Fault early warning method and system based on stress wave pulse train peak separation demodulation

Abstract

This invention relates to a fault early warning method and system based on stress wave pulse train peak segmentation and demodulation, comprising: acquiring the stress wave signal of the device; performing envelope detection processing on the acquired stress wave signal to obtain a stress wave envelope detection signal; performing peak segmentation and demodulation processing on the stress wave envelope detection signal using a moth-and-flame algorithm based on a Gaussian mixture model to generate a stress wave pulse signal after peak segmentation and demodulation; compressing the generated stress wave pulse signal to generate a stress wave pulse compression matrix; configuring a warning level threshold matrix THE of the device; and, during fault early warning, performing a positive correspondence comparison between the stress wave pulse compression matrix and the warning level threshold matrix THE to generate fault early warning status information based on the positive correspondence comparison status. This invention improves the accuracy and applicability of stress wave-based fault early warning.

Description

Technical Field

[0001] This invention relates to a fault early warning method and system, and more particularly to a fault early warning method and system based on stress wave pulse train peak splitting and demodulation. Background Technology

[0002] Stress waves are the propagation form of stress and strain disturbances. In deformable solid media, mechanical disturbances manifest as changes in particle velocity and corresponding changes in stress and strain states. When changes in stress and strain states propagate in the form of waves, stress waves are formed.

[0003] Common vibration sensors typically detect a wide frequency range (e.g., 0-15000Hz) by using a flat frequency response (e.g., 100mV / g). Therefore, they are not sensitive to slight changes in machine friction in the early stages of a fault. Only after the fault worsens and the vibration level is excited to be significantly higher than the background can the vibration sensor detect the abnormality.

[0004] Stress wave sensors are a type of piezoelectric acceleration vibration sensor, but they have a very narrow frequency range (e.g., 30,000-45,000 Hz) and a very high frequency response (e.g., resonance). The frequency response gain within the resonance range has been selected and controlled, making them very sensitive to small defects on mechanical surfaces and exhibiting good consistency.

[0005] In the early stages of equipment component failure, the failure frequency is high but the noise is low. When using stress wave analysis, stress wave characteristics can be separated, detected, and analyzed from the very low frequency range of mechanical vibration and audible noise. Therefore, it plays an unparalleled role in monitoring damage to equipment such as gears and bearings.

[0006] However, the stress wave characteristics obtained by existing methods are stress wave pulse train signals coupled by multiple events such as friction, mechanical impact and dynamic load. They are characteristic signals of multiple stress wave pulse coupling and superposition, which cannot truly reflect the essence of a single friction or single mechanical impact event. The characteristics of the pulse peak are distorted, which affects the accuracy of equipment fault early warning and fault diagnosis. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a fault early warning method and system based on stress wave pulse train peak demodulation, which improves the accuracy and applicability of stress wave-based fault early warning.

[0008] According to the technical solution provided by this invention, a fault early warning method based on stress wave pulse train peak segmentation and demodulation is provided, the fault early warning method comprising:

[0009] For a device awaiting fault warning, acquire the stress wave signal of the device;

[0010] The acquired stress wave signal is subjected to envelope detection processing to obtain the stress wave envelope detection signal;

[0011] For the stress wave envelope detection signal, the moth flame algorithm based on the Gaussian mixture model is used to perform peak splitting and demodulation processing of the overlapping peaks of the pulse train, so as to generate the stress wave pulse signal after peak splitting and demodulation processing.

[0012] The generated stress wave pulse signal is compressed to generate a stress wave pulse compression matrix, wherein the stress wave pulse compression matrix includes stress wave energy, pulse energy, and maximum peak height.

[0013] Configure the warning level threshold matrix THE of the device, wherein the warning level threshold matrix THE includes stress wave energy warning threshold, pulse energy warning threshold and maximum peak height warning threshold;

[0014] During fault warning, the stress wave pulse compression matrix is ​​compared with the warning level threshold matrix THE in a positive correspondence, so as to generate fault warning status information based on the positive correspondence comparison status.

[0015] When comparing the stress wave pulse compression matrix with the warning level threshold matrix THE in a positive correspondence, we have:

[0016] The stress wave energy, pulse energy, and maximum peak height are compared one-to-one with the stress wave energy warning threshold, pulse energy warning threshold, and maximum peak height warning threshold, respectively. For each comparison, a comparison information is generated.

[0017] When at least one comparison information matches the warning threshold, the generated fault warning status information is the fault warning status.

[0018] The acquired stress wave signal is filtered using a bandpass filter to generate a filtered and denoised stress wave signal.

[0019] The stress wave filtered and denoised signal is subjected to envelope detection processing to obtain the stress wave envelope detection signal.

[0020] When performing peak-splitting and demodulation of overlapping peaks in a pulse train using the moth flame algorithm based on a Gaussian mixture model, the following steps are included:

[0021] The moth flame algorithm is initialized, wherein during initialization, the population size q of the moths, the maximum number of iterations T0, and the shape constant η of the spiral are configured. For any moth in the moth population, the position information θ of the moth includes N sets of position parameters, where N is the number of peaks obtained based on the stress wave envelope detection signal, and each set of position parameters corresponds to a Gaussian mixture model.

[0022] After initialization, a moth population is randomly generated, and the moth-flame algorithm iterative processing steps are performed based on the randomly generated moth population.

[0023] In each iterative processing step, the fitness value f(θ) of each moth in the moth population is determined based on the position information θ of each moth in the moth population. After determining the fitness value f(θ) of each moth in the moth population, the fitness function f is updated based on the configured flame number. n Update the number of flames, update the position of the moths and the position of the flames using a logarithmic spiral function, and accumulate the number of iterations;

[0024] Repeat the above iterative processing steps until the number of iterations matches the maximum number of iterations T0; once the number of iterations matches the maximum number of iterations T0, determine the optimal moth location within the moth population, where,

[0025] Based on the determined optimal moth position, a set of position parameters within the optimal moth position are used as the weight, mean, and standard deviation of each Gaussian peak, and a corresponding force wave pulse signal is generated based on all Gaussian peaks.

[0026] For any moth with fitness value f(θ), then:

[0027]

[0028] Where n is the number of data points within the stress wave envelope detection signal X(t), x(ti) is the data point corresponding to time ti within the stress wave envelope detection signal X(t), P(x(ti)|θ) is the probability that data point x(ti) belongs to a Gaussian mixture model corresponding to a position parameter, and α k For the k-th weight within the location information θ, u k f is the k-th mean value within the location information θ. σ (k) represents the kth standard deviation within the location information θ.

[0029] In the iterative processing step, the fitness values ​​f(θ) of all moths are sorted in ascending order, and the sorted fitness values ​​f(θ) and the corresponding moth position information θ are configured as the spatial position of the flame.

[0030] Update function f for the number of flames n Then we have:

[0031]

[0032] Updating the moth's position using a logarithmic spiral function, we have:

[0033]

[0034] Where T is the current iteration number, round is the rounding operation, and M is the integer part of the current iteration. α For the α-th moth, F β Let λ be the β-th flame, and λ be a random number within the range [-1, 1]; S(M α F β ) represents the α-th moth Mα and the β-th flame F. β The logarithmic spiral function between.

[0035] Based on the time window W for acquiring the stress wave signal, for each time window W, when generating the stress wave pulse compression matrix, we have:

[0036]

[0037] Where SWE is the stress wave energy of the stress wave pulse signal S(t), SWPE is the pulse energy of the stress wave pulse signal S(t), SWPA is the maximum peak height of the stress wave pulse signal S(t); k is the stress wave pulse number, L is the threshold value that exceeds the minimum value of the stress wave pulse signal S(t) during the time window W, the k-th stress wave pulse is the stress wave pulse signal S(t) between the k-th rise to the threshold value L and the next fall to the threshold value L, and ts k and te k These are the start and end times of the kth stress wave pulse, respectively.

[0038] The warning level threshold matrix THE of the configuration device includes:

[0039] Acquire the stress wave signal of the device in a healthy state, and generate a stress wave pulse compression matrix in the healthy state based on the stress wave signal, wherein,

[0040] The size of the stress wave pulse compression matrix under healthy conditions is 3×R. For each column of the stress wave pulse compression matrix under healthy conditions, it represents the stress wave energy, pulse energy, and maximum peak height based on the stress wave signal under a healthy condition.

[0041] For the stress wave pulse compression matrix under healthy conditions, calculate the mean and standard deviation of each row of elements to form the mean stress wave energy, mean pulse energy, mean maximum peak height, standard deviation of stress wave energy, standard deviation of pulse energy, and standard deviation of maximum peak height under healthy conditions.

[0042] The average stress wave energy under healthy conditions is summed with several times the standard deviation of stress wave energy under healthy conditions, and the sum is used as the early warning threshold for stress wave energy.

[0043] The average stress wave pulse energy under healthy conditions is accumulated with several times the standard deviation of pulse energy under healthy conditions, and the accumulated sum is configured as the pulse energy warning threshold.

[0044] The average maximum peak height under healthy conditions is summed with several times the standard deviation of the maximum peak height under healthy conditions, and the sum is configured as the maximum peak height warning threshold.

[0045] A fault early warning system based on stress wave pulse train peak segmentation and demodulation includes a fault early warning processor, wherein,

[0046] For any device, the stress wave signal of the device is acquired, and the fault early warning processor uses the fault early warning method described above to perform fault early warning.

[0047] The advantages of this invention are as follows: The moth-flame algorithm is used to demodulate the overlapping peaks of the stress wave envelope detection signal to obtain the stress wave pulse signal. The decomposition accuracy when generating the stress wave pulse signal is high, and the local convergence problem caused by improper initial value setting is avoided. It also overcomes the destruction of the original useful data. A stress wave pulse compression matrix is ​​generated from the generated stress wave pulse signal. This matrix enables data compression, improves data transmission speed, and facilitates fault early warning and fault alarm comparison, thereby improving the accuracy and applicability of stress wave-based fault early warning. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of one embodiment of the fault early warning system of the present invention.

[0049] Figure 2 This is a schematic diagram of one embodiment of the Gaussian peak of the present invention.

[0050] Figure 3 This is an embodiment of the present invention that utilizes Gaussian peaks to generate stress wave pulse signals.

[0051] Figure 4 This is a schematic diagram of one embodiment of the limiting threshold L of the present invention. Detailed Implementation

[0052] The present invention will be further described below with reference to specific accompanying drawings and embodiments.

[0053] To improve the accuracy and applicability of stress wave-based fault early warning, in one embodiment of the present invention, the fault early warning method based on stress wave pulse train peak segmentation and demodulation includes:

[0054] For a device awaiting fault warning, acquire the stress wave signal of the device;

[0055] The acquired stress wave signal is subjected to envelope detection processing to obtain the stress wave envelope detection signal;

[0056] For the stress wave envelope detection signal, the moth flame algorithm based on the Gaussian mixture model is used to perform peak splitting and demodulation processing of the overlapping peaks of the pulse train, so as to generate the stress wave pulse signal after peak splitting and demodulation processing.

[0057] The generated stress wave pulse signal is compressed to generate a stress wave pulse compression matrix, wherein the stress wave pulse compression matrix includes stress wave energy, pulse energy, and maximum peak height.

[0058] Configure the warning level threshold matrix THE of the device, wherein the warning level threshold matrix THE includes stress wave energy warning threshold, pulse energy warning threshold and maximum peak height warning threshold;

[0059] During fault warning, the stress wave pulse compression matrix is ​​compared with the warning level threshold matrix THE in a positive correspondence, so as to generate fault warning status information based on the positive correspondence comparison status.

[0060] Specifically, the equipment for fault early warning can be low-speed, heavy-load equipment such as a large-scale rotary table. Alternatively, it can be other equipment suitable for acquiring stress wave signals. The type of equipment for fault early warning can be selected as needed. Generally, stress wave signals are acquired using stress wave sensors. These sensors are typically installed on surfaces near moving parts of the equipment, such as bearings and gearboxes. During equipment operation, friction, mechanical impact, and dynamic loads transmitted through the moving equipment components generate stress waves. These stress wave signals can then be acquired using a stress wave sensor.

[0061] Stress wave sensors can employ common methods for acquiring stress wave signals in this technical field. However, when using a stress wave sensor to acquire a stress wave signal, it is generally necessary to set the sampling frequency and the sampling time window W. The sampling frequency and time window W can be selected according to actual needs. After configuring the working state of the stress wave sensor, a time-dependent stress wave signal R(t) can be acquired using the stress wave sensor, where t is the sampling time.

[0062] The obtained stress wave signal R(t) should be filtered and denoised first. In one embodiment of the present invention, the obtained stress wave signal is filtered using a bandpass filter to generate a filtered and denoised stress wave signal.

[0063] The stress wave filtered and denoised signal is subjected to envelope detection processing to obtain the stress wave envelope detection signal.

[0064] Specifically, the filtering process using a bandpass filter can be expressed as follows:

[0065]

[0066] Where Q(t) is the stress wave signal after filtering and denoising using a bandpass filter, H(jω) is the bandpass filter, K(jω) is the gain of the bandpass filter at its center frequency ω0, ω0 is the center frequency of the bandpass filter, and Q is the quality factor of the bandpass filter. The quality factor Q is related to the bandwidth; the larger the quality factor Q, the smaller the bandwidth of the bandpass filter. j is a complex number, and ω is the frequency. The center frequency ω0 of the bandpass filter can generally be the average of the low and high frequencies of the stress wave signal. The parameters of the bandpass filter can be designed or configured based on the characteristics of the stress wave signal using commonly used techniques in this field, specifically ensuring that the configured bandpass filter can effectively filter and denoise the stress wave signal.

[0067] After obtaining the stress wave filtered and denoised signal, it is necessary to perform envelope detection processing on the stress wave filtered and denoised signal to generate a stress wave envelope detection signal. Specifically, a Hilbert transform can be applied to the stress wave filtered and denoised signal to obtain the stress wave envelope detection signal after the Hilbert transform, which can be expressed as follows:

[0068] X(t) = Hilbert[Q(t)]

[0069] ={x(t1),x(t2),…,x(tn)}

[0070] Where X(t) is the stress wave envelope detection signal, Hilbert is the Hilbert transform, x(t1) is the envelope detection data point at time t1, and x(tn) is the envelope detection data point at time tn.

[0071] The stress wave envelope detection signal X(t) contains n envelope detection data points. The number of envelope detection data points n is generally related to the stress wave filtered and denoised signal Q(t). Of course, in specific implementations, other techniques can also be used to achieve envelope detection. The specific envelope detection method can be selected according to needs, based on the ability to generate a stress wave envelope detection signal. For the envelope detection data point x(t1) at time t1, it specifically refers to the magnitude of the stress wave signal amplitude at time t1; for the envelope detection data point x(tn) at time tn, it specifically refers to the magnitude of the stress wave signal amplitude at time tn, and so on for other cases. These will not be listed and explained here.

[0072] In one embodiment of the present invention, the moth flame algorithm based on Gaussian mixture model performs peak demodulation processing of overlapping peaks in a pulse train, including:

[0073] The moth flame algorithm is initialized, wherein during initialization, the population size q of the moths, the maximum number of iterations T0, and the shape constant η of the spiral are configured. For any moth in the moth population, the position information θ of the moth includes N sets of position parameters, each set of position parameters corresponds to a Gaussian mixture model, and N is the number of peaks obtained based on the stress wave envelope detection signal.

[0074] After initialization, a moth population is randomly generated, and the moth-flame algorithm iterative processing steps are performed based on the randomly generated moth population.

[0075] In each iterative processing step, the fitness value f(θ) of each moth in the moth population is determined based on the position information θ of each moth in the moth population. After determining the fitness value f(θ) of each moth in the moth population, the fitness function f is updated based on the configured flame number. n Update the number of flames, update the position of the moths and the position of the flames using a logarithmic spiral function, and accumulate the number of iterations;

[0076] Repeat the above iterative processing steps until the number of iterations matches the maximum number of iterations T0; once the number of iterations matches the maximum number of iterations T0, determine the optimal moth location within the moth population, where,

[0077] Based on the determined optimal moth position, a set of position parameters within the optimal moth position are used as the weight, mean, and standard deviation of each Gaussian peak, and a corresponding force wave pulse signal is generated based on all Gaussian peaks.

[0078] Specifically, when using the moth-flame algorithm based on a Gaussian mixture model for peak splitting and demodulation of overlapping peaks in a pulse train, the algorithm needs to be initialized. During initialization, the moth population size *q* is typically configured, meaning the number of moths and the corresponding number of flames need to be initialized. The maximum number of iterations *T0* refers to the maximum number of iterations the moth-flame algorithm can perform. The population size *q* and the maximum number of iterations *T0* can generally be selected and determined based on the specific application requirements. Generally, the population size *q* is set between 30 and 50, the maximum number of iterations *T0* is set between 30 and 50, and the spiral shape constant *η* is set between 1 and 1.5.

[0079] For the position information θ of a moth, we have: θ = [α μ f σ ], where α is the weight, α = [α1, α2, ... α N ] T The value range of each sub-weight within the weight α is [0.01, 1]; μ is the mean, μ = [μ1, μ2…μ]. N ] T f σ f is the standard deviation. σ =[fσ (1),f σ (2)…f σ (N)] T .

[0080] For the stress wave envelope detection signal X(t), we have: In practical implementation, the mean value μ of the envelope detection signal is calculated based on n envelope detection data points within the stress wave envelope detection signal X(t). a With the standard deviation f of the envelope detection signal σa Based on the calculated mean value μ of the envelope detection signal. a This allows us to determine the range of values ​​for elements within the mean μ of the location information θ. The range of values ​​for elements within the mean μ is 0 to μ. a Similarly, based on the calculated standard deviation f of the envelope detector signal... σa The standard deviation f within the location information θ can be determined. σ The range of values ​​for the inner element, and the standard deviation f σ The range of values ​​for the inner element is 0 to f. σa .

[0081] As can be seen from the above explanation, with weight α, mean μ, and standard deviation f... σ Within the corresponding value range, the location information θ can be generated by randomly selecting values, including weights α, mean μ, and standard deviation f. σ The corresponding values ​​are the weight α, mean μ, and standard deviation f within the generated location information θ. σ After obtaining the corresponding value, the position information θ for each moth can be generated.

[0082] For the randomly generated position information θ of the moth, [α1,μ1,f σ (1)] constitutes a set of position parameters, [α2,μ2,f] σ (2)] This constitutes a set of position parameters. The situation of other position parameters can be obtained by referring to the explanation here. That is, there are a total of N sets of position parameters. That is, in the position information θ, the weight α, the mean μ and the standard deviation f can be selected in sequence. σ The corresponding elements within the parameters form a set of positional parameters. Specifically, a set of positional parameters corresponds to the parameters of a Gaussian mixture model. This means that a Gaussian mixture model can be formed based on the weights, mean, and standard deviation within the set of positional parameters. Therefore, a Gaussian peak can be generated based on a set of positional parameters.

[0083] For the moth-flame algorithm, iterative processing is required after initialization. During this process, the current iteration count T is compared to the maximum iteration count T0. If the current iteration count T is less than the maximum iteration count T0, the iterative processing step needs to be repeated. In other words, the termination condition of the moth-flame algorithm is when the current iteration count T reaches the maximum iteration count T0, indicating that the iteration count has matched the maximum iteration count T0.

[0084] As can be seen from the above description, after randomly generating the moth population before or at the beginning of the first iteration, the fitness value f(θ) of each moth can be determined based on the position information θ of each moth.

[0085] In one embodiment of the present invention, for any moth's fitness value f(θ), then:

[0086]

[0087] Where n is the number of data points within the stress wave envelope detection signal X(t), x(ti) is the data point corresponding to time ti within the stress wave envelope detection signal X(t), P(x(ti)θ) is the probability that data point x(ti) belongs to a Gaussian mixture model corresponding to a position parameter, and α k For the k-th weight within the location information θ, u k f is the k-th mean value within the location information θ. σ (k) represents the kth standard deviation within the location information θ.

[0088] Specifically, the situation regarding the stress wave envelope detection signal X(t) can be referred to the above description. The number of peaks N can be determined or generated based on the number of maxima in the stress wave envelope detection signal X(t). The number of maxima in the stress wave envelope detection signal X(t) can be obtained by judging using the maxima determination method in this technical field. Weight α k Specifically, it refers to the k-th weight value within a weight α at a given location θ. k For details, please refer to the explanation of the position information θ and weight α above. Similarly, the k-th mean u can be determined. k and the kth standard deviation f σ (k).

[0089] As explained above, the location of the flame cannot be determined after initialization. In the moth-flame algorithm, the number of flames is generally consistent with the number of moths during initialization. In one embodiment of the present invention, for the first iteration after initialization, the fitness values ​​f(θ) of all moths are sorted in ascending order, and the sorted fitness values ​​f(θ) and the corresponding moth position information θ are used as the initial spatial location of the flame.

[0090] After sorting the fitness values ​​f(θ) in ascending order, the position information θ of the first moth after sorting and its corresponding (i.e., the smallest fitness value) fitness value f(θ) are simultaneously assigned to a flame as the initial spatial position of the first flame; the initial spatial positions of other flames can be configured in the same way, that is, the spatial positions of all flames are obtained.

[0091] During the iteration, after configuring the spatial positions of all flames, the flame count update function f is updated. n Then we have:

[0092]

[0093] Updating the moth's position using a logarithmic spiral function, we have:

[0094]

[0095] Where T is the current iteration number, round is the rounding operation, and M is the integer part of the current iteration. α For the α-th moth, F β For the β-th flame, λ is a random number in [-1, 1]; S(M α ,F β ) represents the α-th moth M α With the βth flame F β The logarithmic spiral function between.

[0096] In the above description, the logarithmic spiral function is used to simulate the spiral flight path of a moth. The next position of the moth is determined by the flame it orbits, and the random number λ represents the direction of rotation of the moth. In practice, the method and process of updating the moth's position based on the logarithmic spiral function are consistent with existing methods.

[0097] Furthermore, as explained above, in the first iteration (where T=1), the positions of the moths and the flame are updated based on the initial spatial position of the flame and the position information of the randomly generated moth population. In the second iteration (where T=2), the positions of the flame and moths are updated again based on the positions updated in the first iteration. The details of subsequent iterations can be found in the explanation provided here.

[0098] Update the flame count function f n This allows us to obtain the number of flames after each update. As the iteration progresses, the number of flames gradually decreases, but the number of moths remains unchanged. Due to the reduction in flames, the position of the moth corresponding to the flames that have been reduced in the sequence is updated based on the flame with the worst (i.e., highest) fitness value at the moment of failure. Therefore, during the iteration process, there may be situations where multiple moths correspond to a single flame. The update function f is based on the number of flames. nIt can be seen that when the number of iterations reaches T0, the number of flames is... That is, the number of flames is 1.

[0099] Moths are individuals moving within the search space, and the flame represents the best position a moth has found so far. Each moth circles a flame, and once a better fitness value is found, the flame's position is updated for the next generation. After each iteration, the moths are sorted according to their fitness value f(θ), and the flame's spatial position is updated. The sorted position typically contains the currently found optimal solution. The final output optimal position information is the flame position information with the smallest fitness value f(θ) in the iteration.

[0100] After T0 iterations, the moth-flame algorithm is complete. At this point, the optimal solution for determining the moth's position θ can be searched. As explained above, the flame position is updated incrementally based on the fitness value f(θ) in each iteration; the smaller the fitness value f(θ), the better the result. Each iteration typically yields a currently found optimal solution. This optimal solution is the flame position information corresponding to the minimum fitness value f(θ) after the last iteration. The optimal moth position within the moth population is also the position of the last flame after T0 iterations.

[0101] As explained above, the position of the last flame includes the weight α, mean μ, and standard deviation f. σ The optimal solution for each moth's position information θ is the corresponding N sets of position parameters. The weights α, mean μ, and standard deviation f of the Gaussian mixture model corresponding to these position parameters are then used. σ The weights α, mean μ, and standard deviation f are configured as a fraction of a peak within each stress wave envelope detector signal X(t). σ That is, to determine the parameter information of each peak, each peak is also a Gaussian peak.

[0102] Therefore, each peak point in the stress wave envelope detection signal X(t) conforms to a Gaussian distribution. Based on the N peak parameter information of the determined stress wave envelope detection signal X(t), a stress wave pulse signal can be generated accordingly.

[0103] Specifically, the Gaussian peaks formed by the N peak parameter information are plotted on the same coordinate axis, where the horizontal axis represents the mean and the vertical axis represents the standard deviation. Thus, based on the mean of each Gaussian peak, the N Gaussian peaks can be plotted on the coordinate axis respectively.

[0104] For the resulting N Gaussian peaks, each Gaussian peak is taken from the portion between its intersection with the preceding and following Gaussian peaks; the first Gaussian peak is taken from the portion before its intersection with the second Gaussian peak, and the last Gaussian peak is taken from the portion after its intersection with the preceding Gaussian peak. If there is no intersection between two Gaussian peaks, the entire peak is retained. In this case, a stress wave pulse signal can be generated on the aforementioned coordinate axes.

[0105] Figure 2 The figure shows one embodiment of plotting N Gaussian peaks on the same coordinate axis. Figure 3 The diagram illustrates one embodiment for generating a stress wave pulse signal. For other implementations with N Gaussian peaks, refer to [reference needed]. Figure 2 , Figure 3 As mentioned above, further examples will not be provided here.

[0106] The generated stress wave pulse signal can be represented as S(t), where S(t) = {s(t1), s(t2), ..., s(tn)}, s(t1) is the amplitude of the pulse signal after peak splitting at time t1, s(tn) is the amplitude of the pulse signal after peak splitting at time tn, and so on for other cases, which will not be elaborated here.

[0107] In one embodiment of the present invention, based on the time window W for acquiring the stress wave signal, for each time window W, when generating the stress wave pulse compression matrix, the following applies:

[0108]

[0109] Where SWE is the stress wave energy of the stress wave pulse signal S(t), SWPE is the pulse energy of the stress wave pulse signal S(t), SWPA is the maximum peak height of the stress wave pulse signal S(t); k is the stress wave pulse number, L is the threshold value that exceeds the minimum value of the stress wave pulse signal S(t) during the time window W, the Kth stress wave pulse is the stress wave pulse signal S(t) between the kth time it rises to the threshold value L and the next time it falls back to the threshold value L, and ts k and te k These are the start and end times of the kth stress wave pulse, respectively.

[0110] The limiting threshold L is specifically the value at which the valley values ​​of all individual stress wave pulses are sorted in ascending order, representing a percentage of 0.1. The sorting percentage specifically refers to the value at which the total number of pulses is sorted in ascending order of the valley values, representing a percentage of 0.1 (for example, if a pulse signal has 20 valley values, the limiting threshold is set at the second valley value from the smallest to the largest). The start and end times of the k-th stress wave pulse are specifically the two points obtained by intersecting the k-th peak (i.e., the peak determined based on the maximum value of the stress wave envelope detection signal) with the limiting threshold L, which represent the start and end times of the pulse.

[0111] Figure 4 An embodiment of the limiting threshold L is shown in the figure. Figure 4 In the graph, the horizontal axis represents time, and the vertical axis represents the acquired acceleration signal. Figure 4 The curve in the figure represents the stress wave pulse signal. For details on how to determine the limiting threshold L, please refer to [reference needed]. Figure 4 And the above explanation.

[0112] The stress wave energy SWE of a stress wave pulse signal S(t) is specifically calculated by summing the values ​​of all data points greater than 0 within the stress wave pulse signal S(t). The pulse energy SWPE of the stress wave pulse signal S(t) is specifically calculated by summing the areas of the N stress wave pulses within the stress wave pulse signal S(t). The maximum peak height SWPA of the stress wave pulse signal S(t) is specifically determined by identifying the maximum peak height of all stress wave pulses in the stress wave pulse signal S(t).

[0113] In one embodiment of the present invention, the warning level threshold matrix THE for the configured device includes:

[0114] Acquire the stress wave signal of the device in a healthy state, and generate a stress wave pulse compression matrix in the healthy state based on the stress wave signal, wherein,

[0115] The size of the stress wave pulse compression matrix under healthy conditions is 3×Q. For each column of the stress wave pulse compression matrix under healthy conditions, it is the stress wave energy, pulse energy and maximum peak height based on the stress wave signal under a healthy condition.

[0116] For the stress wave pulse compression matrix under healthy conditions, calculate the mean and standard deviation of each row of elements to form the mean stress wave energy, mean pulse energy, mean maximum peak height, standard deviation of stress wave energy, standard deviation of pulse energy, and standard deviation of maximum peak height under healthy conditions.

[0117] The average stress wave energy under healthy conditions is summed with several times the standard deviation of stress wave energy under healthy conditions, and the sum is used as the early warning threshold for stress wave energy.

[0118] The average stress wave pulse energy under healthy conditions is accumulated with several times the standard deviation of pulse energy under healthy conditions, and the accumulated sum is configured as the pulse energy warning threshold.

[0119] The average maximum peak height under healthy conditions is summed with several times the standard deviation of the maximum peak height under healthy conditions, and the sum is configured as the maximum peak height warning threshold.

[0120] Specifically, the stress wave pulse compression matrix SW under healthy conditions Q , can be represented as:

[0121]

[0122] Among them, SWE1, SWPE1, and SWPA1 are the stress wave energy, pulse energy, and maximum peak height of the stress wave signal under a healthy state, respectively. Other values ​​can be found here. The stress wave energy, pulse energy, and maximum peak height of the stress wave signal under each healthy state can be obtained using the methods described above.

[0123] Health status generally refers to the non-faulty or fault-free operating state of the same equipment. The stress wave pulse compression matrix SW is located in the health status. Q Specifically, this refers to using a stress wave sensor to collect Q stress wave signals R(t) under healthy conditions, and generating N sets of stress wave energy, pulse energy, and maximum peak height of the stress wave signals under healthy conditions based on the Q stress wave signals R(t).

[0124] The stress wave pulse compression matrix SW under the above-mentioned healthy state Q As can be seen from the expression, each row of the stress wave pulse compression matrix under healthy conditions represents one of the stress wave energy, pulse energy, or maximum peak height of the stress wave signal under different healthy conditions. For example, in the above expression, the first row represents the stress wave energy under different healthy conditions, the second row represents the pulse energy under different healthy conditions, and the third row represents the maximum peak height under different healthy conditions. The value of Q can generally be selected as needed, and the specific value is generally based on effectively configuring the warning level threshold matrix THE.

[0125] When calculating the mean, we have: For the calculation of standard deviation, we have:

[0126] Specifically, for the warning level threshold matrix THE, there may be: Where THE1 is the stress wave energy warning threshold, THE2 is the pulse energy warning threshold, and THE3 is the maximum peak height warning threshold. For the pulse energy warning threshold THE2 and the maximum peak height warning threshold THE3, please refer to the explanation of the stress wave energy warning threshold here, which will not be repeated here. Of course, the warning level threshold matrix THE can also be determined using other numerical relationships, which can be selected according to actual needs.

[0127] In one embodiment of the present invention, when comparing the stress wave pulse compression matrix with the early warning level threshold matrix THE in a positive correspondence, the following is obtained:

[0128] The stress wave energy, pulse energy, and maximum peak height are compared one-to-one with the stress wave energy warning threshold, pulse energy warning threshold, and maximum peak height warning threshold, respectively. For each comparison, a comparison information is generated.

[0129] When at least one comparison information matches the warning threshold, the generated fault warning status information is the fault warning status.

[0130] Specifically, the direct correspondence comparison refers to comparing the stress wave energy of the stress wave pulse compression matrix with the stress wave energy warning threshold, comparing the pulse energy with the pulse energy warning threshold, and comparing the maximum peak height with the maximum peak height warning threshold.

[0131] The generated comparison information includes whether it matches or does not match the warning threshold. A match means the comparison value is equal to the warning threshold or the difference between them is within a preset range. For example, if the stress wave energy warning threshold is the same as the stress wave energy warning threshold, or the difference between them is within a preset range, then it is considered a match. Otherwise, it is considered a mismatch. The preset range can be selected according to actual needs, based on meeting the requirements of fault warning. For comparisons of pulse energy with the pulse energy warning threshold, and comparisons of maximum peak height with the maximum peak height warning threshold, please refer to the explanation here; examples will not be provided here.

[0132] Specifically, when at least one comparison information matches the warning threshold, the generated fault warning status information is a fault warning status, meaning it meets the fault warning conditions. When all comparison information does not match the warning threshold, the generated fault warning status information is a no-fault warning status, in which case the fault warning device is in good working order.

[0133] Furthermore, when the generated fault warning status information is in a fault warning state, the equipment can generally continue to operate. Of course, the fault warning type can also be obtained based on the fault warning status information. In addition, an alarm level threshold matrix (THD) can be set, which may include stress wave energy alarm threshold, pulse energy alarm threshold, and maximum peak value alarm threshold.

[0134] In practice, the stress wave energy alarm threshold can be three times the stress wave energy warning threshold. Similarly, the corresponding values ​​of the pulse energy alarm threshold and the maximum peak value alarm threshold can be obtained. Of course, the specific values ​​can be selected according to actual needs, so as to meet the fault alarm requirements of the equipment.

[0135] In summary, a fault early warning system based on stress wave pulse train peak segmentation and demodulation can be obtained. In one embodiment of the present invention, a fault early warning processor is included, wherein...

[0136] For any device, the stress wave signal of the device is acquired, and the fault early warning processor uses the fault early warning method described above to perform fault early warning.

[0137] Specifically, the fault warning processor can be any commonly used processor type, such as a computer terminal, and can be selected according to needs. The specific method and process of fault warning based on stress wave signals can be referred to the above description, and will not be repeated here.

[0138] As explained above, for the stress wave envelope detection signal, this invention uses the moth-flame algorithm to demodulate the overlapping peaks of the signal to obtain the stress wave pulse signal. The decomposition accuracy when generating the stress wave pulse signal is high, and it avoids the local convergence problem caused by improper initial value setting. At the same time, it also overcomes the destruction of the original useful data. For the generated stress wave pulse signal, a stress wave pulse compression matrix is ​​generated. The stress wave pulse compression matrix can be used to realize data compression, improve the data transmission speed, and facilitate fault early warning and fault alarm comparison, thereby improving the accuracy and applicability of stress wave-based fault early warning.

Claims

1. A fault early warning method based on stress wave pulse train demodulation, characterized in that, The fault early warning method comprises: For a device to be faulted early warned, a stress wave signal of the device is acquired; The acquired stress wave signal is subjected to envelope detection processing to obtain a stress wave envelope detection signal; The stress wave envelope detection signal is subjected to peak separation demodulation processing of pulse train overlapping peaks based on a Gaussian mixture model moth-flame algorithm to generate a stress wave pulse signal after the peak separation demodulation processing; The generated stress wave pulse signal is subjected to data compression to generate a stress wave pulse compression matrix, wherein the stress wave pulse compression matrix comprises stress wave energy, pulse energy and maximum peak height; A warning level threshold matrix THE of the device is configured, wherein the warning level threshold matrix THE comprises a stress wave energy warning threshold, a pulse energy warning threshold and a maximum peak height warning threshold; In a fault early warning, the stress wave pulse compression matrix is compared with the warning level threshold matrix THE in a one-to-one correspondence to generate fault early warning state information based on the one-to-one correspondence comparison state; In the peak separation demodulation processing of pulse train overlapping peaks based on the Gaussian mixture model moth-flame algorithm, the following steps are included: The moth flame algorithm is initialized, wherein, in the initialization, a population size of the moth is configured , a maximum iteration number , a shape constant of a spiral line , position information of any moth in the moth population includes N sets of position parameters is a peak number obtained based on a stress wave envelope detection signal, each set of position parameters corresponds to a Gaussian mixture model; After initialization, a moth population is randomly generated, and an iterative processing step of the moth-flame algorithm is performed based on the randomly generated moth population, wherein In each iteration processing step, based on the position information of each moth in the moth population determining the fitness value of the moth , and after determining the fitness value of each moth in the moth population , updating the number of flames based on the configured flame number update function updating the number of flames, updating the position of the moth and the position of the flame using the logarithmic spiral function, and accumulating the number of iterations; The above iterative processing step is repeated until the number of iterations reaches the maximum number of iterations adaptation; when the number of iterations of the iterative processing reaches the maximum number of iterations After the adaptation, the optimal moth position within the moth population is determined, wherein, Based on the determined optimal moth position, a group of position parameters in the optimal moth position are taken as the weight, mean and standard deviation of each Gaussian peak, and a corresponding stress wave pulse signal is generated based on all Gaussian peaks; The warning level threshold matrix THE of the device is configured, comprising: The stress wave signal of the device in a healthy state is acquired, and a stress wave pulse compression matrix in the healthy state is generated based on the stress wave signal in the healthy state, wherein The size of the stress wave pulse compression matrix under the healthy state is For each column of the stress wave pulse compression matrix under the healthy state, the stress wave energy, the pulse energy, and the maximum peak height based on a stress wave signal under the healthy state. For the stress wave pulse compression matrix in the healthy state, the mean and standard deviation of each row element are calculated to form the mean of stress wave energy in the healthy state, the mean of pulse energy in the healthy state, the mean of maximum peak height in the healthy state, the standard deviation of stress wave energy in the healthy state, the standard deviation of pulse energy in the healthy state and the standard deviation of maximum peak height in the healthy state; The mean of stress wave energy in the healthy state is accumulated with a number of times of the standard deviation of stress wave energy in the healthy state to configure the accumulated sum as the stress wave energy warning threshold; The mean of stress wave pulse energy in the healthy state is accumulated with a number of times of the standard deviation of pulse energy in the healthy state to configure the accumulated sum as the pulse energy warning threshold; The mean of maximum peak height in the healthy state is accumulated with a number of times of the standard deviation of maximum peak height in the healthy state to configure the accumulated sum as the maximum peak height warning threshold.

2. The method according to claim 1, characterized in that, When the stress wave pulse compression matrix is compared with the warning level threshold matrix THE in a one-to-one correspondence, the following is performed: The stress wave energy, pulse energy and maximum peak height are respectively compared with the stress wave energy warning threshold, pulse energy warning threshold and maximum peak height warning threshold in a one-to-one correspondence, and for any comparison, a comparison information is generated, wherein When there is at least one comparison information matching the warning threshold, the generated fault early warning state information is a fault early warning state.

3. The method of claim 1, wherein the method further comprises: The acquired stress wave signal is filtered by a band-pass filter to generate a stress wave filtered and denoised signal after filtering; The stress wave envelope detection signal is obtained by performing envelope detection processing on the stress wave filtered and denoised signal.

4. The method of claim 1, wherein the method further comprises: fitness value for any moth then there is: wherein is a stress wave envelope detection signal is the number of data points within is a stress wave envelope detection signal is within is the data point corresponding to the time instant, is a data point is the probability that a position parameter belongs to a Gaussian mixture model, is position information is the th weight within is position information is the th mean value within is position information is the th standard deviation within is position information is the th standard deviation within is position information 5. The method of claim 1, wherein the stress wave pulse train is split into peaks and demodulated, and wherein the method further comprises: determining a peak amplitude of the stress wave pulse train; and determining a peak amplitude ratio of the stress wave pulse train. In the iteration processing step, the fitness values of all the moths are sorted in ascending order, and the sorted fitness values and the moth position information corresponding to the fitness values are stored in the memory. The spatial position is configured as a flame.

6. The method of claim 1, wherein the method further comprises: Flame number update function Then, there is: ； The position of the moth is updated by using a logarithmic spiral function, and thus wherein, is the current iteration number, is a rounding operation, is the th firefly, is the th flame, is a random number in [-1, 1]; is the th firefly and the th flame between them.

7. The method of claim 4, wherein the method further comprises: Time windows based on acquisition of stress wave signals For each time window In generating the stress wave pulse compression matrix, then, there is: wherein is the stress wave energy of the stress wave pulse signal , is the pulse energy of the stress wave pulse signal , is the maximum peak height of the stress wave pulse signal , k is the number of the stress wave pulse, L is the start time of the stress wave pulse, W is the end time of the stress wave pulse, is the limit threshold above the minimum value of the stress wave pulse signal during the time window is the stress wave pulse between the L time the stress wave pulse rises to the limit threshold L and the time the stress wave pulse falls to the limit threshold again, and are the start and end time of the stress wave pulse, respectively.

8. A fault warning system based on stress wave pulse train demodulation, characterized in that, The fault early warning processor comprises a fault early warning method, wherein For any device, the stress wave signal of the device is acquired, and the fault early warning processor adopts the fault early warning method in any one of claims 1 to 7 to perform fault early warning.

Images