Communication cable channel fault detection method based on convolutional neural network

Abstract

The present invention provides a communication cable channel fault detection method based on a convolutional neural network, comprising the following steps: sending a test signal to a communication cable to be detected and receiving a reflected signal; extracting a soft fault signal containing communication cable channel status information according to the reflected signal; inputting the soft fault signal into a soft fault information detection network based on a convolutional neural network to obtain the fault type, fault distance, and fault value of the communication cable that can reflect the severity of the fault. Constructing a soft fault information detection network includes: establishing a soft fault equivalent model of the communication cable to obtain a channel transfer function; constructing a data set using the soft fault history data of the communication cable, taking the soft fault signal as input, taking the fault type, fault distance, and fault value corresponding to the channel transfer function of the communication cable as output, and using the data set to train the soft fault information detection network. The present invention can solve the technical problem that the soft fault of the communication cable cannot be effectively detected.

Description

Technical Field

[0001] The present invention relates to the technical field of communication cable fault detection, and in particular to a communication cable channel fault detection method based on a convolutional neural network. Background Art

[0002] With installation damage, daily damage, natural aging and other situations, communication systems with communication cables as the core often face abnormalities. If a fault occurs in a key cable segment and is not inspected and maintained in time, it may cause the entire communication system to collapse, resulting in huge economic losses. To ensure the stability of communication, the channel status of the cable needs to be monitored from time to time to avoid failures. In addition, most of the commonly used detection methods currently require professional technicians to analyze the measurement signals to determine the health of the cable, which is costly and inefficient, and often takes hours or even days to inspect and repair the fault. Therefore, it is necessary to study efficient and accurate detection technology to diagnose faults as quickly as possible.

[0003] Communication cable faults can be divided into hard faults and soft faults. Hard faults refer to obvious impedance changes, such as open circuits or short circuits. Soft faults refer to very small impedance mismatches between the fault and the wire, such as friction or wear. In the prior art, however, most methods are more complicated in fault information analysis and only target hard faults, and cannot effectively detect soft faults. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a communication cable channel fault detection method based on convolutional neural network to solve the technical problem that the prior art cannot effectively detect soft faults of communication cables.

[0005] The technical solution adopted by the present invention is a communication cable channel fault detection method based on convolutional neural network, comprising the following steps:

[0006] Send a test signal to the communication cable to be tested and receive a reflected signal;

[0007] Extracting a soft fault signal containing communication cable channel status information according to the reflected signal;

[0008] The soft fault signal is input into the soft fault information detection network based on convolutional neural network to obtain the fault type, fault distance and fault value that can reflect the severity of the fault of the communication cable.

[0009] Furthermore, the test signal is generated by:

[0010] generating a plurality of subcarrier symbol sequences to be transmitted;

[0011] Convert multiple subcarrier symbol sequences from the frequency domain to the time domain to obtain discrete signals;

[0012] The discrete signal is inserted into the cyclic prefix and then converted into the time domain to obtain the test signal.

[0013] Furthermore, a soft fault signal containing communication cable channel status information is extracted based on the reflected signal, including: converting the reflected signal into analog-to-digital, serial-to-parallel, removing cyclic prefix, fast Fourier transform and parallel-to-serial to obtain the soft fault signal.

[0014] Furthermore, a soft fault information detection network based on convolutional neural network is constructed, including:

[0015] Establish an equivalent model of soft fault of communication cable and obtain the channel transfer function of communication cable;

[0016] Build a soft fault information detection network architecture based on convolutional neural network;

[0017] The data set is constructed using the soft fault history data of the communication cable.

[0018] The soft fault signal is taken as input, and the fault type, fault distance and fault value corresponding to the channel transfer function of the communication cable are taken as output. The soft fault information detection network is trained and saved using the data set.

[0019] Furthermore, the channel transfer function of the communication cable is as follows:

[0020]

[0021] In the above formula, Γ F It represents the reflection coefficient caused by the soft fault of the communication cable, α represents the attenuation constant of the communication cable, β represents the phase coefficient of the communication cable, l represents the fault distance, and e represents the natural constant.

[0022] Furthermore, the fault types include: series resistance type, series capacitance type, parallel resistance type, parallel capacitance type and no fault type.

[0023] Furthermore, the soft fault information detection network includes a feature extraction network, a classifier, and a regressor;

[0024] The feature extraction network uses a one-dimensional convolutional neural network to extract the features of soft fault signals;

[0025] The classifier consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 3 fully connected layers;

[0026] The regressor consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 6 fully connected layers.

[0027] Furthermore, the loss function of the classifier is:

[0028]

[0029] In the above formula, a i represents the i-th fault type, p i The probability corresponding to a certain type;

[0030] The activation function of the classifier is ReLU.

[0031] Furthermore, the loss function of the regressor is:

[0032]

[0033] In the above formula, is the predicted fault value and fault distance value, y i is the corresponding true value;

[0034] The activation function of the regressor is ReLU.

[0035] Furthermore, the communication cable is a dual-conductor structure cable.

[0036] Furthermore, the data is enhanced using the window slicing method.

[0037] It can be seen from the above technical solution that the beneficial technical effects of the present invention are as follows:

[0038] No prior knowledge such as the cable propagation wave velocity is required, which solves the problem that traditional cable fault detection methods cannot effectively detect soft faults in communication cables. The severity of the fault can be judged and soft faults in communication cables can be effectively detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the prior art description. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0040] Figure 1 This is a schematic diagram of communication cable signal acquisition according to Embodiment 1 of the present invention;

[0041] Figure 2 This is a schematic diagram of a coaxial cable model circuit according to Embodiment 1 of the present invention;

[0042] Figure 3 This is a schematic diagram of an equivalent model circuit of a coaxial cable soft fault according to Embodiment 1 of the present invention;

[0043] Figure 4 This is a block diagram of a soft fault information detection network structure according to Embodiment 1 of the present invention;

[0044] Figure 5 This is a network structure diagram of a soft fault information detection network classifier according to Embodiment 1 of the present invention;

[0045] Figure 6 This is a network structure diagram of a soft fault information detection network regressor according to Example 1 of the present invention. DETAILED DESCRIPTION

[0046] The following embodiments of the technical solution of the present invention are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore only used as examples, and cannot be used to limit the protection scope of the present invention.

[0047] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in this application should have the common meanings understood by those skilled in the art to which the invention belongs.

[0048] Example 1

[0049] This embodiment provides a communication cable channel fault detection method based on a convolutional neural network, comprising the following steps:

[0050] S1. Send a test signal to the communication cable to be tested and receive the reflected signal

[0051] In a specific embodiment, Figure 1 As shown, a multi-carrier test signal with a pilot symbol is sent to the communication cable to be tested by using orthogonal frequency division multiplexing, and a reflected signal with the channel status information of the communication cable is received. It includes:

[0052] S1-1. Generate multiple subcarrier symbol sequences to be sent

[0053] In a specific implementation, an OFDM symbol generator is used to generate a sequence of 64 subcarrier symbols to be transmitted with 8 pilot symbols, the signal bandwidth is 20MHz, and the center frequency of the signal is set to 900MHz. After the data is generated, each subcarrier is modulated independently by orthogonal phase shift keying. Orthogonal phase shift keying uses a finite number of phases and assigns a unique binary digital pattern to each phase. In the frequency domain, the obtained symbol is represented as {X0, X1, ..., X N-1}, corresponding to the frequencies {f0,f1,...,f N-1}. Where N represents the number of subcarriers.

[0054] S1-2, convert multiple subcarrier symbol sequences from the frequency domain to the time domain to obtain discrete signals

[0055] After the symbol sequence obtained in the first step is converted from serial to parallel, a parallel data stream X(f) is obtained. The parallel data stream X(f) is subjected to an inverse fast Fourier transform to convert the signal from the frequency domain to the time domain, and the discrete signal x(n) is obtained as follows:

[0056]

[0057] In the above formula (1), T S represents the signal duration, and Δf represents the frequency interval between two consecutive subcarriers.

[0058] S1-3. Insert the discrete signal into the cyclic prefix and then convert it to the time domain to obtain the test signal

[0059] In order to reduce inter-symbol interference, a cyclic prefix is ​​inserted after the discrete signal x(n). Specifically, the last 16 bits of x(n) are copied to the head of x(n) to form a cyclic prefix. Then, the signal is converted from the frequency domain to the time domain through parallel-to-serial conversion and digital-to-analog conversion to obtain the test signal x(t).

[0060] S1-4, send a test signal to the communication cable to be tested and receive the reflected signal

[0061] The test signal x(t) is injected into the communication cable. When the test signal x(t) encounters an impedance mismatch during transmission in the communication cable, a reflection signal y(t) is generated. Wherein, x(t) represents the transmitted multi-carrier test signal with pilot symbols, y(t) represents the reflected signal generated by the impedance mismatch part in the communication cable, h(t) represents the channel transfer function of the communication cable, and n(t) represents the noise in the channel.

[0062] In a specific implementation, the reflected signal of the test signal may be received by a signal receiver, a vector network analyzer, or a development board based on the AD9361 chip.

[0063] S2. Extracting a soft fault signal containing the communication cable channel status information according to the reflected signal

[0064] The reflected signal y(t) is converted into a frequency domain Y(f) by analog-to-digital conversion, serial-to-parallel conversion, cyclic prefix removal, fast Fourier transform, and parallel-to-serial conversion. The soft fault signal Y(f) is a one-dimensional signal containing the channel status information of the communication cable, which can be provided to the soft fault information detection network based on the convolutional neural network for soft fault detection.

[0065] S3. Input the soft fault signal into a soft fault information detection network based on a convolutional neural network to obtain the fault type, fault distance and fault value of the communication cable that can reflect the severity of the fault.

[0066] Constructing a soft fault information detection network based on convolutional neural network includes the following steps:

[0067] S3-1. Establish an equivalent model of soft fault of communication cable and obtain the channel transfer function of communication cable

[0068] In this embodiment, the communication cable is a dual-conductor structure cable, and a coaxial cable is used as an example: when a soft fault occurs in the communication cable and the impedance change of the communication cable is analyzed, the communication cable is regarded as a distributed parameter line. Figure 2 The general equivalent circuit of the communication cable is shown. The vector form of the input signal in the communication cable can be expressed as:

[0069]

[0070] In the above formula (2), U + (x) represents the input signal, U - (x) represents the reflected signal, e represents the natural constant, γ represents the propagation constant of the cable, and x represents the length of the communication cable. γ=α+jβ, α is the attenuation constant of the cable, and β is the phase constant of the cable.

[0071] The terminal load of the communication cable is Z L When , its voltage reflection coefficient Γ(x,f) and channel transfer function H(f) are expressed as:

[0072]

[0073]

[0074] In the above equations (3) and (4), e represents a natural constant, γ represents a propagation constant of the cable, and x represents the length of the communication cable. L (f) is the terminal reflection coefficient of the cable. Γ L (f)=(Z L -Z C ) / (Z L +Z C ), Z L is the load impedance of the communication cable terminal, Z C is the characteristic impedance of the communication cable.

[0075] When a soft fault occurs at a certain position of the communication cable, the R, L, C, and G parameters at that position will change, thereby changing its characteristic impedance value. In this embodiment, the impedance discontinuity with a sufficiently short length in the cable is treated as a lumped element, and the soft faults of the communication cable are divided into two categories: series faults and parallel faults. The equivalent model of the soft fault of the communication cable is as follows: Figure 3 As shown, Figure 3The upper figure shows a communication cable channel with a series fault, and the lower figure shows a communication cable channel with a parallel fault.

[0076] When the communication cable terminal load matches Z L =Z C When the reflection coefficient Γ of the series and parallel fault models is FS , Γ FP Can be expressed as:

[0077]

[0078]

[0079] In the above formulas (5) and (6), Z C is the characteristic impedance of the communication cable. FS , Z FP are the fault values ​​in the series and parallel fault models respectively. The resistance, capacitance or inductance can be regarded as the lumped element parameter value instead.

[0080] When a soft fault occurs, for a lossy communication cable with a length of L, the channel transfer function H(f)' can be obtained from equations (4), (5), and (6):

[0081]

[0082] In the above formula (7), Γ F represents the reflection coefficient caused by the soft fault of the communication cable, α represents the attenuation constant of the communication cable, β represents the phase coefficient of the communication cable, l represents the fault distance, and e represents the natural constant. In formula (7), there exists l, |Γ F |、 These three unknown parameters, |Γ F | and The fault value Z F Indicates; will |Γ F |、 It is regarded as an unknown variable. At this time, there is a unique solution that can solve the fault distance, fault type and reflection coefficient value reflecting the fault degree.

[0083] When a communication cable is damaged by a soft fault, its channel transfer function can reflect the fault distance, fault type, and reflection coefficient value reflecting the severity of the fault.

[0084] S3-2. Build a soft fault information detection network architecture based on convolutional neural network

[0085] The soft fault information detection network includes a feature extraction network, a classifier and a regressor. In a specific implementation, a soft fault information prediction network based on a one-dimensional convolutional neural network can be built based on the Keras open source artificial neural network library. The following is a specific description:

[0086] like Figure 4 As shown, the input of the soft fault information detection network is a soft fault signal containing communication cable channel status information extracted based on the reflected signal.

[0087] Feature extraction network: In a specific implementation, a one-dimensional convolutional neural network is used to extract the features of the soft fault signal. The output features after the one-dimensional convolutional neural network are:

[0088] output=dropout(ReLU(Conv1D(x)))

[0089] Among them, x represents the input signal, which is the collected reflection signal containing the communication cable channel status. Output represents the output feature after the one-dimensional convolutional neural network; the output feature output is fed to the classifier and regressor for further processing to obtain the classification and prediction results

[0090] The prediction output of the soft fault information detection network can be expressed as:

[0091]

[0092] in, represents the prediction result, including the fault type, fault distance, and fault value that can reflect the severity of the fault. x represents the collected reflection signal containing the communication cable channel status. Represents the weight coefficient matrix of each layer of the network, L represents the soft fault information detection network has L layers, and the output of each layer is expressed by the function To express.

[0093] The loss function of the soft fault information detection network is:

[0094]

[0095]

[0096] in:

[0097] L t is the loss function of the classifier, a i represents the i-th type, including series resistance type, series capacitance type, parallel resistance type, parallel capacitance type and fault-free type, p i The probability corresponding to a certain type.

[0098] L r is the loss function of the regressor, is the predicted fault value and fault distance value, y iis the corresponding true value. Combining the two loss functions, the final loss function is:

[0099] L=α·[L t ,L r ] T

[0100] In the above formula, α is the attenuation factor, which can avoid overfitting.

[0101] Classifier: In a specific embodiment, Figure 5 As shown in the figure, the network of the classifier consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 3 fully connected layers. A MaxPooling 1D layer with a pool size of 2x1 and a stride of 1 is added after every 2 convolutional layers to retain the main features and reduce the amount of calculation. Rectified linear unit (ReLU) is used as the activation function in each convolutional layer. The input size of the network is 128*1. The first and second convolutional layers have 80 filters each. The third and fourth convolutional layers have 100 filters each, and the fifth and sixth convolutional layers have 160 filters each. All filters in the 6 convolutional layers have the same kernel size of 3x1. At the end of the last MaxPooling layer, a global average pooling layer (GAP) is added, and each feature detector has only one weight left in the global average pooling layer. Then there are 2 Dense layers with 64 and 32 neurons respectively, and the activation function is ReLU. The last Dense layer uses the Softmax activation function to reduce the vector length to 4, so that the sum of the output values ​​of the neural network is 1, representing the probability of various fault types.

[0102] Regressor: In a specific implementation, such as Figure 6As shown in the figure, the network of the regressor consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 6 fully connected layers. A MaxPooling 1D layer with a pool size of 2x1 and a stride of 1 is added after every 2 convolutional layers to retain the main features and reduce the amount of calculation. Rectified linear unit (ReLU) is used as the activation function in each convolutional layer. The input size of the network is 128*1. The first and second convolutional layers have 80 filters each. The third and fourth convolutional layers have 100 filters each, and the fifth and sixth convolutional layers have 160 filters each. All filters in the 6 convolutional layers have the same kernel size of 3x1. At the end of the last MaxPooling layer, a global average pooling layer (GAP) is added, and each feature detector has only one weight left in the global average pooling layer. After extracting the common feature information using CNN (convolutional neural network), two parallel estimation networks are connected to the back of the network, which are used to predict the fault length and fault capacitance or resistance value information respectively. Each estimation network is formed by stacking three fully connected layers, with 64, 32, and 1 neurons respectively, all activated by ReLU. Each neuron in the Dense layer is connected to all neurons in the previous layer through learnable weights. The response of the hidden layer unit is calculated by summing the product of each input signal and its corresponding weight, and passing the sum through ReLU, and finally outputting the fault distance and fault value, which includes the capacitance value and the resistance value.

[0103] S3-3. Use the soft fault history data of the communication cable to build a data set, which includes the channel transfer function of the communication cable

[0104] The collected soft fault history data of the communication cable is illustrated as follows: the cable is LMR-195 coaxial cable. There are 8 lengths of the cable to be tested, namely 5.22m, 6.77m, 8m, 9.94m, 11.2m, 13.15m, 14m, and 15m. The impedance mismatch part in the cable is simulated by a matching microstrip line circuit board. There are 4 types of fault boards: series resistance, series capacitance, parallel resistance, and parallel capacitance, each with 30 different impedance values. By replacing the cable to be tested and the fault board, the channel data of the communication cable under different soft fault conditions can be collected. A total of 9604 sets of data were collected in the data set, including 2330 sets of series resistance data, 2328 sets of series capacitance data, 2406 sets of parallel resistance data, 2290 sets of parallel capacitance data, and 250 sets of fault-free data.

[0105] The soft fault history data of the communication cable includes the channel transfer function of the communication cable when the soft fault occurs, as well as the fault type, fault distance, and fault value determined after the soft fault occurs. The data length in the data set is 1×128.

[0106] S3-4. Take the soft fault signal as input, take the fault type, fault distance, and fault value corresponding to the channel transfer function of the communication cable as output, and use the data set to train and save the soft fault information detection network.

[0107] The soft fault signal extracted from the reflected signal and containing the communication cable channel status information is used as the input of the soft fault information detection network, and the fault type, fault distance and fault value corresponding to the channel transfer function of the communication cable are used as the output of the soft fault information detection network. The soft fault information detection network is trained and the trained network is saved.

[0108] By inputting the soft fault signal into the trained soft fault information detection network, the fault type, fault distance and fault value that can reflect the severity of the fault of the communication cable can be obtained.

[0109] By adopting the technical solution of this embodiment, no prior knowledge such as the cable propagation wave velocity is required, which solves the problem that traditional cable fault detection methods cannot effectively detect soft faults in communication cables. The severity of the fault can be judged, and soft faults in communication cables can be effectively detected.

[0110] Example 2

[0111] When the data set is used to train the soft fault information detection network in Example 1, due to the limited number of samples in the data set, overfitting is likely to occur when training the neural network.

[0112] In order to solve the above technical problems, the following technical solutions are adopted:

[0113] The data is augmented using the window slicing method on the dataset.

[0114] In a specific implementation manner, the data length of the data set in Example 1 is 1×128, the slice length is set to 116, and sub-samples of continuous slices are randomly extracted from the original data, and the labels of the slice samples are the same as the original data.

[0115] By adopting the technical solution of this embodiment, the amount of data can be increased based on the original data, thereby avoiding overfitting of the neural network during the training process.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. A communication cable channel fault detection method based on convolutional neural network, characterized in that: The following steps are involved: Send a test signal to the communication cable to be tested and receive a reflected signal; Extracting a soft fault signal containing communication cable channel status information according to the reflected signal; The soft fault signal is input into the soft fault information detection network based on the convolutional neural network to obtain the fault type, fault distance and fault value of the communication cable that can reflect the severity of the fault; Constructing the soft fault information detection network based on the convolutional neural network includes: establishing an equivalent model of the soft fault of the communication cable to obtain the channel transfer function of the communication cable; building a soft fault information detection network architecture based on the convolutional neural network; using the soft fault history data of the communication cable to construct a data set, taking the soft fault signal as input, taking the fault type, fault distance, and fault value corresponding to the channel transfer function of the communication cable as output, and using the data set to train and save the soft fault information detection network; The channel transfer function of the communication cable is as follows: In the above formula, Γ F It represents the reflection coefficient caused by the soft fault of the communication cable, α represents the attenuation constant of the communication cable, β represents the phase coefficient of the communication cable, l represents the fault distance, and e represents the natural constant.

2. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: The test signal is generated in the following manner: generating a plurality of subcarrier symbol sequences to be transmitted; Convert multiple subcarrier symbol sequences from the frequency domain to the time domain to obtain discrete signals; The discrete signal is inserted into the cyclic prefix and then converted into the time domain to obtain the test signal.

3. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: Extracting a soft fault signal containing communication cable channel status information according to the reflected signal includes: converting the reflected signal into analog-to-digital, serial-to-parallel, removing cyclic prefix, fast Fourier transform and parallel-to-serial to obtain the soft fault signal.

4. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: The fault types include: series resistance type, series capacitance type, parallel resistance type, parallel capacitance type and no fault type.

5. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: The soft fault information detection network includes a feature extraction network, a classifier and a regressor; The feature extraction network uses a one-dimensional convolutional neural network to extract the features of the soft fault signal; The classifier consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 3 fully connected layers; The regressor consists of 6 one-dimensional convolutional layers, 3 global average pooling layers and 6 fully connected layers.

6. The communication cable channel fault detection method based on convolutional neural network according to claim 5 is characterized in that: The loss function of the classifier is: In the above formula, a i represents the i-th fault type, p i The probability corresponding to a certain type; The activation function of the classifier is ReLU.

7. The communication cable channel fault detection method based on convolutional neural network according to claim 5 is characterized in that: The loss function of the regressor is: In the above formula, is the predicted fault value and fault distance value, y i is the corresponding true value; The activation function of the regressor is ReLU.

8. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: The communication cable is a double-conductor structure cable.

9. The communication cable channel fault detection method based on convolutional neural network according to claim 1 is characterized in that: The data is augmented using the window slicing method on the dataset.

Images