Classifications

• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
  • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
  • G05B13/0205—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system
  • G05B13/026—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric not using a model or a simulator of the controlled system using a predictor

Definitions

• the present invention relates to a state determination device and an online state monitoring / diagnosis system for determining the presence or absence of a change in the state of an object in equipment diagnosis, medical diagnosis, and the like.
• the state determination is performed assuming that the probability density function of the waveform data measured from the object of state monitoring follows a “normal distribution” in a normal state ([1] Patent Publication 2000-1711291, [2] ] Toshio Toyoda, Tomoya Niho: "Diagnosis of Rotating Machine Abnormality Using Only Normal Vibration Waveforms", Journal of the Institute of Equipment Management, Vol. 11, No. 1, 1999, p.4-11.
• the waveform data and the values of the characteristic parameters follow a normal distribution, but the waveform data and the values of the characteristic parameters do not always follow the normal distribution even in a normal state. If the state change of an object is determined by a statistical test assuming that it follows a normal distribution, an erroneous determination will result.
• the measured waveform data is converted into waveform data of a known probability distribution (for example, normal distribution), or is calculated from the waveform data.
• the state of the object is determined by statistical tests, possibility theory, information theory, and the like.
• the signal of the measured object is a pulsation signal
• the envelope waveform data of the signal from which the noise has been removed is used.
• the singular component in the signal is detected based on the normalized envelope waveform data and the periodic pulse to determine the state.
• the peak waveform data of the pulsation signal from which noise has been removed as described above is obtained, the peak waveform data is normalized, and a unique component in the pulsation signal is detected based on the normalized peak waveform data and the periodic pulse waveform data. To determine the state. 5. Effects of the Invention
• the waveform data and characteristic parameters measured for monitoring the state of an object do not always follow a normal distribution, the presence or absence of a change in the state of the object is statistically determined by assuming that these waveform data or characteristic parameters follow a normal distribution. Judgment by a test or state prediction leads to large errors.
• the measured waveform data is converted into waveform data of a known probability distribution (for example, normal distribution), or the characteristic parameter calculated from the waveform data is converted to a known probability distribution (for example, normal distribution).
• the state of the object is determined by statistical tests, possibility theory, information theory, and so on. Therefore, the method of the present invention is higher than the case where the accuracy of state determination or state prediction follows the conventional normal distribution.
• the conventional judgment method (characteristic parameter, spectrum, probability, etc.) Density function) is difficult to detect.
• the envelope waveform data of the pulsation signal from which noise has been removed is obtained, and a singular component in the signal is detected based on the normalized envelope waveform data and the periodic pulse to determine the state.
• the main signal processing can be realized by hardware, and the load on the numerical processing device (computer) is small, so that abnormality detection can be realized in real time. Also, since the waveform data in the normal state is not used as a reference, even if the pulsation cycle changes, the effect on the detection and judgment results is small. 6. Best mode for carrying out the invention
• Feature parameters used for state determination include time-domain and frequency-domain feature parameters.
• Some frequency domain feature parameters are defined in (Ref. 1).
• Peng Chen, Toshio Toyoda Self-Organization of Feature Parameters in Frequency Domain by Genetic Programming, Transactions of the Japan Society of Mechanical Engineers (C), Vol. 65 No. 633, p p.1946-1953, 1998.
• the feature parameters in the time domain will be described in detail.
• the characteristic parameters in the time domain used to determine whether or not the state of the object has changed are as follows.
• the extracted waveform data X (t) is normalized by the following equation.
• ⁇ 'i is the discrete value of X (t) after A / D conversion
• S are the average value and standard deviation of x' i, respectively.
• N-1 is the standard deviation.
• -l (5) 2 ( ⁇ -1) ⁇ 3
• ⁇ P is the standard deviation value of the maximum value.
• Equations (2) to (13) are conventional feature parameters, but in order to easily perform high-speed calculations by numerical calculation, the “interval feature parameters” are expressed by equations (14) to (21). A new proposal is made.
• X i ⁇ kak can be set arbitrarily.
• k 0.5 1 2.
• kl is the average value of Xi .
• ⁇ hl is the average value of x h .
• higefact is the frequency with which the waveform crosses the ⁇ level per unit time.
• the measured waveform data is not normalized as in equation (1).
• Peak average value of the absolute value of the waveform data ⁇ ⁇ 1 " ⁇ (2 4)
• N P is the total number of peak values. Peak effective value of the absolute value of the waveform data: (25)
• Pd4 ⁇ "Note that many other feature parameters can be defined in addition to the above feature parameters. When applying this method, first try using the above feature parameters. If the effect of state identification is not good, further What is necessary is just to define additional characteristic parameters.
• the measured waveform data is represented by, and the characteristic parameter calculated from the waveform data is represented by.
• the specified known probability distribution function is ⁇ , then x * i or ⁇ can be transformed into a random variable Xi or Pi according to ⁇ using the following equation.
• F xi (x * i) and F Pi () are the cumulative probability distribution (or cumulative frequency distribution) of x * i and ⁇ , respectively, and ⁇ is the inverse function of ⁇ .
• ⁇ is a normal distribution, Weibull distribution, exponential distribution, gamma distribution, etc.
• the original waveform data x * i is divided into four types as shown in Fig. 1. That is, the data Xi + larger than the average value, the data-smaller than the average value, the absolute value data i xi
• the feature parameter p * i is It is calculated by one of X i + and Xi-, I Xi I and X * iA .
• the probability density function f (t) of the normal distribution is expressed by the following equation.
• ⁇ - 1 is the inverse function of ⁇
• ⁇ ⁇ 0 and ⁇ ⁇ ⁇ ) are the standard deviations of x * ik and p * ik converted to normal distribution, respectively, and can be obtained by the following formula.
• the probability density function (or frequency distribution) of the waveform data x * ik and the characteristic parameter p in state k is d) and f (p * ik), respectively, and the probability distribution function (or cumulative frequency distribution) ) Are F xk (x * ik ) and F Pk (p * ik ).
• the waveform data x * ik and the feature parameter p * ik are converted into a normally distributed random variable by the following formula.
• S xk and S Pk are standard deviations of x * ik and p * ik , respectively, and / xk and / z Pk are average values of x * ik and p * ik , respectively.
• the average value according to the normal distribution is obtained by the following equation.
• ⁇ ' the number of p * ik in the j-th set. Since follows approximately the normal distribution, state determination and state prediction are performed using ⁇ .
• the waveform data x * ik is transformed into the normal probability distribution waveform data by formulas (29), (33), (35), and (37), ⁇ and X 'ik. , 'Ik, / i xik, X "ik, ⁇ are called" normally distributed waveform data ".
• ⁇ piko and p ' are the feature parameters p * i converted to normal random variables by Eqs. (30), (34), (36), and (38).
• P 'ik, ⁇ Plk, p'' are referred to as "characteristic parameters of the normal distribution" c
• the characteristic parameters of the normal distribution obtained in state k and state y are p ik and Piy , respectively.
• the average value ⁇ and standard deviation S of J Pj are calculated by the following formula.
• VJ (44) holds, the significance level is determined to be "/ i ik and / Z iy are not equal."
• ta / 2 (J-1) is the percentage point of the probability density function of the t distribution with J-1 degrees of freedom with respect to the lower probability ⁇ / 2.
• F a / 2 (J-1, Jl) is the percentage point of the probability density function of the F distribution with J-1 degrees of freedom for the lower probability a / 2.
• Equation (44) or Equation (45) When the significance level a is changed, the degree of state change from state y to state k is determined by confirming whether equation (44) or equation (45) is satisfied.
• Table 1 shows an example of determining the degree of state change based on the significance level a. In the case of equipment diagnostics, if state k is normal, state y is normal, cautionary, or dangerous, as shown in Table 1. "(shed 2), and" danger "(a 3) set can be assayed as. That is, if Equation (44) or Equation (45) does not hold at the time of ⁇ , it is determined to be “normal”. Equation (44) or Equation (45) Is determined as “caution” if ⁇ ; 2 is satisfied, and “danger” is determined if ⁇ 3 is satisfied.
• the numerical range of ⁇ in Table 1 is an example, and is determined by the importance of equipment.
• the result of the determination follows the determination result of the feature parameter that indicates the largest change in state. For example, when making a judgment using three feature parameters ⁇ ⁇ 2 and ⁇ 3, if the judgment result of Pl is “Caution”, the judgment result of ⁇ 2 is “Normal”, and the judgment result of ⁇ 3 is “Danger”, The final judgment result is “dangerous”.
• the average value of the characteristic parameter of the normal distribution obtained from the waveform data measured at the reference time point is / ii. If the average value of the characteristic parameter of the normal distribution obtained from the waveform data measured at other times is ik , the confidence interval of ⁇ 0 is given by the following equation. (46) where / 2 (J-1) is the percentage point of the probability density function of the t distribution with J-1 degrees of freedom relative to the lower probability ct / 2. Si. Is Pi obtained from waveform data. Is the standard deviation of Is A with probability 1- if it is within the interval shown in equation (46). There is no difference. / ii. Of 99
• the ° / 0 confidence interval is approximately as follows when J> 10:
• the confidence interval of obtained from the measured waveform data can be obtained by the following equation. ⁇ ,. ⁇ 3S 'VJ (48) where S ik is the standard deviation of p ik obtained from the waveform data.
• the state is determined based on whether or not ⁇ 3 ′ (5 1) k is within these sections.
• the probability distribution function P k ( P i ) of the probability density function f k (p of P i is calculated by Equation (5 2).
• the probability distribution function can be obtained for any probability distribution of pi. If pi follows a normal distribution, the N-stage possibility distribution function p k (p is (See Reference 5),
• the probability of the characteristic parameter Pi of the normal distribution obtained in state k and state y is P k (pO and P y ( ⁇ ), and the value of the characteristic parameter of the normal distribution obtained in state y is Given P'i, the possibility that state y is the same as state k is Required.
• the values of i and j in ⁇ Soil, ⁇ Shi 'S!' (56) are determined by user input.
• the probability distribution function of the normal state is P k (p
• the probability distribution function of the attention state is p cl (pi) and pc 2 (Pi)
• the probability distribution function of the dangerous state is (pi)
• p d2 (pi) The probabilities of “normal”, “caution”, and “danger” obtained at the time of actual identification are displayed as shown in Fig. 2. Also, when it is judged as “danger” It is also possible to issue a warning.
• f P be the probability density function of the characteristic parameter of the normal distribution in the reference state of the object. ( Pi ), and let the probability density function of the characteristic parameter of the normal distribution other than the reference state be (Pi).
• the state other than the reference state is called the “test state”. Whether or not the test state is the same as the reference state can be determined by the following ("Kullback-Leibler Information (KI)" and "Information Divergence (ID) J". ⁇ ' ⁇
• Kip and ID status determination method by P is described in detail in [1 0], it is omitted here. ([10] Nobuyoshi Liu, Toshio Toyoda, Peng Chen, Fang Feng, Tomoya Nibo: “Abnormal Diagnosis of Rotating Machinery by Information Divergence", Journal of the Japan Society of Precision Engineering, Vol.66, No.1, 2000, .157 -162.)
• the feature parameter integration method includes the principal component analysis method and the KL expansion method.
• the new feature parameters obtained by the feature parameter integration method are called “integrated feature parameters”.
• integrated feature parameters an example of the principal component analysis method will be described.
• ⁇ ⁇ ⁇ ⁇ + ⁇ ⁇ 2 +-+ a lm p m
• Each principal component Zl to Z » is also called“ integrated feature parameter ”.
• the correlation matrix is determined as follows:
• Figure 3 shows the acceleration waveform data of the vibration measured when a rotating machine is in a normal state (Fig. 3 (a)) and in a rotating shaft misalignment state (Fig. 3 (b)).
• Equation (64) is small in a normal state and large in an abnormal state.
• the waveform data x * i measured from the object are Waveform data x ' ik with more normal distribution.
• the waveform data ik of the normal distribution can be determined by the above-described state determination method using the characteristic parameter of the normal distribution. This method uses normal distribution waveform data ⁇ xiko and X 'ik. Applicable to
• Test state Whether or not is in the same state as the reference state can be determined by the following “Kullback-Leibler Information (KI) J” and “Information Divergence (ID)”.
• Fig. 6 (a) shows the waveform data of the vibration acceleration measured in the normal state of a rotating machine.
• Figure 6 (c) shows waveform data measured when the same rotating machine is unbalanced.
• the state may be determined by the mean and variance tests shown in Equations (44) and (45).
• the feature parameters are obtained by Eqs. (2) to (25), and statistical tests, possibility theory and integration of feature parameters are obtained.
• the state can also be determined by the method.
• the state of the measurement target can be predicted using the conventional state prediction method ([13] Masumi Ishikawa, Hiromichi Muto: Prediction method, measurement and control, 1982.3. [14] Ogawa, M .: Time series analysis and storage prediction, Bull. Math. Stat., 8, 8—72, 1958. : Statistics for prediction, Koyo Shobo, 1987.)
• Figure 7 shows the method of state prediction. At each measurement time (after converting the characteristic parameter values or principal component values measured at Xl to x to follow the normal probability distribution, the prediction curve and its confidence interval are obtained by regression analysis, and at the intersection with the "life limit" Find the “shortest life”, “average life J” and “longest life”.
• Figure 8 shows an example of waveforms measured eight times while a rotating machine changes from a normal state to an unbalanced state. Measurement 1 is normal and measurement 8 is the heaviest unbalanced condition.
• Figure 9 shows the average value of the characteristic parameters in each measurement, obtained by Eqs. (2) to (13).
• the feature parameters have monotonically increasing values as the degree of the abnormal condition increases (for example, p 4 , If there is p 5), the characteristic parameter values monotonically decreasing (e.g., ⁇ 2, ⁇ 8, ⁇ 9, ⁇ ») also.
• there are some feature parameters eg, p., ⁇
• ⁇ whose values are almost unchanged even when the degree of the abnormal state becomes heavy.
• the life expectancy should be selected by selecting a characteristic parameter whose value is monotonically increasing (or monotonically decreasing). Also, when performing life expectancy prediction using the principal components, only the characteristic parameters whose values are monotonically increasing (or only the characteristic parameters whose monotonically decreasing values) should be selected, and the life prediction should be performed by obtaining the principal components.
• Figure 16 (a) shows the measurement and processing flow for implementing the above-described method of converting waveform data and feature parameters into a normally distributed random variable, the state determination method, and the state prediction method.
• the circuit of the waveform data measurement and state determination device for realizing Fig. 16 (a) is shown in Fig. 17.
• FIGS. 12 (a) and 13 (a) show examples in which a large number of abnormal peaks continuously occur.
• Figure 15 shows the flow of online singularity detection and state determination for such local abnormalities (singular components) that occur in the pulsation signal.
• the rotation pulse signal is also called a periodic pulse signal, and is used to determine the timing of each peak value of the pulsation signal.
• the power cutoff frequency fL of the low-pass filter is determined as follows.
• n is the number of rotations of the shaft (r P m)
• z is the number of peaks per rotation (peak / 1 rotation)
• fO is the margin frequency (> n / 60, and observe the noise removal effect after filtering. ).
• envelope waveform data after low-pass filtering or peak waveform data, or moving average waveform data.
• Fig. 10 (c) (d), Fig. 11 (c) (d), Fig. 12 (c) (d), Fig. 13 (c) (d) show the time series with noise removed by a low-pass filter. Waveform data and spectrum.
• FIG. 10 (e), FIG. 11 (e), FIG. 12 (e), and FIG. 13 (e) show rotation pulse waveform data.
• FIGS. 10 (f) and 12 (f) show the envelope waveform data.
• Figures l l (f) and 13 (f) show peak waveform data.
• an abnormal location can be identified by the correspondence between the location where the absolute value Ix (t) I is larger than 2 ⁇ and the rotation pulse waveform data.
• envelope waveform data and peak waveform data is shown above.
• u (t) is used instead of x (t).
• the singular point of the pulsation signal can be detected in the same manner as in the above procedure. If the cut-off frequency (f.) Is determined by equation (67), the moving average score (M) can be determined.
• Fig. 16 (b) shows the flow of measurement and processing to realize the method for detecting and determining the singular component of the pulsation signal described above.
• Fig. 18 shows the circuit of the pulsation signal measurement and state determination device for realizing Fig. 16 (b).
• Figure 16 shows the processing flow of the status judgment device or the online status judgment system.
• noise was removed from the measured waveform data, the characteristic parameters were obtained, and the characteristic parameters were converted to normally distributed random variables.
• the presence or absence of a state change is determined by integration.
• These processes can be realized by a computer or a dedicated device.
• the detection of singular components from the pulsation signal and the state determination can be realized by a low-pass filter and the envelope (or peak value) by hardware.
• the arithmetic unit or computer determines the normalized x (t), I x (t)> o I, The results can be displayed in real time. 7.
• Figure 1 shows four types of measured waveform data: data Xi + larger than the average value, data X smaller than the average value;-, absolute value data after normalization by equation (1), and total waveform data X * M It is a graph which shows the example divided into.
• FIG. 2 is a graph showing an example of the probability distribution function.
• FIG. 3 is a graph showing an example of vibration waveform data in a normal state and a rotational axis misalignment state.
• FIG. 4 is a graph showing the result of principal component analysis of vibration waveform data in a normal state.
• FIG. 5 is a graph showing principal component analysis results for vibration waveform data in the state of a rotational axis misalignment.
• FIG. 6 is a graph showing an example of vibration waveform data in a normal state and an unbalanced state.
• FIG. 7 is a graph showing a state prediction method.
• FIG. 8 is a graph showing an example of vibration waveform data when changing from a normal state to an unbalanced state.
• FIG. 9 is a graph showing characteristic parameter values of the waveform of each measurement.
• FIG. 10 is a graph showing an example of signal processing when one peak is missing due to envelope waveform data.
• FIG. 11 is a graph showing an example of signal processing when one peak is missing due to peak waveform data.
• FIG. 12 is a graph showing an example of signal processing in the case of a large number of peak abnormalities based on envelope waveform data.
• FIG. 13 is a graph showing an example of signal processing in the case of a large number of abnormal peaks based on peak waveform data.
• FIG. 14 is a graph showing how to obtain moving average waveform data.
• FIG. 15 is a flowchart showing the flow of the pulsating signal specific component detection and abnormality diagnosis processing.
• FIG. 16 is a flowchart showing a processing flow of the state determination device or the online state determination system.
• FIG. 17 is a circuit diagram showing an example of a circuit of the signal measurement and state determination device, and the symbols in the figure are as follows. .
• FIG. 18 is a circuit diagram showing an example of a circuit of the pulsation signal measurement and state determination device, and the symbols in the figure are as follows.

Landscapes

• Engineering & Computer Science (AREA)
• Software Systems (AREA)
• Artificial Intelligence (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Evolutionary Computation (AREA)
• Medical Informatics (AREA)
• Health & Medical Sciences (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Automation & Control Theory (AREA)
• Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
• Complex Calculations (AREA)
• Testing Or Calibration Of Command Recording Devices (AREA)

Priority Applications (1)

Applications Claiming Priority (2)

Publications (1)

Family

ID=32820506

Family Applications (1)

Country Status (3)

Cited By (10)

Families Citing this family (6)

Citations (3)

Family Cites Families (1)

• 2004
  • 2004-01-13 WO PCT/JP2004/000163 patent/WO2004068078A1/ja active Application Filing
  • 2004-01-13 CN CNB2004800059264A patent/CN100386603C/zh not_active Expired - Fee Related
  • 2004-01-13 JP JP2005504665A patent/JPWO2004068078A1/ja not_active Withdrawn

Patent Citations (3)

Non-Patent Citations (2)

Also Published As

Similar Documents

Legal Events