CN-122008207-A - Fault diagnosis system and industrial robot state detection and fault diagnosis device

CN122008207ACN 122008207 ACN122008207 ACN 122008207ACN-122008207-A

Abstract

The invention provides a fault diagnosis system and an industrial robot state detection and fault diagnosis device, wherein the fault diagnosis system comprises a signal preprocessing module, a time-frequency characteristic extraction module, a fault diagnosis module and a classifier, wherein the fault diagnosis module adopts a mixed refining attention mechanism to respectively carry out weight distribution on original characteristics, spliced enhanced time-domain characteristics and frequency-domain combined characteristics, filters important time-frequency characteristics, weights and fuses, enriches and highlights associated fault characteristics so as to extract key fault characteristics of a tag sample in multiple dimensions, and the classifier uses a full-connection layer to construct a mapping relation between the extracted characteristics of the fault diagnosis module and fault tags to classify faults, so that a fault type diagnosis result is obtained. The invention obviously improves the diagnosis precision and efficiency of typical faults such as abrasion of the industrial robot reducer, bearing clamping stagnation and the like.

Inventors

CHEN FEI
CAO YUN
XU BINBIN
DING YANAN
LI JIAN
HE LONGBO

Assignees

深圳技术大学

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (10)

1. A fault diagnosis system, comprising: the signal preprocessing module is used for obtaining an original signal, carrying out characteristic enhancement on the original signal to obtain an enhanced time domain signal, respectively converting the original signal and the enhanced time domain signal into frequency domain signals and splicing the frequency domain signals to obtain a frequency domain combined signal; The time-frequency characteristic extraction module is used for extracting characteristics in an original signal, an enhanced time domain signal and a frequency domain combined signal through nerve convolution to obtain the original characteristics, the enhanced time domain characteristics and the frequency domain combined characteristics, wherein the original characteristics are further enhanced and are subjected to characteristic splicing with the enhanced time domain characteristics to obtain spliced enhanced time domain characteristics; the fault diagnosis module adopts a mixed refining attention mechanism to respectively carry out weight distribution on the original features, the spliced enhanced time domain features and the frequency domain combined features, screens out important time-frequency feature weighted fusion, enriches and highlights associated fault features, thereby drawing key fault features of multiple dimensions of the label sample; And the classifier is used for constructing a mapping relation between the extracted features of the fault diagnosis module and the fault labels by using the full connection layer, classifying the faults and obtaining a fault type diagnosis result.
2. The fault diagnosis system according to claim 1, wherein the signal processing module comprises a first Gramian feature enhancement module, a first fourier transform module, a second fourier transform module and a signal splicing module, wherein an input end of the Gramian feature enhancement module inputs an original signal and outputs an enhanced time domain signal, an input end of the first fourier transform module is connected with an output end of the Gramian feature enhancement module, an input end of the second fourier transform module inputs the original signal, and input ends of the signal splicing module are respectively connected with output ends of the first fourier transform module and the second fourier transform module, and an output end of the signal splicing module outputs a frequency domain combined signal.
3. The fault diagnosis system according to claim 2, wherein the first Gramian feature enhancement module processes the input signal in the following manner: (1) Converting the acquired vibration signal into a two-dimensional signal matrix, The vibration signal comprises a series of sample points, the vibration signal being expressed as: l is the length of the sample point, , The one-dimensional original vibration signal is rearranged into a two-dimensional signal matrix X in a segmentation way, To facilitate data conversion, the sample length L is set to , wherein, Is an integer if Even, the original signal matrix is constructed with the dimensions of If (3) Odd, the dimension of the constructed original signal matrix is ; (2) Gramian matrix for calculating row and column vectors of two-dimensional signal matrix X And ; (3) The newly generated Gramian matrix And The characteristic enhancement matrix is regarded as, and is multiplied by the original two-dimensional signal matrix to enhance the characterization of the internal fault characteristics of the matrix, thus obtaining the two-dimensional characteristic enhancement signal matrix : (4) Enhancing a two-dimensional characteristic into a signal matrix The signals are spread in rows to obtain one-dimensional characteristic enhancement signals , 。
4. The fault diagnosis system according to claim 3, wherein the time-frequency feature extraction module comprises three convolution layer processing modules, a second Gramian feature enhancement module and a feature splicing module, wherein three input ends of the three convolution layer processing modules are respectively in one-to-one correspondence with three output signals of the signal preprocessing module, an input end of a first convolution layer processing module inputs an original signal, an output end outputs an original feature, an input end of a second convolution layer processing module inputs an enhanced time domain signal, an input end of the first convolution layer processing module inputs a frequency domain combined signal, an input end of the second Gramian feature enhancement module is connected with an output end of the first convolution layer processing module, and an input end of the feature splicing module is respectively connected with an output end of the second convolution layer processing module and an output end of the second Gramian feature enhancement module.
5. The fault diagnosis system according to claim 4, wherein the convolution layer processing module comprises a one-dimensional convolution operation layer, a batch normalization layer, an activation function and a random discarding layer, which are sequentially arranged according to a signal processing sequence, wherein the one-dimensional convolution operation layer is used for initially extracting features from an input signal, the batch normalization layer is used for relieving gradient diffusion problems and improving generalization capability of a model, the activation function is used for introducing nonlinearity, so that the feature extraction layer can learn and simulate complex nonlinear relations between the signals and the features, and the random discarding layer is used for discarding a part of neurons in an iterative process of training to prevent overlearning of the model on training data and cause overfitting.
6. The fault diagnosis system as claimed in any one of claims 1-5, wherein the fault diagnosis module comprises: the channel attention refining module is used for acquiring channel attention refining weights; The first refining module multiplies the input characteristics with the channel attention refining weights element by element to obtain channel refining characteristics; The spatial attention refining module is used for dividing channel refining characteristics into first channel refining characteristics and second channel refining characteristics, then acquiring first spatial attention refining weights based on the first channel refining characteristics and acquiring second spatial attention refining weights based on the second channel refining characteristics; The second refining module multiplies the first channel refining feature by the first space attention refining weight element by element to obtain a first space refining feature, and multiplies the second channel refining feature by the second space attention refining weight element by element to obtain a second space refining feature; and the fusion module is used for fusing the first space refining feature and the second space refining feature to obtain a final refining feature.
7. The fault diagnosis system according to claim 6, wherein the channel attention refining module processes: extracting global feature information in feature channels using an averaging pooling layer Extracting most significant feature information in a channel of an input feature by using a maximum pooling layer Aggregating information in both using an adaptive weighting mechanism, setting trainable parameters And For adaptively adjusting the average pooled feature and the occupied feature weight of the maximized pooled feature in training, the aggregated refined feature Sending the data into a convolution layer for learning to obtain channel attention refining weight of each channel 。
8. The fault diagnosis system according to claim 7, wherein the processing method of the spatial attention refining module is as follows: by comparing the channel attention refinement weights, it is determined which channels are important channels and which are secondary important channels, and then the important channels are characterized And the next most important features Separated, polymerized by average pooling and maximum pooling And Channel dimension information on the channel is obtained And Extracting features from the aggregated channel dimension information by using a convolution layer to obtain important channel features Is of the spatial concentration weight of (2) And the next most important channel features Spatial attention weighting The calculation formula is as follows: Wherein, the Representing a series of non-linear computing operations, including batch normalization, a ReLU activation function and a sigmoid activation function, Representing a one-dimensional convolution operation with a number of convolution kernels of 7.
9. The industrial robot state detection and fault diagnosis device comprises the fault diagnosis system according to any one of claims 1-8, and is characterized by further comprising a data acquisition system and a state monitoring system, wherein the data acquisition system is used for acquiring various sensor parameter information of the industrial robot and then uploading the sensor parameter information to an upper computer; The state monitoring system is used for acquiring signals acquired by the data acquisition system, constructing a multidimensional health index system through time domain characteristic parameters, observing multichannel signals acquired by different sensors in real time, calculating time domain indexes of different channels and monitoring time domain characteristics of the industrial robot in real time.
10. The apparatus for detecting and diagnosing the state of the industrial robot according to claim 9, wherein the time domain characteristic parameters comprise spectral kurtosis, peak-to-peak value, kurtosis, waveform factors and spectral energy of dimensionless values, and the state monitoring system extracts the dimensionless values with the lowest correlation with the rotating speed through an optimal order method and inputs the dimensionless values into the model to obtain a state monitoring result.

Description

Fault diagnosis system and industrial robot state detection and fault diagnosis device Technical Field The invention relates to the technical field of industrial robot monitoring, in particular to a fault diagnosis system and a state detection and fault diagnosis device comprising the fault diagnosis system and an industrial robot. Background With the development of industrial automation and intelligence, industrial robots are increasingly used in production and manufacturing. Robots often need to operate in complex environments for long periods of time when performing various tasks, which makes their operational status monitoring particularly important. The traditional industrial robot fault diagnosis generally needs professional engineers or technicians to determine the fault cause of mechanical equipment through observation, test and analysis, and the technicians without sufficient experience cannot accurately diagnose the fault state of the machinery in time for maintenance, so that the production efficiency is affected. Therefore, it is necessary to enable inexperienced workers to diagnose and deal with mechanical faults in time by means of a fault diagnosis apparatus. The existing data acquisition equipment and fault diagnosis equipment have certain limitations in terms of volume, weight, power consumption, data processing capacity and the like, and cannot simultaneously acquire data by utilizing different sensors to perform fault diagnosis by combining advanced artificial intelligence and big data technology, and meanwhile cannot meet the current requirements for on-site real-time data acquisition and fault diagnosis. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a fault diagnosis system, and also provides a state detection and fault diagnosis device comprising the fault diagnosis system and an industrial robot, aiming at solving the problem of power frequency interference in current signal analysis and the fault diagnosis challenge in industrial scenes. The fault diagnosis system of the present invention includes: the signal preprocessing module is used for obtaining an original signal, carrying out characteristic enhancement on the original signal to obtain an enhanced time domain signal, respectively converting the original signal and the enhanced time domain signal into frequency domain signals and splicing the frequency domain signals to obtain a frequency domain combined signal; The time-frequency characteristic extraction module is used for extracting characteristics in an original signal, an enhanced time domain signal and a frequency domain combined signal through nerve convolution to obtain the original characteristics, the enhanced time domain characteristics and the frequency domain combined characteristics, wherein the original characteristics are further enhanced and are subjected to characteristic splicing with the enhanced time domain characteristics to obtain spliced enhanced time domain characteristics; the fault diagnosis module adopts a mixed refining attention mechanism to respectively carry out weight distribution on the original features, the spliced enhanced time domain features and the frequency domain combined features, screens out important time-frequency feature weighted fusion, enriches and highlights associated fault features, thereby drawing key fault features of multiple dimensions of the label sample; And the classifier is used for constructing a mapping relation between the extracted features of the fault diagnosis module and the fault labels by using the full connection layer, classifying the faults and obtaining a fault type diagnosis result. Further, the signal processing module comprises a first Gramian feature enhancement module, a first fourier transform module, a second fourier transform module and a signal splicing module, wherein the input end of the Gramian feature enhancement module inputs an original signal and outputs an enhanced time domain signal, the input end of the first fourier transform module is connected with the output end of the Gramian feature enhancement module, the input end of the second fourier transform module inputs the original signal, the input end of the signal splicing module is respectively connected with the output ends of the first fourier transform module and the second fourier transform module, and the output end of the signal splicing module outputs a frequency domain combined signal. Further, the processing mode of the first Gramian feature enhancement module on the input signal is as follows: (1) Converting the acquired vibration signal into a two-dimensional signal matrix, The vibration signal comprises a series of sample points, the vibration signal being expressed as: l is the length of the sample point, ,The one-dimensional original vibration signal is rearranged into a two-dimensional signal matrix X in a segmentation way, To facilitate data conversion, the sample length