Ship twelve-phase generator fault positioning method and system

Abstract

The invention relates to a ship twelve-phase generator fault positioning method and system, and the method comprises the steps: collecting main electric signal data of a twelve-phase generator and a DC power grid under healthy and various internal faults, including winding phase current, d-q axis current, DC side voltage and DC side current; adopting extended Park transform to process phase current time domain data, adopting discrete Fourier transform to process d-q axis current and direct current side voltage current, and extracting a time domain amplitude and a frequency domain frequency doubling component as fault features; dividing a training set and a test set, and performing normalization preprocessing on each dimension of the data; building a BP neural network, training and adjusting parameters; new operation data are collected regularly, fault features are extracted, and the fault features are input into a BP network to complete fault positioning. The method solves the problems of the limitation of the positioning precision of a ship generator fault positioning method and system and the positioning accuracy of manual threshold setting, improves the positioning accuracy of different internal faults, facilitates the real-time diagnosis of the internal faults of the generator, and reduces the loss.

Description

Technical Field

[0001] The invention relates to ship generator fault diagnosis and fault location technology, and in particular to a ship twelve-phase generator fault location method and system based on time-frequency domain characteristics and BP network. Background Art

[0002] The ship's integrated power system combines the traditional ship's mechanical propulsion system and power system into one, and provides power to the propulsion system, high-energy weapon system, communication, navigation and detection system, and daily equipment through the power network, realizing unified energy supply, distribution, use and management, and is a symbol of fully electrified ships. The ship's twelve-phase generator is a key component of its power supply system. Once an internal fault occurs, it is very easy to cause serious harm to the entire power supply system and the safety of life and property of operators. Without affecting production safety, timely troubleshooting and locating multiple types of internal faults occurring in different parts is an important basis for fault isolation and further fault reconstruction and fault recovery. Therefore, designing a fault location system suitable for internal faults of twelve-phase generators in ship power supply systems is crucial to reducing maintenance costs and ensuring personnel safety.

[0003] The fault location system and method include two stages: fault feature extraction and fault location. For fault feature extraction, traditional fault feature extraction includes phase current amplitude, dq axis current double frequency harmonics, DC grid voltage and current double frequency harmonics, etc., and mostly uses a single feature to judge faults. Its limitations make it difficult to meet the growing requirements for fault location accuracy under various fault modes. The use of multi-sensor and multi-signal source comprehensive criteria has become the current technological trend. However, for fault location, the selection of thresholds for each characteristic value when using multiple criteria greatly affects the positioning accuracy, and the method of setting thresholds using artificial experience cannot meet the accuracy requirements. Therefore, the use of data-driven methods represented by machine learning has become a reliable solution to this problem. Summary of the invention

[0004] Aiming at the limitations of ship generator fault location methods and system positioning accuracy, as well as the positioning accuracy problem of artificial threshold setting, a ship twelve-phase generator fault location method is proposed.

[0005] The technical solution of the present invention is:

[0006] A method for locating a fault of a twelve-phase generator of a ship, comprising:

[0007] S1. Collect the main electrical signal data of the twelve-phase generator and DC power grid under healthy and various internal faults from multiple sensors, including the phase current of each winding, dq axis current, DC side voltage and DC side current;

[0008] S2, using extended Park transform to process the phase current time domain data, using discrete Fourier transform to process the dq axis current and DC side voltage and current, extracting the time domain amplitude and frequency domain double frequency component as fault characteristics;

[0009] S3, divide the training set and test set, and perform normalization preprocessing on each dimension of the data;

[0010] S4. Build BP neural network, train and adjust parameters;

[0011] S5. Regularly collect new operating data and extract the aforementioned fault features, and input them into the BP network to complete fault location.

[0012] Preferably, the method specifically comprises the following steps:

[0013] Step S1 is specifically as follows:

[0014] S1-1. Determine the sampling frequency and sampling duration so that the length of the collected data can include at least three generator rotation cycles;

[0015] S1-2. Under the above parameters and the steady-state operation of the generator, collect the main electrical signal data of the 12-phase generator and the DC power grid under healthy and various internal faults, including 12 phase currents of 4 sets of windings, 4 d-axis currents and 4 q-axis currents, DC side voltage and DC side current;

[0016] Step S2 is specifically as follows:

[0017] S2-1, time domain features, obtain the amplitude of the collected phase current data, a total of 12 features;

[0018] S2-2, frequency domain characteristics, extended Park transformation of each phase current; transformation matrix C from three-phase stationary coordinate system to two-phase rotating coordinate system under equal amplitude transformation 3 / 2 for:

[0019]

[0020] The current relationship between the two coordinate systems can be expressed as:

[0021]

[0022] Among them, i α 、i β They represent the currents in the two-phase rotating coordinate system, i s represents the three-phase stationary coordinate system current matrix, i a 、i b 、i c They represent the a, b, and c phase currents in the three-phase stationary coordinate system respectively;

[0023] The essence of the extended Parker transform is to perform spectrum analysis on the Parker transform vector modulus. When the generator is working unbalanced, its vector modulus includes a DC component and a double frequency component of the fundamental frequency. The DC component is formed by the symmetrical components of the three-phase current, and the double frequency component is generated by the asymmetry of the fault. The extended Parker vector method uses the ratio of the double frequency component amplitude of the Parker transform vector modulus to the DC component amplitude as the fault severity of the generator stator winding short circuit, so that it can be quantified.

[0024]

[0025] Wherein, f' represents the fault characteristic frequency, f represents the electrical frequency, n represents the rotation speed, p represents the number of pole pairs, F represents the fault characteristic, and P() represents the amplitude at the corresponding frequency;

[0026] This step obtains a total of 4 features of 4 sets of windings;

[0027] S2-3, frequency domain characteristics, discrete Fourier transform of d-axis current, q-axis current, DC voltage and DC voltage; related theory proves that when an asymmetric fault occurs, a second harmonic is generated, and the component at the double frequency of the corresponding frequency domain increases; this step obtains 4 d-axis currents, 4 q-axis currents and 10 fault characteristics of DC voltage and current of 4 sets of windings;

[0028] Step S3 is specifically as follows:

[0029] S3-1. According to the above fault feature analysis results, the 26 features of each sample are combined into a vector; in order to optimize the training effect of the BP neural network, the dimensions of different vectors are normalized according to the following formula:

[0030]

[0031] Among them, x' j 、x j 、x min 、x max They represent the normalized value of the j-th sample of a dimension x in the feature vector, the original value of the j-th sample, the minimum value of all samples, and the maximum value of all samples respectively;

[0032] In order to facilitate the output of the BP neural network, the fault type is encoded in the form of a vector; the vector dimension is the number of faults to be classified, the corresponding bit of each type of fault is 1, and the remaining elements of the vector are 0;

[0033] S3-2, divide the data samples in proportion to obtain training set and test set;

[0034] Step S4 is specifically as follows:

[0035] S4-1. Determine the BP neural network structure; adopt the "input layer - hidden layer - output layer" three-layer structure that meets the fault location requirements;

[0036] S4-2, determine the BP neural network parameters; the output is the number of fault types, the input is the number of fault features of the sample; use the empirical formula to determine the number of hidden layers:

[0037]

[0038] Where h represents the number of neurons in the hidden layer, p 输入 represents the number of neurons in the input layer, q 输出 represents the number of neurons in the output layer, t is the adjustment value, t∈[1,10];

[0039] S4-3. Determine the activation function and loss function; the activation function uses the Sigmoid function, and the loss function uses the cross entropy loss function;

[0040] S4-4, input the training set samples into the network for training, output the loss as iterative convergence, test the training effect through the test set, and adjust the neural network parameters as appropriate;

[0041] Step S5 is specifically as follows:

[0042] New operating data is collected regularly and the aforementioned fault features are extracted, and input into the BP network to complete fault location.

[0043] The present invention also provides a ship twelve-phase generator fault location system based on time-frequency domain fault characteristics and BP neural network, comprising:

[0044] The first processing module is used to collect the main electrical signal data of the twelve-phase generator and the DC power grid under the steady-state operation of the generator and under various internal faults;

[0045] The second processing module is used to extract fault features from the collected data, and extract the time domain amplitude and the frequency domain double frequency component as fault features;

[0046] The third processing module is used to normalize the dimensions of the data and generate training sets and test sets;

[0047] The fourth processing module is used to build a BP neural network and perform network training, output the changes of training loss during the iteration process, and adjust the network parameters;

[0048] The fifth processing module is used to collect data in real time, extract features, and input them into the BP network to obtain fault location results.

[0049] The beneficial effects of the present invention are:

[0050] (1) Extract fault features from multi-source signals of multiple sensors. Fault features cover both time domain and frequency domain, avoiding the accidental influence of fault features extracted from a single signal or a single domain on the fault results;

[0051] (2) The BP neural network algorithm is used to drive the fault location data. Compared with the traditional location method based on manual experience, it improves the location accuracy of different internal faults while ensuring engineering practicality;

[0052] (3) Form a complete fault location system to facilitate real-time diagnosis of internal generator faults and reduce fault losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method and system for locating a fault of a twelve-phase generator of a ship according to the present invention;

[0054] Figure 2 It is a system processing flow chart of the method and system for locating faults of twelve-phase generators of a ship according to the present invention;

[0055] Figure 3 It is a loss convergence diagram of an embodiment of a method and system for locating a fault of a twelve-phase generator of a ship according to the present invention;

[0056] Figure 4 This is a fault location confusion matrix diagram of an embodiment of the ship twelve-phase generator fault location method and system of the present invention. DETAILED DESCRIPTION

[0057] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0058] like Figure 1 As shown, this embodiment provides a method and system for locating faults of a twelve-phase generator of a ship based on time-frequency domain fault characteristics and BP neural network. The execution steps of the present invention are specifically described below:

[0059] S1-1. Determine the sampling frequency to be 100kHz and the number of single sampling points to be 50k.

[0060] S1-2. Under the above parameters, a simulation program is used to obtain 200 data of the main electrical signals of the twelve-phase generator and DC power grid under 6 types of faults, namely "healthy, stator inter-turn short circuit fault, stator same-bridge phase-to-phase short circuit fault, stator different-bridge phase-to-phase short circuit fault, stator single-phase grounding fault, and rotor inter-turn short circuit fault" under steady-state operation of the generator, including 12 phase currents of 4 sets of windings, 4 d-axis currents and 4 q-axis currents, DC side voltage and DC side current.

[0061] S2-1. Time domain features: obtain the amplitude of the collected phase current data, with a total of 12 features.

[0062] S2-2, frequency domain characteristics, perform extended Park transformation on each phase current to obtain a total of 4 characteristics of the 4 sets of windings.

[0063] S2-3, frequency domain characteristics, discrete Fourier transform is performed on the d-axis current, q-axis current, DC voltage and DC voltage to obtain a total of 10 fault characteristics of 4 d-axis currents, 4 q-axis currents and DC voltage and current of 4 sets of windings.

[0064] S3-1. The 26 features of each sample are combined into a vector, and the dimensions of different vectors are normalized; the fault type is encoded in vector form.

[0065] S3-2. Divide the data samples in proportion to obtain the training set and the test set.

[0066] The final division results are shown in the following table.

[0067] Table 1. Number of samples

[0068]

[0069] S4-1. Determine that the BP neural network adopts a three-layer structure.

[0070] S4-2. Determine the parameters of the BP neural network: the number of neurons in the input layer is 26, the number of neurons in the output layer is 6, and the number of neurons in the hidden layer is 10.

[0071] S4-3. Determine the activation function and loss function. The activation function uses the Sigmoid function, and the loss function uses the cross entropy loss function.

[0072] S4-4, input the training set samples into the network for training, and the output loss converges with iteration as follows: Figure 3 , and test the training effect through the test set and adjust the neural network parameters.

[0073] Under different fault types, 30 samples of new running data are simulated and collected, fault features are extracted, input into BP network, and the confusion matrix is ​​used to output the fault location effect. Figure 4 , with an accuracy rate of 93.3%.

[0074] This embodiment adopts a ship twelve-phase generator fault location system based on time-frequency domain fault characteristics and BP neural network, such as Figure 2 As shown, including:

[0075] The first processing module is used to collect the main electrical signal data of the twelve-phase generator and the DC power grid under the steady-state operation of the generator and under various internal faults;

[0076] The second processing module is used to extract fault features from the collected data, and extract the time domain amplitude and the frequency domain double frequency component as fault features;

[0077] The third processing module is used to normalize the dimensions of the data and generate training sets and test sets;

[0078] The fourth processing module is used to build a BP neural network and perform network training, output the changes of training loss during the iteration process, and adjust the network parameters;

[0079] The fifth processing module is used to collect data in real time, extract features, and input them into the BP network to obtain fault location results.

[0080] The above-mentioned embodiment only expresses one implementation mode of the present invention, and its description is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A method for locating a fault of a twelve-phase generator of a ship, characterized in that: The following steps are involved: S1. Collect the main electrical signal data of the twelve-phase generator and DC power grid under healthy and various internal faults from multiple sensors, including the phase current of each winding, dq axis current, DC side voltage and DC side current; S2, using extended Park transform to process the phase current time domain data, using discrete Fourier transform to process the dq axis current and DC side voltage and current, extracting the time domain amplitude and frequency domain double frequency component as fault characteristics; S3, divide the training set and test set, and perform normalization preprocessing on each dimension of the data; S4. Build BP neural network, train and adjust parameters; S5. Regularly collect new operating data and extract the aforementioned fault features, and input them into the BP network to complete fault location.

2. The method for locating faults of a twelve-phase generator of a ship according to claim 1, characterized in that: The specific steps include: Step S1 is specifically as follows: S1-1. Determine the sampling frequency and sampling duration so that the length of the collected data can include at least three generator rotation cycles; S1-2. Under the above parameters and the steady-state operation of the generator, collect the main electrical signal data of the 12-phase generator and the DC power grid under healthy and various internal faults, including 12 phase currents of 4 sets of windings, 4 d-axis currents and 4 q-axis currents, DC side voltage and DC side current; Step S2 is specifically as follows: S2-1, time domain features, obtain the amplitude of the collected phase current data, a total of 12 features; S2-2, frequency domain characteristics, extended Park transformation of each phase current; transformation matrix C from three-phase stationary coordinate system to two-phase rotating coordinate system under equal amplitude transformation 3 / 2 for: The current relationship between the two coordinate systems can be expressed as: Among them, i α 、i β They represent the currents in the two-phase rotating coordinate system, i s represents the three-phase stationary coordinate system current matrix, i a 、i b 、i c They represent the a, b, and c phase currents in the three-phase stationary coordinate system respectively; The essence of the extended Parker transform is to perform spectrum analysis on the Parker transform vector modulus. When the generator is working unbalanced, its vector modulus includes a DC component and a double frequency component of the fundamental frequency. The DC component is formed by the symmetrical components of the three-phase current, and the double frequency component is generated by the asymmetry of the fault. The extended Parker vector method uses the ratio of the double frequency component amplitude of the Parker transform vector modulus to the DC component amplitude as the fault severity of the generator stator winding short circuit, so that it can be quantified. Wherein, f' represents the fault characteristic frequency, f represents the electrical frequency, n represents the rotation speed, p represents the number of pole pairs, F represents the fault characteristic, and P() represents the amplitude at the corresponding frequency; This step obtains a total of 4 features of 4 sets of windings; S2-3, frequency domain characteristics, discrete Fourier transform of d-axis current, q-axis current, DC voltage and DC voltage; related theory proves that when an asymmetric fault occurs, a second harmonic is generated, and the component at the double frequency of the corresponding frequency domain increases; this step obtains 4 d-axis currents, 4 q-axis currents and 10 fault characteristics of DC voltage and current of 4 sets of windings; Step S3 is specifically as follows: S3-1. According to the above fault feature analysis results, the 26 features of each sample are combined into a vector; in order to optimize the training effect of the BP neural network, the dimensions of different vectors are normalized according to the following formula: Among them, x' j 、x j 、x min 、x max They represent the normalized value of the j-th sample of a dimension x in the feature vector, the original value of the j-th sample, the minimum value of all samples, and the maximum value of all samples respectively; In order to facilitate the output of the BP neural network, the fault type is encoded in the form of a vector; the vector dimension is the number of faults to be classified, the corresponding bit of each type of fault is 1, and the remaining elements of the vector are 0; S3-2, divide the data samples in proportion to obtain training set and test set; Step S4 is specifically as follows: S4-1. Determine the BP neural network structure; adopt the "input layer-hidden layer-output layer" three-layer structure that meets the fault location requirements; S4-2, determine the BP neural network parameters; the output is the number of fault types, the input is the number of fault features of the sample; use the empirical formula to determine the number of hidden layers: Where h represents the number of neurons in the hidden layer, p 输入 represents the number of neurons in the input layer, q 输出 represents the number of neurons in the output layer, t is the adjustment value, t∈[1,10]; S4-3. Determine the activation function and loss function; the activation function uses the Sigmoid function, and the loss function uses the cross entropy loss function; S4-4, input the training set samples into the network for training, output the loss as iterative convergence, test the training effect through the test set, and adjust the neural network parameters as appropriate; Step S5 is specifically as follows: New operating data is collected regularly and the aforementioned fault features are extracted, and input into the BP network to complete fault location.

3. A ship twelve-phase generator fault location system, characterized in that: A method for locating a fault of a twelve-phase generator of a ship as claimed in any one of claims 1 or 2, comprising: The first processing module is used to collect the main electrical signal data of the twelve-phase generator and the DC power grid under the steady-state operation of the generator and under various internal faults; The second processing module is used to extract fault features from the collected data, and extract the time domain amplitude and the frequency domain double frequency component as fault features; The third processing module is used to normalize the dimensions of the data and generate training sets and test sets; The fourth processing module is used to build a BP neural network and perform network training, output the changes of training loss during the iteration process, and adjust the network parameters; The fifth processing module is used to collect data in real time, extract features, and input them into the BP network to obtain fault location results.

Images