Online fault diagnosis method for dual-active bridge converter based on multi-source data fusion

Abstract

The invention discloses a multi-source data fusion dual-active bridge converter online fault diagnosis method, which comprises the following steps: S1, establishing a simulation model of a dual-active bridge converter to be tested, and selecting a diagnosis signal by analyzing the relationship between different open-circuit fault states of the dual-active bridge converter and circuit signal change, collecting diagnosis signals of the dual-active bridge converter in different open-circuit fault states to construct a multi-source data sequence set; s2, performing spatial reconstruction on the multi-source data sequence set to obtain a multi-scale reconstruction data matrix; s3, the multi-scale reconstruction data matrix is divided into two continuous matrixes in time, after the two continuous matrixes are lifted to a high dimension, the two continuous matrixes are combined with the matrix before dimension rising, and two high-dimensional matrixes are obtained; s4, establishing a Koopman model according to the high-dimensional matrix, and extracting modal parameters of the Koopman model; and S5, inputting the Koopman modal parameters into the trained wavelet time scattering convolutional network, and obtaining a diagnosis result.

Description

Technical Field

[0001] The invention relates to the technical field of power electronic converter fault diagnosis, and in particular to an online fault diagnosis method for a dual active bridge converter using multi-source data fusion. Background Art

[0002] With the rapid development of power electronics technology, Dual Active Bridge Converter (DAB) has become a key component in a variety of application areas, such as power systems, electric vehicles and renewable energy systems. This topology is characterized by the bidirectional flow of electrical energy, allowing efficient energy conversion and exchange between multiple energy sources and loads. However, DAB is often affected by various internal and external factors during operation, resulting in failures, which not only have a negative impact on system performance, but may also pose a threat to the reliability and safety of the equipment.

[0003] Power semiconductor devices play a vital role in converters because they are responsible for controlling the flow and conversion of electrical energy. Although the causes of power equipment failure vary, these failures are closely related to operating conditions and circuit parameters. Investigations have shown that most power electronic devices such as insulated gate bipolar transistors (IGBTs) and their gate drivers are prone to failure, especially open circuit failures, because compared with short circuit failures, short circuit failures will trigger the overcurrent protection of the circuit, while open circuit failures cannot be better protected by integrated gate drive circuits and are difficult to detect and locate. Summary of the invention

[0004] The object of the present invention is to provide an online fault diagnosis method for a dual active bridge converter with multi-source data fusion, which improves the fault tolerance of open circuit fault diagnosis by integrating multiple fault diagnosis signals.

[0005] In order to solve the above technical problems, the present invention provides a multi-source data fusion dual active bridge converter online fault diagnosis method, comprising: S1. Establish a simulation model of the dual active bridge converter to be tested, select diagnostic signals by analyzing the relationship between different open circuit fault states of the dual active bridge converter and circuit signal changes, collect diagnostic signals of the dual active bridge converter under different open circuit fault states to construct a multi-source data sequence set; S2, spatially reconstruct the multi-source data sequence set to obtain a multi-scale reconstructed data matrix; S3, dividing the multi-scale reconstructed data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, merging them with the matrices before the upgrade to obtain two high-dimensional matrices; S4, establishing a Koopman model according to the high-dimensional matrix, and extracting the modal parameters of the Koopman model; S5. Input the modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis results. According to the above scheme, the diagnostic signal includes the leakage inductance current, the midpoint voltage of the primary bridge arm, and the midpoint voltage of the secondary bridge arm.

[0006] According to the above scheme, step S2 includes: using the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstructed data matrix, and using the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstructed data matrix.

[0007] According to the above scheme, in step S3, two continuous matrices are upgraded to high dimensions through a deep neural network algorithm with a trainable dictionary.

[0008] According to the above scheme, in step S4, the modal parameters of the Koopman model are extracted by EDMD dynamic mode decomposition.

[0009] According to the above scheme, the deep neural network algorithm of the trainable dictionary is trained through a fully connected neural network.

[0010] The present invention also provides a multi-source data fusion dual active bridge converter online fault diagnosis method system, comprising: A diagnostic signal selection module is used to analyze the relationship between different open-circuit fault states of the dual active bridge converter and circuit signal changes according to the established simulation model of the dual active bridge converter to be tested, and select a diagnostic signal; A multi-source data sequence set building module, used for collecting diagnostic signals of the dual active bridge converter under different open circuit fault states to build a multi-source data sequence set; A multi-scale reconstruction data matrix acquisition module is used to spatially reconstruct a multi-source data sequence set to obtain a multi-scale reconstruction data matrix; A high-dimensional matrix acquisition module is used to divide the multi-scale reconstruction data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, they are merged with the matrices before the upgrade to obtain two high-dimensional matrices; Koopman modal parameter extraction module, used to establish the Koopman model and extract the modal parameters of the Koopman model; The diagnosis result acquisition module is used to input the Koopman modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis result.

[0011] According to the above scheme, the diagnostic signal includes the leakage inductance current, the midpoint voltage of the primary bridge arm, and the midpoint voltage of the secondary bridge arm.

[0012] According to the above scheme, the method for the multi-scale reconstruction data matrix acquisition module to obtain the multi-scale reconstruction data matrix includes: taking the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstruction data matrix, and taking the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstruction data matrix.

[0013] According to the above scheme, the high-dimensional matrix acquisition module upgrades two continuous matrices to high dimensions through a deep neural network algorithm with a trainable dictionary.

[0014] Beneficial Effects The present invention constructs a multi-source data sequence set according to diagnostic signals under different open-circuit faults, and performs diagnosis based on the multi-source data sequence set, thereby realizing the integration of different fault characteristics, so that the method can more accurately identify different types of open-circuit faults; by adopting Koopman modal parameters for analysis, not only the static structure of the dual active bridge converter system can be provided, but also the dynamic behavior of the system can be captured, and the dynamic system analysis has stronger explanatory power and practicality; a wavelet time scattering convolutional network is adopted, in which different orders of wavelet scattering correspond to the extraction of different order features of the signal, and suitable filters can be adaptively selected to match the frequency spectrum characteristics of different signals, so that each fault can be more accurately identified. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a multi-signal fault diagnosis flow chart combining the Koopman operator and the wavelet time scattering convolution network according to the first embodiment of the present invention; Figure 2 (a) is a schematic diagram of a topological structure simulation of a bidirectional active full-bridge DC-DC converter according to a first embodiment of the present invention; Figure 2 (b) is a diagram of an IGBT driving circuit according to the first embodiment of the present invention; Figure 2 (c) is a diagram showing an output drive waveform of an IGBT drive circuit according to the first embodiment of the present invention; Figure 3 is a mean square error variation curve of the deep neural network with a trainable dictionary according to the first embodiment of the present invention; Figure 4 (a) is a diagram of modal frequency and attenuation rate and a distribution diagram of characteristic roots under fault category F2 of the first embodiment of the present invention; Figure 4 (b) is a diagram of modal frequency and attenuation rate and a distribution diagram of characteristic roots under fault category F3 of the first embodiment of the present invention; Figure 5 (a) to (d) are scattering transformation diagrams of lev=1 under fault categories F1 to F4 of the first embodiment of the present invention; Figure 6(a) is a curve diagram of the training accuracy change of the wavelet time scattering convolutional network in Example 1 of the present invention; Figure 6 (b) is a loss variation curve of the wavelet time scattering convolutional network of the first embodiment of the present invention; Figure 7 is the confusion matrix of the test set of the wavelet time scattering convolutional network according to the first embodiment of the present invention; Figure 8 It is a flow chart of an online fault diagnosis method of a dual active bridge converter with multi-source data fusion according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution of the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings of the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure.

[0017] Embodiment 1: See also Figure 8 , a multi-source data fusion dual active bridge converter online fault diagnosis method, comprising: S1. Establish a simulation model of the dual active bridge converter to be tested. By analyzing the relationship between different open circuit fault states of the dual active bridge converter and the change of circuit signals, select diagnostic signals, collect diagnostic signals of the dual active bridge converter under different open circuit fault states to construct a multi-source data sequence set; the multi-source data sequence set is expressed as ,in ; S2. Perform spatial reconstruction on the multi-source data sequence set to obtain a multi-scale reconstructed data matrix. The multi-scale reconstructed data matrix is ​​expressed as ; S3, dividing the multi-scale reconstructed data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, merging them with the matrices before the upgrade to obtain two high-dimensional matrices; S4, establishing a Koopman model according to the high-dimensional matrix, and extracting the modal parameters of the Koopman model; S5. Input the modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis results. In this embodiment, the simulation circuit diagram of the simulation model of the dual active bridge converter is established. Figure 2 , Figure 2(b) The IGBT drive circuit model is used to generate a pulse modulation (PWM) signal with a drive voltage of 15V / -8V. Extended phase shift control is achieved based on characteristic parameter modeling. Figure 2 (c) is the driving waveform output by the IGBT driving circuit. The open circuit fault is simulated by disconnecting the PWM signal of the IGBT. The fault status and category of the DAB converter are shown in Table 1: Table 1 Fault status and categories

[0018] Furthermore, a simulation model of the dual active bridge converter is established in Pspice.

[0019] Further, the diagnostic signal includes leakage inductance current, primary bridge arm midpoint voltage, and secondary bridge arm midpoint voltage; Due to the symmetrical structure of the DAB converter, the primary output voltage is , and the secondary output voltage and the transformer leakage current The waveform is the same, and the IGBT fault in the symmetrical position cannot be accurately located. Combined with the bridge arm midpoint voltage , and leakage current As a diagnostic signal, all IGBT open circuit fault diagnosis of DAB converter is studied; Therefore, the bridge arm midpoint voltage is extracted , and leakage current ,In this embodiment, 500 points are collected for each sample, and 3 time series are spatially reconstructed.

[0020] Furthermore, the step S2 includes: using the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstructed data matrix, and using the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstructed data matrix.

[0021] Furthermore, in step S3, two continuous matrices are promoted to high dimensions by a deep neural network algorithm with a trainable dictionary; the deep neural network algorithm is trained by a fully connected neural network; Multi-scale reconstruction data matrix The two continuous matrices divided in the time dimension are expressed as:

[0022] in , is the data collection time step Where -dimensional column vector; In this example, a 3-layer fully connected neural network is trained to train the dictionary function:

[0023]

[0024] in, is the network weight, is the bias, is the activation function, is a dimension-raising function; The loss function of the fully connected neural network is defined as:

[0025] In this embodiment, the dimensionality increase process is realized by a deep neural network algorithm. The neural network consists of two hidden layers, each with 50 nodes, and the activation function is the tanh function. The best effect is achieved after 66 steps. The best verification performance is when the mean square error is hour, Figure 3 This is the mean square error curve during neural network training. The regression coefficients of the network training set, validation set, and test set are all around 0.96, indicating that the output value is highly correlated with the actual value. The deep neural network training algorithm with a trainable dictionary can be summarized as follows: Step 1: Status data collection; Step 2: Initialize the Koopman operator and the parameters of the fully connected neural network , input training data, set maximum training time, batch size, maximum training error ; Step 3: Fix the parameters ,optimization ; Step 4: Fix , optimization parameters ; Repeatedly update through optimization algorithm and parameters (i.e. loop steps 3~4); Step 5: Determine whether the maximum training time has been reached or the training error has been met : Yes, end training and save the network parameters of the deep neural network No, go to step 3.

[0026] Furthermore, in step S4, the modal parameters of the Koopman model are extracted by EDMD dynamic mode decomposition; After two continuous matrices are lifted to high dimensions, they are expressed as , , merge the matrices before dimensionality increase, and get the high-dimensional matrix represented as: , ; The finite-dimensional approximate estimation expression of the Koopman operator is:

[0027] In the formula is the generalized inverse, and The eigenvalue of Provides an approximation of the Koopman eigenvalue; Figure 4 The characteristic root distribution of the Koopman operator under the partial open circuit fault state and the attenuation rate and frequency of the mode are shown. The greater the energy, the darker the color of the corresponding eigenvalue. Koopman mode analysis not only provides the static structure of the data, but also captures the dynamic behavior of the system. In the given time series data, each mode obtained in the calculation is associated with a fixed oscillation frequency and attenuation / growth rate. Its modal analysis has stronger explanatory power and applicability in dynamic system modeling and analysis, especially for the complex dynamic behavior contained in time series data.

[0028] The wavelet time scattering convolutional network used in this example is See the scattering transformation diagram of Figure 5 ; Automatically obtain features through wavelet scattering transform, where the number of scattering paths is 7, and the scattering sequence is transposed and reshaped; Training of wavelet temporal scattering convolutional networks can be found in Figure 1 , including: the samples are divided into training set and test set in a ratio of 7:3; the classifier based on scattering convolutional network follows an end-to-end framework. In order to update and optimize the model parameters more effectively, the first-order moment estimation and the second-order moment estimation are used at the same time. The learning rate is adaptively adjusted according to the nature of the gradient. The learning rate is changed at specific intervals by using a piecewise learning rate schedule. The learning rate is reduced after every 5 training cycles to help optimize the weights more slowly in the later stages of training. The mini-batch size is set to 15, which is the number of samples used in each iteration of the gradient descent. The maximum training cycle is 110; In the wavelet time scattering convolutional network, each network group includes convolution, BN, activation, and pooling layers. Finally, the input vector is processed by the Softmax layer to classify the fault. The accuracy and loss change curves of the training set and the confusion matrix of the test set during the iteration process are shown in the figure below. Figure 6 and Figure 7 As shown; The DAB converter fault diagnosis model established based on the wavelet time scattering convolutional network in this embodiment can directly process multiple diagnostic signals, perform a series of convolution, modulus operation and downsampling operations through the wavelet scattering path to extract the denoised deep features, and input them into the scattering convolutional neural network to complete the fault diagnosis; The wavelet operator can be expressed as: ,in, is the maximum size, is the scaling function, Represents the scattering path; the wavelet mode consists of two parts: the first is the translation invariant , called the zero scattering order, and the other is the nonlinear translation covariate ; Repeat until When the order is:

[0029]

[0030]

[0031]

[0032] From the above iterative formula, it can be seen that The covariant part of the order Through a low pass filter The average of the translation invariants , after wavelet The convolution module of ; Assume the subscript sequence of wavelet is , represents a scattering path, Representative The scattering coefficient of the order also indicates the number of wavelet filters; the scattering expression is from the 0th order to the All scattering coefficients of order are composed, that is:

[0033] The obtained multi-scale information division sample set is input into the classifier. While realizing the automatic acquisition of high-order dynamic information, it can weaken the intra-class difference and adaptively reduce the influence of noise. A scattering convolutional network structure is constructed, and the trained fault diagnosis model is tested using the test set to realize the fault diagnosis of the dual active converter.

[0034] This embodiment proposes an online fault diagnosis method for dual active bridge converters with multi-source data fusion, using the Koopman operator and wavelet time scattering network to solve the problem of multiple signals in converter fault diagnosis. Compared with the existing data-driven methods, the proposed method has higher diagnostic accuracy, lower data requirements and more stable diagnostic performance. The Koopman operator can directly examine the evolution of related functions in the state space from observables and keep the related low-frequency spatiotemporal features unchanged, which simplifies the fault analysis of multiple signals to a certain extent. In addition, different orders of wavelet scattering correspond to the extraction of different order features of the signal, and can adaptively select appropriate filters to match the spectral characteristics of different signals, so as to more accurately identify each fault.

[0035] Embodiment 2: Based on the first embodiment, this embodiment provides a multi-source data fusion dual active bridge converter online fault diagnosis method system, including: A diagnostic signal selection module is used to analyze the relationship between different open-circuit fault states of the dual active bridge converter and circuit signal changes according to the established simulation model of the dual active bridge converter to be tested, and select a diagnostic signal; A multi-source data sequence set building module, used for collecting diagnostic signals of the dual active bridge converter under different open circuit fault states to build a multi-source data sequence set; A multi-scale reconstruction data matrix acquisition module is used to spatially reconstruct a multi-source data sequence set to obtain a multi-scale reconstruction data matrix; A high-dimensional matrix acquisition module is used to divide the multi-scale reconstruction data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, they are merged with the matrices before the upgrade to obtain two high-dimensional matrices; Koopman modal parameter extraction module, used to establish the Koopman model and extract the modal parameters of the Koopman model; The diagnosis result acquisition module is used to input the Koopman modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis result.

[0036] Furthermore, the diagnostic signal includes a leakage inductance current, a primary bridge arm midpoint voltage, and a secondary bridge arm midpoint voltage.

[0037] Furthermore, the multi-scale reconstruction data matrix acquisition module, the method for acquiring the multi-scale reconstruction data matrix includes: taking the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstruction data matrix, and taking the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstruction data matrix.

[0038] Furthermore, the high-dimensional matrix acquisition module promotes two continuous matrices to high dimensions through a deep neural network algorithm with a trainable dictionary.

[0039] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0040] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-source data fusion dual active bridge converter online fault diagnosis method, characterized in that: include: S1. Establish a simulation model of the dual active bridge converter to be tested, select diagnostic signals by analyzing the relationship between different open circuit fault states of the dual active bridge converter and circuit signal changes, collect diagnostic signals of the dual active bridge converter under different open circuit fault states to construct a multi-source data sequence set; S2, spatially reconstruct the multi-source data sequence set to obtain a multi-scale reconstructed data matrix; S3, dividing the multi-scale reconstructed data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, merging them with the matrices before the upgrade to obtain two high-dimensional matrices; S4, establishing a Koopman model according to the high-dimensional matrix, and extracting the modal parameters of the Koopman model; S5. Input the modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis results.

2. The multi-source data fusion dual active bridge converter online fault diagnosis method according to claim 1, characterized in that: The diagnostic signals include leakage inductance current, primary bridge arm midpoint voltage, and secondary bridge arm midpoint voltage.

3. The multi-source data fusion dual active bridge converter online fault diagnosis method according to claim 1, characterized in that: The step S2 includes: using the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstructed data matrix, and using the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstructed data matrix.

4. The multi-source data fusion dual active bridge converter online fault diagnosis method according to claim 1, characterized in that: In step S3, two continuous matrices are upgraded to high dimensions through a deep neural network algorithm with a trainable dictionary.

5. The multi-source data fusion dual active bridge converter online fault diagnosis method according to claim 1, characterized in that: In the step S4, the modal parameters of the Koopman model are extracted by EDMD dynamic mode decomposition.

6. The multi-source data fusion dual active bridge converter online fault diagnosis method according to claim 4 is characterized in that: The deep neural network algorithm of the trainable dictionary is trained through a fully connected neural network.

7. A multi-source data fusion dual active bridge converter online fault diagnosis method system, characterized in that: include: A diagnostic signal selection module is used to analyze the relationship between different open-circuit fault states of the dual active bridge converter and circuit signal changes according to the established simulation model of the dual active bridge converter to be tested, and select a diagnostic signal; A multi-source data sequence set building module, used for collecting diagnostic signals of the dual active bridge converter under different open circuit fault states to build a multi-source data sequence set; A multi-scale reconstruction data matrix acquisition module is used to spatially reconstruct a multi-source data sequence set to obtain a multi-scale reconstruction data matrix; A high-dimensional matrix acquisition module is used to divide the multi-scale reconstruction data matrix into two continuous matrices in time, and after the two continuous matrices are upgraded to high dimensions, they are merged with the matrices before the upgrade to obtain two high-dimensional matrices; Koopman modal parameter extraction module, used to establish the Koopman model and extract the modal parameters of the Koopman model; The diagnosis result acquisition module is used to input the modal parameters into the trained wavelet time scattering convolutional network to obtain the diagnosis result.

8. The multi-source data fusion dual active bridge converter online fault diagnosis method system according to claim 7, characterized in that: The diagnostic signals include leakage inductance current, primary bridge arm midpoint voltage, and secondary bridge arm midpoint voltage.

9. The multi-source data fusion dual active bridge converter online fault diagnosis method system according to claim 7, characterized in that: The multi-scale reconstruction data matrix acquisition module obtains the multi-scale reconstruction data matrix by using the number of data sequences from different sources in the diagnostic signal as the rows of the multi-scale reconstruction data matrix and using the length of the data sequence from a single source in the diagnostic signal as the columns of the multi-scale reconstruction data matrix.

10. The multi-source data fusion dual active bridge converter online fault diagnosis method system according to claim 7, characterized in that: The high-dimensional matrix acquisition module promotes two continuous matrices to high dimensions through a deep neural network algorithm with a trainable dictionary.

Images