CN113469060A - Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method - Google Patents
Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method Download PDFInfo
- Publication number
- CN113469060A CN113469060A CN202110753051.1A CN202110753051A CN113469060A CN 113469060 A CN113469060 A CN 113469060A CN 202110753051 A CN202110753051 A CN 202110753051A CN 113469060 A CN113469060 A CN 113469060A
- Authority
- CN
- China
- Prior art keywords
- data
- layer
- fault diagnosis
- model
- neural network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Signal Processing (AREA)
- Acoustics & Sound (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention relates to a multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method, which comprises the following steps: s1, data acquisition; s2, preprocessing data; s3, taking data collected by the simulation test platform as source domain data, and taking data collected by the online monitoring system as target domain data; s4, building a multi-sensor information fusion 1D-CNN model, and putting source domain data into the source domain 1D-CNN model for training; s5, performing on-line diagnosis on the bearing of the target domain; and S6, generating a fault diagnosis result. The invention has the beneficial effects that: by collecting vibration signals of different positions of an aeroengine bearing in different fault states, adopting a multi-channel input 1D-CNN model, fusing and putting the data collected by vibration acceleration sensors in different positions into the model for training, carrying out on-line diagnosis on a target domain bearing, carrying out fault diagnosis and analysis on the bearing of a rotary mechanical part of the aeroengine, accurately finishing the identification of fault types, and avoiding the process of manually excavating characteristics in the traditional method.
Description
Technical Field
The invention relates to the field of fault diagnosis of electromechanical systems, in particular to a multi-sensor fusion convolutional neural network aeroengine bearing fault diagnosis method.
Background
According to the statistics report of national civil aviation safety information in 2018, nearly 40% of common aviation accidents which occur every year are caused by the mechanical problems of equipment system failure, faults, abrasion and falling of key parts and the like, and the engine is used as the key power component of the aviation aircraft, so that mechanical damage is more easily caused to internal component parts after the engine is used for a period of time due to the complex mechanical structure, the high temperature and high pressure and other severe operating environments. The faults of the aircraft engine are frequently generated on shafting parts (such as gears, bearings and the like) forming a rotating mechanism of the aircraft engine, once the surface failure, the fracture and the damage of the parts and the like are generated, the engine is very easy to generate huge vibration and noise, the operation efficiency is reduced, and the whole unit is seriously damaged, so that huge economic loss is caused. If the occurrence of the fault is not accurately detected in real time, great hidden dangers are generated to the safety and the efficiency of the air operation. Therefore, how to monitor the running state of the aircraft engine, diagnose the existing fault information in time and accurately and predict the occurrence of the fault has great research significance for the safety guarantee of the air flight.
The aero-engine mechanical faults can be generally divided into gas circuit faults, accessory faults and rotary machine faults, wherein the rotary machine faults are difficult to accurately identify through analysis and solution by a dynamic model method based on a traditional physical mechanism due to the reasons of various fault types, unobvious fault characteristics and the like, and the aero-engine fault diagnosis also advances to the intelligent and automatic direction along with the application of an artificial intelligence technology initiated by the rapid development of an informatization technology in the 21 st century.
Vibration signal analysis is the most widely used research method in fault diagnosis of rotary machines. The traditional vibration signal analysis is usually based on manual signal processing and statistical methods, and technologies such as frequency spectrum and envelope spectrum analysis, wavelet analysis, time frequency analysis, order analysis, high-order statistic analysis, sparse decomposition and the like are adopted to analyze and solve original signals, so that the attenuation of noise interference components is realized, and related signals of the health state of components are enhanced, so that the frequency of corresponding fault characteristics is provided, and the effective diagnosis of the fault is realized. In order to improve the fault diagnosis efficiency, the fault diagnosis method of the complex mechanical equipment based on data driving is natural.
The convolutional neural network in deep learning can process mass data, potential spatial features of the data can be effectively extracted, the data volume generated in the operation process of the aircraft engine is huge, and the convolutional neural network is very suitable for being adopted for analysis. The one-dimensional convolutional neural network (1D-CNN) can directly conduct feature mining on time domain signals, extracts spatial features of signal data, plays an important role in improving the accuracy of bearing fault diagnosis of the aeroengine, and meanwhile, a multi-sensor information fusion input model can obtain a better recognition effect.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, provides a multi-sensor fusion convolutional neural network aeroengine bearing fault diagnosis method, efficiently utilizes massive aeroengine bearing fault data resources, and provides a feasible means for fault diagnosis of rotary parts of an aeroengine.
The method for diagnosing the bearing fault of the aeroengine with the multi-sensor fusion convolutional neural network comprises the following steps of:
s1, data acquisition: arranging a plurality of vibration acceleration sensors at the bearing of the aeroengine, and acquiring acceleration signals of different positions and directions of equipment;
s2, preprocessing data: normalizing, slicing and labeling the acquired original state parameters, and converting the original state parameters into data types which can be identified by 1D-CNN;
s3, taking data collected by the simulation test platform as source domain data, and taking data collected by the online monitoring system as target domain data;
s4, building a multi-sensor information fusion 1D-CNN model, putting source domain data into the source domain 1D-CNN model for training in an off-line training process, inputting the source domain data, performing parallel training through a plurality of convolution layers and a maximum pooling layer, performing average pooling layer, finally combining flatting layers Flatten into single-channel one-dimensional wave data, performing error back propagation through a Softmax output layer to optimize the model, completing training when the model reaches convergence, and storing parameters of the source domain model; wherein the Softmax output layer consists of a full link layer and a Softmax active layer;
s5, performing on-line diagnosis of the target domain bearing, wherein in the on-line diagnosis process, the source domain model parameters are put into the target domain 1D-CNN model, and when target domain data are input into the target domain 1D-CNN model, feature extraction is performed through the 1D-CNN model;
and S6, generating a fault diagnosis result, and visually evaluating the prediction effect of the model by using the confusion matrix and the classification scatter diagram.
Preferably, the method comprises the following steps: in step S2, the normalization process employs maximum-minimum normalization, and the formula is:
in the formula: x is the number ofmaxIs the maximum value of the sample data, xminIs the minimum value of sample data, x' is the normalized result, and the value interval is [0,1 ]]；
The slicing operation is to divide the acquired long signal waves at every N points to obtain a plurality of sections of short signal wave data with the same length;
and the labeling processing operation is to add corresponding fault labels to different fault data after slicing in a form of 0-M, wherein M is the total number of categories.
Preferably, the method comprises the following steps: in step S4, the error back-propagation algorithm means that the error term of a neuron at layer i is the weighted sum of the error terms of all neurons at layer 1+1 connected to the neuron, and then the weighted sum is multiplied by the gradient of the neuron activation function.
Preferably, the method comprises the following steps: in step S4, the input layer feature map set is a multi-channel one-dimensional array.
Preferably, the method comprises the following steps: in step S4, in the convolutional layer, the convolutional core performs convolution on the output of the previous layer, extracts the spatial features of the local region, and obtains a feature map with a width of N × depth of D × height of 1; the process adopts a nonlinear activation function to construct output characteristics, and the mathematical model of the process is described as follows:
in the formula:
represents the weight of the ith filter kernel at layer l,
denotes the bias of the ith filter kernel at layer l, xl(j) Represents the input of the jth neuron of the ith layer,
represents the input of the j-th neuron at the l +1 level, the symbol · represents the dot product of the kernel and the local region, f represents the nonlinear activation function,
is the output after nonlinear activation.
Preferably, the method comprises the following steps: after convolution operation, the nonlinear activation function ReLU performs nonlinear transformation on the logic value output of each convolution, and transforms the originally linear inseparable multidimensional characteristic to another space, and the calculation formula is as follows:
preferably, the method comprises the following steps: in step S4, the pooling layer reduces the data length by downsampling to reduce the amount of calculation, and takes the maximum value or the average value of the perceptual domain as the output feature map by pooling the maximum value or the average value.
Preferably, the method comprises the following steps: in step S4, the fully-connected layer expands the output of the flattened layer into a one-dimensional vector as its input, and then establishes a fully-connected network between the input and output to integrate the local information distinguished by the convolutional or pooling layers.
Preferably, the method comprises the following steps: in step S4, the Softmax output layer discriminates the tags using the Softmax classifier, and the output result is the probability value of each category, and the tag corresponding to the maximum probability value is taken as the recognition result.
Preferably, the method comprises the following steps: the identification result evaluation criteria are accuracy rate, precision rate and recall rate; accuracy refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples for a given test data set; the accuracy rate P is the ratio of the number of correctly classified as a-tags in the sample to the total number of correctly classified as a-tags; the recall rate R is the ratio of the number of correctly classified A labels in the sample to the number of actual A classes of the sample; the correlation calculation is as follows:
in the formula: TP is the number correctly classified as A, FP is the number classified as A but the true label is not A, and FN is the number of true labels as A but the classification is wrong.
The invention has the beneficial effects that: the invention collects the vibration signals of different positions of the bearing of the aircraft engine in different fault states, adopts a multi-channel input 1D-CNN model, fuses the data collected by the vibration acceleration sensors at different positions and puts the data into the model for training, carries out on-line diagnosis of the bearing in a target domain, carries out fault diagnosis and analysis on the bearing of the rotating mechanical part of the aircraft engine, accurately completes the identification of the fault type, avoids the process of manually digging features in the traditional method, and realizes the end-to-end information processing.
Drawings
FIG. 1 is a diagram of a multi-sensor information fusion concept.
FIG. 2 is a flow chart of a multi-sensor fusion convolutional neural network aeroengine bearing fault diagnosis method.
FIG. 3 is a schematic structural diagram of a bearing fault diagnosis model of the present invention.
FIG. 4 is a graph of model accuracy and loss values.
FIG. 5 is a graph of accuracy versus loss function after analysis of data collected from 1 to 4 sensor placement numbers.
FIG. 6 is a schematic diagram of an online fault diagnosis result confusion matrix.
Detailed Description
The present invention will be further described with reference to the following examples. The following examples are set forth merely to aid in the understanding of the invention. It should be noted that, for those skilled in the art, it is possible to make various improvements and modifications to the present invention without departing from the principle of the present invention, and those improvements and modifications also fall within the scope of the claims of the present invention.
Example one
The embodiment of the application provides a method for diagnosing faults of a multi-sensor fusion convolutional neural network aeroengine bearing, which comprises the following steps: data acquisition, data preprocessing, data storage, source domain 1D-CNN model offline training and target domain bearing online diagnosis, wherein:
a data acquisition part: arranging a vibration acceleration sensor according to actual requirements, and acquiring acceleration signals of different positions and directions of equipment.
A data preprocessing part: the acquired original state parameters need to be normalized, sliced and labeled, and converted into data types which can be identified by the 1D-CNN.
Further, the normalization process adopts maximum and minimum normalization, and the formula is as follows:
in the formula: x is the number ofmaxIs the maximum value of the sample data, xminIs the minimum value of sample data, x' is the normalized result, and the value interval is [0,1 ]]。
And the slicing operation is to segment every N points of the acquired long signal waves to obtain multiple sections of short signal wave data with the same length.
The labeling processing operation is to add corresponding fault labels to different fault data after slicing, generally in a form of 0 to M, where M is the total number of categories.
A data storage section: and an SQL Server database technology is utilized to establish an aeroengine fault database management system, so that data interaction and effective storage are realized.
A source domain 1D-CNN model offline training part: and building a neural network model by adopting Python language, and putting the processed source domain data into the built 1D-CNN model for training in the off-line training process. The method comprises the steps of inputting source domain data (off-line experimental data), performing parallel training through 4 convolutional layers and a maximum pooling layer, performing average pooling, finally merging the source domain data into single-channel one-dimensional wave data in a flattening layer (Flatten), performing error back propagation through a Softmax output layer (the layer consists of a full connection layer and a Softmax activation layer) to optimize a model, and storing source domain model parameters when the model reaches convergence.
Further, the error back propagation algorithm means that the error term of one neuron at the layer 1 is the weighted sum of the error terms of all neurons at the layer 1+1 connected with the neuron, and then the weighted sum is multiplied by the gradient of the neuron activation function.
Further, the input layer feature mapping group is a multi-channel one-dimensional array (the number of channels is the number of sensors).
Further, in the convolutional layer, the convolutional core performs convolution on the output of the previous layer, extracts the spatial features of the local region, and obtains a feature map with a width of N × a depth of D × a height of 1. The process generally adopts a nonlinear activation function to construct output characteristics, and the mathematical model of the process is described as follows:
in the formula: wherein
Representing the weight of the ith filtering kernel at the l layer;
representing the bias of the ith filter kernel at the l layer; x is the number ofl(j) Represents the input of the jth neuron of the ith layer;
represents the input of the jth neuron at the l +1 layer, namely the output of the jth neuron at the l layer. The symbol · represents the dot product of the kernel and the local region, f represents the nonlinear activation function,
is the output after nonlinear activation.
Further, after convolution operation, the activation function (generally ReLU) performs nonlinear transformation on the output of the logic value of each convolution, transforms originally linear inseparable multidimensional features into another space, and enhances the linear separability of the features, and the calculation formula is as follows:
further, the purpose of the pooling layer is to reduce network parameters and reduce data length by downsampling to reduce the amount of computation, and it is common to adopt maximum pooling or average pooling, and take the maximum or average of the perceptual domain as the output feature map.
Further, the fully-connected layer unwinds the output of the flattened layer into a one-dimensional vector as its input, and then establishes a fully-connected network between the input and output, integrating the local information that the convolutional or pooling layers have distinguished.
Further, the output layer often uses a softmax classifier to distinguish the tags, the output result is a probability value of each category, and the tag corresponding to the maximum probability value is taken as the identification result.
Furthermore, the model identification evaluation criteria adopted by the invention are accuracy, precision and recall rate. The accuracy rate refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples for a given test data set, and is presented by a visualization tool carried by the model; the precision ratio (P) is the ratio of the number of correctly classified A labels in the sample to the total number of correctly classified A labels; recall (R) is the ratio of the number of correctly classified as a-labels in the sample to the number of actual a classes of the sample. The correlation calculation is as follows:
in the formula: TP is the number correctly classified as A, FP is the number classified as A but the true label is not A, and FN is the number of true labels as A but the classification is wrong.
The target domain bearing online diagnosis part comprises the following steps: in the online diagnosis process, model parameters which are trained and stored are placed into a target domain 1D-CNN model, when target domain data (online collected data) are input into a network model, feature extraction is carried out through the 1D-CNN model, and finally a fault diagnosis result is output.
In the 1D-CNN, tagged slice data is directly imported into a model for training, potential nonlinear features in original data can be automatically extracted through a convolution-pooling layer in an alternating mode, self-adaptive feature learning is completed in a full connection layer, the process of manually mining features in the traditional method is omitted, and end-to-end information processing is achieved.
Example two
Fig. 2 is a flowchart of a method for diagnosing a bearing fault of an aeroengine with a multi-sensor fusion convolutional neural network, which includes: data acquisition and preprocessing, offline training and online diagnosis. The method specifically comprises the following steps:
s1, data acquisition: collecting relevant data of deep groove ball bearings tested by a helicopter transmission system test bed and a main reducing test bed, wherein the sampling frequency is 10000Hz, and the sampling time is 3 minutes, namely, 1800000 data points are respectively sampled by each fault type; the bearing fault types comprise an outer ring fault, an inner ring fault, a rolling body fault, a joint fault and a normal bearing;
s2, preprocessing data: the data is normalized, sliced and labeled, and is converted into a data type which can be used for supervised learning, and the data structure is shown in the following table 1:
TABLE 1 Label of Gear data set
| Number of samples (training set/testing set) | Sample length | Class of failure | Classification label |
| 2400/600 | 1000 | Normal bearing | 0 |
| 2400/600 | 1000 | |
1 |
| 2400/600 | 1000 | |
2 |
| 2400/600 | 1000 | Failure of rolling |
3 |
| 2400/600 | 1000 | |
4 |
S3, taking the data collected in the 1 st minute as model training data, and setting the data in the 2 nd and 3 th minutes as online diagnosis verification data;
s4, building a multi-sensor information fusion 1D-CNN model, substituting training data into the model for training, and storing model parameters after training is completed;
s5, inputting the online diagnosis verification data into the stored model for inspection;
and S6, generating a diagnosis result, and visually evaluating the prediction effect of the model by using the confusion matrix and the classification scatter diagram.
EXAMPLE III
According to the principle of the 1D-CNN neural network model and the multi-sensor information fusion, the specific structural parameters of the aeroengine bearing fault diagnosis model provided by the invention are shown in the table 2.
The 1D-CNN for diagnosing the bearing fault of the aircraft engine, which is established by the invention, consists of four sub-models with the same parameters, wherein each sub-model comprises four groups of convolution-maximum pooling layers and an average pooling layer. And outputting and converging the last layer of each sub-model, inputting the same Flatten layer, flattening, and finally outputting the recognition result on a Softmax layer. The raw bearing data is transformed into a set of feature maps (multi-channel one-dimensional array) after being convolutional-layered, and then downsampled through maximum pooling, thereby reducing the number of parameters. After repeating the operations for 3 times, connecting the characteristics of the last maximum pooling layer to an average pooling layer, flattening the data into a single-channel one-dimensional array structure through a Flatten layer, transferring the single-channel one-dimensional array structure to a Softmax activation layer through a full connection layer (FC _1), and finally obtaining the probability value of each fault category, wherein the category with the highest probability can be regarded as a recognition result, and the specific structure of the model is shown in FIG. 3.
TABLE 21 detailed parameters of D-CNN model
| Layer number | Layer classes | Size/step/number of cores | Remarks for |
| 1 | |
8×1/2/16 | |
| 2 | |
2×1/1/16 | Maximum pooling |
| 3 | |
4×1/2/64 | |
| 4 | |
2×1/1/64 | Maximum pooling |
| 5 | |
4×1/2/256 | Relu |
| 6 | |
2×1/1/256 | Maximum pooling |
| 7 | |
2×1/1/512 | Relu |
| 8 | |
2×1/1/512 | Maximum pooling |
| 9 | |
2×1/1/512 | Average pooling |
| 10 | Flatten | 256 | / |
| 11 | |
256 | Relu |
| 12 | |
5 | Softmax |
The Relu function is used as an activation function, so that the overfitting phenomenon is reduced; in order to control the learning rate of the network, network parameters are updated using an SGD (random gradient descent) optimization algorithm, the learning rate being set to 0.01; a Dropout regularization method is introduced into the full-connection layer, so that overfitting of training data is avoided, and the rate is 0.2. The neural network training parameters are set as: the maximum iteration time epoch is 30, the small Batch size is 32, and the training mode is as follows: GPU, and sample set as per 4: the ratio of 1 is divided into training samples and test samples. As shown in fig. 4, it can be seen from the graph that, in the process of 30 times of model iteration, the accuracy of the training set rapidly increases and gradually converges to 100%, the loss function of the training set rapidly decreases and converges to 0, while the accuracy of the test set increases faster, and the loss function of the test set is at a lower level at the beginning of the test and decreases faster, indicating that the model has significant effect on feature mining and learning of the multi-sensor information-fused aero-engine fault data.
The performance of the neural network algorithm (multisensor 1D-CNN) adopted by the invention is further verified, and the specific accuracy and the model training speed are compared with the training effect of the feedforward neural network and the SVM, which is shown in Table 3.
TABLE 3 accuracy and model training speed
| Model name | Rate of accuracy | Speed of training |
| SVM | 63.08％ | 2s 12ms/step |
| Feedforward neural network | 81.10％ | 2s 74ms/step |
| Multisensor 1D-CNN | 100.00％ | 1s 12ms/step |
In view of quantification, the accuracy of the aircraft engine fault diagnosis method provided by the invention can reach 100%, the accuracy is improved by 36.92% compared with a Support Vector Machine (SVM), the accuracy is improved by 18.9% compared with a feedforward neural network, the training speed is higher, and the existing fault characteristics can be accurately identified according to the bearing vibration signal.
The multi-sensor information fusion 1D-CNN model established by the invention is improved and optimized on the basis of single sensor data so as to improve the identification accuracy. In order to embody the advantages of multi-sensor information fusion, 1 to 4 sensors are used for collecting data to construct 4 1D-CNNs: 1D-CNN-1S, 1D-CNN-2S, 1D-CNN-3S, 1D-CNN-4S. Fig. 5 shows the performance trend of the model after the data collected from 1 to 4 sensor arrangement numbers are analyzed, and specific accuracy values are shown in table 4.
TABLE 4 accuracy
| Number of sensors | Rate of accuracy | Number of |
| 1 | 86.29％ | 38 |
| 2 | 94.58％ | 27 |
| 3 | 99.96％ | 12 |
| 4 | 100.00％ | 14 |
From the above, when only 1 acceleration sensor is arranged to collect the unidirectional vibration acceleration of the engine, the fault identification accuracy of the 1D-CNN model is only 86.29%, the convergence rate is relatively slow, and the accuracy fluctuates to a certain extent during the initial training of the model. When 2 acceleration sensors are installed at different positions of the equipment, the model identification accuracy rate gradually rises, and the accuracy rate convergence speed is also rapidly accelerated, so that the effect of comprehensively analyzing the vibration signals captured at different positions on distinguishing fault types is improved to a certain extent. When the number of sensors reaches 3, the accuracy rapidly rises to 99.96%, and almost all the classes can be identified. When the number of the sensors arranged at different positions and in different directions of the device reaches 4, the model reaches the recognition accuracy of 100 percent, and the data measured by the 4 acceleration sensors arranged in the invention has extremely high relevance with the bearing fault category. The economic factors are comprehensively considered, the number of the sensors installed on the control equipment is 3-4, and the required diagnosis effect can be ensured.
In order to further check the effectiveness of the method provided by the invention, the data of the 2 nd minute and the 3 rd minute are respectively used as two verification sets (each verification set comprises 2995 pieces of unknown label data), the data are randomly scrambled and are used as unknown data to be placed in a model, the identification result of each data is obtained, the result is compared with a real label, and a diagnosis result confusion matrix is drawn, as shown in fig. 6.
As can be seen from fig. 6, when the new data of the two unknown tags are input for verification, the predicted tags and the real tags of the two data are completely corresponding to each other, no classification error exists, the recognition accuracy rate reaches 100%, and the accuracy rate and the recall rate are further calculated to be 1, which indicates that the effect of the method provided by the invention on the aspect of diagnosis of the bearing fault of the aircraft engine meets the requirements.
According to the method, vibration signals of different positions of an aircraft engine bearing in different fault states are collected, a 1D-CNN model is input through multiple channels, labeled slice data are directly led into the model to be trained in the 1D-CNN model, potential nonlinear features in original data can be automatically extracted through a convolution-pooling layer in an alternating mode, self-adaptive feature learning is completed on a full connection layer, target domain bearing online diagnosis is performed, fault diagnosis and analysis are performed on a bearing of a rotary mechanical part of the aircraft engine, fault type identification is accurately completed, a process of manually excavating features in a traditional method is omitted, and end-to-end information processing is achieved.
In an actual application scene, a worker can install the acceleration sensor at a specified position of the aircraft engine, collect vibration signals in the operation process of the acceleration sensor, fuse the data collected by the sensors at different positions and place the data into the 1D-CNN model provided by the invention, so that whether the current equipment has faults or not and the types of the faults can be diagnosed, and an accurate and reliable basis is provided for maintenance workers.
Claims (10)
1. A multi-sensor fusion convolutional neural network aeroengine bearing fault diagnosis method is characterized by comprising the following steps:
s1, data acquisition: arranging a plurality of vibration acceleration sensors at the bearing of the aeroengine, and acquiring acceleration signals of different positions and directions of equipment;
s2, preprocessing data: normalizing, slicing and labeling the acquired original state parameters, and converting the original state parameters into data types which can be identified by 1D-CNN;
s3, taking data collected by the simulation test platform as source domain data, and taking data collected by the online monitoring system as target domain data;
s4, building a multi-sensor information fusion 1D-CNN model, putting source domain data into the source domain 1D-CNN model for training in an off-line training process, inputting the source domain data, performing parallel training through a plurality of convolution layers and a maximum pooling layer, performing average pooling layer, finally combining flatting layers Flatten into single-channel one-dimensional wave data, performing error back propagation through a Softmax output layer to optimize the model, completing training when the model reaches convergence, and storing parameters of the source domain model; wherein the Softmax output layer consists of a full link layer and a Softmax active layer;
s5, performing on-line diagnosis of the target domain bearing, wherein in the on-line diagnosis process, the source domain model parameters are put into the target domain 1D-CNN model, and when target domain data are input into the target domain 1D-CNN model, feature extraction is performed through the 1D-CNN model;
and S6, generating a fault diagnosis result, and visually evaluating the prediction effect of the model by using the confusion matrix and the classification scatter diagram.
2. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S2, the normalization process employs maximum-minimum normalization, and the formula is:
in the formula: x is the number ofmaxIs the maximum value of the sample data, xminIs the minimum value of sample data, x' is the normalized result, and the value interval is [0,1 ]]；
The slicing operation is to divide the acquired long signal waves at every N points to obtain a plurality of sections of short signal wave data with the same length;
and the labeling processing operation is to add corresponding fault labels to different fault data after slicing in a form of 0-M, wherein M is the total number of categories.
3. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, the error back-propagation algorithm means that the error term of a neuron at layer i is the weighted sum of the error terms of all neurons at layer i +1 connected to the neuron, and then the weighted sum is multiplied by the gradient of the neuron activation function.
4. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, the input layer feature map set is a multi-channel one-dimensional array.
5. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, in the convolutional layer, the convolutional core performs convolution on the output of the previous layer, extracts the spatial features of the local region, and obtains a feature map with a width of N × depth of D × height of 1; the process adopts a nonlinear activation function to construct output characteristics, and the mathematical model of the process is described as follows:
in the formula:
represents the weight of the ith filter kernel at layer l,
denotes the bias of the ith filter kernel at layer l, xl(j) Represents the input of the jth neuron of the ith layer,
represents the input of the j-th neuron at the l +1 level, the symbol · represents the dot product of the kernel and the local region, f represents the nonlinear activation function,
is the output after nonlinear activation.
6. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 5, characterized in that: after convolution operation, the nonlinear activation function ReLU performs nonlinear transformation on the logic value output of each convolution, and transforms the originally linear inseparable multidimensional characteristic to another space, and the calculation formula is as follows:
7. the multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, the pooling layer reduces the data length by downsampling to reduce the amount of calculation, and takes the maximum value or the average value of the perceptual domain as the output feature map by pooling the maximum value or the average value.
8. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, the fully-connected layer expands the output of the flattened layer into a one-dimensional vector as its input, and then establishes a fully-connected network between the input and output to integrate the local information distinguished by the convolutional or pooling layers.
9. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 1, characterized in that: in step S4, the Softmax output layer discriminates the tags using the Softmax classifier, and the output result is the probability value of each category, and the tag corresponding to the maximum probability value is taken as the recognition result.
10. The multi-sensor fusion convolutional neural network aircraft engine bearing fault diagnosis method of claim 9, wherein: the identification result evaluation criteria are accuracy rate, precision rate and recall rate; accuracy refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples for a given test data set; the accuracy rate P is the ratio of the number of correctly classified as a-tags in the sample to the total number of correctly classified as a-tags; the recall rate R is the ratio of the number of correctly classified A labels in the sample to the number of actual A classes of the sample; the correlation calculation is as follows:
in the formula: TP is the number correctly classified as A, FP is the number classified as A but the true label is not A, and FN is the number of true labels as A but the classification is wrong.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110753051.1A CN113469060A (en) | 2021-07-02 | 2021-07-02 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110753051.1A CN113469060A (en) | 2021-07-02 | 2021-07-02 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113469060A true CN113469060A (en) | 2021-10-01 |
Family
ID=77877838
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110753051.1A Withdrawn CN113469060A (en) | 2021-07-02 | 2021-07-02 | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113469060A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114088405A (en) * | 2021-11-10 | 2022-02-25 | 中国人民解放军陆军炮兵防空兵学院 | Engine fault diagnosis method of CNN fault diagnosis model based on spectrogram |
| CN114186617A (en) * | 2021-11-23 | 2022-03-15 | 浙江大学 | Mechanical fault diagnosis method based on distributed deep learning |
| CN114186337A (en) * | 2021-11-30 | 2022-03-15 | 大连理工大学 | Gas compressor rotating stall prediction method based on multi-source data fusion |
| CN114282579A (en) * | 2021-12-30 | 2022-04-05 | 浙大城市学院 | Aviation bearing fault diagnosis method based on variational modal decomposition and residual error network |
| CN114509158A (en) * | 2022-01-04 | 2022-05-17 | 东南大学 | Acoustic-vibration fused blade crack fault detection method and application |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
| CN114757242A (en) * | 2022-06-16 | 2022-07-15 | 中国空气动力研究与发展中心低速空气动力研究所 | Helicopter noise enhancement method and detection method based on cyclic wiener filtering |
| CN114997246A (en) * | 2022-07-27 | 2022-09-02 | 浙江大学 | Cavitation state identification method driven by fluid mechanical vibration data |
| CN115049852A (en) * | 2022-08-15 | 2022-09-13 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN115270896A (en) * | 2022-09-28 | 2022-11-01 | 西华大学 | Intelligent diagnosis method for identifying loosening fault of main bearing of aircraft engine |
| CN116226469A (en) * | 2023-05-09 | 2023-06-06 | 华南理工大学 | Intelligent diagnosis method and system for energy storage equipment faults |
| CN116702030A (en) * | 2023-05-31 | 2023-09-05 | 浙江大学 | Blast furnace state monitoring method and device based on sensor reliability analysis |
| CN116881846A (en) * | 2023-07-17 | 2023-10-13 | 无锡北微传感科技有限公司 | Multi-mode communication iron tower monitoring method based on multi-sensor information fusion |
| CN116894190A (en) * | 2023-09-11 | 2023-10-17 | 江西南昌济生制药有限责任公司 | Bearing fault diagnosis method, device, electronic equipment and storage medium |
| WO2023206860A1 (en) * | 2022-04-27 | 2023-11-02 | 山东瑞美油气装备技术创新中心有限公司 | Method and apparatus for determining mechanical device fault |
| CN117171528A (en) * | 2023-08-23 | 2023-12-05 | 中国建筑第二工程局有限公司 | High-altitude overhanging structure health monitoring method based on deep convolution fusion memory network |
| CN117330315A (en) * | 2023-12-01 | 2024-01-02 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117780679A (en) * | 2024-02-26 | 2024-03-29 | 无锡精恩风机有限公司 | Centrifugal fan fault diagnosis method and online monitoring system |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635677A (en) * | 2018-11-23 | 2019-04-16 | 华南理工大学 | Combined failure diagnostic method and device based on multi-tag classification convolutional neural networks |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN112161784A (en) * | 2020-09-07 | 2021-01-01 | 华南理工大学 | Mechanical fault diagnosis method based on multi-sensor information fusion migration network |
| CN112254964A (en) * | 2020-09-03 | 2021-01-22 | 太原理工大学 | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network |
| CN112396109A (en) * | 2020-11-19 | 2021-02-23 | 天津大学 | Motor bearing fault diagnosis method based on recursion graph and multilayer convolution neural network |
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| CN112903294A (en) * | 2021-01-07 | 2021-06-04 | 泰华宏业(天津)智能科技有限责任公司 | Rolling bearing fault diagnosis method based on VMD and deep convolution neural network |
| CN112949823A (en) * | 2021-02-05 | 2021-06-11 | 杭州电子科技大学 | Industrial process performance diagnosis method based on one-dimensional multi-scale depth convolution neural network |
| CN112989712A (en) * | 2021-04-27 | 2021-06-18 | 浙大城市学院 | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning |
| CN113033309A (en) * | 2021-02-25 | 2021-06-25 | 北京化工大学 | Fault diagnosis method based on signal downsampling and one-dimensional convolution neural network |
-
2021
- 2021-07-02 CN CN202110753051.1A patent/CN113469060A/en not_active Withdrawn
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635677A (en) * | 2018-11-23 | 2019-04-16 | 华南理工大学 | Combined failure diagnostic method and device based on multi-tag classification convolutional neural networks |
| CN110866365A (en) * | 2019-11-22 | 2020-03-06 | 北京航空航天大学 | Mechanical equipment intelligent fault diagnosis method based on partial migration convolutional network |
| CN112254964A (en) * | 2020-09-03 | 2021-01-22 | 太原理工大学 | Rolling bearing fault diagnosis method based on rapid multi-scale convolution neural network |
| CN112161784A (en) * | 2020-09-07 | 2021-01-01 | 华南理工大学 | Mechanical fault diagnosis method based on multi-sensor information fusion migration network |
| CN112396109A (en) * | 2020-11-19 | 2021-02-23 | 天津大学 | Motor bearing fault diagnosis method based on recursion graph and multilayer convolution neural network |
| CN112577747A (en) * | 2020-12-07 | 2021-03-30 | 东南大学 | Rolling bearing fault diagnosis method based on space pooling network |
| CN112903294A (en) * | 2021-01-07 | 2021-06-04 | 泰华宏业(天津)智能科技有限责任公司 | Rolling bearing fault diagnosis method based on VMD and deep convolution neural network |
| CN112949823A (en) * | 2021-02-05 | 2021-06-11 | 杭州电子科技大学 | Industrial process performance diagnosis method based on one-dimensional multi-scale depth convolution neural network |
| CN113033309A (en) * | 2021-02-25 | 2021-06-25 | 北京化工大学 | Fault diagnosis method based on signal downsampling and one-dimensional convolution neural network |
| CN112989712A (en) * | 2021-04-27 | 2021-06-18 | 浙大城市学院 | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114088405A (en) * | 2021-11-10 | 2022-02-25 | 中国人民解放军陆军炮兵防空兵学院 | Engine fault diagnosis method of CNN fault diagnosis model based on spectrogram |
| CN114186617A (en) * | 2021-11-23 | 2022-03-15 | 浙江大学 | Mechanical fault diagnosis method based on distributed deep learning |
| CN114186337A (en) * | 2021-11-30 | 2022-03-15 | 大连理工大学 | Gas compressor rotating stall prediction method based on multi-source data fusion |
| WO2023123593A1 (en) * | 2021-12-30 | 2023-07-06 | 浙大城市学院 | Variational mode decomposition and residual network-based aviation bearing fault diagnosis method |
| CN114282579A (en) * | 2021-12-30 | 2022-04-05 | 浙大城市学院 | Aviation bearing fault diagnosis method based on variational modal decomposition and residual error network |
| CN114509158A (en) * | 2022-01-04 | 2022-05-17 | 东南大学 | Acoustic-vibration fused blade crack fault detection method and application |
| CN114684217A (en) * | 2022-03-16 | 2022-07-01 | 武汉理工大学 | Rail transit health monitoring system and method |
| CN114684217B (en) * | 2022-03-16 | 2024-03-01 | 武汉理工大学 | Rail transit health monitoring system and method |
| WO2023206860A1 (en) * | 2022-04-27 | 2023-11-02 | 山东瑞美油气装备技术创新中心有限公司 | Method and apparatus for determining mechanical device fault |
| CN114757242A (en) * | 2022-06-16 | 2022-07-15 | 中国空气动力研究与发展中心低速空气动力研究所 | Helicopter noise enhancement method and detection method based on cyclic wiener filtering |
| CN114997246A (en) * | 2022-07-27 | 2022-09-02 | 浙江大学 | Cavitation state identification method driven by fluid mechanical vibration data |
| CN115049852A (en) * | 2022-08-15 | 2022-09-13 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN115049852B (en) * | 2022-08-15 | 2022-12-02 | 中国长江三峡集团有限公司 | Bearing fault diagnosis method and device, storage medium and electronic equipment |
| CN115270896A (en) * | 2022-09-28 | 2022-11-01 | 西华大学 | Intelligent diagnosis method for identifying loosening fault of main bearing of aircraft engine |
| CN116226469A (en) * | 2023-05-09 | 2023-06-06 | 华南理工大学 | Intelligent diagnosis method and system for energy storage equipment faults |
| CN116226469B (en) * | 2023-05-09 | 2023-08-08 | 华南理工大学 | Intelligent diagnosis method and system for energy storage equipment faults |
| CN116702030A (en) * | 2023-05-31 | 2023-09-05 | 浙江大学 | Blast furnace state monitoring method and device based on sensor reliability analysis |
| CN116702030B (en) * | 2023-05-31 | 2024-01-23 | 浙江大学 | Blast furnace state monitoring method and device based on sensor reliability analysis |
| CN116881846A (en) * | 2023-07-17 | 2023-10-13 | 无锡北微传感科技有限公司 | Multi-mode communication iron tower monitoring method based on multi-sensor information fusion |
| CN116881846B (en) * | 2023-07-17 | 2024-04-05 | 无锡北微传感科技有限公司 | Multi-mode communication iron tower monitoring method based on multi-sensor information fusion |
| CN117171528A (en) * | 2023-08-23 | 2023-12-05 | 中国建筑第二工程局有限公司 | High-altitude overhanging structure health monitoring method based on deep convolution fusion memory network |
| CN117171528B (en) * | 2023-08-23 | 2024-03-08 | 中国建筑第二工程局有限公司 | High-altitude overhanging structure health monitoring method based on deep convolution fusion memory network |
| CN116894190A (en) * | 2023-09-11 | 2023-10-17 | 江西南昌济生制药有限责任公司 | Bearing fault diagnosis method, device, electronic equipment and storage medium |
| CN116894190B (en) * | 2023-09-11 | 2023-11-28 | 江西南昌济生制药有限责任公司 | Bearing fault diagnosis method, device, electronic equipment and storage medium |
| CN117330315A (en) * | 2023-12-01 | 2024-01-02 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117330315B (en) * | 2023-12-01 | 2024-02-23 | 智能制造龙城实验室 | Rotary machine fault monitoring method based on online migration learning |
| CN117780679A (en) * | 2024-02-26 | 2024-03-29 | 无锡精恩风机有限公司 | Centrifugal fan fault diagnosis method and online monitoring system |
| CN117780679B (en) * | 2024-02-26 | 2024-05-03 | 无锡精恩风机有限公司 | Centrifugal fan fault diagnosis method and online monitoring system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| Shi et al. | Planetary gearbox fault diagnosis using bidirectional-convolutional LSTM networks | |
| CN112989712B (en) | Aeroengine fault diagnosis method based on 5G edge calculation and deep learning | |
| Gou et al. | Aeroengine control system sensor fault diagnosis based on CWT and CNN | |
| WO2023123593A1 (en) | Variational mode decomposition and residual network-based aviation bearing fault diagnosis method | |
| Xiang et al. | Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder | |
| CN104712542B (en) | A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults | |
| Zou et al. | Bearing fault diagnosis based on combined multi-scale weighted entropy morphological filtering and bi-LSTM | |
| CN109211548A (en) | A kind of mechanical failure diagnostic method | |
| CN111678699B (en) | Early fault monitoring and diagnosing method and system for rolling bearing | |
| KR102321607B1 (en) | Rotating machine fault detecting apparatus and method | |
| CN110060368A (en) | Mechanical method for detecting abnormality based on potential feature coding | |
| Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
| Son et al. | Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge | |
| CN116610998A (en) | Switch cabinet fault diagnosis method and system based on multi-mode data fusion | |
| CN116861369A (en) | Air compressor health assessment method and system for industrial internet heterogeneous data fusion | |
| Liu et al. | A deep support vector data description method for anomaly detection in helicopters | |
| CN115375026A (en) | Method for predicting service life of aircraft engine in multiple fault modes | |
| Lu et al. | Early fault warning and identification in condition monitoring of bearing via wavelet packet decomposition coupled with graph | |
| Xu et al. | New RUL prediction method for rotating machinery via data feature distribution and spatial attention residual network | |
| BENKEDJOUH et al. | Deep Learning for Fault Diagnosis based on short-time Fourier transform | |
| CN114330096A (en) | Deep learning-based rail transit motor bearing fault degree classification model construction method | |
| Zurita et al. | A review of vibration machine diagnostics by using artificial intelligence methods | |
| Bond et al. | A hybrid learning approach to prognostics and health management applied to military ground vehicles using time-series and maintenance event data | |
| CN112560252B (en) | Method for predicting residual life of aeroengine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20211001 |
|
| WW01 | Invention patent application withdrawn after publication |