Ion tube fault diagnosis method

Abstract

The application discloses an ion tube fault diagnosis method, comprising the following steps: classifying and marking faults according to Lissajous figures of the ion tube; constructing a convolutional neural network, and training the convolutional neural network by using the classified and marked Lissajous figures; and constructing a two-stage classification framework by using the trained convolutional neural network, and diagnosing faults of the ion tube. The application adopts the convolutional neural network to diagnose faults of the ion tube, and the designed method has the advantages of high efficiency, accuracy and intelligence.

Description

Technical Field

[0001] This invention relates to a method for diagnosing ion tube faults. Background Technology

[0002] Ion tubes are dielectric barrier discharge devices used in air purification. The operating status of the ion tube is a key factor affecting air purification efficiency. In actual use, ion tube malfunctions are difficult to detect in a timely manner, significantly increasing user concerns about their reliability. To improve equipment reliability, manufacturers of air purification equipment using ion tubes typically increase the frequency of manual maintenance. This increases maintenance and operating costs, and manual maintenance also suffers from high safety risks, low efficiency, and a high rate of misdiagnosis.

[0003] As a typical dielectric barrier discharge device, the Lissajous figure is the main characterization of the ion tube's operating characteristics. The Lissajous figure of an ion tube can be obtained through experimental measurement, and then its fault status can be determined manually. Manual interpretation is generally labor-intensive and prone to subjective errors, and also faces difficulties in batch processing and achieving automation and intelligent fault diagnosis of ion tubes.

[0004] Convolutional neural networks (CNNs) based on deep learning have been widely used in fault classification and diagnosis. By constructing a CNN that matches the characteristics of the object being diagnosed, and then training the network with labeled samples, intelligent fault diagnosis can be achieved. However, constructing a suitable CNN and a suitable sample set for classification are key to designing a CNN-based fault diagnosis method, and the structure, parameters, and training method of the CNN will also affect the accuracy of fault diagnosis. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an ion tube fault diagnosis method that is highly reliable and easy to automate and intelligently diagnose ion tube faults.

[0006] The technical solution of this invention to solve the above problems is: a method for diagnosing ion tube faults, comprising the following steps:

[0007] S1. Based on the Lissajous figure of the ion tube, classify and mark the faults;

[0008] S2. Construct a convolutional neural network and train it using a Lissajous figure with classification labels;

[0009] S3. A two-level classification framework is constructed using a trained convolutional neural network to diagnose faults in ion tubes.

[0010] The advantages of this invention are as follows: For the problem of ion tube fault diagnosis, compared with traditional methods, this invention provides a highly reliable ion tube fault diagnosis method that is easy to automate and automates. The method's strengths lie in its use of a two-level classification framework built on a convolutional neural network to classify and identify the Lissajous figures of the ion tubes, thereby obtaining the fault status of the ion tubes. This achieves automated and intelligent fault diagnosis, improving efficiency and accuracy while reducing maintenance costs. Attached Figure Description

[0011] Figure 1 A flowchart of the ion tube fault diagnosis method in this invention.

[0012] Figure 2 A typical Lissachus diagram of the novel tube in this invention is shown in the figure.

[0013] Figure 3 The typical Lissajous figure shows the breakdown tube in this invention.

[0014] Figure 4 The typical Lissachus diagram of the semi-new tube in this invention is shown in the figure.

[0015] Figure 5 A typical Lissajous figure shows the open tube in this invention.

[0016] Figure 6 The first convolutional neural network structure diagram in this invention.

[0017] Figure 7 The principle diagram of the second-level classification in this invention.

[0018] Figure 8 The second type of convolutional neural network structure diagram in this invention.

[0019] Figure 9 The present invention employs a two-level classification framework constructed using a convolutional neural network. Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0021] The flowchart of the ion tube fault diagnosis method of the present invention is as follows: Figure 1 As shown, it includes the following steps:

[0022] S1. Based on the Lissajous figure of the ion tube, classify and mark the faults.

[0023] The ion tube refers to a high-energy ion tube used to generate negative ions. It consists of an inner electrode, an outer electrode, and a dielectric material. The outer electrode is a stainless steel mesh, the inner electrode is an aluminum plate, and the dielectric material is 1mm thick quartz glass. The ion tube is cylindrical, with the dielectric material being a closed-structure glass tube. The inner electrode is tightly attached to the inside of the glass tube, and the outer electrode is wrapped around the outside of the glass tube. A high voltage is generated by a power frequency transformer and applied to the inner and outer electrodes to achieve dielectric barrier discharge. The Lissajous figure of the ion tube refers to a closed curve constructed by the amount of charge flowing through the ion tube and the applied voltage during operation. The Lissajous figure directly reflects the key information of the ion tube's state.

[0024] Ion tubes are classified and labeled according to their Lissajous figures. The fault categories include: brand new tubes, breakdown tubes, semi-new tubes, and open tubes. A brand new tube refers to a newly manufactured ion tube. A breakdown tube is one where the glass dielectric has cracked, causing breakdown discharge under high voltage. A semi-new tube is one that has accumulated more than 100 hours of operation and can still function normally. An open tube is one where the top of the glass tube is broken, creating an open space, but the discharge area is not cracked and can still discharge. Lissajous figures are obtained for each of these four types of ion tubes. Figure 2 The Lissajous figure shows a brand new tube, labeled as a brand new tube; Figure 3 The Lissajous figure for the breakdown tube is shown in the diagram and is labeled as the breakdown tube; Figure 4 The Lissajous figure shows a semi-new tube, labeled as a semi-new tube. Figure 5 The Lissajous figure shows an open tube, labeled as an open tube.

[0025] S2. Construct a convolutional neural network and train it using a Lissajous figure with classification labels.

[0026] The constructed convolutional neural network includes two types of LeNet convolutional neural networks. The first type of convolutional neural network performs first-level classification to distinguish between broken tubes and non-broken tubes, where non-broken tubes include brand new tubes, semi-new tubes, and open tubes. The second type of convolutional neural network performs second-level classification to distinguish between brand new tubes, semi-new tubes, and open tubes.

[0027] The first type of convolutional neural network structure is as follows: Figure 6As shown, the structure from input to output includes: input layer, convolutional layer, pooling layer, fully connected layer 1, fully connected layer 2, and output layer. The input layer takes an image of the ion tube as input (Lissajous figure), with an image size of 32×32 pixels. The convolutional layer contains 6 convolutional kernels, each 5×5 in size, and uses the ReLU nonlinear activation function for nonlinear calculation. The pooling layer uses max pooling with a pooling window size of 2×2 to downsample the feature map. Fully connected layer 1 flattens the pooled feature map into a one-dimensional vector. Fully connected layer 2 uses the Softmax activation function to provide fault classification prediction, converting the output into a probability distribution for classification. The output layer outputs a binary classification result: "Breakdown(0)" for a broken tube and "Workable(1)" for a non-broken tube.

[0028] right Figure 6 The first type of convolutional neural network shown was trained using 300 Lissajous figures labeled as broken-down tubes and non-broken-down tubes. The neural network was set up and trained using Keras and TensorFlow, employing the Adam optimizer and classification cross-entropy loss function to train the network and evaluate errors during training. The model's performance was then evaluated on the test set.

[0029] Figure 7 This is a schematic diagram of the second-level classification principle in this invention. The purpose of the second-level classification is to further diagnose the state of the non-breakdown ion tube based on the first-level classification. Therefore, this classification process first normalizes the Lissajous figure of the non-breakdown ion tube, and then... Figure 8 The second type of convolutional neural network shown is used for classification, and the classification results are obtained for open tubes (represented by the number "0"), semi-new tubes (represented by the number "1"), and brand new tubes (represented by the number "2").

[0030] The second type of convolutional neural network structure is as follows: Figure 8As shown, the structure from input to output includes: input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, fully connected layer 1, fully connected layer 2, and output layer. The input layer takes an image of the ion tube as Lissajous figure, with a size of 32×32 pixels. Convolutional layer 1 consists of 6 convolutional kernels of size 5×5, which use the ReLU activation function to introduce a nonlinear transformation. Pooling layer 1 uses max pooling to reduce the spatial dimension of the feature map and uses a 2×2 pooling window. Convolutional layer 2 consists of 16 convolutional kernels of size 5×5, which use the ReLU activation function to reduce the interdependence of network model parameters. Pooling layer 2 flattens the obtained feature map into a one-dimensional vector so that it can be input into the fully connected layer. Fully connected layer 1 is used to learn the complex patterns and high-level features of the input data, connecting each neuron of the previous layer to the next layer to complete the classification task. Fully connected layer 2 uses a Softmax classifier to convert the output into a probability distribution for each category. The output layer outputs three categories of non-breakdown ion tubes: open tube "broken(0)", semi-new tube "dirty(1)", and brand new tube "new(2)".

[0031] right Figure 8 The second type of convolutional neural network shown was trained using 200 Lissajous figures labeled as open tubes, semi-new tubes, and novel tubes. The neural network was set up and trained using Keras and TensorFlow, and the Adam optimizer and classification cross-entropy loss function were used to train the network and evaluate the error during the training process. The performance of the model was evaluated on the test set.

[0032] S3. A two-level classification framework is constructed using a trained neural network to diagnose faults in ion tubes.

[0033] A two-level classification framework is constructed using a pre-trained convolutional neural network to classify the Lissajous figures of the ion tubes to be diagnosed. The two-level classification framework is as follows: Figure 9 As shown in the diagram. The specific classification process is as follows: The Lissajous figure of the ion tube to be diagnosed is input into the first type of convolutional neural network for first-level classification, determining whether the ion tube is a breakdown tube. If it is determined to be a breakdown tube, the diagnosis ends; if it is determined to be a non-breakdown tube, it proceeds to the second-level classification. The Lissajous figures of the ion tubes to be diagnosed that have entered the second-level classification are normalized. The normalized Lissajous figures are then input into the second type of convolutional neural network for second-level classification, determining whether the ion tube to be diagnosed is a brand new tube, a semi-new tube, or an open tube. Through the two-level classification framework, the ion tube to be diagnosed is determined to be one of the following: a breakdown tube, a brand new tube, a semi-new tube, or an open tube, thus achieving fault diagnosis of the ion tube.

Claims

1. An ion tube failure diagnosis method characterized by comprising: Comprise the following steps: S1. According to the Lissajous figure of ion tube, fault classification and marking are carried out; The ion tube refers to a high-energy ion tube for generating negative ions, which is composed of an inner electrode, an outer electrode and a medium, wherein the outer electrode is a stainless steel mesh, the inner electrode is an aluminum plate, the medium is 1mm thick quartz glass, the ion tube adopts a cylindrical shape, the medium is a glass tube with a closed structure, the inner electrode is tightly attached to the inner side of the glass tube, and the outer electrode is wrapped outside the glass tube, and a high voltage is generated by a power frequency transformer and loaded on the inner electrode and the outer electrode to realize dielectric barrier discharge of the ion tube; The Lissajous figure of the ion tube refers to a closed curve graph constructed by the charge quantity flowing through the ion tube and the applied voltage when the ion tube is working, and the Lissajous figure of the ion tube is the key information directly reflecting the state of the ion tube; According to the Lissajous figure of the ion tube, the ion tube is classified and marked, and the fault classification includes: new tube, breakdown tube, half new tube, open tube, new tube refers to a new ion tube just out of the factory, breakdown tube refers to an ion tube with a crack in the glass medium, which forms a breakdown discharge under high voltage, half new tube refers to an ion tube with a cumulative working time of more than 100 hours and still can work normally, open tube refers to an ion tube with a broken glass tube top end, forming an open space, but the discharge area has no crack, and the ion tube can still discharge; The Lissajous figure of the above four types of ion tubes is obtained respectively and marked; S2. Construct a convolutional neural network, and train the convolutional neural network using the classified Lissajous figure; S3. Construct a two-level classification framework using the trained convolutional neural network to diagnose the fault of the ion tube; The Lissajous figure of the ion tube to be diagnosed is classified by using the trained convolutional neural network to construct a two-level classification framework, and the specific classification process is: the Lissajous figure of the ion tube to be diagnosed is input into the first convolutional neural network for first-level classification to determine whether the ion tube to be diagnosed is a breakdown tube, if it is determined to be a breakdown tube, the diagnosis is ended, if it is determined to be a non-breakdown tube, it enters the second-level classification; The Lissajous figure of the ion tube to be diagnosed in the second-level classification is normalized, and the normalized Lissajous figure is input into the second convolutional neural network for second-level classification to determine whether the ion tube to be diagnosed is a new tube, a half new tube or an open tube; Through the two-level classification framework, the ion tube to be diagnosed will finally be determined as one of the breakdown tube, the new tube, the half new tube and the open tube, and the fault diagnosis of the ion tube is realized.

2. The ion tube failure diagnosis method according to claim 1, characterized by, In step S2, the constructed convolutional neural network comprises two LeNet convolutional neural networks, the first convolutional neural network performs first-level classification to distinguish between breakdown tubes and non-breakdown tubes, wherein the non-breakdown tubes include new tubes, half new tubes and open tubes; The second convolutional neural network performs second-level classification to distinguish between new tubes, half new tubes and open tubes. The first convolutional neural network from input to output structure is: input layer, convolutional layer, pooling layer, full connection layer 1, full connection layer 2, output layer, wherein the input layer inputs the Lissajous figure image of the ion tube, the image size is 32*32 pixels; the convolutional layer contains 6 convolution kernels, each convolution kernel size is 5*5, and the nonlinear calculation is realized by using the ReLU nonlinear activation function; the pooling layer adopts the maximum pooling layer, the pooling window size is 2*2, and the feature map is down-sampled; the full connection layer 1 is used to flatten the feature map after the pooling into a one-dimensional vector; the full connection layer 2 uses the activation function Softmax to provide the classification prediction of the fault, and converts the output into a probability distribution for classification; the output layer outputs the binary classification result: the breakdown tube "Breakdown (0)", the non-breakdown tube "Workable (1)"; The training sample of the first convolutional neural network is 300 Lissajous figures labeled as breakdown tubes and non-breakdown tubes, the neural network is set and trained through Keras and TensorFlow, the Adam optimizer and the classification cross-entropy loss function are used to train the network and evaluate the error in the training process, and finally the performance of the model is evaluated on the test set; The second convolutional neural network from input to output structure is: input layer, convolutional layer 1, pooling layer 1, convolutional layer 2, pooling layer 2, full connection layer 1, full connection layer 2, output layer, wherein the input layer inputs the Lissajous figure image of the ion tube, the image size is 32*32 pixels; the convolutional layer 1 includes 6 convolution kernels with a size of 5*5, and the ReLU activation function is used to introduce nonlinear transformation; the pooling layer 1 adopts the maximum pooling layer to reduce the spatial dimension of the feature map, and uses a 2*2 pooling window; the convolutional layer 2 includes 16 convolution kernels with a size of 5*5, uses the ReLU activation function, and reduces the mutual dependence of network model parameters; the pooling layer 2 flattens the obtained feature map into a one-dimensional vector for input into the full connection layer; the full connection layer 1 is used to learn the complex mode and high-level features of the input data, connects each neuron of the previous layer to the next layer, and completes the classification task; the full connection layer 2 uses the Softmax classifier to convert the output into a probability distribution for each class; the output layer outputs three categories of non-breakdown ion tubes: open tube "broken (0)", half-new tube "dirty (1)", and brand-new tube "new (2)"; The training sample of the second convolutional neural network is 200 Lissajous figures labeled as open tube, half-new tube and brand-new tube, the neural network is set and trained through Keras and TensorFlow, the Adam optimizer and the classification cross-entropy loss function are used to train the network and evaluate the error in the training process, and the performance of the model is evaluated on the test set.

Images