RU2687177C1 - Method of detecting and classification a signal in control systems - Google Patents

Abstract

FIELD: monitoring systems; data processing.

SUBSTANCE: use for signal detection and classification in control systems. Summary of invention consists in the fact that detection and classification of signal are based on signal detection mechanism using method of accumulation and determination of characteristics of random signal, wherein the signal after each iteration is measured using an identification tester, obtained estimates of the identification parameter are compared with a predetermined threshold value, when it is reached, the iteration is terminated.

EFFECT: simple algorithm of operation and high accuracy of signal classification.

1 cl, 7 dwg, 1 tbl

Description

The invention relates to the field of signal processing methods in control systems of pipelines and lengthy structures.

A known method for detecting defects in pipelines [US Pat. 2439551 of the Russian Federation IPC G01N 29/04. Method for detection of defects in pipelines / Alekseev, SP and etc.]. The method consists in installing acoustic sensors, recording acoustic oscillations, determining the location of a defect in a controlled section of the pipeline and recording it, while acoustic oscillations are fixed on a mode of radial oscillations of circular hollow cylinders on a transverse piezoelectric effect in This collector electrodes deposited on the side surfaces of the hollow cylinder, inside the pipeline is placed a diagnostic module, also equipped with acoustic sensors, through determining the nonlinear properties of the controlled medium by determining the function relating the response of the medium pressure with the pressure disturbance, the emitting and receiving transducers mounted on the distances l / l _σ = 1 from each other emit acoustic signals at frequencies of 140 and 150 kHz, determined normalized histograms probability density for each signal, by a polynomial approximation, determine the analytical expression for each histogram, calculate the nonlinearity function and the values of the moment functions, which are akterizuyut changing the shape of the normal distribution law, to change the shape of the distribution law define foreign inclusions in a controlled environment [1].

The essence of the proposed technical solution is illustrated in the drawing (Fig. 1). The block diagram of the analogue shown in FIG. 1, includes: microprocessor 1, buffer device 2, external memory 3, control device 4, adjustable preamplifier 5, power amplifier 6, radiator 7, receiver 8, band-pass filter 9, preamplifier 10, ADC 11, normalization device 12, device determine histogram 13, block 14 for calculating the coefficients of a polynomial, device 15 for estimating the accuracy of the approximation, device 16 specifying the degree of approximating polynomial, decision-making device 17, block 18 for dividing, integrating device 19, block 20 for extracting the coefficient in the first two terms of the expansion 20, the unit 21 calculating the quotient.

In the microprocessor 1, radiating signals and operation parameters are generated. Through the buffer device 2 information from the microprocessor 1 is fed to the external memory 3 and the control device 4. The control device 4 controls the operation of the external memory and adjustable preamplifier 5. Power amplifier 6 amplifies the signal and delivers it to the emitter 7.

The signals from the receiver 8 are fed to the band-pass filter 9 and the pre-amplifier 10, which form a pre-processing unit. The signals are digitized by means of ADC 11. Next, the digital signal is fed to the normalization device 12 and the histogram determination device 13. The device 16 sets the degree of approximating polynomial and controls the operation of the polynomial coefficient calculation unit 14. Next, the approximation accuracy 15 is evaluated and the data is transmitted to the decision device 17 which is controlled by the control unit 4. In this case, if the accuracy does not satisfy the specified threshold, the degree of the polynomial increases. The increase occurs until the accuracy of the approximation is satisfactory. In block 18, the expression for the probability density of the emitted signal, which is read from the external memory 3, and the resulting probability density are divided in block 14.

The integrator 19 represents the obtained result on the coefficient extraction block with the first two terms of the decomposition 20, and in block 21 the quotient is calculated. The result obtained through the buffer element is displayed on the microprocessor 1.

The signals from the ADC 11 are recorded on the hard disk of the microprocessor 1. Each report is encoded in 14-bit format. The algorithm receives data on the basis of which the normalized probability density histograms are determined for each signal. Then, using a polynomial approximation, an analytical expression of the probability density for each histogram is determined. Depending on the study (detection of defects in pipelines or searching for leaks of the product being transported), the nonlinearity function or the values of moment functions that characterize the change in the shape of the distribution law (in the case of detection of foreign inclusions in the medium) are calculated.

The first disadvantage of this analogue method is the absence of a classifier, which would allow to attribute the received accumulated signal to a certain form of distribution. The second disadvantage is the intersection of the confidence intervals of the recorded moments of distributions for signals of different forms, which does not allow to guarantee the belonging of the received signal to one particular form (reference).

Closest to the claimed method [2], indicated in the patent prototype [US Pat. 2016135127 Method for detecting and classifying changes in the parameters of the shell of the pipeline and its environment / Epifantsev BN, Komarov VA, Nigrey NN, Ischak E.R.]. FIG. 2 shows the layout of the prototype.

In the prototype, vibroacoustic signals excite successive impacts on the transmission medium at intervals exceeding the correlation interval of the vibroacoustic noise existing in it, readings of recorded reactions to each impact at the other end of the monitored section are summed up with previously obtained similar readings, the resulting signal modulus is normalized and taken as distribution density of time intervals of samples from the beginning to the end of the signal generated in the adder. At each reception and accumulation of the impact, estimates of the mean, standard deviation, asymmetry and excess are calculated by their distribution, and then their regression lines are constructed. The first patent classifier estimates the distances and angles between all pairwise regression line combinations in order to classify the current impact to a specific form. The distance between the regression lines is determined by the formula (1) as the modulus of the difference between the average values of these lines:

where d is the distance between the regression lines, p1 - the values of the regression line of the first parameter, p2 - the values of the regression line of the second parameter.

The angle is calculated based on the scalar product of vectors, which are regression lines (2):

- вектор линии регрессии первого параметра,

- вектор линии регрессии второго параметра.where α is the angle between the regression lines,

- the vector of the regression line of the first parameter,

- vector of the regression line of the second parameter.

On the basis of the calculated parameters d and α, a decision is made to assign the signal to a specific form. For this, it is necessary to have an appropriate database of reference parameter values, which requires individual calibration.

Summarizing the properties of the analogue and the prototype, it can be stated that they use indirect methods of measuring the distribution, which leads to a complication of the structure and algorithms of signal processing.

The aim of the patented method is to simplify the processing algorithms and increase their efficiency through the use of a direct method for measuring the shape of distributions.

The acoustic method of controlling extended objects is based on the analysis of random signals. When using the S-tester as an object control method, the received pulse can be attributed to a certain class of signals using the scale of the identification parameter conversion to a certain type of signal. The essence of the proposed solution is to use a special tool - an identification tester (IT), the static characteristic of which (for example, IT, the so-called S-type) is described in table 1.

Table 1 is a scale that relates the ordinal numbers (Rank) of samples with output numerical parameters (Mean S) IT and the qualitative characteristics of the following types of random signal distributions: two-modal (2mod), arcsine (asin), uniform (even), trapezoidal (trap ), triangular (simp), normal (gaus), two-sided exponential (lapl) and Cauchy (kosh). The initial count (const) has a zero value and refers to constant signals in time. The names of the distributions are taken from the dictionary of the names of random signals taken in the field of statistical measurements [3].

The bottom line of Table 1 presents an example of measuring a certain signal, whose readings for the S-tester were Sy = 24.36. Interpolation of these readings suggests that the signal under investigation has a distribution close to normal (gaus) in accordance with the obvious rule:

Name Sy≈Id ^* [min (ΔS _yg ; ΔS _yl )],

where ΔS _yg = abs (Sy-S _g ) = abs (32-24,36) = 7,64; ΔS _yl = abs (Sy-S _l ) = abs (24.36-15) = 9.36. It follows that: Name Sy≈Id ^* [min (7.64 / gaus; 9.36 / lapl)] = GAUS. Here, the symbol Id * N denotes the operation inverse to the operation Id [.] Of identifying the distribution on the scale.

Our studies have established the following fundamental properties of the identification scale (table 1):

1) the scale covers the full range of all possible existing and conceivable forms of distributions;

2) the values of the samples do not depend on the shift parameters and the scale of the analyzed signals;

3) despite the limited number of samples (9), the accuracy of identification of distributions can be improved by interpolating the value of the identification parameter (S) between adjacent samples.

Thus, IT can be interpreted as a measuring transducer with a scale similar to that of a conventional analog indicating device, but supplemented with category (quality) marks.

Therefore, IT is a universal classifier that can be used to recognize any selective signal implementations.

The principle of the S-tester is based on measuring the steepness of the ranked function of the signal in its central part relative to the median. FIG. Figure 3 presents examples of the ranked functions of sample realizations of random signals with Cauchy normal, uniform, and two-modal distributions. It can be seen that the mean steepness (Mean S) in the central area gradually increases in the direction from Cauchy to a two-distance distribution, which serves as the basis for digitizing the form (names) of the distributions.

The algorithm for calculating the identification parameter S and its evolution in the system is as follows.

1) Selective signal implementation is ranked in ascending order.

2) From the ranked function, by uniformly sampling, 9 values of C (1), C (2), ..., C (8), C (9) are selected, and the fifth value of C (5) should coincide with the median of the sample under study.

3) After that, the identification parameter S is calculated by the formula (3).

where C (i) is the i-oe value of the ranked function. In this case, if the noise component prevails in the signal, then the IT readings will be in the first half (at the beginning) of the scale.

4) With each iteration of accumulation, the influence of the noise component will decrease, and the regular component will increase, which leads to an increase in the value of S.

5) In this case, the value of the identification parameter S will tend to one of the distributions corresponding to the regular component. For example, if the probe pulse has a rectangular shape, then the IT reading will tend to S = 100 (2mod distribution). If the probe pulse has a triangular shape, then the IT reading will tend to S = 75 (even the distribution, Table 1).

6) The accumulation iterations stop when the IT readings do not reach a certain threshold level (S1).

To verify the correct functioning of the proposed algorithms, a virtual device was developed (in the Lab VIEW simulation environment) with a virtual device whose block diagram is shown in FIG. 4, 5 and 6 (formalized block diagram).

The control panel (Fig. 4) of the virtual instrument consists of three displays that display the graph of the probing signal (Signal Graph), the graph of the output signal (Out Graph) and the dependence of the current value of the identification parameter (S) on the iteration number. The controls are the switch (DistrName) of specifying the type of random signal (noise), the switch (SignalName) of specifying the shape of the probing signal, the setting unit (#Sample) of the number of counts of the probing signal, the setting unit (Ampl) of the probing signal. To display the values of measured parameters, the IdP Array window is used. A judgment identifying the output waveform is displayed in the String window. FIG. 4 shows an example of the operation of the system when:

1) has a rectangular shape (SignalName = squ);

2) white noise with a normal distribution is set (DistrName = gaus);

3) the number of samples of the probing signal is 100 (# Sample = 100);

4) the signal-to-noise ratio is 1: 5, i.e. the effective noise value is 5 times the effective value of the probing signal.

At the same time, the following estimates of the output parameters were obtained:

1) the number of iterations of the accumulation mode = 1420 (1st row of the IdP Array window);

2) the ideal value of the probe pulse = 100 (2nd line of the IdP Array window);

3) threshold value of the shape of the output signal (probe pulse + residual noise) = 90, 0431 (3rd row of the IdP Array window);

4) real signal-to-noise ratio = 0.2087 (4th line of the IdP Array window);

5) the control value of the probe pulse shape parameter = squ (window String), the coincidence of which with the probe pulse value (SignalName = squ), indicates the correctness and accuracy of the system operation.

FIG. 5 of the structure of the program code of the system are shown: main, executable module, module of setting parameters, module of measured parameters and module of decision making.

The executable module is designed as a For-Loop type loop, inside which are: a probing signal generator, a noise generator, a signal adder, a cumulative register, two S-type identification testers, a threshold device, and a signal-to-noise ratio meter. The probe signal generator is multifunctional and allows you to set 5 types of pulsed and periodic signals. The noise generator allows you to set 14 types (by distribution) of random signals.

The module of setting parameters is located to the left of the main module and includes controls for the model.

The module of measured parameters is located to the right of the main module and includes an S-type identification tester, a dividing device and a display device for output IdP Array parameters.

The decision module is located on the right-bottom of the main module, consists of 2 comparators and 2 switches, with the help of which the inference function is implemented (output judgment).

The program code of the system operates in the “Specified error” mode, in which the output information is the number of accumulation cycles. However, the program code could be easily adapted to the “Specified number of accumulation cycles” mode, in which the output information is the error in estimating the shape of the probing signal.

FIG. 7 shows an example of pulses received from a pipeline. The ranked amplitude functions for the healthy state 7 (a) and defective 7 (b) are different, which makes it possible to classify them. The normal state of the pipeline is described by the Laplace distribution (S≈20), and the defective state - by the Gauss distribution (S≈28).

LITERATURE

1. Pat. 2439551 Method for detection of defects in pipelines / Alekseev, SP and etc.

2. Pat. 2016135127 Method for detecting and classifying changes in the parameters of the shell of the pipeline and its environment / Epifantsev B.N., Komarov V.A., Nigrey N.N., Ishchak E.R.

3. Encyclopedic dictionary on electronics, optoelectronics and hydroacoustics / V.G. Dozhdikov, Yu.S. Lifanov, M.I. Saltan; by ed. V.G. Dozhdikova. - Ed. 3rd, pererabot. and add. - Moscow: Energy, 2008. - 611 p.

Claims (2)

1. Method of detecting and classifying a signal in control systems based on the mechanism of detecting a signal using the method of accumulating and determining the characteristics of a random signal, characterized in that in order to simplify the algorithm of operation and improve the accuracy of classification, the signal after each iteration is measured using an identification tester , the obtained estimates of the identification parameter are compared with some predetermined threshold value, upon reaching which the iterations stop. 2. A method for detecting and classifying a signal in control systems according to claim 1, which allows to recognize a defective and defect-free state of an object, characterized in that the identification tester, made in the form of a software module, measures the average steepness of the central portion of the ranked signal function, which is logically related to type (form) of distribution.

Images