A generator rotor inter-turn short circuit on-line monitoring system and method based on a detection coil method

Abstract

This invention belongs to the field of generator fault monitoring technology, and provides an online monitoring system and method for inter-turn short circuits in generator rotors based on the detection coil method. It employs a flexible substrate and a planar helical detection coil integrated on it as a flexible magnetic flux sensor, completely eliminating the safety hazards of traditional rigid magnetic flux sensors. The flexible magnetic flux sensor requires no rotor removal or drilling; it is fixed to the generator slot wedge or stator teeth using a high-temperature adhesive, significantly reducing the installation time per unit. This invention effectively filters out main magnetic flux interference and electromagnetic noise through an intelligent algorithm combining improved wavelet threshold function denoising and a BP neural network, featuring high safety, convenient installation, and high detection accuracy.

Description

Technical Field

[0001] This invention belongs to the field of generator fault monitoring technology, specifically relating to an online monitoring system and method for inter-turn short circuits in generator rotors based on the detection coil method, which is particularly suitable for online identification, location and early warning of inter-turn short circuit faults in rotor windings of salient-pole synchronous generators. Background Technology

[0002] Inter-turn short circuits in generator rotor windings are one of the most common electrical faults in generator sets. When an inter-turn short circuit occurs in the rotor winding, it leads to an increase in excitation current and intensified unit vibration. In severe cases, it may cause rotor grounding or even burnout. Currently, power plants mostly use offline detection methods (such as AC impedance testing and RSO testing). However, the intervals between generator shutdowns are generally long, making it difficult to guarantee whether an inter-turn short circuit fault will occur during this period. Therefore, online monitoring is necessary.

[0003] However, existing online monitoring methods have the following drawbacks: magnetic flux probes are mostly made of epoxy resin, which are large and heavy, requiring invasive operations during installation, and some even require damage to the generator casing. They are also difficult to disassemble, and once they fall off, they can easily cause mechanical damage to the high-speed generator body. Existing signal processing methods mostly use simple amplitude comparison methods, which are easily affected by main magnetic flux interference, electromagnetic noise and excitation current fluctuations, resulting in low detection accuracy and high false alarm rate. Summary of the Invention

[0004] The present invention aims to overcome the shortcomings of the prior art and provide a generator rotor inter-turn short circuit online monitoring system and method based on the detection coil method, so as to achieve high-precision, interference-resistant rotor inter-turn short circuit online monitoring and fault location.

[0005] This invention is achieved through the following technical solution: In a first aspect, the present invention provides an online monitoring system for inter-turn short circuits in a generator rotor based on the detection coil method, comprising: A flexible flux sensor, comprising a flexible substrate and a planar helical detection coil integrated on the flexible substrate, is attached to the stator side of a generator by an adhesive and is used to collect the differential flux signal of the rotor air gap. The signal conditioning and acquisition unit is used to amplify, filter, and perform analog-to-digital conversion on the magnetic flux differential signal, and output the original voltage signal sequence. The data processing and diagnosis unit is used to denoise the original voltage signal sequence using an improved wavelet threshold function, perform spectral analysis on the denoised signal, and extract feature vectors. The feature vectors are then input into a pre-trained BP neural network model, which outputs the probability of each inter-turn short circuit state category. Based on the probability of each inter-turn short circuit state category, the unit diagnoses whether an inter-turn short circuit fault exists. If the diagnosis result indicates that an inter-turn short circuit fault exists, the unit locates the stator slot with the inter-turn short circuit fault and assesses the degree of fault. The human-machine interaction unit is used to display the output results of the data processing and diagnostic unit and issue early warning information according to the degree of fault.

[0006] Preferably, the flexible substrate and the planar helical detection coil are encapsulated with a protective layer.

[0007] Furthermore, the flexible magnetic flux sensor integrates a temperature sensor, which is used to monitor the ambient temperature in real time and feed it back to the signal conditioning and acquisition unit; the signal conditioning and acquisition unit amplifies, filters, thermally compensates for and converts the magnetic flux differential signal according to the ambient temperature, and outputs the original voltage signal sequence.

[0008] Secondly, the present invention provides an online monitoring method for inter-turn short circuits of a generator rotor based on the probe coil method, comprising, based on the generator rotor inter-turn short circuit online monitoring system based on the probe coil method as described above: A flexible magnetic flux sensor is installed on the stator side of the generator to collect the differential magnetic flux signal in the rotor air gap; The signal conditioning and acquisition unit amplifies, filters, and performs analog-to-digital conversion on the magnetic flux differential signal to output the original voltage signal sequence. The data processing and diagnosis unit uses an improved wavelet threshold function to denoise the original voltage signal sequence. The denoised signal is then subjected to spectral analysis to extract feature vectors. The feature vectors are input into a pre-trained BP neural network model to output the inter-turn short circuit state category probability. Based on the inter-turn short circuit state category probability, the existence of an inter-turn short circuit fault is diagnosed. If the diagnosis result indicates the existence of an inter-turn short circuit fault, the stator slot with the inter-turn short circuit fault is located and the degree of fault is assessed. The human-computer interaction unit displays the results output by the data processing and diagnostic unit and issues early warning information based on the degree of fault.

[0009] Preferably, the step of using an improved wavelet threshold function to denoise the original voltage signal sequence specifically involves: For the original voltage signal sequence f[n] Perform J-level discrete wavelet decomposition to obtain the wavelet coefficients of each level:

[0010] in, This represents the discrete form of the wavelet basis functions. j For scale parameters, k These are translation parameters; An improved wavelet threshold function is used to quantize the wavelet coefficients at each level:

[0011] Among them, threshold Tj Determined by the following formula:

[0012] in, N j For the first j The number of layer wavelet coefficients; The denoised signal is obtained through wavelet inverse reconstruction: .

[0013] Furthermore, the step of performing spectral analysis on the denoised signal and extracting feature vectors specifically involves: For the denoised signal f^ [n] Perform a Fast Fourier Transform:

[0014] Extracting the fundamental amplitude and the amplitude of each harmonic :

[0015] Constructing feature vectors:

[0016] Where M is the harmonic order, f 0 represents the frequency of the fundamental wave.

[0017] Furthermore, the BP neural network model employs a three-layer BP neural network, with the number of nodes in the input layer... I=M+1 Hidden layer node count H, output layer node count O， Each node in the output layer corresponds to a short-circuit state category between turns. Hidden layer output:

[0018] Output layer output:

[0019] Where φ(.) represents the hidden layer activation function (taking the tanh function). ; ψ(.) represents the output layer activation function (using the softmax function). ; w ip Indicates the input layer's first... i The first neuron to the hidden layer p The connection weights of each neuron, where i =1,2,…, I ( I(Input feature dimension) p =1,2,…, H ( H (Number of hidden layer nodes); b p Indicates the hidden layer number 1 p The bias term for each neuron is a scalar used to adjust the activation threshold of the neuron. w pq : indicates the hidden layer p The nth neuron to the output layer q The connection weights of each neuron, where q =1,2,…, O ( O To output the number of categories, here O =3); b q : indicates the output layer q Bias terms for each neuron; The output of the BP neural network model represents the probability of each inter-turn short circuit state category, and the category corresponding to the highest probability is taken as the diagnosis result.

[0020] Furthermore, if the diagnosis result indicates the presence of an inter-turn short-circuit fault, the stator slot containing the inter-turn short-circuit fault is located, specifically as follows: If the diagnosis indicates an inter-turn short circuit fault, extract the denoised signal. f^ [ n The waveform is used to identify the peak value of the induced electromotive force corresponding to each stator slot; let the peak value of the induced electromotive force corresponding to the s-th stator slot be... P s The corresponding stator slots of the same pole are s′ Calculate the relative rate of change:

[0021] like Δ s >δ If so, it is assumed that an inter-turn short circuit has occurred in the s-th stator slot. δ This is the threshold for the rate of change.

[0022] Furthermore, the assessment of the degree of failure specifically includes: The degree of failure is determined by the output of the BP neural network model and Δ s Overall decision:

[0023] Where α represents the weights output by the BP neural network model, and β represents... Δ sThe weight, This represents the probability of an inter-turn short-circuit fault in the rotor output of the BP neural network model.

[0024] Furthermore, the issuance of early warning information based on the severity of the fault specifically includes: Level 1 warning: 0.2 ≤ Severity < 0.4, the system records and issues a notification; Level 2 warning: 0.4 ≤ Severity < 0.7, triggering an audible and visual alarm, and recommending a shutdown for inspection; Level 3 warning: Severity ≥ 0.7, upload to the central control center, and recommend emergency shutdown.

[0025] Compared with the prior art, the present invention has the following beneficial effects: This invention relates to an online monitoring system for inter-turn short circuits in generator rotors based on the probe coil method. It employs a flexible substrate and a planar helical probe coil integrated thereon as a flexible flux sensor. This flexible flux sensor is lightweight, thin, and soft; even if it detaches, it will not cause serious damage to the high-speed rotating generator rotor, completely eliminating the safety hazards of traditional rigid flux sensors. The flexible flux sensor requires no rotor removal or drilling; it is fixed to the generator slot wedge or stator teeth using a high-temperature adhesive. The installation time for a single unit can be reduced from 2-3 days to less than 4 hours. This invention utilizes an intelligent algorithm combining improved wavelet threshold function denoising and a BP neural network to effectively filter out main flux interference and electromagnetic noise, maintaining a fault identification rate of over 95% even in environments with a signal-to-noise ratio below 20dB. Therefore, this system features high safety, convenient installation, and high detection accuracy. Furthermore, this system is compatible with both the new flexible flux sensor and the power plant's existing older flux sensors, offering strong compatibility, protecting the power plant's existing investment, and facilitating technology promotion.

[0026] Furthermore, the flexible flux sensor integrates a temperature sensor to monitor the ambient temperature in real time and feed it back to the signal conditioning and acquisition unit. The signal conditioning and acquisition unit then performs thermal drift compensation on the signal acquired by the flexible flux sensor, thereby eliminating the impact of ambient temperature changes on the measurement accuracy of the flexible flux sensor and ensuring the reliability of fault diagnosis.

[0027] This invention presents an online monitoring method for inter-turn short circuits in generator rotors based on the probe coil method. It employs a flexible substrate and a planar helical probe coil integrated thereon as a flexible flux sensor. Even if the coil detaches, it will not cause serious damage to the high-speed rotating generator rotor, completely eliminating the safety hazards of traditional rigid flux sensors. Furthermore, the flexible flux sensor is fixed to the generator stator side using a high-temperature adhesive, eliminating the need for rotor removal or drilling, making installation convenient. This invention improves the wavelet threshold function to ensure continuity and no constant deviation at the threshold, effectively preserving signal abrupt changes and effectively filtering out main flux interference and electromagnetic noise, thus improving detection accuracy. Simultaneously, fault diagnosis using a BP neural network model can accurately capture the complex correspondence between the extracted feature vector after denoising and the inter-turn short circuit state categories of the rotor. This effectively solves the problems of complex inter-turn short circuit fault characteristics and the difficulty in diagnosis due to main flux interference and residual electromagnetic noise, significantly improving the accuracy and reliability of fault diagnosis. This model can directly output the probability of each inter-turn short circuit state category, clearly determining whether an inter-turn short circuit fault exists, and also provides accurate data support for fault severity assessment and stator slot fault location.

[0028] Furthermore, when the diagnosis indicates the presence of an inter-turn short-circuit fault, the output of the BP neural network model is compared with... Δ s By comprehensively assessing the severity of a fault, the BP neural network model can output the probability of short-circuit state categories between each turn, accurately reflecting the overall characteristics and approximate level of the fault. Meanwhile, Δs can directly quantify the differences in the induced electromotive force of the stator slots, intuitively reflecting the degree of impact of the fault on the electromagnetic induction on the stator side. The two complement each other and work synergistically, effectively making up for the inability to accurately quantify the severity of the fault when relying solely on the output of the BP neural network model. At the same time, it avoids the shortcomings of relying solely on Δs, which is susceptible to external electromagnetic interference and cannot be combined with the overall characteristics of the fault. It achieves accurate quantification and comprehensive assessment of the fault severity. It can grasp the overall attributes of the fault through the BP neural network model and accurately define the severity of the fault with the quantitative characteristics of Δs, ensuring the accuracy and objectivity of the fault severity assessment and providing a scientific and reliable basis for subsequent fault handling. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a block diagram of the overall structure of the system of the present invention.

[0031] Figure 2 The following is a schematic diagram of the flexible magnetic flux sensor of the present invention: (a) cross-sectional view; (b) front view.

[0032] Figure 3 This is a flowchart of the fault diagnosis process based on the BP neural network of this invention.

[0033] Figure 4 The following are comparison diagrams showing the effects of the improved wavelet threshold function denoising in this invention: (a) Noisy signal; (b) Traditional wavelet denoising result; (c) Improved wavelet threshold function denoising result.

[0034] Figure 5 Typical waveforms obtained by testing the method of the present invention on a fault simulation device (comparison of normal operating conditions and fault operating conditions); (a) normal operating conditions; (b) fault operating conditions. Detailed Implementation

[0035] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0036] It should be noted that the process equipment or apparatus not specifically mentioned in the following embodiments are all conventional equipment or apparatus in the art.

[0037] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Furthermore, unless otherwise stated, the numbering of each method step is merely a convenient tool for identifying each method step, and not intended to limit the order of the method steps or define the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0038] like Figure 1 As shown, the online monitoring system for inter-turn short circuits of a generator rotor based on the detection coil method of the present invention includes: A flexible flux sensor, comprising a flexible substrate and a planar helical detection coil integrated on the flexible substrate, is attached to the stator side of a generator by an adhesive and is used to collect the differential flux signal of the rotor air gap. The signal conditioning and acquisition unit is used to amplify, filter, and perform analog-to-digital conversion on the magnetic flux differential signal, and output the original voltage signal sequence. The data processing and diagnosis unit is used to denoise the original voltage signal sequence using an improved wavelet threshold function, perform spectral analysis on the denoised signal, and extract feature vectors. The feature vectors are then input into a pre-trained BP neural network model, which outputs the probability of each inter-turn short circuit state category. Based on the probability of each inter-turn short circuit state category, the unit diagnoses whether an inter-turn short circuit fault exists. If the diagnosis result indicates that an inter-turn short circuit fault exists, the unit locates the stator slot with the inter-turn short circuit fault and assesses the degree of fault. The human-machine interaction unit is used to display the output results of the data processing and diagnostic unit and issue early warning information according to the degree of fault.

[0039] This invention relates to an online monitoring system for inter-turn short circuits in generator rotors based on the probe coil method. It employs a flexible substrate and a planar helical probe coil integrated thereon as a flexible flux sensor. This flexible flux sensor is lightweight, thin, and soft; even if it detaches, it will not cause serious damage to the high-speed rotating generator rotor, completely eliminating the safety hazards of traditional rigid flux sensors. The flexible flux sensor requires no rotor removal or drilling; it is fixed to the generator slot wedge or stator teeth using a high-temperature adhesive. The installation time for a single unit can be reduced from 2-3 days to less than 4 hours. This invention utilizes an intelligent algorithm combining improved wavelet threshold function denoising and a BP neural network to effectively filter out main flux interference and electromagnetic noise, maintaining a fault identification rate of over 95% even in environments with a signal-to-noise ratio below 20dB. Therefore, this system features high safety, convenient installation, and high detection accuracy.

[0040] The flexible magnetic flux sensor structure of this invention is as follows: Figure 2 As shown, the planar spiral detection coil is made of copper foil through laser processing, with a thickness of 0.1-0.5 mm. The flexible substrate has a thickness of 0.3 mm-0.5 mm and is made of polyimide material. Both the flexible substrate and the planar spiral detection coil are encapsulated with a protective layer. In use, the flexible flux sensor is attached to the surface of the stator teeth or slot wedges using a high-temperature adhesive, eliminating the need for drilling or invasive installation.

[0041] In some embodiments of the present invention, the flexible magnetic flux sensor integrates a temperature sensor for real-time monitoring of ambient temperature and feedback to the signal conditioning and acquisition unit. Based on the real-time temperature value, the signal conditioning and acquisition unit calculates the thermal drift of the flexible magnetic flux sensor output caused by the current temperature through an algorithm, and then corrects the synchronously acquired magnetic flux differential signal in real time to achieve thermal drift compensation, thereby eliminating the influence of ambient temperature changes on the measurement accuracy of the flexible magnetic flux sensor and ensuring the reliability of fault diagnosis.

[0042] In some embodiments of the present invention, the signal conditioning and acquisition unit includes a signal amplification circuit, an anti-aliasing filter circuit, a multi-channel synchronous data acquisition card, and an FPGA controller connected in sequence. The signal amplification circuit amplifies the magnetic flux differential signal; the anti-aliasing filter circuit filters the amplified signal, removing high-frequency noise and interference, preventing frequency aliasing during signal sampling, and ensuring that the signal entering the multi-channel synchronous data acquisition card is clean, effective, and alias-free. The multi-channel synchronous data acquisition card converts the filtered analog signal into a raw voltage signal sequence and realizes synchronous acquisition of multiple flexible magnetic flux sensor signals; the FPGA controller controls the multi-channel synchronous data acquisition card to perform multi-channel synchronous data acquisition and real-time preprocessing of the raw voltage signal sequence.

[0043] When a temperature sensor is present, the temperature signal collected by the temperature sensor and the magnetic flux differential signal enter the signal conditioning and acquisition unit together. Finally, the multi-channel synchronous data acquisition card completes the analog-to-digital conversion, converting the temperature signal into a digital temperature value. The FPGA controller calculates the thermal drift error of the flexible magnetic flux sensor at the current temperature according to the preset temperature-drift mathematical model / calibration curve. The thermal drift error is used to digitally correct the magnetic flux differential signal collected at the same time to achieve temperature compensation. The temperature-compensated signal is then sent to the data processing and diagnostic unit to ensure that misjudgment is not caused by temperature drift.

[0044] In some embodiments, the sampling frequency of the multi-channel synchronous data acquisition card is not less than 200kHz, the resolution is not less than 12 bits, and the input signal range is 0-5V or 4-20mA.

[0045] In some embodiments of the present invention, the data processing and diagnostic unit includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following functions: 1) Signal denoising and feature vector construction based on improved wavelet threshold function; 2) Diagnosis of inter-turn short-circuit faults; 3) Fault stator slot location based on characteristic waveforms; 4) Short circuit severity assessment and early warning.

[0046] In some embodiments of the present invention, the human-computer interaction unit includes a touch screen, an Ethernet communication interface, and a USB interface, used to display voltage waveforms, fault diagnosis results, and historical data queries collected by the flexible magnetic flux sensor in real time.

[0047] In some embodiments of the present invention, the data processing and diagnostic unit is connected to the central control center via Ethernet to transmit diagnostic results to the central control center for unified management and monitoring.

[0048] Based on the system described above, this invention provides an online monitoring method for inter-turn short circuits in generator rotors based on the detection coil method, such as... Figure 3 As shown, it includes the following steps: Step S1: Install the flexible magnetic flux sensor on the stator side of the generator to collect the differential magnetic flux signal of the rotor air gap; Step S2: The magnetic flux differential signal is amplified, filtered, and converted from analog to digital using the signal conditioning and acquisition unit to output the original voltage signal sequence; Step S3: The data processing and diagnosis unit uses an improved wavelet threshold function to denoise the original voltage signal sequence, performs spectral analysis on the denoised signal, and extracts feature vectors; inputs the feature vectors into a pre-trained BP neural network model, outputs the inter-turn short circuit state category probability, and diagnoses whether an inter-turn short circuit fault exists based on the inter-turn short circuit state category probability. Step S4: If the diagnosis result indicates that there is an inter-turn short circuit fault, locate the stator slot with the inter-turn short circuit fault and assess the degree of fault. Step S5: Use the human-machine interaction unit to display the results output by the data processing and diagnosis unit and issue warning information according to the degree of fault.

[0049] In some embodiments of the present invention, the online monitoring method for inter-turn short circuits of generator rotor based on the detection coil method specifically includes the following implementation process: Step S1: Signal Acquisition A flexible flux sensor is installed on the stator side of the generator. The differential flux signal in the generator's air gap is acquired in real time by the flexible flux sensor. The differential flux signal is then amplified, filtered, and converted from analog to digital to obtain the original voltage signal sequence. f(t) .

[0050] Step S2: Adaptive wavelet denoising For the original voltage signal sequence f(t) Multi-level discrete wavelet decomposition is performed, and an improved wavelet threshold function is used to quantize the high-frequency wavelet coefficients of each level to remove electromagnetic noise and excitation ripple interference, thus reconstructing the denoised signal. f^ (t) .

[0051] The improved wavelet threshold function is:

[0052] Among them, threshold T j Determined by the following formula:

[0053] in, For the first j Layer wavelet coefficients,N j For the first j The number of layer wavelet coefficients, j For scale parameters, k The shift parameter is used. This wavelet threshold function is continuous and has no constant deviation at the threshold, effectively preserving the signal's abrupt change characteristics, effectively filtering out main magnetic flux interference and electromagnetic noise, and improving detection accuracy.

[0054] Step S3: Extraction of fundamental and harmonic components The denoised signal is synchronously detected to extract the fundamental component and harmonic components related to the rotor speed, and a feature vector is constructed. F =[ A 0, A 1,..., A n ],in, A i For the first i The amplitude of the subharmonic.

[0055] Step S4: Fault Feature Identification eigenvectors F Input a pre-trained BP neural network model and output the inter-turn short circuit state category probability. Based on the inter-turn short circuit state category probability, diagnose whether an inter-turn short circuit fault exists.

[0056] The training samples for the BP neural network model are derived from historical fault data from the field and data generated by the simulation model.

[0057] Step S5: Locating the faulty stator slot If the diagnosis indicates an inter-turn short circuit fault, then extract the denoised signal. f^(t) The waveform is used to identify the peak value of the induced potential corresponding to each stator slot. The difference between the peak values ​​of the induced potential of adjacent stator slots with the same pole is calculated. If the difference between the peak value of the induced potential of a certain stator slot and the peak value of the induced potential of its corresponding stator slot with the opposite pole exceeds a preset rate of change threshold, then the stator slot is considered to have an inter-turn short circuit.

[0058] Step S6: Fault severity assessment and early warning The severity of the fault is comprehensively assessed based on the percentage decrease in the peak value of the induced electromotive force in the faulty stator slot, combined with the inter-turn short-circuit probability output by the BP neural network model. If the assessment result exceeds the first-level warning threshold, an audible and visual alarm is issued through the human-machine interface unit and uploaded to the central control center via the Ethernet interface.

[0059] Example A planar spiral detection coil was fabricated by laser marking using a 0.2mm thick copper foil attached to a 0.3mm thick polyimide film. The coil has 50 turns, a wire width of 0.3mm, and a wire spacing of 0.1mm. A high-temperature (180℃) resistant polyimide protective layer was applied to the outside of the coil, resulting in a flexible flux sensor. The fabricated flexible flux sensor was attached to a stator slot wedge using a high-temperature adhesive and led out to a junction box via a high-temperature shielded cable, connecting to a signal conditioning and acquisition unit. A temperature sensor was also connected to the tail of the flexible flux sensor to monitor its temperature drift.

[0060] Online monitoring was performed on a certain analog power unit using the system of this invention. The sampling frequency was 200kHz. The acquired raw voltage signal sequence was first subjected to wavelet denoising. The results show that after denoising using the improved wavelet threshold function of this invention, the signal-to-noise ratio was improved by about 8dB compared with the traditional soft thresholding method, and the spike abrupt changes in the waveform were completely preserved. Figure 4 ).

[0061] The temperature compensation process used in this embodiment is implemented in the following way: Step 1: Temperature Calibration and Model Establishment Temperature characteristic calibration experiments are conducted on the flexible magnetic flux sensor before it leaves the factory or during the initial installation phase. The flexible magnetic flux sensor is placed in a temperature-controlled chamber, and the response output amplitude to the same standard magnetic field source (or fixed signal source) is measured at different temperatures T (e.g., 0℃ to 80℃). V out (T).

[0062] Establish a temperature-gain relationship model. This typically exhibits a linear or quadratic relationship.

[0063] in, T 0 is the reference temperature (usually taken as 25℃). V out ( T 0) is the standard output at the reference temperature, and β1 and β2 are the temperature drift coefficients of the flexible flux sensor (obtained through calibration fitting).

[0064] Step 2: Real-time temperature acquisition The ambient temperature of the flexible magnetic flux sensor is monitored in real time by a temperature sensor integrated at the tail of the flexible magnetic flux sensor. T real .

[0065] Step 3: Dynamic gain compensation In the signal conditioning and acquisition unit, the compensation coefficient K is calculated in real time. comp。

[0066]

[0067] The acquired raw signals V raw Provide compensation:

[0068] In this way, regardless of changes in ambient temperature, the compensated signal amplitude is corrected to the reference temperature. T The level below 0 eliminates the influence of temperature drift on the amplitude. The above calculation method is implemented in the FPGA controller of the signal conditioning and acquisition unit.

[0069] The online monitoring method in this embodiment specifically includes: (1) Signal acquisition The differential magnetic flux signal in the air gap of the generator is acquired in real time by a flexible magnetic flux sensor. After the signal conditioning and acquisition unit amplifies, filters, performs temperature compensation, and performs analog-to-digital conversion on the differential magnetic flux signal, the original voltage signal sequence is obtained. f[n] , n = 0 , 1 ,…, N 1 Where N is the number of sampling points and the sampling frequency is... f s ≥200 kHz.

[0070] (2) Adaptive wavelet denoising For the original voltage signal sequence f[n] Perform J-level discrete wavelet decomposition (DWT) to obtain the wavelet coefficients of each level:

[0071] in, This represents the discrete form of the wavelet basis functions. j For scale parameters, k These are the translation parameters.

[0072] An improved wavelet threshold function is used to quantize the wavelet coefficients at each level:

[0073] Among them, threshold T j Determined by the following formula:

[0074] in, N j For the first j The number of layer wavelet coefficients.

[0075] Finally, the denoised signal is obtained through inverse wavelet reconstruction.

[0076] (3) Spectrum analysis and feature extraction For the denoised signal f^ [n] Perform Fast Fourier Transform (FFT):

[0077] Extract the amplitude of the fundamental component and each harmonic component. and :

[0078] Constructing feature vectors:

[0079] Where M is 10 (i.e., the 10th harmonic is extracted). f 0 represents the fundamental frequency (corresponding to the rotor rotation frequency).

[0080] (4) Neural network fault diagnosis A three-layer backpropagation neural network is used, with the number of nodes in the input layer being... I=M+1 Hidden layer node count H=15, output layer node count O= 3 (These correspond to different inter-turn short circuit probabilities, including normal state probability, minor short circuit probability, and severe short circuit probability). Minor short circuit and severe short circuit refer to two states with different numbers of short-circuited turns; the severe short circuit state has more short-circuited turns than the minor short circuit state.

[0081] Hidden layer output:

[0082] Output layer output:

[0083] Where φ(.) represents the hidden layer activation function (taking the tanh function). ; ψ(.) represents the output layer activation function (using the softmax function). ; w ip Indicates the input layer's first... i The first neuron to the hidden layer p The connection weights of each neuron, where i =1,2,…, I ( I (Input feature dimension)p =1,2,…, H ( H (Number of hidden layer nodes); b p Indicates the hidden layer number 1 p The bias term for each neuron is a scalar used to adjust the activation threshold of the neuron. w pq : indicates the hidden layer p The nth neuron to the output layer q The connection weights of each neuron, where q =1,2,…, O ( O To output the number of categories, here O =3); b q : indicates the output layer q Bias terms for each neuron.

[0084] The network output is y=[y1,y2,y3] T This represents the probability of each category, and the category corresponding to the highest probability is taken as the diagnostic result. If y2>0.5 or y3>0.5, then a fault is considered to exist.

[0085] (5) Location and severity assessment of faulty stator slots If the diagnosis indicates a fault, from the denoised signal f^ [ n Extract the peak value of the induced potential corresponding to each stator slot from the []. Let the peak value of the induced potential corresponding to the s-th stator slot be []. P s The corresponding stator slots with the same pole (i.e., stator slots that differ by one pair of poles) are: s′ Calculate the relative rate of change:

[0086] like Δ s >δ ( δ If we take 15%, then we consider that an inter-turn short circuit has occurred in the s-th stator slot.

[0087] Figure 5 The waveforms under normal and fault conditions show that, under fault conditions, the difference in the peak value of the induced electromotive force between adjacent stator slots with the same pole increases.

[0088] The degree of failure is determined by the neural network output and Δ s Overall decision:

[0089] Where α=0.6 and β=0.4 are empirical weights.

[0090] (6) Tiered early warning Three warning thresholds are set based on the severity of the fault: Level 1 Warning (Attention): 0.2 ≤ Severity < 0.4, the system will record and issue a notification.

[0091] Level 2 Warning: 0.4 ≤ Severity < 0.7, triggers audible and visual alarm, and recommends arranging for system shutdown and inspection.

[0092] Level 3 Warning (Emergency): Severity ≥ 0.7. Immediately upload to the central control center and recommend emergency shutdown.

[0093] When Severity < 0.2, it is considered a normal fluctuation and no warning is triggered.

[0094] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. An online monitoring system for inter-turn short circuits in a generator rotor based on the probe coil method, characterized in that, include: A flexible flux sensor, comprising a flexible substrate and a planar helical detection coil integrated on the flexible substrate, is attached to the stator side of a generator by an adhesive and is used to collect the differential flux signal of the rotor air gap. The signal conditioning and acquisition unit is used to amplify, filter, and perform analog-to-digital conversion on the magnetic flux differential signal, and output the original voltage signal sequence. The data processing and diagnostic unit is used to denoise the original voltage signal sequence using an improved wavelet threshold function, perform spectral analysis on the denoised signal, and extract feature vectors. The feature vector is input into a pre-trained BP neural network model, which outputs the probability of each inter-turn short circuit state category. Based on the probability of each inter-turn short circuit state category, it is diagnosed whether there is an inter-turn short circuit fault. If the diagnosis result is that there is an inter-turn short circuit fault, the stator slot with the inter-turn short circuit fault is located and the degree of fault is assessed. The human-machine interaction unit is used to display the output results of the data processing and diagnostic unit and issue early warning information according to the degree of fault.

2. The online monitoring system for inter-turn short circuits of a generator rotor based on the detection coil method according to claim 1, characterized in that, The flexible substrate and the planar spiral detection coil are encapsulated with a protective layer.

3. The online monitoring system for inter-turn short circuits of generator rotors based on the detection coil method according to claim 2, characterized in that, The flexible magnetic flux sensor integrates a temperature sensor, which is used to monitor the ambient temperature in real time and feed it back to the signal conditioning and acquisition unit. The signal conditioning and acquisition unit amplifies, filters, thermally compensates for and converts the magnetic flux differential signal according to the ambient temperature, and outputs the original voltage signal sequence.

4. A method for online monitoring of inter-turn short circuits in generator rotors based on the probe coil method, characterized in that, The generator rotor inter-turn short circuit online monitoring system based on the probe coil method according to any one of claims 1 to 3 includes: A flexible magnetic flux sensor is installed on the stator side of the generator to collect the differential magnetic flux signal in the rotor air gap; The signal conditioning and acquisition unit amplifies, filters, and performs analog-to-digital conversion on the magnetic flux differential signal to output the original voltage signal sequence. The data processing and diagnosis unit uses an improved wavelet threshold function to denoise the original voltage signal sequence. The denoised signal is then subjected to spectral analysis to extract feature vectors. The feature vectors are input into a pre-trained BP neural network model, which outputs the inter-turn short circuit state category probability. Based on the inter-turn short circuit state category probability, the existence of an inter-turn short circuit fault is diagnosed. If the diagnosis result indicates the existence of an inter-turn short circuit fault, the stator slot with the inter-turn short circuit fault is located and the degree of fault is assessed. The human-computer interaction unit displays the results output by the data processing and diagnostic unit and issues early warning information based on the degree of fault.

5. The online monitoring method for inter-turn short circuits in generator rotors based on the detection coil method according to claim 4, characterized in that, The method of denoising the original voltage signal sequence using an improved wavelet threshold function is as follows: For the original voltage signal sequence f[n] Perform J-level discrete wavelet decomposition to obtain the wavelet coefficients of each level: in, This represents the discrete form of the wavelet basis functions. j For scale parameters, k These are translation parameters; An improved wavelet threshold function is used to quantize the wavelet coefficients at each level: Among them, threshold T j Determined by the following formula: in, N j For the first j The number of layer wavelet coefficients; The denoised signal is obtained through wavelet inverse reconstruction: 。 6. The online monitoring method for inter-turn short circuits of generator rotor based on the detection coil method according to claim 5, characterized in that, The step of performing spectral analysis on the denoised signal and extracting feature vectors specifically involves: For the denoised signal f^ [n] Perform a Fast Fourier Transform: Extracting the fundamental amplitude and the amplitude of each harmonic : Constructing feature vectors: Where M is the harmonic order, f 0 represents the frequency of the fundamental wave.

7. The online monitoring method for inter-turn short circuits of generator rotor based on the detection coil method according to claim 6, characterized in that, The BP neural network model uses a three-layer BP neural network, with the number of nodes in the input layer... I=M+1 Hidden layer node count H, output layer node count O， Each node in the output layer corresponds to a short-circuit state category between turns. Hidden layer output: Output layer output: Where φ(.) represents the hidden layer activation function (taking the tanh function). ; ψ(.) represents the output layer activation function (using the softmax function). ; w ip Indicates the input layer's first... i The first neuron to the hidden layer p The connection weights of each neuron, where i =1,2,…, I ( I (Input feature dimension) p =1,2,…, H ( H (Number of hidden layer nodes); b p Indicates the hidden layer number 1 p The bias term for each neuron is a scalar used to adjust the activation threshold of the neuron. w pq : indicates the hidden layer p The nth neuron to the output layer q The connection weights of each neuron, where q =1,2,…, O ( O To output the number of categories, here O =3); b q : indicates the output layer q Bias terms for each neuron; The output of the BP neural network model represents the probability of each inter-turn short circuit state category, and the category corresponding to the highest probability is taken as the diagnosis result.

8. The online monitoring method for inter-turn short circuits of generator rotor based on the detection coil method according to claim 5, characterized in that, If the diagnosis result indicates the presence of an inter-turn short circuit fault, then the stator slot with the inter-turn short circuit fault is located, specifically as follows: If the diagnosis indicates an inter-turn short circuit fault, extract the denoised signal. f^ [ n The waveform is used to identify the peak value of the induced electromotive force corresponding to each stator slot; let the peak value of the induced electromotive force corresponding to the s-th stator slot be... P s The corresponding stator slots of the same pole are s′ Calculate the relative rate of change: like Δ s >δ If so, it is assumed that an inter-turn short circuit has occurred in the s-th stator slot. δ This is the threshold for the rate of change.

9. The online monitoring method for inter-turn short circuits of a generator rotor based on the detection coil method according to claim 8, characterized in that, The assessment of the degree of failure specifically includes: The degree of failure is determined by the output of the BP neural network model and Δ s Overall decision: Where α represents the weights output by the BP neural network model, and β represents... Δ s The weight, This represents the probability of an inter-turn short-circuit fault in the rotor output of the BP neural network model.

10. The online monitoring method for inter-turn short circuits of a generator rotor based on the detection coil method according to claim 9, characterized in that, The issuance of early warning information based on the severity of the fault is specifically as follows: Level 1 warning: 0.2 ≤ Severity < 0.4, the system records and issues a notification; Level 2 warning: 0.4 ≤ Severity < 0.7, triggering an audible and visual alarm, and recommending a shutdown for inspection; Level 3 warning: Severity ≥ 0.7, upload to the central control center, and recommend emergency shutdown.

Images