 versus 3.58 g s peak on 4

5 G5 & G6 G2 G1 G3 & G4 Figure 2. Photo of the Inboard End of Roller Mill Gearbox identifying measurement points 1 through 6. January 9). The PeakVue spectra and time waveform acquired on February 6, 22 for measurement point G2 are presented in Figure 3. Similar data acquired the previous month, January 9, 22, are presented in Figure 4. The major difference between these two data sets is the significant increase in the peak impacting g-levels, 51 g s on Feb. 6 vs 3.6 g s on Jan. 9, 22. The signature in the spectra in Figs. 3 and 4 are representative of an inner race fault wherein the fault frequency for the inner race will be side banded with running speed. This is the case since the impacts introduced by the rolling elements passing over the defective inner race will be amplitude modulated at running speed due to passing in and out of the load zone once per rev of the shaft (inner race). The PeakVue data acquired from measurements points G1 (Figure 5) and G5 (Figure 6) on February 5, 22, are similar to that obtained from measurement point G2 for the same date. The important differences between the PeakVue data are the maximum 5

6 RMS Acceleration in G-s Acceleration in G-s IGB - Roller Mill Peak Vue RM-PV -G2 Gearbox Point BPFI Frequency in Hz Route Spectrum 6-Feb-2 12:43:59 (PkVue-HP 1 Hz) OVERALL= 8.28 A-DG RMS = 8.28 LOAD = 1. RPM = 893. (14.89 Hz) Route Waveform 6-Feb-2 12:43:59 (PkVue-HP 1 Hz) RMS = PK(+) = 51.9 CRESTF= Time in msecs Freq: Ordr: Spec: Dfrq: Figure 3. PeakVue data acquired on February 6, 22 from measurement point G RMS Acceleration in G-s IGB - Roller Mill Peak Vue RM-PV -G2 Gearbox Point BPFI Route Spectrum 9-Jan-2 12:43:1 (PkVue-HP 1 Hz) OVERALL=.448 A-DG RMS =.4387 LOAD = 1. RPM = 892. (14.87 Hz) Acceleration in G-s Frequency in Hz Time in msecs Route Waveform 9-Jan-2 12:43:1 (PkVue-HP 1 Hz) RMS =.8999 PK(+) = 3.58 CRESTF= 3.94 Freq: Ordr: Spec: Dfrq: Figure 4. PeakVue data acquired on January 9, 22 from measurement point G2. 6

7 RMS Acceleration in G-s Acceleration in G-s IGB - Roller Mill Peak Vue RM-PV -G1 Gearbox Point Frequency in Hz Route Spectrum 6-Feb-2 12:43:19 (PkVue-HP 1 Hz) OVERALL= 3.76 A-DG RMS = 3.74 LOAD = 1. RPM = 893. (14.89 Hz) Route Waveform 6-Feb-2 12:43:19 (PkVue-HP 1 Hz) RMS = 6.71 PK(+) = CRESTF= Time in msecs Freq: Ordr: Spec: Dfrq: Figure 5. PeakVue data acquired on February 6, 22 from measurement point G1. RMS Acceleration in G-s IGB - Roller Mill Peak Vue RM-PV -G5 Gearbox Point Route Spectrum 6-Feb-2 12:46:24 (PkVue-HP 1 Hz) OVERALL= 2.46 A-DG RMS = 2.45 LOAD = 1. RPM = 535. (8.92 Hz) Acceleration in G-s Frequency in Hz Route Waveform 6-Feb-2 12:46:24 (PkVue-HP 1 Hz) RMS = 4.51 PK(+) = CRESTF= Time in msecs Freq: Ordr: Spec: Figure 6. PeakVue data acquired on February 6, 22 from measurement point G5. 7

8 impacting levels are reduced from 5 g s at G2 to 24 g s at G1 and from 5 g s at G2 to 15 g s at G5. Since stress waves attenuate with propagation away from the initiation site (see Ref. 1), the most probable source for the impacting is the inner race for bearing No. 3. The problem is the inner race fault frequency for the FAG NU 234E (believed to be bearing No. 3) does not agree (the BPFI for the FAG bearing is 8.3 orders) with what is observed (8.7 orders) from Figure 3. The amplitude of the impacts has proven to be a reliable indicator of the impacting source; therefore, we conclude the bearing in place at position No. 3 differs from what was believed (probable 15 roller versus 14 roller) to be in place. When the bearing was replaced, it was positively identified to be a 15 rolling element bearing. 3.4 Velocity Vibration Data The PeakVue data presented in the previous section identified a significant increase in the peak values from the PeakVue time waveform data acquired on February 2, 22 relative to the data acquired in the previous months on January 9, 22 (51 g s peak on February 6 versus 3.58 g s peak on January 9). The PeakVue spectral data (see Figure 3) identified a repetitive event occurring at 129 Hz (8.7 orders) with amplitude modulation (side banding in spectra) at running speed. The velocity vibration spectra data for selected dates for measurement point G2H are presented in Figure 7. The data from November 1, 21 represents pre fault or reference spectra. The data from April 3, 22 was acquired after bearing replacement. The sets of data form February 22 were taken in the interval of time where PeakVue was seeing very significant impacting at the believed inner race fault at 8.7 orders of inlet shaft. The intermediate data sets in Figure 7 clearly have a family of harmonic activity at the same 8.7 orders. Additionally, the activity at 8.7 orders are sidebanded at running speed as would be expected for an inner race fault. The time data captured corresponding to the spectral data in Figure 7 are presented in Figure 8. The key features are the presence of short term transient events occurring at once per rev in the data acquired on February 9 and 13 of 22. The postulate is the inner race fault is exciting a structural sub resonance each time the fault 8

9 IGB - Roller Mill Normal Vibration RM-NV -G2H Gearbox Point 2 Horizontal Max Amp.32 Plot Scale X Apr-2 14:49:48 PK Velocity in In/Sec X X X Frequency in Orders Feb-2 1:52:17 9-Feb-2 19:36:43 1-Nov-1 11:5:6 Ordr: Freq: Sp 3: Figure 7. Velocity spectra for measurement point G2A: reference spectra, 1- November 1, 21; fault, February 9, 22; Fault, February 13, 22; post bearing replacement, April 3, 22. IGB - Roller Mill Normal Vibration RM-NV -G2H Gearbox Point 2 Horizontal Apr-2 14:49 Acceleration in G-s Feb-2 1:52 9-Feb-2 19: Nov-1 11: Time in msecs Figure 8. Time Wave form in g s corresponding to spectra data of Figure 7. 9

10 RMS Acc in G-s RMS Acc in G-s Acc in G-s 1 8 IGB - Roller Mill Peak Vue RM-PV -G2 Gearbox Point Ref Peak=.25 g's Fault=1.5 _ g's Alert= g's Days: 31-May-1 To 7-Feb May-1 1 X X Sidebands Frequency in Hz Peak=5 g's Time in Seconds Jan-2 7-Feb Trend Display Overall Value Route Spectrum 7-Feb-2 14:56:29 (PkVue-HP 1 Hz) OVERALL= 7.83 A-DG RMS = 7.81 LOAD = 1. RPM = 893. (14.88 Hz) Route Waveform 7-Feb-2 14:56:29 (PkVue-HP 1 Hz) RMS = PK(+) = CRESTF= 4.18 Freq: Ordr: Spec: Dfrq: Figure 9. Graph of trended overall (PeakVue data) from May 31, 21 through February 7, 22 (a), PeakVue spectra on February 7, 22 (b), and PeakVue time data block on February 7, 22 (c) from measurement point G2. passes through the load zone. This is postulated to be the source for the increased spectral activity in the 5 to 65 order range observed in Figure Severity Assessment The parameter which was trended during data accrual for the PeakVue data was the digital overall value. In Ref. 1, it is noted that no aproiri absolute value can be assigned to this parameter regarding fault severity, but it is a recommended trend parameter. The recommended procedure is to establish a baseline value and set the ALERT level at 3 to 4 times the baseline value and FAULT at twice the ALERT. For measurement point G2, the reference overall is found to be.25 g s (using data acquired in the May July, 21 time frame). Hence set ALERT at.75 g s and FAULT at 1.5 g s. The trended data from 31-May-1 through 7-Feb-2 with PeakVue spectra and time waveform taken at the time for last data point are presented in Figure 9. On the trended 1

11 Peak g Trend Data For Measurement Point G g's Peak Peak G's Alert Fault Days Figure 1. Trended peak g-level from PeakVue data from May 1, 21 through April 22 for measurement point G2. data, note that the ALERT level was not exceeded on 9-Jan-2 but the FAULT level was exceeded by a factor of 5 approximately 1 month later. This rapid increase is not typical but implies the fault is serious. The recommended trend parameter is the Peak value from the PeakVue time waveform. This was not trended, but the trend has been constructed from the stored data and is presented in Figure 1. The ALERT level recommended in Ref. 1 for an inner race fault for this speed machine is 3 g s peak and 6 g-for the fault level. On 9-June-2, the Peak g-level had exceeded the ALERT level but not the FAULT level. Approximately one month later, the peak g-level had exceeded the FAULT level by a factor of 8. The peak g-level exceeded the FAULT g-level by approximately a factor of 13 at time the machine was shut down and the bearing replaced. After bearing replacement, the peak g-level (around 2 g s) reduced to a value below the ALERT level for measurement point G2. 11

12 RMS Acc in G-s PK In/Sec Acc in G-s Jun-1 IGB - Roller Mill Normal Vibration RM-NV -G2H Gearbox Point 2 Horizontal 21-Sep Days: 31-May-1 To 25-Feb X X X X X X X Frequency in Orders Time in Seconds Jan Feb-2 25-Feb-2 Trend Display Overall Value Route Spectrum 25-Feb-2 13:2:27 OVERALL=.4193 A-DG PK =.978 LOAD = 1. RPM = 892. (14.87 Hz) Route Waveform 25-Feb-2 13:2:27 RMS =.4152 PK(+/-) = 2.23/2.9 CRESTF= 5.37 Ordr: Freq: Spec: Dord: Figure 11. Graph of trended overall (RMS acceleration data) from May 31, 21 through February 25, 22, velocity spectra data on February 25, 22, and Acceleration time data block on February 25, 22 The large peak g-levels experienced on this machine are indicative of a serious friction (spalling) event occurring. The negative to the call for a spalling event as the potential source is the lack of randomness in the PeakVue time waveform (which generally is the case with spalling). Spalling would explain the high g-levels Severity Assessment from Velocity Spectral Data. The trended digital overall from the acceleration spectral data from May 31, 21 through February 25, 22 are presented in Figure 11. The velocity spectral data with corresponding time waveform from the last trended data point are included in Figure 11. The overall trend value increased by a factor of two past January 9, 22 which is not large within it self, but the impacting like feature at 1X observed in the time waveform coupled with activity in the spectral data in the 5 to 65 order range are significant. This higher frequency activity 12

13 often is associated with a bearing in it s latter stage (stage 4) of failure. 2 The element missing from the classic stage 4 classification in the absence of running speed and harmonics thereof. The inner race fault frequency with many harmonics are present. The severity assessment from the velocity spectral data alone would be in the beginning of stage Defective Bearing. A photograph of the defective bearing is presented in Figure 12. The bearing was identified to be a KOY NU234 which is a 15 rolling element bearing (the fault frequency for the inner race is at 8.7 orders which agrees with what was observed). The fault is a deep groove across the bearing face with evidence of spalling occurring immediately downstream of the fault (this explains the high g-levels occurring from spalling immediately following a roller passing over the fault which would be a periodic event). Figure 12. Photograph of Defective Inner Race showing fault indicating spalling following impacting. 2 Illustrated Vibration Diagnostic Chart, Technical Associates of Charlotte, P.C. 13

14 3.6 Post Bearing Replacement Introduction. After the defective bearing was replaced, PeakVue data was acquired on all measurement points on the motor/gearbox. The expectation was the peak g-levels would reduce to their pre-fault conditions of less than 1 g on each measurement point. This was not the case. The largest peak g reading were acquired on measurement point G3 which is located at the bearings supporting the first intermediate shaft below the beveled gear pair (see Figures 1 and 2). The PeakVue data acquired following the bearing replacement will be presented and discussed in the next subsection. This will be followed by a presentation and discussion of data acquired from strain gauges placed on the inlet shaft in 1997 and motor current data analysis. The strain gauge data and motor current data suggests excitation of a torsional resonance as the probable source for the observed activity post bearing replacement PeakVue Data for Measurement Point G3. The measurement point G3 would be most active for activity associated with Bearing #6 (see Figure 1) and/or the beveled gear set on the first intermediate shaft. The shaft is turning 532 RPM. Using the recommendation for alert/fault levels for the peak-g levels (see Ref. 1) in the PeakVue time waveform, the alert level for the shaft turning at 535 RPM is set at 2 g s peak and the fault level at twice the alert or 4 g s peak. The overall alert level is set at 3 times reference and fault at twice alert. The trended overall pre and post bearing replacement are presented in Figure 13. From the overall levels in the May to September, 21 time frame, the reference g-levels are established to be less than.15 g s. Based on recommendations from Ref. 1, the ALERT level is set at.45 g s and the FAULT level at.9 g s for the overall value. The overall level following bearing replacement (3-APR-2) exceeded the fault level (see Figure 13). The peak g-level from the PeakVue time waveforms are presented in the trending format from May 1 through 3-Apr-2 in Figure 14. The bearing was replaced at approximately 275 days after initiation of the trended data (31-May-1). The levels did decrease after bearing replacement, but the peak g-level of 7.73 g s on 3-Apr-2 soon 14

15 RMS Acc in G-s RMS Acc in G-s Acc in G-s Frequency in Hz IGB - Roller Mill Peak Vue RM-PV -G3 Gearbox Point Fault=.9 g's Ref Level=.15 g's Alert=.45 g's Days: 21-Sep-1 To 3-Apr Time in Seconds 3-Apr-2 Trend Display Overall Value Route Spectrum 3-Apr-2 1:5: (PkVue-HP 1 Hz) OVERALL=.5154 A-DG RMS =.5179 LOAD = 1. RPM = 535. (8.92 Hz) Route Waveform 3-Apr-2 1:5: (PkVue-HP 1 Hz) RMS =.6927 PK(+) = 3.92 CRESTF= 5.66 Date: 3-Apr-2 Time: 14:17:52 Ampl:.978 Figure 13. Graph of trended overall (PeakVue data) from May 31, 21 through April 3, 22 (a), PeakVue spectra on April 3, 22 (b), and PeakVue time data block on April 3, 22 (c) from measurement point G3. after startup are twice the fault level setting of 4 g s (the alert level consistent with recommendation in Ref. 1 is set at 2 g s). The PeakVue data acquired on 2-Apr-2 are presented in Figure 15. In the spectral data the activity is dominated by the 1 st intermediate shaft frequency (8.87 Hz) and many harmonics. The PeakVue time waveform shows two very repetitive impacts per turn of the 1 st intermediate shaft * located at precisely ¼ turn of the shaft apart. The PeakVue time wave form from this measurement point for September 21, 21, January 9, 22, April 3, 22 and April 3, 22 are presented in Figure 16. The impacting pattern clearly changed following the replacement of the faulty bearing in March, 22. The pattern observed in April 22 are similar to what was observed in the * The pattern in the PeakVue time waveform is what would be expected for two cracked teeth on the 37 tooth beveled gear located 9 or 1 teeth apart. Past experience on this gear box suggests a torsional resonance is the more probable source. 15

16 Peak g Trend Data For Measurement Point G Peak G's 15 Alert g's Peak Fault Days Figure 14. Trended peak g-level from PeakVue data from May 1, 21 through April 3, 22 for measurement point G time frame (see Figure 17 for representative PeakVue time waveform data from February, March, and June of 1997). In 1997, it was postulated the most probable explanation for the impacting patterns of Figure 17 was a torsional resonance at a frequency of 4 times the first intermediate shaft (35.5 Hz) speed. On October 3, 1997, a strain gauge was placed on the inlet shaft to the gear box and dynamic stress measurements acquired. The spectra from this measurement are presented in Figure 18. The most active response is at 35.5 Hz 16

17 RMS Acceleration in G-s IGB - Roller Mill Peak Vue RM-PV -G3 Gearbox Point 3 Route Spectrum 3-Apr-2 14:17:48 (PkVue-HP 1 Hz) OVERALL=.9783 A-DG RMS =.9787 LOAD = 1. RPM = 535. (8.92 Hz) Acceleration in G-s Frequency in Hz Time in msecs Route Waveform 3-Apr-2 14:17:48 (PkVue-HP 1 Hz) RMS = 1.23 PK(+) = 7.73 CRESTF= 6.56 Freq: Ordr: Spec: Figure 15. PeakVue data from measurement point G3 acquired on April 3, 22 (soon after startup following bearing replacement). which is at four times first intermediate shaft speed. This data (as well as visual inspection) supports the postulate of torsional resonance being responsible for the impacting present in the PeakVue data and occurring at the beveled gear set. Torsional activity should introduce dynamic variations in torsional load experienced by the motor. This will introduce variations in current (amplitude variations) supplied to the motor. Therefore, it was decided to carry out spectral analysis on the motor current. The objective is to determine the presence of amplitude modulation of the motor current which will occur as side banding around the line frequency (6 Hz). The motor current spectral data acquired on November 13, 1997 are presented in Figure 19 (note the amplitude scale is logarithmic). The periodic sidebands are identified on the spectral data plot and are basically the same as those identified in the strain gauge data of Figure 18. The conclusion from the comparison of the torsional spectral data (Figure 18) 17

18 with the current spectral data (Figure 19 is that the current spectral data provides confirmatory evidence of torsional vibration. Current spectral data was acquired on April 3, 22 and are presented in Figure 2. The spectral data for Figure 2 are basically the same as that of Figure 19. Therefore, the conclusion is the probable source of the impacting observed in April 22 is the same (torsional resonance) as that experienced in Summary and Conclusions 4.1 Summary In this paper, analysis of vibration data using both a) velocity spectral analysis and b) peak value (PeakVue) analysis acquired from a large rolling mill gearbox were presented. The gear box was well instrumented with accelerometers place in the proximity of each bearing support location IGB - Roller Mill Peak Vue RM-PV -G3 Gearbox Point Apr-2 1:5 Acceleration in G-s Apr-2 14:17 9-Jan-2 12: Sep-1 9: Time in Seconds Figure 16. PeakVue time data blocks for measurement point G3 acquired on September 21, 21, January 9, 22, April 3, 22, and April 3,

19 INL - Roller Mill Peak Vue RM-PV -G3 Gearbox Point Jun-97 13:32 Acceleration in G-s Mar-97 14: Feb-97 13: Time in Seconds Figure 17. PeakVue time data blocks for measurement point G3 acquired on February 13, 1997, March , and January 11, Figure 18. Strain gauge spectral data (gauge placed on inlet shaft) acquired on October 3,

20 RMS Amplitude in Amps Hz=4 x first int shaft Hz=2 x input shaft 3 x first int shaft 2 x first int shaft IGB - Roller Mill Normal Vibration RM-NV -MC2 Motor Current Hz=first int shaft 4.7 Hz 1.18 Hz 1.18 Hz 4.7 Hz 8.9 Hz=first int shaft 2 x first int shaft 3 x first int shaft Hz=2 x input shaft Hz=4 x first int shaft Spectrum Display 13-Nov-97 12:41 RMS = 3.96 LOAD = 19% RPM = 891. (14.85 Hz) Frequency in Hz Label: ROLLER MIL-C2 / Freq: Ordr: Spec: Dfrq: Figure 19. Motor current spectral data acquired on November 13, RMS Amplitude in Amps Hz=4 x first int shaft 2 x input shaft 3 x first int shaft IGB - Roller Mill Normal Vibration RM-NV -MC2 Motor Current 2 2 x first int shaft 14.9 Hz=1 x input shaft 1 x first int shaft 4.7 Hz 1.1 Hz 1.1 Hz 4.7 Hz 1 x first int shaft 14.9 Hz=1 x input shaft 2 x first int shaft 3 x first int shaft 2 x input shaft 35.5 Hz=4 x first int shaft Spectrum Display 3-Apr-2 1:15 RMS = LOAD = 1% RPM = 892. (14.87 Hz) Frequency in Hz Freq: Ordr: Spec: Dfrq: Figure 2. Motor current spectral data acquired on April 3,

21 Data has been acquired and trended from this gearbox for several years. On February 6, 22, the overall trend parameter for PeakVue spectral data form measurement point G2 showed an increase of about a factor of ten from the previous months data. Close examination of the spectral PeakVue data showed a classic defective inner race bearing fault with a fault frequency of 8.7 orders of the input shaft speed. There are three bearings side-by-side supporting this input shaft. There are two measurement points mounted over the two outer bearings. Using the three bearings which were believed to be installed, the conclusion would be the first (bearing No. 1) bearing would be the bearing with the fault since it s the only one of the three having a BPFI about 8.7 orders. This conclusion is consistent with what was observed in the velocity spectral data based on activity level. The peak g-levels observed in the PeakVue data provided strong evidence that the faulted bearing was Bearing No. 3. If this was to be the case, then the conclusion was the bearing in place differed from what was thought to be in place (had to have one additional roller). Additionally, the PeakVue data, namely the high peak g-levels being experienced, implied the fault was serious and should be replaced soon. The bearing was replaced in March 22 and verified to have the additional rolling element (15 versus 14) with a serious inner race fault. In addition to a mechanical defect on the inner race, visual inspection verified spalling was occurring in the proximity of the mechanical defect. Post data acquired following bearing replacement showed no evidence of any bearing defects within the gearbox. However, the PeakVue data did show a defect (based on impacting g-levels) was present in the proximity of measurement point G3. This measurement point would be most sensitive to problems within bearing number 6 or the input beveled gear set. Based on previous experience with this gear box, the most probable source for the impacting occurring at this position was postulated to be impacting between the gears within the beveled gear set introduced by torsional resonance. Motor current spectral analysis and previous strain gage spectral analysis support the torsional resonance postulate. 21

22 4.2 Conclusions When faults are the source of stress wave activity, they often can be detected and fault identified at varying distance from the source. The location of the fault can be identified by comparing peak g-levels detected form the various locations where PeakVue spectral data identify the fault. The location where the largest peak g-level occurs is the location of the fault. The severity level of the fault can be judged from the peak g-level by a) their absolute level and b) their trended value over time. For severity assessment, it is necessary that the methodology employed to capture the stress wave activity maintains the absolute g-level independent of the analysis bandwidth, machine speed, or the duty cycle of the impacting event (PeakVue captures the true peak g-level emerging form the sensor-accelerometer). The peak g values should be extracted from batch data which contains a minimum of 15 revs of the machine. Stress wave analysis on the reference gear box identified a probable torsional problem in the gear box post bearing replacement. Motor current analysis provided confirmatory data supporting the torsional resonance postulate. The use of stress wave analysis, PeakVue, provided the primary tool for fault detection and severity assessment on this gear box (in the author s experience, for gear boxes in general). The normal velocity spectral analysis did provide confirmatory data of the inner race fault, but it provided negative data in identifying which of the three bearings was defective. 22