Intelligent fault diagnosis method based on deep learning

Abstract

The invention belongs to a diagnosis method, and particularly relates to an intelligent fault diagnosis method based on deep learning. An intelligent fault diagnosis method based on deep learning comprises the following steps: step 1, constructing a neural network; 2, training and fixing parameters; 3, improving the precision; 4, forming a convolutional layer; step 5, constructing a pooling layer; and step 6, carrying out fault diagnosis on the equipment. Compared with the prior art, the method has the following beneficial effects that the method has three characteristics of sparse communication, weight sharing and pooling, dimensionality sampling can be effectively reduced, the dimensionality of data in time and space is simplified, the number of training parameters is reduced, the complexity of the network is reduced, and algorithm overfitting can be avoided. Therefore, the data training efficiency is improved, and the fault diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention relates to a diagnostic method, and in particular to an intelligent fault diagnosis method based on deep learning. Background Art

[0002] Due to the rapid development of industrialization and the high-speed operation of machinery and equipment, the probability of equipment failure has greatly increased. However, traditional fault diagnosis methods require a lot of time and cost, and are increasingly difficult to adapt to today's environment. Therefore, fault diagnosis methods based on machine learning have emerged. This method is efficient and accurate, and has now become a research hotspot in various industries.

[0003] In recent years, due to the rapid development of sensor technology, Internet technology, and big data technology, machine learning methods have been widely used in the field of fault diagnosis. However, since traditional machine learning methods rely on the scalar characteristics of the time-frequency domain signal of the data and identify fault features through shallow models, this method has limited generalization ability and is limited to specific scenarios. The fault intelligent diagnosis method based on deep learning can solve this problem well. This method has good data mining and adaptive capabilities, does not require prior information and a certain fault diagnosis model, and can directly extract effective features from the original signal without being restricted by the scene. However, when applying this method, the dependence on the amount of data must be resolved, because too much data will bring more redundant information, which will interfere with the results of fault identification. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning-based intelligent fault diagnosis method to address the defects of the prior art.

[0005] The specific technical solution adopted by the present invention is as follows: a fault intelligent diagnosis method based on deep learning, which includes the following steps:

[0006] Step 1: Build a neural network;

[0007] Step 2: Training and fixing parameters;

[0008] Step 3: Improve accuracy;

[0009] Step 4: Form a convolutional layer;

[0010] Step 5: Construct the pooling layer;

[0011] Step 6: Diagnose equipment faults.

[0012] The above-mentioned intelligent fault diagnosis method based on deep learning, wherein the step 1 comprises:

[0013] The neural network includes a convolutional layer, a pooling layer, and a fully connected layer, and these three layers receive signals in sequence.

[0014] An intelligent fault diagnosis method based on deep learning as described above, wherein step two includes,

[0015] Training and fixing parameters are to train using pre-set training data to obtain the weight parameters of the model.

[0016] An intelligent fault diagnosis method based on deep learning as described above, wherein step three includes,

[0017] By using the gradient descent method to minimize the loss function, the weight parameters in the network are adjusted layer by layer in reverse, and the accuracy of the network is improved through frequent iterative training

[0018] An intelligent fault diagnosis method based on deep learning as described above, wherein step four includes,

[0019] Generate different feature maps through convolution operations and activation operations to form a convolutional layer; each neuron in the feature map is connected to only a local area of the input through an activation function, and all neurons in the feature map share the same kernel.

[0020] An intelligent fault diagnosis method based on deep learning as described above, wherein step five includes,

[0021] Adopt a downsampling method to reduce the feature space of the feature map, compress the amount of data and parameters, and reduce overfitting;

[0022] The convolutional layer and the max pooling layer alternately form the low hidden layer of the convolutional neural network, and the high layer is the fully connected layer corresponding to the hidden layer of the traditional multi-layer perceptron and the logistic regression classifier.

[0023] An intelligent fault diagnosis method based on deep learning as described above, wherein the input of the first fully connected layer is the feature image obtained by feature extraction by the convolutional layer and the subsampling layer;

[0024] Since the spatial connection of the samples is local, each neuron senses local features, and then these different local neurons obtained by sensing are integrated at a higher layer to reduce the number of neuron connections;

[0025] Parameter sharing is carried out between neurons, multiple filters are used to deconvolve the samples to obtain multiple feature maps, and the neurons in the hidden layer share weights to detect exactly the same features at different positions;

[0026] The last output layer is a classifier, which can use logistic regression to classify the input, and finally obtain a classification result, and the fault diagnosis of the device is realized through this result.

[0027] The present invention has the following beneficial effects compared with the prior art: The method has three characteristics: sparse connectivity, shared weights, and pooling, which can effectively reduce dimensional sampling, simplify the dimensions of data in terms of time and space, reduce the number of training parameters, lower the complexity of the network, and avoid overfitting of the algorithm. Thus, the training efficiency of data can be improved, and the accuracy of fault diagnosis can be enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the diagnostic process of the fault intelligent diagnosis method based on deep learning in an embodiment of the present invention;

[0029] Figure 2 It is the accuracy curve when testing with test data in an embodiment of the present invention when the amount of data in the test set is different;

[0030] Figure 3 It is a schematic diagram of the implementation process of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0031] The present invention will be further described and explained below in conjunction with the drawings and specific embodiments.

[0032] A fault intelligent diagnosis method based on deep learning according to the present invention will be described in detail below in conjunction with the drawings and specific embodiments for the mechanical energy of the present invention.

[0033] A Figure 1 A fault intelligent diagnosis system based on deep learning as shown, the system includes: a convolutional layer, a pooling layer, and a fully connected layer, wherein the low hidden layer is alternately composed of a convolutional layer and a pooling layer, and the high layer is the fully connected layer corresponding to the hidden layer of the traditional multi-layer perceptron and the logistic regression classifier.

[0034] Preprocess the data to be analyzed, construct a sample set, input the training sample set into the network model for training, calculate the error between the output value and the true label according to the network parameters, and thus automatically adjust the network parameters. At the beginning, the network parameters are randomly initialized, so a large error is generated. Through network training, the objective function is iteratively optimized, thereby adjusting the network parameters to gradually reduce the error and finally achieve the expected training goal. Thus, a fault diagnosis model is obtained.

[0035] To construct a deep learning network, the core is to construct a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is composed of convolutional operations and activation operations to generate different feature maps. Each neuron in the feature map is connected to a local area through an activation function. The neurons reduce the model parameters by sharing a kernel, and through convolution and training, the key features and information of the input samples are extracted.

[0036] The pooling layer is constructed by downsampling to reduce the feature map parameters and ensure the extraction of high-level features.

[0037] The convolutional layer and the pooling layer alternate. The effectiveness of the algorithm for feature recognition, and the two of them constitute the low hidden layer to obtain the feature representation of the data.

[0038] In a convolutional neural network, each layer consists of multiple feature maps. Each feature map extracts a kind of feature of the input through a convolutional filter, and each feature map has multiple neurons. After the input sample statistics and the filter are convolved, this local feature is extracted. Once this local feature is extracted, its positional relationship with other features is also determined. The input of each neuron is connected to the local receptive field of the previous layer. Each feature extraction layer is followed by a computational layer for local averaging and secondary extraction, also called the feature mapping layer. Each computational layer of the network consists of multiple feature mapping planes, and the weights of all neurons on the plane are equal. Usually, the mapping from the input layer to the hidden layer is called a feature mapping, that is, the feature extraction layer is obtained through the convolutional layer, and the feature mapping layer is obtained after pooling.

[0039] Then a fully connected layer is constructed to connect the convolutional layer and the output layer, convert the two-dimensional feature information output by the convolutional network into one-dimensional feature information, and send the feature representation obtained by training the convolutional neural network to the output layer for classification output.

[0040] All training samples in the training set and the corresponding fault status labels are sent into the intelligent fault diagnosis system based on the convolutional neural network for model training, and the fault status of the test set samples is predicted. It is found that as the data volume of the sample set increases, the diagnostic accuracy is higher. It can be seen that the present invention is not sensitive to the sample data volume and still has good robustness for samples with a large data volume.

Claims

1. An intelligent fault diagnosis method based on deep learning, characterized in that, The steps include: Step 1: Build a neural network; Step 2: Training and fixing parameters; Step 3: Improve accuracy; Step 4: Form a convolutional layer; Step 5: Construct the pooling layer; Step 6: Diagnose equipment faults.

2. The intelligent fault diagnosis method based on deep learning according to claim 1, characterized in that: The step 1 comprises: The neural network includes a convolutional layer, a pooling layer, and a fully connected layer, and these three layers receive signals in sequence.

3. The intelligent fault diagnosis method based on deep learning according to claim 2, characterized in that: The step 2 comprises: The training and fixed parameters are trained using the preset training data to obtain the weight parameters of the model.

4. The intelligent fault diagnosis method based on deep learning according to claim 3, characterized in that: The step three comprises: The weight parameters in the network are reversely adjusted layer by layer by minimizing the loss function using the gradient descent method, and the accuracy of the network is improved through frequent iterative training.

5. The intelligent fault diagnosis method based on deep learning according to claim 4, characterized in that: The step 4 includes: Different feature maps are generated through convolution operations and activation operations to form a convolution layer; each neuron in the feature map is connected to only a local area of ​​the input through the activation function, and all neurons in the feature map share the same kernel.

6. The intelligent fault diagnosis method based on deep learning according to claim 5, characterized in that: The step five comprises: Use downsampling to reduce the feature space of feature maps, compress the amount of data and parameters, and reduce overfitting; Convolutional layers and maximum pooling layers alternately form the low hidden layers of the convolutional neural network, and the high layers are fully connected layers corresponding to the hidden layers and logistic regression classifiers of the traditional multi-layer perceptron.

7. The intelligent fault diagnosis method based on deep learning according to claim 6, characterized in that: The input of the first fully connected layer is the feature image obtained by feature extraction by the convolutional layer and the subsampling layer; Since the spatial connection of the sample is local, each neuron senses the local features, and then these sensed different local neurons are integrated at a higher level to reduce the number of neuron connections; Parameters are shared between neurons, and multiple filters are used to deconvolve samples to obtain multiple feature maps. The neurons in the hidden layer share weights and detect exactly the same features at different positions. The last output layer is a classifier, which can use logistic regression to classify the input and finally get a classification result, through which the fault diagnosis of the equipment can be realized.

Images