CN-121980413-A - Mechanical intelligent fault diagnosis method and device based on multi-mode sensing signals

CN121980413ACN 121980413 ACN121980413 ACN 121980413ACN-121980413-A

Abstract

The invention discloses a mechanical intelligent fault diagnosis method and device based on multi-mode sensing signals, which belong to the field of mechanical equipment fault diagnosis and comprise the steps of collecting and preprocessing multi-mode data, namely collecting three-channel original signals of two vibration channels and one current channel of mechanical rotating equipment under normal and multiple fault working conditions, and intercepting equal-length fragments to form a sample; the method comprises the steps of constructing and training a convolutional neural network model, carrying out deep feature learning on a multi-mode sample, inputting a multi-mode signal to be detected into the trained model, carrying out fault diagnosis reasoning, and outputting a corresponding diagnosis result. The invention integrates multi-sensor information and a deep learning algorithm, and realizes high-accuracy fault detection and intelligent health state assessment of the mechanical rotating component.

Inventors

Xiong Lechao
XU LIJUN
WANG YANHUA
ZHANG ZHEN
YAO HUI
ZHU LIJIAN

Assignees

上海新力动力设备研究所

Dates

Publication Date: 20260505
Application Date: 20251223

Claims (9)

1. A mechanical intelligent fault diagnosis method based on multi-mode sensing signals is characterized by comprising the following steps: collecting multi-mode sensing signal data of a mechanical rotating part which normally operates under different operating conditions, intercepting the collected multi-mode sensing signals into a plurality of pieces of data fragments with the same length according to the same time length to form a sample, and constructing a multi-mode sample data set with health status labels; In the model training process, parameters of the convolutional neural network are adjusted through forward propagation and backward propagation, so that the model can learn characteristic modes under various health states according to multi-mode input; Intercepting and processing multi-mode sensing signal data of the machine to be detected according to the equal time length, inputting the data into a trained convolutional neural network model, analyzing and judging the input data by utilizing the model, and outputting a diagnosis result corresponding to the current working condition of the machine, wherein the diagnosis result comprises a plurality of types of predefined health or fault states.
2. The intelligent mechanical fault diagnosis method based on the multi-mode sensing signals according to claim 1, wherein the multi-mode sensing signals come from a rotating part of the machine, the multi-mode sensing signals comprise two-channel bidirectional vibration signals collected by vibration sensors arranged at a bearing part and a gear box part, and single-phase current signals collected by current sensors arranged at a motor power supply circuit, and all the channel signals are collected synchronously.
3. The mechanical intelligent fault diagnosis method based on the multi-mode sensing signal according to claim 1, wherein the convolutional neural network model comprises five convolutional pooling units, one adaptive pooling layer and two full-connection layers which are sequentially connected, wherein: Each convolution pooling unit comprises a convolution block and a maximum pooling layer which are connected in sequence.
4. The mechanical intelligent fault diagnosis method based on the multi-mode sensing signals according to claim 3, wherein each convolution block comprises a one-dimensional convolution operation layer, a batch normalization layer and a nonlinear activation function layer which are sequentially connected, and the calculation form of a first convolution block is as follows: Wherein the method comprises the steps of Represent the first Layer input feature map (L) The flow path of the liquid is provided with a channel, Is the first Layer number The weights of the individual convolution kernels are, A convolution operation is represented and is performed, As a bias term, σ () is a nonlinear activation function.
5. A mechanical intelligent fault diagnosis method based on multi-mode sensing signals as set forth in claim 3, characterized in that Layer maximizing pooling layer output The method comprises the following steps: Wherein, the Representing the set of indices within the current pooling window, Represent the first And the characteristic value with the index m in the characteristic diagram output by each convolution block.
6. The method for mechanical intelligent fault diagnosis based on multi-modal sensing signals as set forth in claim 3, wherein the adaptive pooling layer receives deep features outputted by the last largest pooling layer and maps the deep features to feature vectors z with fixed lengths: Wherein the method comprises the steps of For the deep features output by the last convolution pooling unit, Representing that the extracted deep features are mapped to feature vectors with fixed length, wherein K is a fixed dimension and is a positive integer.
7. The mechanical intelligent fault diagnosis method based on the multi-mode sensing signals according to claim 3, wherein the mechanical intelligent fault diagnosis method is characterized in that: The first full-connection layer receives the fixed-length feature vector output by the self-adaptive pooling layer, performs linear transformation and nonlinear mapping processing on the feature vector to realize further fusion and compression of high-level semantic features, and transmits the processed feature representation as input to the second full-connection layer for classification and discrimination of the subsequent multi-class working condition states; the second full-connection layer is used for outputting classification results of the multiple working condition states; the output of the two full-connection layers is as follows: Wherein, the In order to activate the function, Is a probability distribution vector of various working condition states, The bias term for the first fully connected layer, The bias term for the second fully connected layer, For the weight matrix of the first fully connected layer, Is the feature vector extracted by the convolution layer and the pooling layer.
8. The method for mechanical intelligent fault diagnosis based on multi-modal sensing signals as claimed in claim 4, wherein the activation function of the nonlinear activation function layer is Mish functions.
9. The mechanical intelligent fault diagnosis device based on the multi-mode sensing signals is characterized by comprising a data acquisition and preprocessing module, a data set storage module and an intelligent fault diagnosis module, wherein: The data acquisition and preprocessing module is used for acquiring a double-channel bidirectional vibration signal and a single-phase current signal of the mechanical rotating component, intercepting the acquired double-channel bidirectional vibration signal and single-phase current signal into a plurality of sections of data fragments with the same length according to the equal time length to form multi-mode sample data, attaching a label to the data, and outputting the data to the data set storage module or the intelligent fault diagnosis module; the data set storage module stores multi-mode sample data with labels; The intelligent fault diagnosis module is used for storing the convolutional neural network model, calling the multi-mode sample data in the data set storage module to train the convolutional neural network model, and outputting a diagnosis result of the current working condition of the machine based on the multi-mode sample data output by the data acquisition and preprocessing module after training is completed. The mechanical intelligent fault diagnosis device based on the multi-mode sensing signal according to claim 9, wherein the convolutional neural network model comprises five convolutional pooling units, one adaptive pooling layer and two fully connected layers which are sequentially connected, wherein: each convolution pooling unit comprises a convolution block and a maximum pooling layer which are sequentially connected; Each convolution block comprises a one-dimensional convolution operation layer, a batch normalization layer and a nonlinear activation function layer which are sequentially connected, wherein the activation function of the nonlinear activation function layer adopts Mish functions.

Description

Mechanical intelligent fault diagnosis method and device based on multi-mode sensing signals Technical Field The invention relates to a mechanical intelligent fault diagnosis method and device based on multi-mode sensing signals, and belongs to the field of mechanical equipment fault diagnosis. Background Rotary machines (e.g., motors, gearboxes, bearings, etc.) in machines are critical components in industrial processes, the operating state of which is directly related to the safety and efficiency of the production. However, these rotating parts operate at high speed for a long period of time and are subjected to alternating load, and various failures such as bearing wear, gear damage, motor short circuit and the like are liable to occur. Once critical components fail, equipment downtime, production interruption, and even safety accidents and economic losses can result. Therefore, the fault of the rotary machine can be effectively detected and diagnosed in time, and the method has important significance for guaranteeing the stable operation of the machine. Traditional mechanical fault diagnosis methods often rely on a single sensor signal and manually extracted features. For example, only the bearing vibration signal or the motor current signal is used, the characteristic frequency or the statistical characteristic is determined empirically, and then the classification diagnosis is performed by means of a traditional machine learning method such as a Support Vector Machine (SVM). The single mode signal method can play a role in some cases, but due to the complex working conditions of the industrial field, multiple factors can influence the characteristics of the single signal, so that the diagnosis accuracy is reduced. For example, when fault signals are inundated with noise or different faults behave similarly on a single signal, diagnosis based on a single mode signal may be difficult to distinguish. In addition, the SVM and other methods need to manually select features, the feature extraction process is complicated, the expert experience is greatly dependent, and the new fault mode is difficult to adapt in time. In recent years, with the development of sensor technology and data acquisition technology, industrial sites can acquire various forms of sensing data at the same time, and in addition, the improvement of computing capability and deep learning are widely applied in the field of fault diagnosis. The deep learning models such as Convolutional Neural Network (CNN) can automatically learn fault characteristics from original signals, and the end-to-end intelligent diagnosis is realized, so that the method has higher robustness and accuracy compared with the traditional method. However, most current studies apply deep learning models only for single modality signals, underutilizing complementary information from different sensing sources. Therefore, an intelligent fault diagnosis method capable of fusing various sensing signals is needed to improve the fault detection accuracy and generalization capability of complex machinery. The invention patent refers to a method and a system for diagnosing bearing faults (inventor: zhu Wangchun, zhou Canwen, jun, yang Dihuan, chen Yuxia, wang Jing; applicant: guilin university of electronic technology; application (patent) No. CN202311363625. X) The invention provides a bearing fault diagnosis method and system based on an embedded multi-layer small convolution structure deep learning network model. However, the fault diagnosis mode provided by the invention only faces a single bearing object, ignores the complex space-time coupling relation of multi-mode data of each key component in the transmission chain of the whole system, and cannot perform real-time fault diagnosis on multiple key components. Disclosure of Invention The invention solves the problems of overcoming the defects of the prior art, providing a mechanical intelligent fault diagnosis method and device based on multi-mode sensing signals, realizing comprehensive and reliable fault monitoring and diagnosis of mechanical rotating parts and improving the detection accuracy. The technical scheme of the invention is as follows: In a first aspect, a mechanical intelligent fault diagnosis method based on multi-mode sensing signals is provided, including: collecting multi-mode sensing signal data of a mechanical rotating part which normally operates under different operating conditions, intercepting the collected multi-mode sensing signals into a plurality of pieces of data fragments with the same length according to the same time length to form a sample, and constructing a multi-mode sample data set with health status labels; In the model training process, parameters of the convolutional neural network are adjusted through forward propagation and backward propagation, so that the model can learn characteristic modes under various health states according to multi-mode input; Intercepting and processing mu