CN114993669B - Multi-sensor information fusion transmission system fault diagnosis method and system - Google Patents

Abstract

The invention provides a transmission system fault diagnosis method and system based on multi-sensor information fusion, which belong to the technical field of transmission system fault diagnosis, and comprise the steps of obtaining vibration signals, motor stator current signals and sound signals collected by sensors distributed in a transmission system, and carrying out sparse resonance decomposition; inputting the acquired preprocessing data into respective convolutional neural networks in a parallel mode to learn characteristics, and extracting fault characteristics in different types of signals; inputting the obtained feature data into a multi-task learning module in a parallel mode for processing; the multi-task learning module comprises a supervised classification task and a characteristic measurement learning task, so that fault classification is realized. The obtained fault classification is combined with the acoustic array sensor, the positioning of the abnormal part of the sound is analyzed, the accurate judgment of the fault is finally realized, the method can be used as the effective verification of the fault diagnosis, and the fault diagnosis accuracy is further improved.

Description

Multi-sensor information fusion transmission system fault diagnosis method and system

Technical Field

The invention relates to the technical field of transmission system fault diagnosis, in particular to a transmission system fault diagnosis method and system based on multi-sensor information fusion.

Background

The transmission system is a key part for the operation of mechanical equipment and bears the key role of transmitting kinetic energy. High failure rates in the drive train are a significant cause of machine downtime. Therefore, real-time monitoring and fault diagnosis of the transmission system are of great significance to guarantee the operation of equipment.

Research shows that when a transmission system is abnormal, the transmission system not only shows abnormal vibration, but also causes the magnetic flux of a stator of the motor to change, finally changes a series of current parameters including the stator current and the like, and generates abnormal sound changes. Therefore, the vibration signal, the current signal, and the sound signal have complementarity in the failure diagnosis. Currently, the real-time monitoring and fault diagnosis of a transmission system mainly comprises a method based on vibration signals, but the installation position and the installation mode of signal acquisition equipment influence the effect of the method; compared with a vibration signal, the current signal and the sound signal are easy to obtain, but the fault information is weak. Therefore, a fault diagnosis method for multi-sensor fusion is designed, aiming at learning complementary diagnosis characteristic information in vibration, current and sound signals. And then, the distance reduction in the class of the similar samples is restrained by measuring the learning auxiliary task, so that the discrimination capability of the learned features is enhanced, and the performance of the fault classification main task is improved. And finally, combining the obtained fault classification with the positioning of the acoustic array sensor on the acoustic abnormal part to finally realize the accurate judgment of the fault.

Disclosure of Invention

Aiming at the defects, the invention provides the transmission system fault diagnosis method with multi-sensor information fusion, which extracts and fuses useful complementary features of multi-sensor signals, improves the fault classification and diagnosis capability of the transmission system by combining the positioning function of the acoustic array sensor on fault sound, and provides the fault diagnosis method with engineering practical value.

To achieve the above object, the present invention provides the following technical solutions:

the method for diagnosing the faults of the multi-sensor information fusion transmission system comprises the following steps:

step 1: acquiring vibration signals, generator stator current signals and sound signals collected by sensors distributed in a transmission system, and performing sparse resonance decomposition after segmenting the acquired signals;

and 2, step: inputting the preprocessed data obtained in the step 1 into respective convolutional neural networks in a parallel mode to learn characteristics, and extracting fault characteristics in different types of signals;

and step 3: inputting the feature data obtained in the step 2 into a multi-task learning module in a parallel mode for processing; the multi-task learning module comprises two tasks, namely a supervised classification task and a characteristic metric learning task, so that fault classification is realized.

And 4, step 4: combining the fault classification obtained in the step 3 with the sound array sensor, analyzing the positioning of the abnormal part of the sound, and finally realizing accurate judgment of the fault;

wherein, the step 3 comprises the following steps:

step 31: taking the fusion characteristic representation obtained in the step 2 and a training set label as input to participate in a multi-task module as a sample center of a center loss function; the two tasks are related tasks, the distance reduction in the class of the same type of samples is restrained through an auxiliary task of metric learning, namely the function of sample aggregation is realized, and the output is as follows:

wherein, W and b are weight matrix and offset vector respectively, train the model through Adam optimizer, make the cross entropy loss converge to the minimum, its formula is:

wherein M is the number of fault categories, N is the total number of samples, y _ic Is a sign function, whose value takes 0 or 1 if the class of sample i and the sample are observedClass c is 1 if the same, otherwise is 0 _ic The same prediction probability is used for the class of the observation sample i and the class c of the sample;

the calculation formula of the measurement auxiliary task Center-Loss is as follows:

wherein, c _yi ∈R ^d As the y-th of the extracted fusion feature _i Class center, c _yi As the sample center of the feature changes; x is the number of _i Representing the fused features obtained before the fully connected layer, m representing the batch size of the training sample;

the classification loss function of the classification main task is used for restricting the increase of the inter-class distance of different types of samples, so that the classification accuracy is improved, the characteristics are more obvious, and the generalization performance of the model is enhanced;

the net final loss function is:

L＝αL _{cross-entropy} +βL _center-loss

wherein α and β are weight coefficients of the cross entropy loss function and the central loss function in the training, respectively, and the following is better influence of the quantization cross entropy loss function and the central loss function on the classification result, wherein α + β =1, and α, β ≧ 0.

The technical scheme of the invention is further improved as follows: the step 1 comprises the following steps:

step 11: marking the obtained multi-sensor data according to the number of fault types, dividing the multi-sensor data into a plurality of non-overlapping and equal-length sample fragment sets with the length of L by a sliding window dividing mode, dividing sensor signals with different sources in the same mode, wherein the number of each type of fault samples is the same;

step 12: and performing resonance sparse decomposition on each obtained sample segment, and using the obtained data for input data of the convolutional neural network.

The technical scheme of the invention is further improved as follows: the step 2 comprises the following steps:

step 21: inputting the input data obtained in the step 1 into respective convolutional neural networks in parallel, and extracting features, wherein the extracted features are used as the input of a later fusion layer;

step 22: setting the number of network layers of different convolutional neural networks according to different characteristics of data, wherein each convolutional neural network comprises different convolutional layers and different maximum pooling layers, and extracting fault characteristic representation of signals respectively;

step 23: the learned fault signatures of the signals are concatenated together along the direction of the variable axis, fusing the fused signature representations of the multi-sensor signals.

The technical scheme of the invention is further improved as follows: the step 4 comprises the following steps:

step 41: and positioning the abnormal sounding part with fault information through the acoustic array sensor.

Step 42: and (4) combining the fault classification obtained in the step (3) with the acoustic array sensor, positioning and analyzing the abnormal part of the sound, and outputting an accurate fault position.

The multi-sensor information fusion transmission system fault diagnosis system comprises:

the vibration acquisition module is used for acquiring vibration signals in the transmission system;

the current acquisition module is used for acquiring a field motor stator current signal;

the sound acquisition module is used for acquiring sound signals in the transmission system;

the vibration feature extraction convolutional neural network is used for extracting the features of the vibration signals;

the current feature extraction convolutional neural network is used for extracting the features of the current signals;

the voice feature extraction convolutional neural network is used for extracting the features of the voice signals;

the cascade module is used for carrying out cascade processing on the vibration, current and sound characteristics;

the multi-task classification module is used for judging and classifying the data after the cascade processing;

the acoustic array sensing module is used for acquiring signals of the abnormal part;

and the fault output module is used for analyzing and outputting an accurate fault position by combining the signal of the abnormal part with the signal for judging and classifying.

Compared with the prior art, the transmission system fault diagnosis method based on multi-sensor information fusion has the following beneficial effects:

the invention provides a transmission system fault diagnosis method based on multi-sensor information fusion, which utilizes redundancy and complementarity among vibration signals, current signals and sound signals to automatically learn useful and complementary features from original signals, and restricts the decrease of the similar sample intra-class distance by measuring the auxiliary task of learning so as to enhance the discrimination capability of the learned features. Meanwhile, the capability of the acoustic array sensor for positioning the abnormal sounding part with fault information is utilized, and the obtained fault classification is combined with the positioning of the acoustic array sensor for the abnormal sounding part, so that the fault is accurately judged, and a new way is provided for the field of fault diagnosis of the transmission system. The invention improves the performance of a fault classification task, and meanwhile, the acoustic array sensor has the function of positioning abnormal parts of transmission fault sound, and can be used as effective verification of fault diagnosis to further improve the accuracy of fault diagnosis.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of a transmission system fault diagnosis method of multi-sensor information fusion in accordance with the present invention;

FIG. 2 is a block diagram of the multi-sensor signal feature fusion and fault diagnosis framework of FIG. 1.

Detailed Description

The technical solution of the present invention will be clearly and completely described by the following detailed description. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

This embodiment describes the present invention in detail with reference to the attached drawings:

as shown in FIG. 1, a flow chart of a method for diagnosing faults of a transmission system with multi-sensor information fusion is provided, which comprises the following steps:

step 1: as shown in fig. 2, the method includes acquiring collected vibration signal data, motor rotor current signal data and acoustic array sensor sound data, performing resonance sparse decomposition after sliding window segmentation is completed, and completing signal data preprocessing, and includes the following specific operation steps:

step 11: marking the obtained multi-sensor data according to the number of fault types, dividing the multi-sensor data into a plurality of non-overlapping and equal-length sample fragment sets with the length of L by a sliding window dividing mode, dividing sensor signals with different sources in the same mode, wherein the number of each type of fault samples is the same;

step 12: and performing resonance sparse decomposition on each obtained sample segment, and using the obtained data for input data of a convolution network.

Step 2: as shown in fig. 2, the preprocessed data obtained in step 1 is input into the convolutional neural network in a parallel manner, different signals are input into different convolutional neural networks to learn features, and fault features in different types of signals are extracted, and the specific operation steps are as follows:

step 21: inputting the input data obtained in the step 1 into different convolutional neural networks in parallel to perform feature extraction, wherein the extracted features are used as the input of a later fusion layer;

step 22: setting the number of network layers of different convolutional neural networks according to different characteristics of data, wherein each convolutional neural network comprises different convolutional layers and different maximum pooling layers, and extracting fault characteristic representation of signals respectively;

step 23: cascading together fault features of the learned signals along the direction of the variable axis, fusing the fused feature representations of the multi-sensor signals;

and step 3: as shown in fig. 2, inputting the feature data obtained in step 2 into a multi-task learning module designed for processing in a parallel manner; the multi-task learning module comprises two tasks which are respectively a supervised classification task and a characteristic measurement learning task, so that fault classification is realized, and the specific operation steps are as follows:

step 31: and (3) taking the fusion feature representation obtained in the step (2) and a training set label as input to participate in a multi-task module as a sample center of a center loss function. The two tasks are related tasks, the distance reduction in the class of the same type of samples is restrained through an auxiliary task of metric learning, namely the function of sample aggregation is realized, and the output is as follows:

wherein, W and b are respectively a weight matrix and a bias vector, and the Adam optimizer is used for training the model to ensure that the cross entropy loss is converged to the minimum, and the formula is as follows:

where M is the number of fault categories, y _ic Is a sign function, the value of which is 0 or 1, if the type of the observed sample i is the same as the sample type c, the value is 1, otherwise, the value is 0 _ic The same prediction probability is used for the class of the observation sample i and the class c of the sample;

the calculation formula of the measurement auxiliary task Center-Loss is as follows:

wherein, c _yi ∈R ^d As the y-th of the extracted fusion feature _i Class center, c _yi Varying with the sample center of the feature. x is the number of _i Representing the fusion features obtained before the fully connected layer, and m represents the batch size of the training sample.

And the classification loss function of the classification main task is used for restricting the increase of the inter-class distance of different types of samples, so that the classification accuracy is improved, the characteristics are more remarkable, and the generalization performance of the model is enhanced.

The net final loss function is:

L＝αL _{cross-entropy} +βL _center-loss

wherein α and β are weight coefficients of the cross entropy loss function and the central loss function in the training, and α + β =1, and α and β ≧ 0, for better quantifying the influence of the cross entropy loss function and the central loss function on the classification result.

And 4, step 4: as shown in fig. 2, the fault classification obtained in step 3 and the positioning of the acoustic array sensor on the acoustic abnormal portion are combined to finally realize the accurate judgment of the fault, and the specific operation steps are as follows:

step 41: and positioning the abnormal sounding part with fault information by using the acoustic array sensor.

Step 42: and (4) combining the fault classification obtained in the step (3) and the positioning of the sound array sensor on the sound abnormal part, thereby realizing the accurate judgment of the fault.

The system of the transmission system fault diagnosis method based on the multi-sensor information fusion comprises a vibration acquisition module, a current acquisition module, a sound acquisition module, a vibration feature extraction convolutional neural network, a current feature extraction convolutional neural network, a sound feature extraction convolutional neural network, a cascade module, a multi-task classification module, a sound array sensing module and a fault output module.

The vibration acquisition module acquires vibration signals in the transmission system; the current acquisition module acquires a field motor stator current signal; the sound acquisition module acquires sound signals in the transmission system; extracting the characteristics of the vibration signals by using a vibration characteristic extraction convolutional neural network; carrying out feature extraction on the current signal by using a current feature extraction convolutional neural network; extracting the characteristics of the sound signal by a sound characteristic extraction convolution neural network; the cascade module carries out cascade processing on the vibration, current and sound characteristics; the multi-task classification module judges and classifies the data after the cascade processing; the acoustic array sensing module collects signals of abnormal parts; and the fault output module is used for analyzing and outputting an accurate fault position by combining the signal of the abnormal part with the signal for judging and classifying.

Claims (4)

1. The method for diagnosing the faults of the multi-sensor information fusion transmission system is characterized by comprising the following steps of:

step 1: acquiring vibration signals, motor stator current signals and sound signals acquired by sensors distributed in a transmission system, and segmenting the acquired signals and then carrying out sparse resonance decomposition;

step 2: inputting the preprocessing data obtained in the step 1 into respective convolutional neural networks in a parallel mode to learn characteristics, and extracting fault characteristics in different types of signals;

and step 3: inputting the feature data obtained in the step 2 into a multi-task learning module in a parallel mode for processing; the multi-task learning module comprises two tasks, namely a supervised classification task and a feature metric learning task, so that fault classification is realized;

and 4, step 4: combining the fault classification obtained in the step 3 with the sound array sensor, analyzing the positioning of the abnormal part of the sound, and finally realizing accurate judgment of the fault;

wherein, the step 3 comprises the following steps:

step 31: taking the fusion feature representation and the training set label obtained in the step 2 as input to participate in a multi-task learning module as a sample center of a center loss function; the two tasks are related tasks, the distance reduction in the class of the similar samples is restrained through the characteristic metric learning task, namely the function of sample aggregation, and the calculation formula of the central loss function of the characteristic metric learning task is as follows:

wherein x is _i Represents the fusion features obtained by the ith observation sample before the full connection layer, m represents the batch size of the training sample, c _yi ∈R ^d As the y-th of the extracted fusion feature _i Class center, c _yi As the sample center of the feature changes;

the classification loss function of the supervised classification task is used for restricting the increase of the inter-class distance of different types of samples, so that the classification accuracy is improved, the characteristics are more obvious, the generalization performance of the model is enhanced, and the output of the supervised classification task is as follows:

wherein, W and b are respectively a weight matrix and a bias vector, the model is trained by an Adam optimizer, so that the cross entropy loss is converged to the minimum, and the calculation formula of the cross entropy is as follows:

wherein M is the number of fault categories, N is the total number of samples, y _ic Is a sign function, the value of which is 0 or 1, if the type of the ith observation sample is the same as the sample type c, the value is 1, otherwise, the value is 0 _ic The same prediction probability is used for the class of the ith observation sample and the class c of the sample;

the final penalty function of the multi-task learning module is therefore:

L＝αL _{cross-entropy} +βL _center-loss

wherein α and β are weight coefficients of the cross entropy loss function and the central loss function in the training, respectively, and α + β =1, and α and β are greater than or equal to 0, for better quantifying the influence of the cross entropy loss function and the central loss function on the classification result.

2. The multi-sensor information-fused drive train fault diagnosis method according to claim 1, characterized in that: the step 1 comprises the following steps:

step 11: marking the obtained multi-sensor data according to the number of fault types, dividing the multi-sensor data into a plurality of non-overlapping equal-length sample fragment sets with the length of L by a sliding window segmentation mode, segmenting sensor signals with different sources in the same mode, wherein the number of each type of fault samples is the same;

step 12: and performing resonance sparse decomposition on each obtained sample segment, and using the obtained data for the input data of the convolutional neural network.

3. The multi-sensor information-fused drive train fault diagnosis method according to claim 1, characterized in that: the step 2 comprises the following steps:

step 21: inputting the input data obtained in the step 1 into respective convolutional neural networks in parallel, and extracting features, wherein the extracted features are used as the input of a later fusion layer;

step 22: setting the number of layers of different convolutional neural networks according to different characteristics of data, wherein each convolutional neural network comprises different convolutional layers and different maximum pooling layers, and extracting fault characteristic representation of signals respectively;

step 23: and cascading the learned fault characteristics of the signals together along the direction of the signal variable axis to fuse the fused characteristic representation of the multi-sensor signals.

4. The multi-sensor information-fused drive train fault diagnosis method according to claim 1, characterized in that: the step 4 comprises the following steps:

step 41: positioning an abnormal sounding part with fault information through an acoustic array sensor;

step 42: and (4) combining the fault classification obtained in the step (3) with the acoustic array sensor, positioning and analyzing the abnormal part of the sound, and outputting an accurate fault position.

Images