A transformer coil fault detection system based on ultrasonic reflection echo

Abstract

This invention relates to the field of non-destructive testing technology for power equipment, and discloses a transformer coil fault detection system based on ultrasonic reflected echoes. It primarily addresses the problem of severe multipath interference in ultrasonic signals caused by the multi-layered structure inside the transformer, making it difficult to accurately extract fault features of specific coils. This system transmits and receives ultrasonic waves through components on the outer wall of the tank; it calculates the theoretical time of flight using a three-dimensional acoustic mapping model of the transformer and constructs a target time window; it performs time-domain truncation on the original echo, directionally stripping away interference to extract the effective reflected echo from the specific target coil; it extracts the acoustic feature parameters and obtains the feature difference vector by differential analysis with baseline features; finally, it determines the type of physical deformation of the coil and outputs the quantified deformation degree, achieving non-invasive, high-precision fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for power equipment, and discloses a transformer coil fault detection system based on ultrasonic reflection echo. Background Technology

[0002] When using ultrasound for internal transformer testing, the complex interfaces between insulating oil, insulating paperboard, support bars, and multi-layered coils inside the transformer cause severe attenuation, refraction, and multipath reflection of the ultrasonic signal during penetration and reflection. This complex acoustic physics environment results in the received ultrasonic echo signal being mixed with a large number of interference waves from non-target structures, leading to a very low signal-to-noise ratio and highly aliased waveforms in the original signal, making it difficult to directly distinguish the true reflection characteristics of the target coil.

[0003] Existing signal processing and diagnostic methods lack effective spatiotemporal isolation mechanisms when dealing with aliased ultrasonic echoes, failing to selectively isolate the reflected waves from a specific target coil amidst cluttered background interference. This results in severe distortion of the extracted feature parameters. Furthermore, traditional analysis methods often lack standardized baseline comparison mechanisms and intelligent classification models, leaving diagnostic results at a qualitative warning level. They cannot accurately determine the specific type of coil physical deformation, nor can they output reliable quantitative deformation levels, significantly limiting the engineering practical value of non-invasive testing.

[0004] In summary, existing ultrasonic testing technologies for transformers have significant bottlenecks and shortcomings in resisting internal acoustic multipath interference, directional extraction of target coil echoes, and in-depth quantitative analysis of fault deformation. Therefore, there is an urgent need in this field for a transformer coil fault detection solution that can overcome interference from complex media interfaces, achieve accurate target echo interception, and provide intelligent, high-precision quantitative assessment. Summary of the Invention

[0005] The purpose of this invention is to provide a transformer coil fault detection system based on ultrasonic reflected echoes. Its core innovation lies in overcoming the technical bottleneck of traditional ultrasonic testing, which is easily affected by multipath interference from the complex internal media of transformers. It innovatively introduces a three-dimensional acoustic mapping model of the transformer to calculate the theoretical flight time and construct a dedicated target time window. By precisely extracting the original aliased signal in the time domain, it achieves directional stripping of the effective reflected echoes from a specific target coil. Building upon this breakthrough, the invention further extracts acoustic features of specific dimensions and performs differential operations with healthy baseline data to generate a feature difference vector. Then, it uses an intelligent classification model to perform deep mapping on this difference vector. This not only enables non-invasive and accurate diagnosis of the physical deformation type of the coil but also calculates and outputs the quantitative deformation displacement of the structural interface based on the time offset. Ultimately, it constructs a high-precision non-destructive testing solution from targeted extraction of complex echoes to quantitative assessment of fault depth.

[0006] The objective of this invention can be achieved through the following technical solutions: A transformer coil fault detection system based on ultrasonic reflected echoes includes the following modules: An ultrasonic transceiver and acquisition module includes a non-invasive ultrasonic transceiver component coupled to the outer wall of the transformer tank, used to transmit detection ultrasonic waves into the transformer and acquire the ultrasonic echo signal formed after the detection ultrasonic waves are reflected at the interface of the multi-layer structure inside the transformer. The echo directional stripping module is used to acquire a three-dimensional acoustic mapping model of the transformer's internal structure. It then uses the real-time temperature of the transformer's internal insulating oil to correct the propagation speed of ultrasonic waves in the oil. Based on the three-dimensional acoustic mapping model and the corrected propagation speed, it calculates the theoretical flight time of the probed ultrasonic wave from the transmitting end to the preset target coil and back to the receiving end. A target time window is then constructed based on this theoretical flight time. This target time window is used to perform time-domain truncation of the ultrasonic echo signal to strip away multipath interference echoes generated by non-target structures and extract the effective reflected echo specific to the target coil. The feature extraction and difference module is used to extract the acoustic feature parameters of the effective reflected echo and perform difference processing on the acoustic feature parameters and the baseline reference features of the corresponding coil obtained in advance to obtain the feature difference vector. The fault diagnosis and quantitative evaluation module is used to determine the physical deformation type of the target coil based on the feature difference vector and output the corresponding quantitative deformation degree.

[0007] Preferably, the ultrasonic transceiver and acquisition module transmits a broadband linear frequency modulated pulse signal; When acquiring the ultrasonic echo signal, the ultrasonic transceiver and acquisition module specifically performs the following: receiving the original echo signal reflected from the multi-layer structure interface, performing autocorrelation processing on the original echo signal using the built-in pulse compression algorithm to generate a high-resolution echo signal with improved signal-to-noise ratio, and outputting the high-resolution echo signal as the ultrasonic echo signal to the echo directional stripping module for subsequent time-domain truncation.

[0008] Preferably, the ultrasonic transceiver and acquisition module includes an ultrasonic transceiver component with a magnetic base. The ultrasonic transceiver assembly is attached to the smooth surface of the outer wall of the transformer tank via the magnetic base and acoustic coupling agent, so as to penetrate the metal tank wall barrier and transmit the detection ultrasonic waves into the transformer.

[0009] Preferably, the three-dimensional acoustic mapping model obtained by the echo directional stripping module specifically includes the three-dimensional coordinates of the transformer tank wall, the physical width value of the insulating oil gap, the spatial position coordinates of the insulating paperboard, and the theoretical acoustic impedance parameters of the preset target coil.

[0010] Preferably, when the echo directional stripping module combines the real-time temperature correction of the insulating oil to adjust the propagation speed of sound and calculates the theoretical flight time, it specifically performs the following: Obtain the real-time temperature value of the insulating oil inside the transformer; substitute the real-time temperature value into the preset oil temperature-sound speed correction equation to calculate the dynamic propagation speed of the ultrasonic wave after correction at the current temperature. Combining the linear spatial distance from the transmitter to the preset target coil in the three-dimensional acoustic mapping model with the dynamic propagation speed of sound, the theoretical flight time after temperature compensation is obtained by dividing the distance by the speed.

[0011] Preferably, the echo-directed stripping module performs the following specific actions when constructing the target time window: Obtain the physical radial thickness parameter of the preset target coil; Divide the physical radial thickness parameter by the preset equivalent sound velocity parameter of the coil material to calculate the one-way sound path time span required for the ultrasonic wave to completely penetrate inside the preset target coil. Using the theoretical flight time as the starting reference point, the sum of twice the one-way acoustic path length and the preset detection ultrasonic pulse width is used to generate the target time window with absolute start and absolute end times.

[0012] Preferably, the acoustic feature parameters of the effective reflected echo extracted by the feature extraction and differential module include the following three dimensions: time-of-flight offset, peak amplitude attenuation rate, and center frequency shift.

[0013] Preferably, the baseline reference features obtained by the feature extraction and difference module are the initial acoustic dataset corresponding to the target coil, which is measured and stored by the transformer under factory fault-free conditions. The feature extraction and difference module performs an absolute difference operation by subtracting the currently extracted acoustic feature parameters from the corresponding initial parameters in the initial acoustic dataset, and then concatenates the absolute difference values ​​in a fixed dimensional order to generate the feature difference vector.

[0014] Preferably, the fault diagnosis and quantitative evaluation module is pre-configured with a support vector machine classification model trained under supervised learning. The fault diagnosis and quantitative evaluation module feeds the feature difference vector as input features into the support vector machine classification model, and the support vector machine classification model maps and outputs the physical deformation type label corresponding to the target coil. The physical deformation type label includes coil bulging, coil denting, coil loosening, or coil local warping.

[0015] Preferably, when the fault diagnosis and quantitative evaluation module outputs the corresponding quantitative deformation degree, it specifically performs the following: Extract the time-of-flight offset parameter contained in the feature difference vector; Based on the assumption of rigid displacement of the local interface inside the transformer, the actual physical displacement distance of the target coil structure interface is calculated by multiplying the time-of-flight offset parameter by the corrected dynamic propagation speed of sound and dividing by two. The actual physical displacement distance is then output as the quantified deformation degree.

[0016] The beneficial effects of this invention are: The system of this invention precisely corrects the dynamic propagation speed of ultrasonic waves by incorporating the real-time temperature of the insulating oil inside the transformer. It also calculates the theoretical flight time using a three-dimensional acoustic mapping model inside the transformer to construct a directional target time window, effectively overcoming the sound speed drift error caused by oil temperature fluctuations during equipment operation. Simultaneously, this high-precision time window is used to perform time-domain truncation of the echo signal, accurately extracting the effective echo specific to the target coil from complex internal multipath reflections and reverberation. This effectively eliminates interference signals generated by non-target structures, fundamentally ensuring the authenticity and accuracy of subsequent acoustic feature extraction.

[0017] This invention employs a non-invasive magnetic ultrasonic transceiver to transmit broadband linear frequency modulated pulse signals. A built-in pulse compression algorithm at the receiving end performs autocorrelation processing on the original echo signal, effectively overcoming the severe energy attenuation problem encountered by ultrasonic waves when penetrating the thick metal walls and multi-layered insulating oil gaps of transformers. This design not only ensures the detection signal has the physical penetration capability to reach the deep inner coils, but also significantly improves the signal-to-noise ratio and axial spatial resolution of the echo without increasing the peak power of the hardware transmission, providing extremely pure, high-resolution data support for identifying minute deformations in the coils.

[0018] This invention extracts feature difference vectors from multiple dimensions, including time domain, frequency domain, and amplitude, and feeds them into a pre-trained support vector machine classification model. This not only automatically and intelligently maps specific physical deformation types such as coil bulges, dents, or loosening, but also creatively combines the assumption of local rigid displacement to directly convert time-of-flight offset in the time domain into the actual physical displacement distance of the target interface. This mechanism effectively breaks through the subjective limitations of traditional ultrasonic testing, which heavily relies on manual image interpretation and experience-based judgment. It directly transforms ambiguous acoustic signals into intuitive equipment deformation indicators, providing a scientific and reliable decision-making basis for transformer condition early warning and precise maintenance. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a transformer coil fault detection system based on ultrasonic reflected echo according to the present invention; Figure 2 This is a schematic diagram of a transformer coil fault detection system based on ultrasonic reflected echo according to the present invention. Figure 3 This is a physical diagram illustrating the target time window construction and time domain truncation of a transformer coil fault detection system based on ultrasonic reflection echoes according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example: Figures 1-3As shown, this embodiment provides a transformer coil fault detection system based on ultrasonic reflected echoes. Addressing the technical challenge of complex interfaces between insulating oil, insulating cardboard, wooden support strips, and multiple layers of copper and aluminum coils within large power transformers, which easily lead to severe attenuation, refraction, and multipath reverberation interference in ultrasonic signal propagation, this system employs a non-intrusive hardware and software integrated architecture. The system consists of four core modules: an ultrasonic transceiver and acquisition module, an echo directional stripping module, a feature extraction and differential module, and a fault diagnosis and quantitative assessment module.

[0022] As the front end of the system, the ultrasonic transceiver and acquisition module undertakes the core task of penetrating physical barriers to excite signals and acquire raw echoes with high fidelity.

[0023] Structure and Implementation of Non-Intrusive Hardware Coupling Components: Since transformers are strictly prohibited from being opened or drilled during operation, this system must employ non-intrusive testing. In this embodiment, the ultrasonic transceiver component uses a broadband piezoelectric ceramic transducer, preferably made of PZT-5A material with high piezoelectric constant and large electromechanical coupling coefficient. To overcome the total acoustic impedance reflection caused by the air gap between the steel tank wall and the transducer surface, the ultrasonic transceiver component integrates a high-strength magnetic base at its bottom. The specific implementation process is as follows: The testing personnel first select a smooth area on the outer wall of the transformer tank, directly opposite the preset target coil, as the testing point; they then gently polish the anti-rust paint film and dirt from the testing point surface using sandpaper or an angle grinder, exposing a smooth metal base; subsequently, a high-temperature resistant, acoustically impedance-matched industrial-grade acoustic coupling agent is evenly applied between the transducer radiating surface and the metal base; finally, the strong magnetic force of the magnetic base is used to firmly attach the ultrasonic transceiver component to the outer wall of the tank. This mechanical structure ensures that the contact stress between the transducer and the tank wall remains constant under high-frequency, minute vibrations, allowing ultrasonic energy to penetrate the metal tank wall barrier with maximum transmittance and radiate detection ultrasonic waves into the insulating oil inside the transformer.

[0024] Waveform Design and Pulse Compression Autocorrelation Algorithm: Ultrasonic waves inevitably encounter significant energy attenuation when penetrating 10-20mm thick steel plates, tens of millimeters of oil gaps, and multiple layers of insulating paper. Using conventional high-voltage narrow pulses, the energy is concentrated in a very short time, easily interfering with the secondary relay protection equipment of the transformer body due to instantaneous high voltage. Using wide pulses, although the energy increases, causes the echo to widen on the time axis, severely reducing axial resolution and causing echo overlap at the coil's front and rear interfaces. To resolve this physical contradiction, in this embodiment, an arbitrary waveform generator within the ultrasonic transceiver module is responsible for generating and exciting a broadband linear frequency modulated pulse signal. The frequency of this signal changes linearly with time during the pulse duration. When the ultrasonic transceiver component receives the chaotic "raw echo signal" reflected back from the internal multi-layer structure, the signal is sent to the built-in high-speed analog-to-digital converter for digitization. Subsequently, the built-in digital signal processor or field-programmable gate array calls the digital filter to perform autocorrelation pulse compression calculations.

[0025] Let the transmitted broadband linear frequency modulated pulse discrete-time signal be... The original aliased echo signal acquired by the ADC, which includes background noise and multipath interference, is The high-resolution echo signal with improved signal-to-noise ratio is generated after processing by the built-in pulse compression algorithm. The following autocorrelation integral formula satisfies the equivalentity of the frequency domain and time domain: ,in, This represents the output high-resolution echo signal matrix, whose time axis resolution is greatly compressed, and the waveform exhibits an extremely sharp Sinc function peak shape. The time delay sliding step size variable represents the time delay during the correlation function calculation process. t represents the absolute time independent variable of the system ADC sampling clock. This represents the actual received raw broadband echo signal, which in the time domain appears as a long, overlapping cluttered cluster with a low signal-to-noise ratio. This represents the complex conjugate of the ideal broadband linear frequency modulated pulse signal used for initial excitation at the transmitter. In digital signal processing implementations, the pre-set transmit waveform is typically stored in a ROM, and its conjugate is retrieved and convolved in real time.

[0026] According to matched filter theory, the integration operation reaches its maximum output when the reflected waveform in the received signal matches the local reference signal. This processing method can instantly compress the energy of a chirp signal, which lasts for tens of microseconds, into an extremely narrow time window of less than one microsecond without increasing the peak transmit power. It gives the signal-to-noise ratio of the original aliased signal a processing gain of tens of decibels (dB), effectively removes irrelevant white noise, and separates the aliased reflected waves into independent, sharp pulse peaks, serving as the physical basis for subsequent precise time-domain truncation.

[0027] In the echo directional stripping module, after obtaining the high-resolution echo signal, the next challenge is that, on the long echo time axis, countless pulse peaks represent different interfaces of the box wall, oil gap, cardboard, and multi-layer coil. The system must use logical operations to accurately determine which waveform segment belongs to the real "target coil".

[0028] The echo-directed stripping module has a built-in database storing the factory physical structure parameters of various transformer models under test. During implementation, the system first retrieves and reconstructs a three-dimensional acoustic mapping model of the transformer's internal structure. Specifically, this model is not simply a geometric drawing, but a data structure containing acoustic attributes. Its contents include: the three-dimensional coordinates of the transformer tank wall, determining the spatial origin of the ultrasonic probe's oil inlet interface; the physical width of the insulating oil gap, determining the longest path for sound waves to propagate in pure oil; the spatial coordinates and thickness of the insulating cardboard, calculating the acoustic attenuation and delay of the solid insulation; and the theoretical acoustic impedance parameters of the preset target coil, such as the density of copper (8900 kg / m³). 3 The product of this and the speed of sound (4700 m / s) is used to evaluate the reflection coefficient, as well as the linear spatial distance between the target coil and the probe origin.

[0029] In this field, the vast majority of detection errors stem from the drift in the sound velocity of transformer insulating oil. Under no-load and full-load conditions, the oil temperature of a transformer can fluctuate drastically from 20°C to 80°C. Since transformer oil is a liquid medium, its bulk modulus of elasticity is extremely sensitive to temperature; for every 1°C increase in temperature, the propagation speed of ultrasound in the oil decreases by approximately 3.5 to 4.0 m / s. Using a fixed sound velocity would result in positioning deviations of several centimeters, leading to complete misidentification of the echo interface. Therefore, this module uses a PT100 platinum resistance temperature sensor integrated into the transformer body or other communication interfaces to read the real-time temperature of the insulating oil. Based on this, it performs extremely rigorous dynamic corrections to the sound velocity, thereby calculating the theoretical flight time of the ultrasound wave from the probe to the surface of the target coil.

[0030] The theoretical flight time for an ultrasonic wave to travel through multiple layers of dielectric material in a transformer, from the transmitting end to the interface in front of the preset target coil, and back to the receiving end. The specific calculation expression is as follows: in, The theoretical flight time serves as the baseline variable for the start of the time window; 2 represents the distance multiplication constant. Because the detection method uses pulse reflection, the ultrasonic wave must complete both the "outward journey" and the "return journey," therefore the total time must be multiplied by 2. , These represent the physical thickness of the high-carbon steel plate of the fuel tank wall, read from the 3D model, and the fixed longitudinal wave velocity at room temperature in this material, respectively. It is typically calibrated to 5920 m / s. The physical width of the pure transformer insulating oil gap, read from the inner wall of the oil tank to the surface of the insulating cardboard, is taken from the model. The initial sound velocity of the transformer's #25 insulating oil under standard reference temperature conditions is preferably 1425 m / s, measured at 20°C in this embodiment. The sound velocity-temperature negative correlation temperature coefficient is specific to insulating oil. In this embodiment, through extensive laboratory calibration, the preferred value is 3.85. This represents the current actual temperature of the insulating oil, fed in from the external transformer monitoring system bus or additional sensors. The denominator in the formula... This is the most crucial modified dynamic propagation speed of sound in the embodiment; and The corresponding preset standard reference temperature is 20℃ in this embodiment; , These represent the cumulative thickness of the insulating paperboard and the sound velocity within the paperboard, respectively. The sound velocity in the paperboard is affected by oil penetration, and in this embodiment, an equivalent value of 2200 m / s is used.

[0031] The control system can accurately calculate at what time the reflected echo from the target coil "should" arrive at the receiver under the current ambient temperature, thus providing the system with an absolutely precise "time anchor point".

[0032] Physical closure control and time-domain interception execution of the target time window: Simply finding the starting point is insufficient. To ensure complete extraction of the target coil's echo characteristics while discarding echoes from deep structures on the coil's back side, the system must... Based on this, a strictly limited "target time window" is constructed by extending backward. In actual execution, the module extracts the physical radial thickness parameters of the preset target coil from the 3D model again, and calculates the time span required for the sound wave to completely penetrate inside the coil in combination with the material.

[0033] Absolute termination time The formula for the control mechanism of the target time window is as follows: ,in, The absolute termination and closure time node reached when the target time window is extended backward along the time axis; The radial physical thickness dimension of the target coil; The equivalent longitudinal wave velocity of the ultrasonic wave propagating inside a specific metal conductor of the target coil is given; copper is typically represented by 4700 m / s. This is the preset margin parameter for the ultrasonic pulse width. Because the waveform after pulse compression is not an infinitely narrow needle, but a Sinc wave with a main lobe width, this parameter ensures that the main lobe of the reflected waveform at the coil interface is not rigidly cut off, thus guaranteeing the integrity of the feature.

[0034] When the control logic generates a closed interval Then, the signal processor will perform data interception. For the entire high-resolution echo signal, in... All previous data points are forced to zero or discarded; currently... All subsequent data points are also forcibly set to zero. Ultimately, only the data segment falling within this time window is fully retained. This segment effectively eliminates background interference from non-target structures, resulting in a valid reflected echo specifically belonging to the target coil.

[0035] In the feature extraction and difference module, the system in this embodiment does not rely on manual observation of waveform morphology by the naked eye, but uses an algorithm to perform three-dimensional digital feature analysis on the extracted "effective reflected echo".

[0036] Time-domain flight time offset ( The system performs a Hilbert transform on the valid echo to extract the waveform envelope curve, and uses a peak-finding algorithm to locate the absolute timestamp corresponding to the highest point of the envelope. This parameter is directly related to the change in the physical spatial position of the coil interface.

[0037] Peak amplitude attenuation rate ( ): Extract the voltage amplitude value of the highest peak value. When the coil becomes mechanically loose, the adhesion between the coil surface and the insulating cardboard decreases, and the intrusion of tiny oil gaps causes a drastic change in the acoustic impedance reflection coefficient of the interface, resulting in a significant reduction in the energy of the reflected echo. The magnitude of the attenuation rate directly characterizes the degree of loss of coil clamping force.

[0038] Center frequency shift ( The system performs a Fast Fourier Transform on the intercepted effective time-domain echo, converting it to the frequency domain to determine the center frequency of the energy centroid. According to acoustic theory, when the coil undergoes local deformation or material fatigue, the intrinsic resonant frequency of its acoustic medium will shift microscopically, causing the spectrum of the reflected wave to drift to lower or higher frequencies.

[0039] Baseline feature comparison and feature difference vector generation: In order to eliminate the influence of initial tolerances caused by transformer manufacturing process, this system introduces the concept of "relative health diagnosis".

[0040] During the factory testing phase of the target transformer, or in the initial commissioning state after maintenance and confirmation of no faults, the system will perform a full scan according to the same steps, and store the extracted time-domain, amplitude, and frequency-domain data of each coil in a non-volatile memory as the initial acoustic dataset.

[0041] During routine online testing, the module subtracts the three acoustic parameters extracted in real time from the corresponding normal parameters in the initial acoustic dataset to obtain the absolute difference. Then, these differences are strictly calculated according to... The sequence of features is concatenated into a matrix or array structure to generate a high-dimensional feature difference vector, which serves as a standardized input interface for subsequent artificial intelligence diagnostic modules.

[0042] The fault diagnosis and quantitative assessment module is the core processing unit that transforms the detection results from qualitative analysis to quantitative automated assessment.

[0043] Traditional ultrasonic equipment can only alarm "there is an abnormality," while this system can clearly identify "what kind of abnormality." In this embodiment, a pre-trained support vector machine classification model is embedded within the fault diagnosis and quantitative assessment module. During the research and development phase, technicians constructed a scaled-down model of an equal-scale transformer, artificially creating various coil defects, and combined this with a large amount of historical acoustic data from the repair of real power grid transformer covers, collecting tens of thousands of sets of feature difference vector sample datasets.

[0044] Before factory configuration, the Support Vector Machine (SVM) classification model undergoes rigorous offline supervised learning training. A scaled-down model of a transformer is constructed, artificially introducing specific defects. Acoustic data from real historical power grid equipment undergoing repair is used to acquire a large number of raw ultrasonic echo signals with known physical morphologies. The system extracts time-of-flight offset, peak amplitude attenuation rate, and center frequency shift from these sample signals using a preset algorithm as three-dimensional input features. Corresponding labels for real physical types such as coil bulges, coil dents, coil loosening, and localized coil warping are manually assigned, constructing a standard training sample library containing tens of thousands of data sets. This sample library is input into the initial SVM model, and a radial basis function (RBF) kernel is introduced to map low-dimensional features to a high-dimensional space. Through iterative calculations, the weight parameters are continuously adjusted until the optimal classification hyperplane that maximizes the interval between different fault categories is found. This completes the model training and stores it in the system's non-volatile memory for online retrieval.

[0045] The SVM model is trained using this data containing known fault types. SVM maps low-dimensional, seemingly nonlinear, inseparable feature difference vectors to a high-dimensional space by introducing a radial basis function kernel, and iteratively finds the optimal classification hyperplane that maximizes the margin between different fault categories.

[0046] In actual testing, when the system feeds real-time feature difference vectors into the SVM model, the model automatically matches and outputs the corresponding physical deformation type label of the target coil based on the coordinate position of the vector in multi-dimensional space. The types of labels that this system can identify with high confidence clearly include: coil bulging – typically manifested as significant timing advance; coil denting – significant timing lag; coil loosening – small timing shift but extremely large amplitude attenuation; and localized coil warping – a non-linear combination of frequency shift and attenuation rate.

[0047] Absolute quantitative evaluation algorithm for physical displacement: For defects such as "bulges" or "dents" that indicate structural displacement, maintenance personnel need to know extremely precise offset dimensions to assess whether the safety margin of insulation electrical distance has been exceeded. To this end, this module creatively utilizes a conversion mechanism from "time domain microseconds to spatial millimeters".

[0048] Based on the assumption that the deformation of the local interface inside the transformer in the early stage of a fault usually conforms to the physical assumption of rigid displacement, the system extracts the "time-of-flight offset" parameter from the feature difference vector and restores it to the actual physical displacement distance of the target coil structure interface in space. Its mathematical formula is as follows: ,in, This refers to the "quantified deformation degree" value that the system ultimately outputs on the terminal display screen, which is the specific physical size of the convexity or concavity of the coil interface; according to the symbol convention: a negative value represents that the interface is closer to the probe - that is, bulging outwards, and a positive value represents that the interface is farther away from the probe - that is, concave inwards; This refers to the real-time timestamp deviation value extracted separately from the aforementioned feature difference vector; To retrieve the real-time temperature of the insulating oil that has been passed through the system The current actual liquid sound velocity is dynamically corrected. If the factory nominal sound velocity is used instead of the real-time corrected sound velocity, the final displacement quantization calculation will have a serious error of tens of millimeters due to the nearly 100 meters per second deviation in sound velocity at high temperatures. This is the core physical basis for the high-precision quantization of this invention.

[0049] Through this formula, the system can not only inform users that "the equipment has malfunctioned", but also directly output a specific quantitative assessment report on the operation and maintenance terminal, such as "Warning: the third wire cake area of ​​the C phase low voltage winding has bulged outward by 2.45mm".

[0050] To demonstrate the full feasibility of all modules and parameters in this application in a real-world engineering environment, we will take the testing of the A-phase high-voltage winding of a 220kV large power transformer that has been in service for 5 years and has a rated operating oil temperature of 65℃ as an example to demonstrate the complete working cycle of the entire system from startup to the issuance of a quantitative report: Preparation for communication handshake: The operator arrives at the site with a portable testing terminal and, without power interruption, attaches the ultrasonic probe with a strong magnetic base to the corresponding oil tank wall of transformer A. The system main control board automatically hands over to the substation backend via the IEC-61850 protocol to read the real-time temperature of the transformer's insulating oil. .

[0051] The system retrieves a three-dimensional acoustic mapping model of the internal structure of phase A. Assuming a reference temperature of 20℃ with an oil sound velocity of 1420 m / s and an attenuation coefficient of 3.85, the system instantly calculates that the actual oil sound velocity at the current high temperature of 65℃ has significantly decreased to 1246.75 m / s. Subsequently, combining the straight-line distance of 300 mm from the probe with parameters of the enclosure wall and cardboard, the system accurately calculates the theoretical flight time for the probe to reach the surface of the target high-voltage coil. for .

[0052] The system reads that the thickness of the high-voltage coil is 35mm. Substituting this into the copper sound velocity and allowing for pulse width margin, the absolute termination time is calculated. for At this point, the system locks and generates a span of [missing information] in memory space. Target time window The hardware probe emits a 500kHz-2MHz wideband frequency-modulated pulse, and the ADC captures a pulse that lasts for a long time. The mixed electrical signal. The digital DSP performs autocorrelation compression on this signal, significantly improving the signal-to-noise ratio of the echo signal, narrowing the main lobe of the waveform, and highlighting the peak characteristics. The time-domain truncation program starts and removes... All the clutter reflected from the oil layer before and Then came the complete reverberation from the iron core and low-voltage windings. At this point, a section only [length missing] was successfully extracted from the massive amount of noise. The target effective echo.

[0053] The system has this The waveform was subjected to peak detection and FFT calculation. It was found that the peak arrival time was earlier than the factory database reading. This deviation, along with other attenuation deviations, is concatenated into the vector space to generate a feature difference vector.

[0054] Intelligent diagnosis and rigidity restoration: The difference vector is fed into a pre-installed SVM model, and the model achieves a high-scoring match to the "coil bulge" category. Finally, the system performs absolute displacement conversion: the advanced time difference is... Multiplying by the currently precisely corrected sound speed of 1246.75m / s and dividing by 2, the abstract microsecond-level acoustic signal is instantly restored to the real physical distance - the calculation shows that the coil interface at that point has undergone a substantial outward bulge deformation of approximately 1.99mm.

[0055] The system interface highlights the fault location in 3D and prints a detailed inspection report stating, "Middle section of A-phase high-voltage winding, physical deformation type determined as bulge; absolute quantitative deformation degree: 1.99 mm, approaching the insulation oil gap alarm threshold, it is recommended to carry out shutdown and cover removal maintenance."

[0056] The transformer coil fault detection system based on ultrasonic reflection echo disclosed in this embodiment not only improves the structure and method in terms of hardware topology, detection signal modulation and acoustic three-dimensional mapping; but also overcomes the technical bottleneck of traditional ultrasonic equipment that can only perform qualitative detection by relying on the above four core mathematical models and computing architecture. It provides a full-chain non-destructive testing technology solution that can achieve targeted signal extraction, intelligent fault classification and sub-millimeter absolute quantitative restoration of physical displacement in complex multiphase medium high-temperature environments.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An ultrasonic echo-based transformer coil fault detection system, characterized in that, Includes the following modules: An ultrasonic transceiver and acquisition module includes a non-invasive ultrasonic transceiver component coupled to the outer wall of the transformer tank, used to transmit detection ultrasonic waves into the transformer and acquire the ultrasonic echo signal formed after the detection ultrasonic waves are reflected at the interface of the multi-layer structure inside the transformer. The echo directional stripping module is used to acquire a three-dimensional acoustic mapping model of the transformer's internal structure. It then uses the real-time temperature of the transformer's internal insulating oil to correct the propagation speed of ultrasonic waves in the oil. Based on the three-dimensional acoustic mapping model and the corrected propagation speed, it calculates the theoretical flight time of the probed ultrasonic wave from the transmitting end to the preset target coil and back to the receiving end. A target time window is then constructed based on this theoretical flight time. This target time window is used to perform time-domain truncation of the ultrasonic echo signal to strip away multipath interference echoes generated by non-target structures and extract the effective reflected echo specific to the target coil. The feature extraction and difference module is used to extract the acoustic feature parameters of the effective reflected echo and perform difference processing on the acoustic feature parameters and the baseline reference features of the corresponding coil obtained in advance to obtain the feature difference vector. The fault diagnosis and quantitative evaluation module is used to determine the physical deformation type of the target coil based on the feature difference vector and output the corresponding quantitative deformation degree.

2. A transformer winding fault detection system based on ultrasonic reflection echoes according to claim 1, characterized in that, The ultrasonic transceiver and acquisition module emits a broadband linear frequency modulated pulse signal; When acquiring the ultrasonic echo signal, the ultrasonic transceiver and acquisition module specifically performs the following: receiving the original echo signal reflected from the multi-layer structure interface, performing autocorrelation processing on the original echo signal using the built-in pulse compression algorithm to generate a high-resolution echo signal with improved signal-to-noise ratio, and outputting the high-resolution echo signal as the ultrasonic echo signal to the echo directional stripping module for subsequent time-domain truncation.

3. The transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, The ultrasonic transceiver and acquisition module includes an ultrasonic transceiver component with a magnetic base. The ultrasonic transceiver assembly is attached to the smooth surface of the outer wall of the transformer tank via the magnetic base and acoustic coupling agent, so as to penetrate the metal tank wall barrier and emit the detection ultrasonic waves into the transformer.

4. The transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, The three-dimensional acoustic mapping model obtained by the echo directional stripping module specifically includes the three-dimensional coordinates of the transformer tank wall, the physical width of the insulating oil gap, the spatial coordinates of the insulating paperboard, and the theoretical acoustic impedance parameters of the preset target coil.

5. The transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, When the echo directional stripping module combines the real-time temperature correction of the insulating oil to adjust the propagation speed of sound and calculates the theoretical flight time, it specifically performs the following: Obtain the real-time temperature value of the insulating oil inside the transformer; substitute the real-time temperature value into the preset oil temperature-sound speed correction equation to calculate the dynamic propagation speed of the ultrasonic wave after correction at the current temperature. Combining the linear spatial distance from the transmitter to the preset target coil in the three-dimensional acoustic mapping model with the dynamic propagation speed of sound, the theoretical flight time after temperature compensation is obtained by dividing the distance by the speed.

6. The transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, When constructing the target time window, the echo-directed stripping module specifically performs the following: Obtain the physical radial thickness parameter of the preset target coil; Divide the physical radial thickness parameter by the preset equivalent sound velocity parameter of the coil material to calculate the one-way sound path time span required for the ultrasonic wave to completely penetrate inside the preset target coil. Using the theoretical flight time as the starting reference point, the target time window is generated by extending twice the one-way sound path time span along the time axis and the sum of the preset detection ultrasonic pulse width, thus generating the target time window with absolute start and absolute end times.

7. The transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, The acoustic feature parameters of the effective reflected echo extracted by the feature extraction and differential module include the following three dimensions: time-of-flight offset, peak amplitude attenuation rate, and center frequency shift.

8. A transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, The baseline reference features obtained by the feature extraction and difference module are the initial acoustic dataset corresponding to the target coil, which is measured and stored by the transformer under factory fault-free conditions. The feature extraction and difference module performs an absolute difference operation by subtracting the currently extracted acoustic feature parameters from the corresponding initial parameters in the initial acoustic dataset, and then concatenates the absolute difference values ​​in a fixed dimensional order to generate the feature difference vector.

9. A transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, The fault diagnosis and quantitative evaluation module is pre-configured with a support vector machine classification model trained under supervised learning. The fault diagnosis and quantitative evaluation module feeds the feature difference vector as input features into the support vector machine classification model, and the support vector machine classification model maps and outputs the physical deformation type label corresponding to the target coil. The physical deformation type label includes coil bulging, coil denting, coil loosening, or coil local warping.

10. A transformer coil fault detection system based on ultrasonic reflected echo as described in claim 1, characterized in that, When the fault diagnosis and quantitative evaluation module outputs the corresponding quantitative deformation degree, it specifically performs the following: Extract the time-of-flight offset parameter contained in the feature difference vector; Based on the assumption of rigid displacement of the local interface inside the transformer, the actual physical displacement distance of the target coil structure interface is calculated by multiplying the time-of-flight offset parameter by the corrected dynamic propagation speed of sound and dividing by two. The actual physical displacement distance is then output as the quantified deformation degree.

Images