DE19954066C1 - Harmonics analysis method for rotary vibration of machine component, has sampled values obtained over full period fed to digital low-pass filter with cascaded filter stages - Google Patents

Abstract

Analysis method uses sampling of the vibration signal for the rotating component over a full period at a fixed sampling rate, with low-pass filtering of the sample values using a filter threshold dependent on the rotation rate. The digital low-pass filter has cascaded filter stages, with the filter threshold for each stage having a fixed ratio to the filter threshold for the preceding stage.

Description

The invention relates to a method for order analysis of Vibrations of a rotating part, especially one Machine, in terms of amplitude and phase.

Modern motor drives make the demands very high level of noise and vibration minimization Expectations. In the course of improving product quality highly accurate measurements and analyzes of rotary or translatory measured quantities necessary.

The measurement technology required for this must offer the possibility Even at variable speeds, vibrations as harmonics to determine a shaft speed and correct amplitude and Deliver phase measurements. For changes or Modifications to the product or drive immediately and quickly To be able to judge the measurements including Results can also be evaluated online.

The processes to be examined are usually harmonic Vibrations of a speed of the machine to be examined, d. H. so-called orders. On the basis of these regulations, the Mechanical engineer draw conclusions about the cause of the pull the respective vibration. So z. B. the 0.5 order a 4-stroke internal combustion engine due to irregularities in the  Firing order caused. Such an order analysis is done with a Fourier transform (FFT) with the following illustration carried out according to amount and phase.

The aim of the calculation is to always have the same order on the map the same spectral line, the true frequency response to maintain and amplitudes and phases as little as possible distort.

So-called window processes for the Ord Performance analysis are ruled out for the above reasons. There is therefore only the possibility of having a coherent system a sampling over integer periods of the to be examined To build up the signal curve, but the sampling rate is continuous must be adjusted to the speed of the machine. During the actual measurement over a machine cycle however, the sampling rate should be retained since it is equidistant Support points in the signal for further processing of the same Condition are.

Under the above conditions it follows that the current Sampling rate in the known prior art only based on the previous period can be estimated. This but differs from the current period, so that possibly the last scan of a cycle is not exactly with that Period end matches.

In principle, a moving filter, ie a so-called tracking filter, which, for. B. from JS Mitchell, Introduction to machinery analysis and monitoring, 2nd ed., Tulsa, Oklahoma, USA, PennWell Books 1993 , page 75, are used, and scanning is carried out over integer machine cycles with subordinate Fourier transformation.

The function of such moving filters that a filter ver position over several speed decades and an evaluation of high Enabling machine orders requires a filter structure, over its entire area with very high resolution, must be quasi-analog adjustable. However, this is with justifiable hardware expenditure or only to a limited extent possible.

Vibration measurements on rotating machines to assess their condition can be carried out using the so-called order analysis. Here an evaluation of the periodic frequency components in the vibration signal is carried out, which are directly related to the speed in order to, for. B. to draw conclusions about the causes of errors. Reference is made here, for example, to "measurement technology and measurement signal processing", KW Bonfig, ed., Renningen-Malsheim: expert-Verlag 1996 , pages 122-127, DE 37 25 123 A1 or DE 40 32 299 C2.

An order analysis also takes place with acoustic measuring methods specifically in automotive development. According to technical Messen atm, 1978, issue 4, pages 141-146, is an order known analysis in which from a continuous change total noise, a proportion is extracted which is one corresponds to the prescribed order. An analysis is carried out for this Made orders about the speed and it comes in digital tracking filter for use. The tracking filter is at Speed changes automatically to the respective frequency of the selected order. The one for filter tuning required control frequency is determined by pulses that the Speed are proportional.

The use of a digital tracking filter for Schwin tion analysis is also previously known from US 5,445,028.  

Ultimately, DD 158 581 shows one method and one Circuit arrangement for vibration monitoring rotating Machines, which are about early damage detection to reach. After the solution there, the changes the level total values and the vectorial changes of the complex harmonic vibrations in the time interval Reference level values and complex harmonic reference components compared and evaluated. For the evaluation is the Mandatory use of a computer system. From DD 158 581 tracking filters are also known.

From the above, it is therefore an object of the invention further developed procedure for order analysis of Vibrations of a rotating part, especially one Specify machine, with the possibility of the procedure exists, the corresponding same order always on the to map the same spectral line in each case without amplitudes or phases are falsified, as well as the goal of maintaining the Width of the frequency response of a correspondingly implemented one Measuring arrangement.

Furthermore, a simple post-processing of the received Measured values take place in such a way that the known fast Fourier transformation to vibration assessment can be carried out.

The object of the invention is achieved with a Ver drive according to its definition according to the teaching of the patent claim 1, the subclaims at least expedient Refinements and developments include.

It is returned to a moving filter grabbed, which are adjusted in its filter corner frequency can without time gaps due to otherwise given Ein oscillation times arise. Such a filter can be implemented on the hardware side  implemented by a digital signal processor that is operated with suitable software.

The scanning of those recorded in a manner known per se Vibration signals occur over entire periods with a fixed Input sampling rate, with digital low-pass filtering in Connection is carried out, the filter corner frequency from the current speed in correspondingly varied is derived.

Because the filter corner frequency at the beginning of a machine cycle depending on the speed at the last one before established revolution and for the duration of this Cycle is held invariant, the result of the fil Equidistant support over the machine cycle receive the signal curve.

The filtering becomes a in a signal post-processing step Interpolation to correct a discrepancy between the current and for setting the filter corner frequency underlying speed downstream, then from the temporal course of the filtered samples by itself known fast Fourier transform the associated one Spectrum is won.

In one embodiment of the invention, any one becomes measuring analog value using an analog-to-digital converter digitized fixed sampling rate and at least one that digital tracking filter forming signal processor supplied. The Signal processor receives the duration of the last expired Shaft revolution and thus sets the above Corner frequency of a low-pass filter. To an unwanted one oscillate the filter when switching the corner frequency avoid, the filter is implemented cascaded and only the filter output switched. Due to the fact that  each filter outlet with the next filter inlet of the kaska Died arrangement is connected, the measurement signal passes through all Filter levels.

The filter that is preferably implemented with signal processors works So with a corner frequency that to the already expired Period belongs. The signal adaptation necessary for this reason is carried out by appropriate post-processing. Around Avoiding subsampling is a matter of experience values or expected acceleration values of the machine a corresponding oversampling worked.

At the beginning of each machine cycle, the digital filter Once set, the filter constants until the next Cycle are preserved. So the sampling rate is pro Cycle constant and at the filter output lie over time equidistant support points.

In the case when data reduction directly at the exit of everyone Filter stage is made and the corner frequency relative to Input sampling rate is set, all filter stages of the accompanying digital filter with the associated parameters be constructed identically. A single Koef efficiency set used for the corresponding filter levels, with only the exit lists, i.e. H. the so-called pipes, be filed separately.

For example, the cutoff frequency can be set to 1/6 of the input sampling rate can be set, then the data reduction by the factor 2 takes place.

Since the filtering described takes place in one stage, the selected based on the period of the previous cycle has been used for exact adaptation to the currently measured A corresponding interpolation is necessary for the period.  

As a result, the time window determined by the period is completely filled with equidistant values. In the case when the interpolation is carried out on 2 ⁿ points, the desired Fourier analysis can then be carried out. For reasons of runtime, real transformations are available for the FFT.

With the method presented and the associated digital tracking filter can be extremely precise in a particularly simple manner Online measurement and analysis, especially of rotary ones Measured variables take place, with the required filter particularly simply as z. B. FIR filter can be realized.

The invention is illustrated below with the aid of an embodiment game and figures are explained in more detail.

Here show:

Fig. 1 is a block diagram of the measuring system;

Fig. 2 is a schematic diagram of the digital filter symbolized by switching the cutoff frequency;

Fig. 3 shows the basic structure of the filter cascade and

Compensation FIGS. 4 to 6, the structure of the digital filter, the flow of data and measures symbolically represented at run time.

The illustration in Fig. 1 which illustrates the basic structure of a measurement system for Trim analysis of vibrations of a rotating part, is based on a rotating shaft, the vibration characteristics are to be examined. A corresponding sensor block, which also contains an analog-digital converter unit, is connected to the filter input. The adjustment values of the filter for its corner frequency are obtained from the signals of a speed sensor. A sampling rate reduction is carried out on the output side of the filter assembly, ie a corresponding post-processing takes place, followed by a Fourier analysis unit.

The speed of the shaft is recorded via a time measurement, at the timekeeping the mark revolving around the wave which starts after a predetermined number of over used brands, which corresponds to a cycle, stopped again becomes. In this way, the cycle time z. B. in seconds to be determined. By specifying the data to be recorded The number of marks passing the sensor is the resolution can be determined for the later spectrum.

Fig. 2 shows the basic structure of a filter implemented by a signal processor with a cascaded structure, with only the filter output being switched to prevent the filter from settling when the corner frequency is switched. The digital filter is preset once at the beginning of each machine cycle, the setting remaining constant until the beginning of the next cycle. This means that the sampling rate per cycle is constant and there are equidistant reference points over time at the filter output.

A plurality of digital filters are cascaded, as indicated in FIG. 3, the output of the overall filter being switched through to the associated individual filter. When filtering, stages 1 to n are run through completely, with a filter settling only at the start of the measurement, but not when the output is switched.

According to the exemplary embodiment, an FIR filter can be used, the cut-off frequency f _{g of} the filter being 1/6 of the sampling frequency f _a (f _g / f _a = 1/6). In this way, the maximum frequency is cut, ie a low-pass filter is built up, as a result of which the sampling rate of the filter output can be reduced. In the configuration shown, one output value is generated for every two input values. The filter is always called up when there are two new input values. Then a new value is produced for the pipe of the next stage, so that a hierarchical structure of the filter calls can be derived from this, which can be implemented for any filter depth. The computing time for the filtering is independent of the selected filter depth.

According to the exemplary embodiment, the filter only has to be hardware once be present on the side and can be reduced accordingly Data records are always called up again. The required code is very compact and characteristic filter values are easy to adjust.

A small, limited number of input values is for each Save filter level after each filtering of the previously going to partially replace the filter, with the oldest values be removed from memory. The length of this storage pipe depends on the desired filter key data and is approximately 7 to 41 values.

The structure of the digital filter, as illustrated in FIG. 4, has the shape of quasi-isosceles triangles, the height corresponding to the number of stages to be filtered. Each additional filter stage halves the output sampling rate and the cutoff frequency.

The structure presented has the advantage that between two Samples exactly one filter call is made, independently on the number of filter levels. This allows filters to be any Cut-off frequency implemented with corresponding data reduction become.

According to FIG. 4, no filter calculation call is provided after the 8th calculation step in the first filter level. The filter call would then be insertable for an insertable fifth level. Due to the strictly symmetrical filter structure, any filter depth can be implemented without affecting the runtimes, which is a significant advantage. Furthermore, whenever the filter has been run through completely, no further filter call is necessary between two scans. Here could z. B. cyclic initialization can be accommodated.

Fig. 5 symbolizes the data flow, a filter is called when two new input values are available and the corresponding time between two samples is available.

Because all filter-determining data regardless of the filter level exist only once, the term is by the Filter levels always the same relative to the input sampling rate. At the presented embodiment with data reduction by Factor 2 per level, the runtime doubles in each Step. By choosing coefficients symmetrical to the center the delay chain creates an exactly linear phase response. These requirements make it very simple and accurate Phase compensation or runtime compensation by shifting the timeline possible.

In addition to these filter-related runtimes, the scans according to FIG. 3 can be adapted in an equally simple manner in accordance with the given filter cascading.

The runtime compensation, such as feasible, is made clear with the illustration according to FIG. 6.

The signal processor is supported by a special subroutine the current filter level operated as a parameter. This Filter level causes the FIR program to select a specific one Access the pipe and the generated value in the next pipe to file. Before leaving this program procedure  the current filter level with the cycle time of the machine compared. In the case when the values match signal is passed on. The interrupt input of the digital signal processor used is via the Controlled the start and end of each machine cycle. The then triggered routine recognizes the period and sets an appropriate identifier for the required filter level.

It should be noted at this point that the higher-level program control can have a linear program structure as well as a procedural program structure. The procedural program structure offers advantages in program length for many filter levels. For time-critical applications, a combination of both possible program designs, e.g. B. a two-stage, procedural structure can be selected. The need for the linear structure is 2 ⁿ - 1 lines for n filter levels. The need for the procedural structure is only (n. 3) - 4 lines.

The timing of the measurement is as follows. At the start Every machine cycle generates a trigger signal which generates the period for the digital filter, whereby for the the corresponding period, the corresponding filter level is set is.

During the current period, the initial values of the Filters into the memory block of another digital signal processor written. The address of this memory block is located the second processor stuck by every new machine cycle between two possible addresses. During the current cycle, the values of the previous one Cycle transformed and saved and the respective Memory block cleared for a new transfer. The maximum time offset between measurement and storage is thus 2 cycles.

Claims (9)

1. Procedure for order analysis of vibrations of a rotating part, in particular a machine, with respect to the Amplitude and phase, with a sample of vibration signals over entire periods with a fixed input Sample rate executed and the samples obtained from digital filtering, in particular special low pass filtering, with a depending on the current speed subjected to certain filter corner frequency , the digital filter continues to be a variety of kaska Dated filter stages includes an input signal be run through one after the other, the filter cut-off frequency a subsequent filter stage in a fixed predetermined Relationship to the filter corner frequency of the previous filter stage stands. 2. The method according to claim 1, characterized in that the digital filter is designed as a low-pass filter and for that predetermined ratio the value ½ is selected. 3. The method according to claim 1 or 2, characterized in that the filter cutoff frequency at the beginning of a machine cycle in Dependence on the speed of the last previous one Revolution and for the duration of the machine cycle Is kept invariant such that the result of the filtering over the machine cycle equidistant support points of the Can be obtained. 4. The method according to claim 3, characterized in that an interpolation for filtering as signal postprocessing Correction of a deviation between the current and that for the setting of the filter corner frequency based on the rotation number is downstream.   5. The method according to any one of the preceding claims, characterized in that from the course of the filtered samples by fast Fourier transformation (FFT) obtained the associated spectrum becomes. 6. The method according to claim 5, characterized in that the fast Fourier transformation and a storage the frequency spectrum obtained after the completion of a machine cycle carried out during the subsequent machine cycle become. 7. The method according to any one of the preceding claims characterized in that the result of the filtering in a filter stage of a scan with output sample rate. 8. The method according to claim 7, characterized by a hierarchically staggered calculation of the filter values of the individual filter stages such that a calculation of the nth filter stage is carried out each time after 2 ⁿ new input values are present. 9. The method according to claim 8, characterized in that into the hierarchically staggered calculation depending on the number of filter stages used is calculation-free Sections are inserted in such a way that the signal propagation time through the filter regardless of the number of filter stages used is.