CN112101174A - LOF-Kurtogram-based mechanical fault diagnosis method - Google Patents

Abstract

A fault diagnosis method for a rotating machine based on LOF-Kurtogram comprises the steps of firstly, utilizing a sliding time window with a fixed length to divide a section of mechanical monitoring signal into a plurality of data sections; extracting characteristic index vectors of a time domain, a time-frequency domain and the like of each data segment; setting an initial value of a parameter k in an LOF algorithm, calculating a local abnormal factor value of each data segment, screening abnormal segments based on a 3 sigma criterion, and removing and cleaning the abnormal segments; and then inputting the cleaned data into Kurtogram, obtaining a filtered data segment, solving an envelope spectrum of the filtered data segment, searching fault characteristic frequency, and finishing fault diagnosis. According to the invention, the traditional Kurtogram algorithm is improved based on the local abnormal factor, the diagnosis of data under large impact noise caused by environmental interference is realized, the anti-noise processing capability of mechanical fault data is improved, and the method has a more ideal effect on the diagnosis of mechanical fault.

Description

LOF-Kurtogram-based mechanical fault diagnosis method

Technical Field

The invention belongs to the field of mechanical monitoring and fault diagnosis, and particularly relates to a LOF-Kurtogram-based mechanical fault diagnosis method.

Background

Industrial equipment contains a large number of rotating machine parts such as bearings, gears and the like, which play a role in supporting and transmitting torque in the rotating machine and are very important for safe and reliable operation of equipment. However, the rotating parts have frequent faults, once the rotating parts have faults, the equipment cannot normally operate, the machine is stopped, and when the rotating parts have faults, the machine set is damaged, so that serious casualties are caused. Therefore, the timely and accurate diagnosis of the rotary machine fault is very important to prevent major accidents and improve the economic benefit of equipment production. The method is an important means for diagnosing the fault of the rotary machine by acquiring the vibration signal based on the vibration sensor, analyzing the signal and judging whether the fault exists in the equipment. When rotating parts such as bearings and gears have faults, monitoring signals acquired by the sensors are represented as periodic impact, and the signal processing method judges whether the faults exist according to whether the signals contain periodic impact frequency components. In an industrial field, the environment is noisy, a large amount of noise is mixed in an acquired monitoring signal, periodic components are submerged in the noise, the periodic components are difficult to find by directly calculating the frequency spectrum of the periodic components, and misjudgment on faults is easily caused.

Spectral kurtosis (Kurtogram) is a common signal processing fault diagnosis method, and the method utilizes the characteristic that kurtosis is sensitive to impact, the kurtosis value of a signal containing impact is often larger than that of a signal without impact, and the extraction of an impact signal is completed based on the extraction of a frequency band with a large kurtosis value. However, Kurtogram is only suitable for extracting impact signals without significant noise, and in a real environment, for example, due to the influence of human tests, environmental interference and the like, the acquired signals often contain one or more significant noise, and the acquired signals have larger kurtosis values, so that target signals finally extracted by Kurtogram do not contain periodic impact, and faults cannot be accurately diagnosed.

Disclosure of Invention

In order to achieve the purpose, the invention adopts the technical scheme that:

a fault diagnosis method for monitoring rotary machinery based on LOF-Kurtogram comprises the following steps:

1) obtaining a section of rotating machinery monitoring signal as an original signal x (t), wherein t is 1, …, N and N are the number of sampling points of the section of signal, and dividing x (t) into a plurality of sections by using a sliding window with the window length l, and the number of the sections is S;

2) extracting time domain characteristics and time-frequency domain characteristics from each segment to form a characteristic index set; wherein the content of the first and second substances,

in f_iRepresenting a characteristic index vector of the ith data segment during sliding, wherein the characteristic index comprises a mean value, a maximum value, a minimum value, a peak-to-peak value, a variance, a kurtosis, a root-mean-square, a form factor, a peak factor, a pulse factor, a square root amplitude value, a margin factor, a skewness and a wavelet energy component ratio;

3) setting the k value, searching and calculating the feature point set

K is a near neighborhood of any point in the tree;

4) calculating local reachable density of all feature points, wherein any point is

Where o ' is a neighborhood of o, k _ distance (o ') represents the kth distance of point o ', k _ distance (o) represents the kth distance of point o, d (o, o ') represents the euclidean distance of points o and o ', where the value of k is taken to be 5; n is a radical of_k(o) is a neighborhood of point o, i.e., all points within the kth distance of o, including points at the kth distance;

5)

the local anomaly factor value calculation expression for point o is as follows:

formula (III) lrd_k(o) and lrd_k(o ') the local achievable densities of point o and point o', respectively;

6) calculating according to equation (2)

The local abnormal factor value of the feature point used in (1) is recorded as

Detecting abnormal data segments according to a threshold T, from the sequence

Select all greater than T

And determining the position of a noise point in the original mechanical monitoring data according to the window width l and the data segment sequence number i. The threshold value T is calculated here using the 3-sigma principle, i.e.

In the formula

Indicating the calculated standard deviation.

7) Eliminating a data segment where a noise point is located, and inputting the residual data into a Kurtogram of short-time Fourier transform for calculation;

8) obtaining a center frequency fc and a bandwidth B according to a spectrum kurtosis result, and reconstructing to obtain a corresponding time domain signal y (t);

9) and calculating the envelope spectrum of the reconstructed time domain signal to obtain the fault characteristic frequency and finish fault diagnosis.

Compared with the prior art, the invention has the beneficial effects that:

the method adopts a sliding window to divide the acquired data into a plurality of segments, extracts time domain and time frequency characteristics, converts the segments into a multi-dimensional characteristic point set, calculates local abnormal factors of the segments, detects large noise points through a 3-sigma threshold value, prevents the noise points from having adverse effects on the spectral kurtosis to cause the failure of spectral kurtosis diagnosis, and highlights the periodicity of repeated transient impact after eliminating the noise points. The invention solves the problem that the fault diagnosis of the traditional Kurtogram method is invalid in the presence of noise, and finally can successfully detect repeated transient impact and periodic components to finish the fault diagnosis even in the presence of environmental noise by the LOF-Kurtogram.

Drawings

FIG. 1 is a flow chart of the present invention.

Fig. 2 is a time domain waveform diagram of a rolling bearing fault according to the embodiment, wherein (a) is a simulation signal waveform without noise, (b) is a local time domain waveform of data after large impact noise is added, (c) is a local time domain waveform of data after gaussian white noise is added, and (d) is an entire time domain waveform of a signal to be diagnosed.

FIG. 3 is a graph of the noise recognition result of the embodiment, (a) is a graph of the local abnormal factor value of the data after adding noise in the embodiment, (b) is a graph of the data detection result of the embodiment based on the local abnormal factor value, and (c) is a graph of the time domain waveform of the data after washing the impact data

FIG. 4 is a graph showing the results of Kurtogram analysis of data after local abnormal factor cleaning.

Fig. 5 is a LOF-Kurtogram final diagnosis result graph, (a) a Kurtogram original signal is input, (b) a target signal detected after Kurtogram filtering, and (d) an envelope spectrum of the target signal is detected.

Fig. 6 is a diagram showing the results of a fault analysis based on the conventional Kurtogram method.

Fig. 7 is a fault diagnosis result diagram based on the conventional Kurtogram method, wherein (a) a Kurtogram original signal is input, (b) a target signal detected after Kurtogram filtering, and (d) an envelope spectrum of the target signal is detected.

Detailed Description

The present invention is described in further detail below with reference to the attached drawings.

As shown in fig. 1, a method for detecting abnormal segments of mechanical monitoring data based on nuclear estimation LOF includes the following steps:

a fault diagnosis method for monitoring rotary machinery based on LOF-Kurtogram comprises the following steps:

1) obtaining a section of rotating machinery monitoring signal as an original signal x (t), wherein t is 1, …, N and N are the number of sampling points of the section of signal, and dividing x (t) into a plurality of sections by using a sliding window with the window length of l without overlapping, and the number of the sections is marked as S;

2) extracting time domain characteristics and time-frequency domain characteristics from each segment to form a characteristic index set; wherein the content of the first and second substances,

in f_iRepresenting a characteristic index vector of the ith data segment during sliding, wherein the characteristic index comprises a mean value, a maximum value, a minimum value, a peak-to-peak value, a variance, a kurtosis, a root-mean-square, a form factor, a peak factor, a pulse factor, a square root amplitude value, a margin factor, a skewness and a wavelet energy component ratio;

3) setting the k value, searching and calculating the feature point set

K is a near neighborhood of any point in the tree;

4) calculating local reachable density of all feature points, wherein any point is

Where o ' is a neighborhood of o, k _ distance (o ') represents the kth distance of point o ', k _ distance (o) represents the kth distance of point o, d (o, o ') represents the euclidean distance of points o and o ', where the value of k is taken to be 5; n is a radical of_k(o) is a neighborhood of point o, i.e., all points within the kth distance of o, including the kth distanceA point on;

5)

the local anomaly factor value calculation expression for point o is as follows:

formula (III) lrd_k(o) and lrd_k(o ') the local achievable densities of point o and point o', respectively;

6) calculating according to equation (2)

The local abnormal factor value of the feature point used in (1) is recorded as

Detecting abnormal data segments according to a threshold T, from the sequence

Select all greater than T

And determining the position of a noise point in the original mechanical monitoring data according to the window width l and the data segment sequence number i. The threshold value T is calculated here using the 3-sigma principle, i.e.

In the formula

Represents the calculated standard deviation;

7) eliminating a data segment where a noise point is located, and inputting the residual data into a Kurtogram of short-time Fourier transform for calculation;

8) obtaining a center frequency fc and a bandwidth B according to a spectrum kurtosis result, and reconstructing to obtain a corresponding time domain signal y (t);

9) and calculating the envelope spectrum of the reconstructed time domain signal to obtain the fault characteristic frequency and finish fault diagnosis.

The invention is described in detail below with reference to embodiments that use simulation experiments of bearing fault signals to validate the method of the invention.

The bearing is an important part of rotating mechanical equipment, faults occur frequently, repeated transients are main components of the bearing during faults, and the following models are adopted to simulate bearing fault signals:

wherein H (1,2, … H) represents the number of impacts, A_hIs the amplitude of the signal, which is 0.6-1.6 m.s²U (t) is a step function, outer ring fault characteristic frequency f_oTaken as 50Hz and damping coefficient beta_wTaken as 900, resonant frequency f_reTaking 3000, setting the sampling frequency to 12KHz, setting the total sampling time to 4s, obtaining the local time domain waveform of the original signal after simulation as shown in FIG. 2(a), seeing the periodic impact caused by the bearing defect, adding the local time domain signal after noise point as shown in FIG. 2(b), adding the local time domain waveform of Gaussian white noise as shown in FIG. 2(c), and finally obtaining the whole time domain signal to be analyzed in FIG. 2 (d).

The method is used for detecting the noise point, firstly the segmentation length l of the sliding window is set to be 500, the sliding window is not overlapped, the data is divided into 96 sections, then the time domain and the time-frequency domain characteristics of each section are extracted, 96 characteristic point sets are formed, and each characteristic point can be described by using the time domain and the time-frequency domain characteristics. Then, a near-neighborhood point of each feature point is calculated and searched, and on the basis, the k distance and the local reachable density of each feature point are calculated in sequence, wherein the parameter k is set to be 5. Then sequentially calculating the local abnormal factor value of each point of the original signal; based on the obtained local anomaly factor value, feature points exceeding the threshold are detected by the 3 σ criterion, and as a result, as shown in fig. 3(a), the feature points exceeding the threshold are found from the original time domain signalThe noise is shown in fig. 3(b), and then the data segment is cleaned from the original data, and the cleaned data is shown in fig. 3 (c). The cleaned data are input into the spectral kurtosis method, and the obtained result is shown in fig. 4, according to fig. 4, the frequency band center frequency of the target signal is selected to be 2062.5Hz, and the number of layers is 5.5. Fig. 5(a), (b), (c) show the input raw signal, the Kurtogram filtered signal and the final envelope spectrum, respectively, from which the fault frequency (f) can be clearly seen_o50Hz) and multiples thereof.

In order to further verify the superiority of the method, the noise point detection result of the method is compared with the noise point detection result based on the traditional Kurtogram algorithm, and in the fault diagnosis method based on the traditional Kurtogram algorithm, the target frequency center frequency is 5400Hz, the number of layers is layer 6, and the result is shown in fig. 6. In fig. 7, (a) is input original data, which contains large impact noise, and (b) and (c) are respectively a target signal and a corresponding envelope spectrum after Kurtogram processing and filtering, and it can be seen that a fault frequency (50Hz) peak value is not obvious, which cannot be clearly identified from the graph, and a bearing fault cannot be diagnosed, so that the traditional Kurtogram cannot successfully diagnose the bearing fault.

The two methods are compared, noise point detection based on the traditional Kurtogram algorithm cannot eliminate large impact noise influence caused by the environment, and under the impact noise influence, the traditional Kurtogram cannot detect periodic impact caused by a rotary machine fault and cannot diagnose the fault; the method accurately detects and then removes the large impact noise, can successfully detect the periodic signal, and can clearly identify the fault characteristic frequency amplitude in the envelope spectrum, thereby indicating that the method improves the capability of diagnosing the rotary machine fault. Therefore, the rotary machine fault diagnosis method has ideal fault diagnosis effect. It should also be noted that modifications and variations are possible without departing from the inventive concept, and are intended to be included within the scope of the invention.

Claims (1)

1. A fault diagnosis method for monitoring rotary machinery based on LOF-Kurtogram comprises the following steps:

1) obtaining a section of rotating machinery monitoring signal as an original signal x (t), wherein t is 1, …, N and N are the number of sampling points of the section of signal, and dividing x (t) into a plurality of sections by using a sliding window with the window length of l without overlapping, and the number of the sections is marked as S;

2) extracting time domain characteristics and time-frequency domain characteristics from each segment to form a characteristic index set; wherein the content of the first and second substances,

in f_iRepresenting a characteristic index vector of the ith data segment during sliding, wherein the characteristic index comprises a mean value, a maximum value, a minimum value, a peak-to-peak value, a variance, a kurtosis, a root-mean-square, a form factor, a peak factor, a pulse factor, a square root amplitude value, a margin factor, a skewness and a wavelet energy component ratio;

3) setting the k value, searching and calculating the feature point set

K is a near neighborhood of any point in the tree;

4) calculating local reachable density of all feature points, wherein any point is

Where o ' is a neighborhood of o, k _ distance (o ') represents the kth distance of point o ', k _ distance (o) represents the kth distance of point o, d (o, o ') represents the euclidean distance of points o and o ', where the value of k is taken to be 5; n is a radical of_k(o) is a neighborhood of point o, i.e., all points within the kth distance of o, including points at the kth distance;

5)

the local anomaly factor value calculation expression for point o is as follows:

formula (III) lrd_k(o) and lrd_k(o ') the local achievable densities of point o and point o', respectively;

6) calculating according to equation (2)

The local abnormal factor value of the feature point used in (1) is recorded as

Detecting abnormal data segments according to a threshold T, from the sequence

Select all greater than T

Determining the position of a noise point in the original mechanical monitoring data according to the window width l and the data segment sequence number i; the threshold value T is calculated here using the 3-sigma principle, i.e.

In the formula

Represents the calculated standard deviation;

7) eliminating a data segment where a noise point is located, and inputting the residual data into a Kurtogram of short-time Fourier transform for calculation;

8) obtaining a center frequency fc and a bandwidth B according to a spectrum kurtosis result, and reconstructing to obtain a corresponding time domain signal y (t);

9) and calculating the envelope spectrum of the reconstructed time domain signal to obtain the fault characteristic frequency and finish fault diagnosis.

Images