The capture and analysis of stress waves provides significant improvement in condition monitoring of critical rotating machinery

Stress waves accompany metal-to-metal impacting, fatigue cracking, scuffing, abrasive wear and other defects commonly encountered with rotating machinery faults. These stress waves are short-term (fractional to a few milliseconds) transient events which introduce ripples on the surface of the machinery as they propagate away from the initiating event. The important features to capture for fault diagnostics is the magnitude of the events and rate (periodic or random) of occurrence. In this study, the stress waves are captured by (a) separating them from the normal vibrations through the use of high pass filters, (b) capturing the magnitude through recording the peak values (identified as Peak Value measuring methodology) over repeated continuous small time increments consistent with the analysis bandwidth and (c) identifying periodicity through spectral analysis. The analysis methodology has been implemented into a standard data collector/analyzer employed in routine condition monitoring of rotating machinery based on vibration analysis.

This methodology has proven to be a very valuable tool to add to the analyst tool box. Representative results showing bearing faults and cracked gear teeth are presented. The faults demonstrated are:
o Outer race bearing defect in a critical pinion stand gear box on a tandem mill
o Cracked gear teeth on output shaft of a double speed reduction precision tension bridle gear box
o Outer race bearing defect on a mandrel drive motor
o Outer race bearing defect on a centrifugal water pump
o Inner race bearing defect on a Basic Oxygen Furnace (BOF) vessel

The conclusion from this study is that this methodology significantly improves the analyst capability to detect and classify severity of bearing and gearing defects commonly experienced in the steel industry.

1.0 Introduction
Representative results from vibration analysis employing standard vibration and stress wave analysis for two gear boxes, a motor, a pump and a basic oxygen furnace (BOF) vessel are presented. The faults detected and presented are from defective bearings and cracked teeth in a gear box. The standard vibration analysis employs the velocity spectral data methodology. The stress wave analysis employs the peak value measuring methodology implemented on a standard portable data collector/analyzer.

For very slow speed machinery such as the BOF vessel and gear boxes, the stress wave analysis employing the peak value measuring methodology is demonstrated to be a valuable tool for the purposes of (a) detecting faults, otherwise missed with standard velocity spectral analysis and (b) assist in establishing the severity level. In the next section, a brief discussion of stress waves and reasons for the appearance in rotating equipment are presented. This will be followed by a brief discussion of the analysis methodology employed. Representative case studies from two gear boxes, a motor, a pump, and a BOF vessel are presented in Section 3. The conclusions drawn from this study are presented in the final section.

2.0 Stress wave activity and analysis methodology

2.1 Stress Wave Activity in Metallic Systems
Stress waves are set up in a metallic system when events such as impacting, fatigue cracking, scuffing, abrasive wear, etc., occur. Stress waves in metal appear as both longitudinal and bending waves. Bending waves introduce ripples on the surface (hence excites an accelerometer attached to the surface) as they propagate away from the initiating site at the speed of sound. The stress waves are of short duration (fractional to a few milliseconds) and hence appear in the output of an accelerometer as short-term transient events.

2.2 Analysis Methodology
The velocity spectral analysis was carried out using standard guidelines for selecting analysis bandwidth, resolution, etc. Stress waves, which are short-term transient events, are characterized by broad frequency content. This makes possible the separation of the stress waves from the normal vibration through the use of high pass filters. For gearing systems, the general rule for selecting the high pass filter is greater than three times gear mesh since excessive wear on teeth will often manifest itself at three times gear mesh. For other systems, the high pass filter is typically selected at 30 to 50 times shaft speed (greater than 3 to 5 harmonics of inner race bearing defect).

The spectral analysis bandwidth for peak value measuring is selected following the same general rules as used in normal spectral analysis. The selection of the spectral analysis bandwidth, Fmax, defines the sampling rate to be employed (the sampling rate typically is set at 2.56 times Fmax).

In normal spectral analysis, the signal is routed through a high order, low pass (anti-aliasing) filter to remove maximum energy associated with frequencies greater than one half the sampling rate (avoids aliasing). The analog signal which emerges from the anti-aliasing filter is converted to a digital representation by sampling the analog signal at the discrete time intervals established by the analysis bandwidth. Each digital representation is the amplitude of the signal emerging from the antialiasing filter at the instant in time the analog signal was digitized. The anti-aliasing filter removed any significant variation of signal (possibly short-term transient event) which could have occurred between sample times. Hence normal spectral analysis generally are not responsive to stress wave activity.

Stress waves accompany many mechanical faults found in rotating machinery. The specific frequencies found within stress waves generally are not of interest. What is most useful is the detection and quantification of the events relative to (a) rate of occurrence and (b) magnitude of individual events. In this methodology, the detection and specification of magnitude of individual events and identification of rate of occurrence are accomplished by:
o Separate stress waves from normal vibration with high pass filter
o Find absolute peak value of signal over each time increment specified by analysis bandwidth. The block of data acquired for further processing, e.g., FFT, consists of peak values for each increment of time in lieu of instantaneous values used for normal analysis
o Perform spectral analysis, FFT, for the block of peak values which will identify any periodic event rate. Random event rates will increase the spectral noise floor. The time waveform consists of the peak value block of data used for FFT analysis

3.0 CASE STUDY

3.1 Introduction
A typical case is selected for presentation and illustration of the importance of peak value stress wave analysis and normal spectral vibration analysis for the detection and specification of severity level of faults found in machinery in the steel-making industry. The case consists of a BOF vessel with defective bearing.

3.2 Basic Oxygen Furnace (BOF) Vessel
The basic oxygen furnace vessel, see 'Figure14', is one of several very slow speed machines in the steel industry which are critical to operations. These generally are very large machines with bearings that are large, expensive, long lead times, etc. Advanced warning of impending failure of specific components becomes very useful information relative to scheduling repairs, etc.

In machinery rotating at speeds in excess of 300 RPM, normal velocity spectral analysis has proven to be an effective tool for early detection of commonly occurring faults such as bearing defects. Additionally, the incorporation of stress wave analysis has also proven to be helpful for these class of machines and are capable of detecting other faults such as cracked teeth in gear boxes, etc. For very slow machinery (less than 10 RPM), normal velocity spectral analysis provides little or no reliable detection capability. On the slow speed machinery, stress wave analysis provides the only reliable detection methodology.

Stress waves are transient events characterized by reasonably high frequency (1 to 10 kHz). To reliably capture these events in a digital manner, the time domain data must be sampled at a high rate (order of 50,000 samples per second). For a vessel like the BOF vessel which is turning at 0.5 RPM, very large data blocks would be required to store data (one complete revolution would require 12 megabytes). The BOF only turns about 160, hence only 3 megabytes are required for data storage for one swing.

Once the data is captured, the analyst has the task of "analysis". The use of spectral analysis is not an option here. What has been done is to capture the data (usually at a less than desirable sampling rate such as 12K samples per second) and store in a spectrum or transient capture instrument. Then the data can be displayed as a "compressed time trace" where the entire block of data is displayed on a single screen where only max. and min. values within the horizontal resolution are displayed. The number of points over which the max/min values are selected is simply determined by dividing the total number of data points by the horizontal pixel resolution, e.g., if we have 106 data points and 103 pixels, the max/min values are selected from 103 data points for each pixel.

An example of capturing a large block of data and displaying a compressed time waveform from a sensor on the bearing housing of a BOF vessel is presented in Figure 15. This data was acquired at a sampling rate of 2.56 time 5KHz or 12.8 K samples/sec (which is about 1/4th the generally accepted required sampling rate). In this data, it is clear that there are impacting events occurring at some reasonably repetitive rate. The task for the analyst then is to determine the most probable source for these events and then estimate severity. For this task, it would be helpful to have historical data to compare with, trend, etc. The large blocks of data from which this compressed time waveform was extracted makes trending difficult due to large storage requirements, etc.

At a certain facility where two BOF vessels are present, the vessels were turned through two complete turns (at nominal rate of 0.5 RPM) and the peak value measuring time waveform captured. One of the two vessels showed an indication of a defective inner race. The time waveform covering the two revolutions are presented in 'Figure 16(a)'. The portion of the peak value measuring time waveform through the defective regions are expanded and presented in 'Figure 16(b)'. Here the time between impacts which would indicate inner race are highlighted. The conclusion is that there is an inner race defect which has initiated (a) trending and (b) scheduling replacement.

In Peak value measuring methodology, currently implemented, a sampling rate of 100,000 samples per second is used to find the peak values. For a specified analysis bandwidth of 20Hz and 1600 lines, the time waveform is 4096 data points for 1600/20 or 80 seconds of elapsed time. This clearly encompasses one work swing of the BOF vessel. Since we are dealing with only 4x103 data points versus 4x106 data points required for the raw data, trending, order tracking, etc., becomes a very practical capability.

4.0 SUMMARY AND CONCLUSIONS
Emphasis has been placed in this paper on the analysis of stress waves employing the peak value analysis methodology for detecting and classifying severity level of faults typically experienced in machinery commonly encountered in the steel industry. The speed range for the machinery monitored for this study ranged from 0.5 to 900 RPM. Both stress wave and normal vibration analysis were employed for machinery greater than 100 RPM. Only stress wave analysis was considered for machinery less than 10 PRM.

The implementation of the peak value stress wave analysis was accomplished using the same instrumentation as was employed for the normal vibration analysis and required minimal effort (no additional hardware or software) from the operator. The primary source of information from normal velocity spectral analysis is the velocity spectral data. In the peak value data analysis, both the spectral and peak value time waveform data are important for fault identification and severity evaluation.

For the faults such as cracked teeth in gear boxes and bearing faults on the BOF vessel (0.5 RPM), the analysis was the only methodology which permitted detection and severity evaluation. For other faults, both Peak Value measuring methodology and normal velocity spectral analysis contributed to detection and severity evaluation. In some cases, it was Peak Value measuring methodology which was paramount in initiating corrective action while in others it was normalvelocity spectral analysis which initiated the correctiveaction.

The main conclusions are:
o The condition monitoring program should continue with the normal velocity spectral data collection, trending, alarming, etc.,
o The inclusion of stress wave analysis employing a methodology which provides both spectral data and equally important, maintain the true peak values in the time domain should be a part of the standard condition monitoring program

David M. Stobbe
Marc Phillips
James Robinson