CN-122020359-A - Multi-defect identification method for main bearing of shield

CN122020359ACN 122020359 ACN122020359 ACN 122020359ACN-122020359-A

Abstract

The invention discloses a shield main bearing multi-defect identification method which comprises the steps of multi-source signal acquisition, feature extraction, feature fusion and normalization, mode identification and threshold analysis, wherein after vibration signals and current signals are acquired, time domain and frequency domain feature extraction, feature vector set fusion construction and normalization processing are respectively carried out, a convolutional neural network mode identification algorithm is adopted to train and classify the normalized feature vector set, whether the main bearing has defects of cracking and peeling or not is identified, accurate distinction is carried out, and meanwhile, a threshold analysis rule is set to verify and assist in judging a mode identification result. According to the method, the complete identification flow is constructed by fusing the multi-source signal information, and the accuracy and the reliability of multi-defect identification of the main bearing of the shield can be effectively improved by combining pattern identification and threshold analysis, so that powerful support is provided for equipment maintenance and safe operation.

Inventors

XU QINWEI
ZHAO ZHILIN
Zhao Shuangcong
CAO JIANXIN
MEN YANQING
LIU FENGZHOU
LI HU
GAO MINGXIN
XIE HAO
SUN PEIXIN
WANG QI

Assignees

济南轨道交通集团有限公司
山东轨道交通研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. The method for identifying the multiple defects of the main bearing of the shield is characterized by comprising the following specific steps: s1, collecting multi-source signals including vibration signals and current signals in the running process of a main bearing of a shield; S2, extracting time domain features of the acquired vibration signals, and calculating the mean value, variance, root mean square value and kurtosis value of the vibration signals; extracting frequency domain features of the vibration signal, obtaining a spectrogram by utilizing fast Fourier transform, and determining characteristic frequency points with prominent frequency components in the spectrogram and amplitude values thereof; s3, fusing the extracted vibration signal time domain features, the frequency domain features and the current signal time domain features to construct a feature vector set; S4, carrying out normalization processing on the constructed feature vector set to enable feature data to be located in the [0,1] interval; s5, training and classifying the normalized feature vector set by adopting a pattern recognition algorithm; And S6, setting a threshold analysis rule, verifying and assisting in judging the pattern recognition result, and further confirming or judging the defect type as an abnormal state by combining threshold analysis when the confidence of the pattern recognition result is lower than a threshold.
2. The method for identifying the multiple defects of the main bearing of the shield according to claim 1 is characterized in that in the step S1, a high-precision acceleration sensor is adopted for collecting vibration signals, the sampling frequency is not lower than 10kHz, a current transformer is adopted for collecting current signals, and the sampling frequency is not lower than 1kHz.
3. The method for identifying multiple defects of a main bearing of a shield according to claim 1, wherein in S2, when frequency domain feature extraction is performed on a vibration signal, feature frequency points and amplitudes thereof in different frequency bands are respectively determined by performing segmentation processing on a spectrogram.
4. The method for identifying multiple defects of a main bearing of a shield according to claim 3, wherein in S2, extracting frequency domain features of the vibration signal specifically includes: a, fast Fourier transform, namely setting the acquired vibration signal sequence as Sampling frequency is The sampling point number is The frequency spectrum is obtained through fast Fourier transform The calculation formula is as follows: ; calculating frequency and amplitude, calculating frequency spectrum The corresponding frequencies are: The frequency spectrum amplitude adopts an amplitude spectrum, and the calculation formula is as follows: ; c, determining characteristic frequency points and amplitude values thereof, namely firstly calculating the amplitude values of all the frequency points, and then sequencing according to the amplitude values from large to small, and taking the characteristic frequency points before taking the characteristic frequency points The frequency points corresponding to the amplitude values are taken as characteristic frequency points, and the corresponding amplitude values form a frequency domain characteristic vector.
5. The method for identifying multiple defects of a main bearing of a shield according to claim 1, wherein the step S3 specifically comprises the following steps: S31, defining characteristic dimension, and setting a vibration signal time domain characteristic vector as Wherein As the mean value of the vibration signal, In order to be the standard deviation of the vibration signal, For the root mean square value of the vibration signal, Is the kurtosis value of the vibration signal, and the frequency domain characteristic vector of the vibration signal is Wherein Is characterized by the frequency point at which the frequency of the signal is higher, Is the amplitude of the corresponding characteristic frequency point, and the time domain characteristic vector of the current signal is Wherein Is the average value of the current signal, Is the standard deviation of the current signal, As the root mean square value of the current signal, Is the peak value of the current signal; s32, feature fusion is carried out to construct a feature vector set, namely, vibration signal time domain feature vectors are used for Sum current signal time domain eigenvector Sequentially splicing to form a complete feature vector The method comprises the following steps: 。
6. the method for identifying multiple defects of the main bearing of the shield according to claim 1, wherein the step S4 specifically comprises the following steps: s41, defining feature dimension and sample number, and setting up a constructed feature vector set Wherein each feature vector Comprises The dimensions of the features are such that, Is the number of samples; s42, calculating the maximum value and the minimum value of each characteristic dimension, namely, for each characteristic dimension Calculate its maximum value in all samples And minimum value ; S43, performing linear transformation on each characteristic dimension, namely, performing linear transformation on each characteristic vector Is defined by each feature dimension of (1) Performing linear transformation according to the following formula to obtain a normalized feature vector set : Wherein, the method comprises the steps of, Represent the first The first feature vector The original values of the individual feature dimensions, Represent the first The first feature vector Normalized values for the individual feature dimensions.
7. The method for identifying multiple defects of a main bearing of a shield according to claim 1, wherein in S5, a convolutional neural network based on deep learning is adopted by a pattern recognition algorithm, and the structure of the pattern recognition algorithm comprises an input layer, a plurality of convolutional layers, a pooling layer, a full-connection layer and an output layer.
8. The method for identifying multiple defects of the main bearing of the shield according to claim 7, wherein the step S5 specifically comprises the following steps: S51, training process, namely defining a loss function as a cross entropy loss function: ; Wherein, the Is the actual label of the sample and, The sample that is network prediction belongs to Probability of class; updating network weights by using random gradient descent or Adam optimization algorithm, wherein the learning rate is as follows The iteration number is epoch; s52, in the classification process, the normalized feature vector is input into a trained CNN model, probability distribution that each sample belongs to different defect categories is output, and the category corresponding to the maximum probability value is taken as a classification result.
9. The method for identifying multiple defects of a main bearing of a shield according to claim 1, wherein the step S6 specifically comprises the following steps: S61, defining a confidence threshold and a feature threshold, wherein the confidence threshold of the pattern recognition result is set as the confidence threshold Setting corresponding characteristic threshold values according to historical data and expert experience for each type of typical defects of the main bearing; s62, judging the confidence coefficient of the pattern recognition result, namely taking the maximum probability value as the confidence coefficient of the pattern recognition result for the probability distribution of each sample belonging to various defects output by the pattern recognition algorithm The method comprises the following steps: wherein Representing that the sample belongs to The probability of a class defect is determined, Is the total number of defect categories; S63, threshold analysis logic, if the confidence level is high Directly adopting the defect category corresponding to the pattern recognition result as a final judgment result, and if the confidence coefficient is high The threshold analysis auxiliary decision flow is triggered.
10. The method for identifying multiple defects of a main bearing of a shield according to claim 9, wherein in S63, the trigger threshold analysis auxiliary determination process specifically includes: S631, for each type of typical defects, respectively checking whether the corresponding characteristic value of the current sample exceeds the characteristic threshold value of the type of defects, S632, judging that the type of defects is the type of defects if the characteristic value of the sample meets the characteristic threshold value condition of the type of defects, comprehensively judging by combining expert experience or further diagnostic rules if the characteristic threshold value condition of a plurality of defect types is met, judging that the type of defects are abnormal if the characteristic threshold value condition of any type of defects is not met, and prompting that the operation condition of the main bearing needs to be further checked and analyzed.

Description

Multi-defect identification method for main bearing of shield Technical Field The invention relates to the technical field of shield machines, in particular to a multi-defect identification method for a main bearing of a shield. Background The shield machine plays a key role in the tunnel excavation process, and the main bearing is used as a core component of the shield machine, and the running state of the main bearing is directly related to the stability and the construction efficiency of the whole equipment. In the long-term operation of the main bearing, various typical defects such as cracks, flaking and the like are easy to occur, and if the defects cannot be identified and processed accurately in time, equipment failure and even serious accidents can be caused. However, the existing main bearing defect identification method often has a certain limitation, is difficult to comprehensively analyze various source signals such as vibration and current at the same time, and still needs to be improved in the aspect of accurately distinguishing different defect types. Therefore, a method for effectively fusing the characteristics of multi-source signals and implementing precise identification of various defects by using an advanced pattern recognition algorithm is needed. Disclosure of Invention The invention aims to solve the defects in the prior art and provides a multi-defect identification method for a main bearing of a shield. In order to achieve the above purpose, the present invention adopts the following technical scheme: A multi-defect identification method for a shield main bearing comprises the following specific steps: s1, collecting multi-source signals including vibration signals and current signals in the running process of a main bearing of a shield; S2, extracting time domain features of the acquired vibration signals, and calculating the mean value, variance, root mean square value and kurtosis value of the vibration signals; extracting frequency domain features of the vibration signal, obtaining a spectrogram by utilizing fast Fourier transform, and determining characteristic frequency points with prominent frequency components in the spectrogram and amplitude values thereof; s3, fusing the extracted vibration signal time domain features, the frequency domain features and the current signal time domain features to construct a feature vector set; S4, carrying out normalization processing on the constructed feature vector set to enable feature data to be located in the [0,1] interval; S5, training and classifying the normalized feature vector set by adopting a pattern recognition algorithm, and recognizing whether the main bearing has typical defects such as cracks, flaking and the like and realizing accurate distinction; And S6, setting a threshold analysis rule, verifying and assisting in judging the pattern recognition result, and further confirming or judging the defect type as an abnormal state by combining threshold analysis when the confidence of the pattern recognition result is lower than a threshold. In the S1, a high-precision acceleration sensor is adopted for collecting vibration signals, the sampling frequency is not lower than 10kHz, a current transformer is adopted for collecting current signals, and the sampling frequency is not lower than 1kHz. In S2, when the frequency domain feature extraction is performed on the vibration signal, the frequency spectrum diagram is processed in a segmentation way, so that the feature frequency points and the amplitudes thereof in different frequency bands are respectively determined, and the vibration characteristics of the main bearing are more comprehensively reflected. As a further technical solution of the present invention, in S2, the extracting the frequency domain feature of the vibration signal specifically includes: a, fast Fourier transform, namely setting the acquired vibration signal sequence as Sampling frequency isThe sampling point number isThe spectrum is obtained by Fast Fourier Transform (FFT)The calculation formula is as follows:; calculating frequency and amplitude, calculating frequency spectrum The corresponding frequencies are: The frequency spectrum amplitude adopts an amplitude spectrum, and the calculation formula is as follows: ; c, determining characteristic frequency points and amplitude values thereof, namely firstly calculating the amplitude values of all the frequency points, and then sequencing according to the amplitude values from large to small, and taking the characteristic frequency points before taking the characteristic frequency points Frequency points corresponding to the amplitude values are taken as characteristic frequency points, and the corresponding amplitude values form a frequency domain characteristic vector, whereinAnd determining the number of the preset characteristic frequency points according to actual experience or requirements. As a further technical scheme of the present invention,