Coded data generation or conversion – Code generator or transmitter – With variable pulse spacing or grouping
Reexamination Certificate
2000-01-04
2002-05-21
Hilten, John S. (Department: 2863)
Coded data generation or conversion
Code generator or transmitter
With variable pulse spacing or grouping
C702S035000, C702S185000
Reexamination Certificate
active
06392584
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to a system and a corresponding method of operation for giving a warning of impending failure of machines based on their vibrations.
2. Description of the Related Art
An ounce of prevention is said to be worth a pound of cure. When it comes to machines failing, however, it is often difficult to prevent a failure without any warning that any failure is imminent.
Most machines operate by generating and applying at least one periodic force to at least one mass in order to cause each mass to rotate, reciprocate, or both. The physical response of the machine to the periodic driving force(s) at any given point can then be determined and analyzed by sensing the vibrations and translations of the machine at that point. Many different types of sensors may be used to sense motions of the machine and generate a signal (usually electrical) that represents the sensed motions. Examples of common include accelerometers, strain gauges, pressure transducers (for indirect measurements via fluids) and even acoustic transducers. Expressed in more mathematical terms, one can attempt at least a partial construction of the machine's transfer function.
The responsive motions of the machine may then be analyzed in one or both of two main ways, namely, in the frequency domain and in the time domain. Frequency-domain techniques—as their name implies—involve converting and analyzing the measured amplitude response signal into a spectrum. This may be done using analog filters, but is most commonly done by converting the analog output signal from the sensor into digital (numerical) form and then using a Fast Fourier Transform (FFT) routine to estimate the spectrum.
It is known that as the internal wear—particularly on bearings—increases, so too does the energy contained in higher frequencies of the machine's response spectrum. Any increase (especially any relatively sudden) increase in the amplitude of the higher-frequency components of the machine's response spectrum will then indicate an increasing likelihood of wear-induced failure of the machine. Less subtle faults such as an actual mechanical failure within the machine will typically be detectible as even larger and more sudden changes in the spectrum, especially in the high-frequency components, since such faults often give rise to impulsive forces within the machine.
One drawback of frequency-domain techniques is that it assumes some knowledge of the “normal” spectrum of the machine. Some machines, for example, operate normally at much higher frequencies than others, and a sensor tuned for a slower machine would need to be calibrated for a different, high-rpm machine. Frequency-based warning systems must therefore often be calibrated for each machine. Proper calibration is often beyond the level skill of those who work with the machines; at the very least, it is a procedure that one would preferably not have to worry about at all.
In time-domain systems, the response signal of the machine is sensed and digitized as before, but rather than analyzing it into a spectrum, its statistical properties are compiled and analyzed. Most machines operating normally usually display the properties of a stationary stochastic system; sufficiently large and/or sudden deviations from the “normal” statistical profile can then be assumed to indicate a highly likely or increasingly likely or, indeed, an existing internal fault or defect that may lead to failure.
Known statistical parameters of any set of data such as the amplitude measurements of the machine response signal taken during some predetermined interval include the mean (first moment) and standard deviation (second moment).
It has been discovered, however, that the fourth moment of the machine response signal is particularly useful in detecting defects and predicting impending failure. Because the kurtosis parameter is even more sensitive to statistically outlying measurements than are the mean and standard deviation, it is particularly useful for detecting impulsive forces, even when these are small and are superimposed on much stronger low-frequency signal components.
U.S. Pat. No. 4,089,055 (Dyer, et al., May 9, 1978) discloses an electronic monitoring apparatus in which a kurtosis coefficient is calculated for at least two frequency bands of a machine. Variations of the kurtosis coefficient are then detected and used to provide an indication of the condition of an object such as a machine.
One shortcoming of existing machine monitors such as Dyer, which rely on statistical analysis, for example of kurtosis, is that they often require calibration for each particular application. A variation in the level of vibration, for example, can cause the measurement resolution of such known systems to be reduced because of the fixed range of their analog-to-digital conversion circuitry. Another problem is that, with a fixed sampling frequency, they are often sensitive to variations in the fundamental frequency component of the machine, for example, the rotation speed, that is, rpm. As is well known (the Nyquist criterion), the sample rate of an analog signal must be at least twice the lowest frequency component from which accurate information is required. Information contained in higher-frequency components will be aliased and will show up as increased noise in lower frequency components. Note that this phenomenon holds true for all sampling processes, even those in which actual processing of the sampled signals is done in the time domain.
What is needed is therefore a monitoring system that able not only to provide a warning of possible defects within a machine (or similar physical system), but is also able to do so without needing special calibration for each application. In other words, the monitoring system should be either wholly self-calibrating, or at least require less calibration than existing systems when used to monitor a wide variety of machines. The improved system should preferably be able to provide adequate resolution over a wide range of vibration amplitudes, and it should relatively insensitive to variations in the frequency changes (such as operating rpm) of the monitored system. This invention provides such a system.
SUMMARY OF THE INVENTION
The invention provides a system for monitoring the status of a machine comprising a sensor (preferably, one or more accelerometers) that generates an output signal corresponding to motions of the machine at a monitoring point. An analog-to-digital converter (ADC) converter is then used to convert the output signal of the sensor into a series of samples forming a digital input signal. A processor then partitions the digital input signal samples into a plurality of input data sets and calculates an alarm parameter for each data set. The processor then generates an alarm signal when the alarm parameter of at least one data set meets a predetermined alarm criterion. In order to enable use of the invention with a wide range of machines, and to make the system according to the invention wholly or at least substantially self-calibrating, the processor preferably also adjusts the sampling rate of the ADC as a predetermined function of the input data sets. An auto-ranging feature, which may be implemented within the processor or as a separate circuit, is preferably also included to scale the output signal of the sensor to fall within a predetermined range of the ADC.
The preferred alarm parameter is kurtosis, that is, the statistical fourth moment of the samples of each respective data set. The alarm criterion is then preferably that the kurtosis value has exceeded a predetermined threshold. The sensor is preferably a two-axis accelerometer pair so that the system is substantially independent of the mounting orientation of the sensor on the machine.
REFERENCES:
patent: 4417337 (1983-11-01), Favin et al.
patent: 5602757 (1997-02-01), Haseley et al.
patent: 5619202 (1997-04-01), Wilson et al.
patent: 5943634 (1999-08-01), Piety et al.
patent: 5952803 (1999-09-01), Canada et al.
Hilten John S.
Slusher Jeffrey
Washburn Douglas N
LandOfFree
System and method for detecting and warning of potential... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for detecting and warning of potential..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for detecting and warning of potential... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2890491