WO2022062161A1

WO2022062161A1 - Large machine set friction fault analysis method and system based on waveform and dimensionless learning

Info

Publication number: WO2022062161A1
Application number: PCT/CN2020/131616
Authority: WO
Inventors: 荆晓远; 陈润航; 王许辉; 张清华; 成明康; 姚永芳; 孔晓辉; 陈俊均; 吴松松
Original assignee: 广东石油化工学院
Priority date: 2020-09-28
Filing date: 2020-11-26
Publication date: 2022-03-31
Also published as: CN112183344A; CN112183344B

Abstract

A large machine set friction fault analysis method and system based on waveform and dimensionless learning, which belong to the technical field of fault detection. The method comprises: extracting a machine fault vibration signal by means of double probes, and preprocessing data; moreover, performing friction fault feature extraction; establishing a fault prediction model by using a machine learning method; and predicting whether there is a fault in an unknown label signal, and determining a fault type, wherein during the preprocessing process of a machine fault vibration signal and data, two probe points are installed, and vibration double-view signals of a large sliding machine set are collected by means of the two probe points; after probes collect data, a discrete Fourier transform is performed in an alignment mode, and Fourier values obtained after the transform are modified; and an adaptive threshold value is set according to a signal situation, and the signal storage amount is reduced, thereby accelerating transmission. In the method, the problem of difficult feature extraction can be effectively solved during the large machine set friction fault diagnosis process; and an effective feature can be extracted, thereby solving the problem of fault prediction.

Description

Analysis method and system of friction fault of large unit based on waveform and dimensionless learning

technical field

The invention belongs to the technical field of fault detection, and in particular relates to a method and system for analyzing friction faults of large units based on waveforms and dimensionless learning.

Background technique

At present, the large-scale equipment has a complex structure, perfect functions, and the internal parts of the equipment are closely connected, which makes the production process achieve high speed and large-scale, which also causes the large-scale equipment to fail and cause huge losses, which also increases It reduces the difficulty of fault diagnosis for large-scale equipment. When an object and another object move along the tangent direction of the contact surface or tend to move relative to each other, there is a force between the contact surfaces of the two objects that hinders their relative movement. This force is called frictional force. This phenomenon or characteristic between the contact surfaces is called "friction", so it is of great significance to detect faults in large-scale equipment based on frictional vibration signals. The existing common methods have problems such as difficulty in feature extraction and incomplete feature extraction.

Through the above analysis, the problems and defects of the prior art are as follows: the existing common methods have problems such as difficulty in feature extraction and incomplete feature extraction. This makes it difficult to diagnose faults in the operation of large-scale equipment.

The difficulty of solving the above problems and defects is as follows:

The feature extraction of friction fault problem is difficult, and the feature extraction is not comprehensive.

The significance of solving the above problems and defects is:

The problem of friction fault detection can be well solved, and the detection accuracy of friction fault can be improved.

SUMMARY OF THE INVENTION

Aiming at the problems existing in the prior art, the present invention provides a method and system for analyzing friction faults of large units based on waveform and dimensionless learning.

The present invention is realized in this way, a kind of large-scale unit friction fault analysis method based on waveform and dimensionless learning, described large-scale unit friction fault analysis method based on waveform and dimensionless learning, including:

Step 1: Use dual probes to extract machine fault vibration signals, and preprocess the data;

Step 2, extracting friction fault features;

Step 3, using the machine learning method to establish a fault prediction model;

Step 4: Predict whether there is a fault in the unknown tag signal, and determine the fault type.

Further, in the step 1, the process of preprocessing the machine fault vibration signal and data is as follows:

1) Install two probe points, and obtain the vibration double-view signal of the large-scale sliding unit through the acquisition of the two probe points. The data acquisition is 32/rms, that is, 32 points are sampled for each revolution of the bearing, and 32 circles of data are collected;

2) After the probe collects the data, the discrete Fourier transform is performed in alignment, the window size is 32*32=1024 points, and the transformed Fourier value is modified; the adaptive threshold is set according to the signal condition, the signal storage capacity is reduced, and the speed is accelerated. transmission.

Further, the discrete Fourier transform formula in described 2) is as follows:

Where n=0, . . ., N-1, N represents the data length.

Further, in the second step, the process of extracting friction fault features is:

(1) Perform two-layer decomposition and transformation of wavelet packet on the fault signal. The wavelet packet analyzes the details of the input signal by using multiple iterative wavelet transforms, obtains the wavelet coefficients at different scales, and sets the scale coefficients of the HH layer of the signal to zero;

(2) Calculate the dimensionless characteristic wave index S _f , take the wave index as one of the extracted features, and the specific calculation formula is as follows:

in

represents the root mean square value of the waveform data,

Indicates the absolute average of waveform data;

(3) Calculate the dimensionless feature peak index, and take the peak index as one of the extracted features. The specific calculation formula is as follows:

where x _max represents the peak value of the waveform,

Represents the root mean square value;

(4) Calculate the dimensionless characteristic pulse index, and take the pulse index as one of the extracted features. The specific calculation formula is as follows:

where x _max represents the peak value of the waveform,

Indicates the absolute average of waveform data;

(5) Calculate the dimensionless characteristic kurtosis index, which indicates the level of the actual kurtosis relative to the normal kurtosis. The kurtosis index reflects the shock characteristics in the vibration signal, and the dimensionless characteristic kurtosis index is used as one of the extracted features. The formula is as follows:

in

(6) Calculate the dimensionless feature margin index, which is generally used to detect the wear of mechanical equipment; if the skewness index does not change much, the ratio of the effective value to the average value increases, indicating that the gap increases due to wear, so the vibration The effective value of the energy index increases faster than the average value, and its margin index also increases. Taking the dimensionless feature margin index as one of the extracted features, the specific calculation formula is as follows:

in

(7) Calculate the dimensionless feature Teager energy operator, take the Teager energy operator as one of the extracted features, and the specific calculation formula is as follows:

where t is the data collection time,

α _t is the offset angle before and after time t;

(8) Calculate the standard deviation. The standard deviation represents the degree of dispersion of the data, and it represents the variability of a single statistic in multiple sampling; it can be understood that the former represents the variability of the data itself, while the latter represents the variability of the data itself. is the variability of sampling behavior, and the specific calculation formula is as follows:

(9) Calculate the standard deviation of the mean. The standard deviation of the mean refers to a standard for measuring the degree of dispersion of the data distribution, which is used to measure the degree to which the data values deviate from the arithmetic mean; the smaller the standard deviation, the more these values deviate from the mean. The less, and vice versa; the size of the standard deviation can be measured by the multiplying relationship between the standard deviation and the average, and the specific calculation formula is as follows:

(10) Calculate the sample circle mean (circle_mean) of the sample, and take the sample circle mean as one of the extracted features. The specific calculation formula is as follows:

Where X is the sample, sin is the sine function, cos is the cosine function, arctan2 is the tangent function, and π is the pi;

in

S=∑ _i sin(angle) C=∑ _i cos(angle), res=arctan2(S, C).

Further, in the third step, the cca dimension is reduced by using the two view features extracted in the second step, and the features of the two views after the dimension reduction are spliced together as an input vector, and a machine learning model is used for training.

Another object of the present invention is to provide a large unit friction fault analysis system based on waveform and dimensionless learning that implements the waveform and dimensionless learning-based large unit friction fault analysis method. Unit friction failure analysis system, including:

The data acquisition module extracts the machine fault vibration signal by using double probes, and preprocesses the data;

Feature extraction module, to perform feature extraction on friction fault signal;

Prediction model building module, by using machine learning methods, to establish a fault prediction model;

The fault prediction module predicts whether there is a fault in the unknown tag signal, and determines the fault type.

Another object of the present invention is to provide a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the following steps :

Use dual probes to extract machine fault vibration signals and preprocess the data;

Feature extraction for friction fault signals;

Build fault prediction models by using machine learning methods;

Predict whether there is a fault in the unknown tag signal and determine the type of fault.

Another object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, causes the processor to perform the following steps:

Extracting machine fault vibration signals by using dual probes and preprocessing the data;

Feature extraction for friction fault signals;

Build fault prediction models by using machine learning methods;

Another object of the present invention is to provide a large-scale equipment for implementing the method for analyzing friction faults of large-scale units based on waveform and dimensionless learning.

Combined with all the above-mentioned technical solutions, the advantages and positive effects possessed by the present invention are:

In the process of diagnosing friction faults of large units, the invention can effectively solve the problem of difficulty in extracting features; meanwhile, it can extract effective features to solve the problem of fault prediction, and at the same time, a new method of feature extraction is proposed. The invention achieves good results on the problem of friction fault diagnosis of large units.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present application more clearly, the following briefly introduces the drawings that need to be used in the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a flowchart of a method for analyzing friction faults of large units based on waveform and dimensionless learning provided by an embodiment of the present invention.

2 is a schematic structural diagram of a large unit friction fault analysis system based on waveform and dimensionless learning provided by an embodiment of the present invention;

In the figure: 1. Data acquisition module; 2. Feature extraction module; 3. Prediction model building module; 4. Fault prediction module.

FIG. 3 is a schematic structural diagram of a two-layer decomposition and transformation of a wavelet packet provided by an embodiment of the present invention.

detailed description

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In view of the problems existing in the prior art, the present invention provides a method and system for analyzing friction faults of large units based on waveform and dimensionless learning. The present invention is described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , the waveform and dimensionless learning-based friction fault analysis method for large units provided by the embodiment of the present invention includes:

S101: Extract the machine fault vibration signal by using dual probes, and preprocess the data.

S102: Extracting friction fault features.

S103: Use a machine learning method to establish a fault prediction model.

S104: Predict whether the unknown tag signal has a fault, and determine the fault type.

In S101 provided by the embodiment of the present invention, the process of preprocessing the machine fault vibration signal and data is as follows:

1) Install two probe points, and obtain the vibration double-view signal of the large-scale sliding unit through the acquisition of the two probe points. The data acquisition is 32/rms, that is, 32 points are sampled for each revolution of the bearing, and the data of 32 revolutions are collected.

The discrete Fourier transform formula is specifically as follows:

Where n=0, . . ., N-1, N represents the data length.

In S102 provided by the embodiment of the present invention, the process of extracting friction fault features is:

(1) Perform two-layer decomposition and transformation of wavelet packet on the fault signal. The wavelet packet is to analyze the details of the input signal by using multiple iterative wavelet transformation. The specific structure is shown in Figure 2. The wavelet coefficients at different scales are obtained, and the scale coefficients of the signal HH layer are set to zero.

in

represents the root mean square value of the waveform data,

Indicates the absolute average of waveform data.

where x _max represents the peak value of the waveform,

Indicates the root mean square value.

where x _max represents the peak value of the waveform,

Indicates the absolute average of waveform data.

in

(6) Calculate the dimensionless feature margin index, which is generally used to detect the wear of mechanical equipment. If the skewness index does not change much, the ratio of the effective value to the average value increases, indicating that the gap increases due to wear, so the effective value of the vibration energy index increases faster than the average value, and the margin index also increases, and the infinite The class feature margin index is one of the extracted features, and the specific calculation formula is as follows:

in

where t is the data collection time,

α _t is the offset angle before and after time t.

(8) Calculate the standard deviation. The standard deviation represents the degree of dispersion of the data and the variability of a single statistic in multiple sampling. It can be understood in this way that the former represents the variability of the data itself, while the latter represents the variability of sampling behavior. The specific calculation formula is as follows:

(9) Calculate the standard deviation of the mean value. The standard deviation of the mean value refers to a standard for measuring the degree of dispersion of the data distribution, which is used to measure the degree to which the data value deviates from the arithmetic mean. The smaller the standard deviation, the less the values deviate from the mean, and vice versa. The size of the standard deviation can be measured by the multiplying relationship between the standard deviation and the average. The specific calculation formula is as follows:

Where X is the sample, sin is the sine function, cos is the cosine function, arctan2 is the tangent function, and π is the pi.

in

S=∑ _i sin(angle) C=∑ _i cos(angle), res=arctan2(S, C).

In S103 provided by the embodiment of the present invention, the cca dimension is reduced by using the two view features extracted in S102, and the features of the two views after the dimension reduction are spliced together as an input vector, and a machine learning model is used for training.

As shown in FIG. 2 , the large-generator friction fault analysis system based on waveform and dimensionless learning provided by the embodiment of the present invention includes:

The data acquisition module 1 extracts the vibration signal of the machine fault by using double probes, and preprocesses the data.

The feature extraction module 2 performs feature extraction on the friction fault signal.

The prediction model building module 3 establishes a fault prediction model by using the machine learning method.

The fault prediction module 4 predicts whether there is a fault in the unknown tag signal, and determines the fault type.

The technical solution of the present invention is further described below by taking the sliding mechanical data of a certain large-scale unit as an example.

A certain large unit equipment provided by the present invention is provided with two probes for data collection, and the data is that 32 points are collected for each rotation of the mechanical bearing, and then the period is 32 rotations. A set of data is 1024 waveform points, and the length of the converted waveform is 1024. Extract the spectral vector and other dimensionless vectors from the data, and splicing them together to form the features of a view; perform feature dimension reduction on a single view through cca, splicing the dimensionality-reduced features, and then bring them into the machine learning model for training; The features are extracted from the fault signal of the label, and the fault prediction result is obtained through the trained model.

In the description of the present invention, unless otherwise stated, "plurality" means two or more; the terms "upper", "lower", "left", "right", "inner", "outer" The orientation or positional relationship indicated by , "front end", "rear end", "head", "tail", etc. are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, not An indication or implication that the referred device or element must have a particular orientation, be constructed and operate in a particular orientation, is not to be construed as a limitation of the invention. Furthermore, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be construed to indicate or imply relative importance.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person skilled in the art is within the technical scope disclosed by the present invention, and all within the spirit and principle of the present invention Any modifications, equivalent replacements and improvements made within the scope of the present invention should be included within the protection scope of the present invention.

Claims

A method for analyzing friction faults of large units based on waveform and dimensionless learning, characterized in that the method for analyzing friction faults for large units based on waveforms and dimensionless learning includes:

Use dual probes to extract machine fault vibration signals and preprocess the data;

Extraction of friction fault features;

Use machine learning methods to build fault prediction models;

Predict whether there is a fault in the unknown tag signal and determine the type of fault.
The friction fault analysis method for large units based on waveform and dimensionless learning as claimed in claim 1, wherein the process of preprocessing the machine fault vibration signal and data is as follows:

1) Install two probe points, and obtain the vibration double-view signal of the large-scale sliding unit through the acquisition of the two probe points. The data acquisition is 32/rms, that is, 32 points are sampled for each revolution of the bearing, and 32 circles of data are collected;

2) After the probe collects data, the discrete Fourier transform is performed in alignment, the window size is 32*32=1024 points, and the transformed Fourier value is modified; the adaptive threshold is set according to the signal condition, the signal storage capacity is reduced, and the speed is accelerated. transmission.
As claimed in claim 2, the friction fault analysis method of large unit based on waveform and dimensionless learning is characterized in that, the discrete Fourier transform formula in described 2) is as follows:

Where n=0, . . ., N-1, N represents the data length.
The friction fault analysis method for large units based on waveform and dimensionless learning according to claim 1, wherein the process of extracting the friction fault features is:

(1) Perform two-layer decomposition and transformation of wavelet packet on the fault signal. The wavelet packet analyzes the details of the input signal by using multiple iterative wavelet transforms, obtains the wavelet coefficients at different scales, and sets the scale coefficients of the HH layer of the signal to zero;

(2) Calculate the dimensionless characteristic wave index S f , take the wave index as one of the extracted features, and the specific calculation formula is as follows:

in
represents the root mean square value of the waveform data,
Indicates the absolute average of waveform data;

(3) Calculate the dimensionless feature peak index, and take the peak index as one of the extracted features. The specific calculation formula is as follows:

where x max represents the peak value of the waveform,
Represents the root mean square value;

(4) Calculate the dimensionless characteristic pulse index, and take the pulse index as one of the extracted features. The specific calculation formula is as follows:

where x max represents the peak value of the waveform,
Indicates the absolute average of waveform data;

(5) Calculate the dimensionless characteristic kurtosis index, which indicates the level of the actual kurtosis relative to the normal kurtosis. The kurtosis index reflects the shock characteristics in the vibration signal, and the dimensionless characteristic kurtosis index is used as one of the extracted features. The formula is as follows:

in

(6) Calculate the dimensionless feature margin index, which is generally used to detect the wear of mechanical equipment; if the skewness index does not change much, the ratio of the effective value to the average value increases, indicating that the gap increases due to wear, so the vibration The effective value of the energy index increases faster than the average value, and its margin index also increases. Taking the dimensionless feature margin index as one of the extracted features, the specific calculation formula is as follows:

in

(7) Calculate the dimensionless feature Teager energy operator, take the Teager energy operator as one of the extracted features, and the specific calculation formula is as follows:

where t is the data collection time,
α t is the offset angle before and after time t;

(8) Calculate the standard deviation. The standard deviation represents the degree of dispersion of the data, and it represents the variability of a single statistic in multiple sampling; it can be understood that the former represents the variability of the data itself, while the latter represents the variability of the data itself. is the variability of sampling behavior, and the specific calculation formula is as follows:

(9) Calculate the standard deviation of the mean. The standard deviation of the mean refers to a standard for measuring the degree of dispersion of the data distribution, which is used to measure the degree to which the data values deviate from the arithmetic mean; the smaller the standard deviation, the more these values deviate from the mean. The less, and vice versa; the size of the standard deviation can be measured by the multiplying relationship between the standard deviation and the average, and the specific calculation formula is as follows:

(10) Calculate the sample circle mean (circle_mean) of the sample, and take the sample circle mean as one of the extracted features. The specific calculation formula is as follows:

Where X is the sample, sin is the sine function, cos is the cosine function, arctan2 is the tangent function, and π is the pi;

in
S=Σ i sin(angle) C=Σ i cos(angle), res=arctan2(S, C).
The method for analyzing friction faults of large units based on waveform and dimensionless learning according to claim 1, wherein the cca dimension reduction is performed by using the extracted two view features, and the features of the two views after dimension reduction are spliced , as the input vector, to train the machine learning model.
A large-generator friction fault analysis system based on waveform and dimensionless learning for implementing the large-generator friction fault analysis method based on waveform and dimensionless learning according to any one of claims 1 to 5, characterized in that the waveform-based friction fault analysis system and dimensionless learning large unit friction fault analysis system, including:

The data acquisition module extracts the machine fault vibration signal by using double probes, and preprocesses the data;

Feature extraction module, to perform feature extraction on friction fault signal;

Prediction model building module, by using machine learning methods, to establish a fault prediction model;

The fault prediction module predicts whether there is a fault in the unknown tag signal, and determines the fault type.
A computer device, characterized in that the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the following steps:

Use dual probes to extract machine fault vibration signals and preprocess the data;

Feature extraction for friction fault signals;

Build fault prediction models by using machine learning methods;

Predict whether there is a fault in the unknown tag signal and determine the type of fault.
A computer-readable storage medium storing a computer program, when the computer program is executed by a processor, the processor causes the processor to perform the following steps:

Extracting machine fault vibration signals by using dual probes and preprocessing the data;

Feature extraction for friction fault signals;

Build fault prediction models by using machine learning methods;

Predict whether there is a fault in the unknown tag signal and determine the type of fault.
A large-scale equipment for implementing the method for analyzing friction faults of large units based on waveform and dimensionless learning according to any one of claims 1 to 5.