MMC fault detection method and system based on sound-light-electricity fusion in wind power generation

Abstract

The invention provides an MMC fault detection method and system based on sound-light-electricity fusion in wind power generation. The system comprises an acoustic monitoring unit, an optical monitoring unit, an electrical signal acquisition unit and an AI algorithm processing module. The acoustic monitoring unit adopts a non-contact acoustic emission and ultrasonic sensor to capture acoustic signals of partial discharge and mechanical vibration of the power module; the optical monitoring unit obtains the temperature field and discharge radiation information of the power device through infrared thermal imaging, ultraviolet detection and fiber grating sensing; the electrical acquisition unit acquires voltage, current and high-frequency electromagnetic signals of the sub-modules; and the AI algorithm processing module performs fusion analysis on the multi-modal data by using a deep learning model, identifies a fault type and evaluates a health state. The method has the advantages of non-intrusive and high-sensitivity on-line monitoring, early discovery and accurate positioning of the early fault of the MMC converter valve can be realized in a strong-interference marine environment, and the operation safety and self-healing capability of the system are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of power electronic equipment fault monitoring, and particularly relates to a method and system for MMC fault detection based on the fusion of sound, light, electricity and electricity in wind power generation. Background Technology

[0002] The rapid development of wind power has placed higher demands on long-distance, large-scale clean energy transmission. Since wind farms are typically far from inland load centers, transmission distances can reach hundreds to thousands of kilometers. Traditional AC transmission suffers from high losses and low efficiency due to limitations in reactive power compensation and stability. Flexible DC transmission technology, with its advantages of low loss, large capacity, and friendliness to weak grids, has become the mainstream solution for long-distance wind power grid connection. Among these technologies, the Modular Multilevel Converter (MMC), as the core topology of third-generation voltage source converter technology, is widely used in wind power transmission projects. MMC overcomes the series withstand voltage problem of devices through the cascading of numerous submodules, achieving hundreds of megawatts to gigawatts in engineering applications such as the ±500 kV Zhangbei DC grid and the ±525 kV Beihai wind power transmission project.

[0003] However, with the increasing scale and voltage level of wind farms, the tens of thousands of IGBT power devices integrated in the MMC valve hall face harsh marine environments and long-term operational stress, posing unprecedented challenges to their operational reliability. Timely detection and location of these device-level faults are of great significance for ensuring the continuous and stable operation of wind power flexible DC systems.

[0004] Current condition monitoring technologies are still insufficient to fully meet the above requirements, mainly due to the following problems and shortcomings: Lack of sensitivity to early faults: Traditional monitoring methods mainly rely on electrical quantity sensors to collect signals such as voltage and current of submodules, which can only detect serious faults on a millisecond time scale. For early device degradation such as loose bonding wires and aging solder joints, as well as hidden defects such as partial discharge inside the package, simple electrical quantity monitoring lacks sufficient spatial resolution and sensitivity, and often fails to capture the early signs of faults in a timely manner.

[0005] Limited monitoring information: Existing monitoring systems have relatively limited functions, mostly focusing on single-dimensional monitoring of electrical or temperature signals, failing to comprehensively perceive anomalies in multiple physical quantities such as acoustics and optics. This single approach leads to insufficient criteria for judging complex faults, easily resulting in misjudgments or missed detections.

[0006] Lack of intelligent analysis and fault-tolerant decision-making: Current monitoring systems typically upload sensor data to a remote control center for analysis by manual or simple threshold judgments, making timely decisions difficult. The lack of advanced intelligent algorithms to fuse multi-source heterogeneous data prevents the full exploitation of correlations between different fault characteristics, resulting in high uncertainty in fault warnings. Summary of the Invention

[0007] To address at least some of the aforementioned problems, this invention provides a method and system for MMC fault detection in wind power generation based on the fusion of acoustic, optical, and electrical spectroscopy.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: The fault detection system based on the fusion of acoustic, optical, and electronic technologies in the wind power MMC system includes an acoustic monitoring unit, an optical monitoring unit, an electrical acquisition unit, and an AI algorithm processing module; The acoustic monitoring unit is used to collect acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit is used to collect the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit is used to acquire the voltage, current and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module is used to perform fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and to perform control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

[0009] A further improvement of the present invention is that the AI ​​algorithm processing module includes a "sudden fault" detection module and an "intelligent feature analysis" module, which performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit and electrical acquisition unit to identify faults, and outputs in two ways: one way directly enters the "sudden fault" detection module to identify short-term high-frequency anomalies; the other way enters the "intelligent feature analysis" module as input for deep feature extraction.

[0010] A further improvement of the present invention is that the acoustic monitoring unit includes one or more acoustic emission sensors and ultrasonic sensors, which are installed on or near the power module package surface of the MMC converter valve to detect acoustic signals generated by cracks, partial discharges and vibrations of cooling components inside the power device.

[0011] A further improvement of the present invention is that the optical monitoring unit includes an infrared thermal imager, an ultraviolet light sensor, and a fiber Bragg grating sensor; the infrared thermal imager is used to monitor the surface temperature distribution of the IGBT chip and power connectors in the MMC submodule to detect local overheating faults; the ultraviolet light sensor is used to detect the partial discharge ultraviolet radiation on the surface of the insulator in the MMC valve hall; the fiber Bragg grating sensor is arranged at key locations in the submodule or busbar to measure temperature changes and realize distributed temperature monitoring.

[0012] A further improvement of the present invention is that the electrical acquisition unit includes a voltage sensor, a current sensor, and an ultra-high frequency partial discharge detection antenna; wherein the voltage sensor and the current sensor are connected to the MMC submodule for real-time acquisition of the submodule capacitor voltage and bridge arm current signals, and the ultra-high frequency partial discharge detection antenna is installed in the converter valve chamber for capturing ultra-high frequency electromagnetic pulse signals generated by partial discharge.

[0013] Fault detection methods based on acoustic-optical-electrical fusion in wind power MMC systems include: The acoustic monitoring unit collects acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit collects the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit acquires the voltage, current, and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and performs control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

[0014] A further improvement of the present invention is that the method specifically includes: First, raw information from three dimensions—electrical, acoustic, and optical—is collected through acoustic monitoring, optical monitoring, and electrical acquisition units. Electrical signals reflect changes in voltage, current, and electromagnetic discharge of the power module; acoustic signals record mechanical vibrations and acoustic emission events; and optical signals monitor temperature distribution and insulation status using infrared, ultraviolet, and fiber optic sensors. The second step involves time synchronization and data alignment, using a unified clock and timestamp mechanism to ensure that all modal data correspond to the same time reference. Following this, the modal preprocessing stage filters, denoises, and standardizes the raw signals. Electrical signals are filtered to remove power frequency interference using bandpass or notch filtering; acoustic signals undergo envelope extraction to enhance weak features; and optical signals are enhanced through radiometric calibration and image enhancement to highlight local hotspots and discharge areas. The processed data is then converted into quantifiable feature vectors during the feature extraction stage: energy, frequency, and waveform are extracted from electrical signals. The system extracts mutation indicators, including pulse counts and energy distribution from acoustic signals, and maximum temperature, hotspot area, and light intensity trends from optical signals. After standardization and alignment, the features of different modalities are uniformly mapped to a shared embedding space, ensuring that each modality expresses similar meanings on the same dimension. Next, the core cross-modal fusion stage begins. The AI ​​algorithm processing module automatically learns the correlations and weights between electrical, acoustic, and optical signals through an attention mechanism. This allows the AI ​​algorithm processing module to adaptively strengthen key modal information according to different fault types, obtaining a fused comprehensive feature representation. The system then determines whether the current equipment is abnormal, identifies the fault type and specific location, and outputs a health score to quantify the fault severity. Finally, the closed-loop and self-learning stage feeds the diagnostic results back to the control system, enabling fault isolation, derated operation, or alarms. Simultaneously, new data is used to update the AI ​​model, allowing the algorithm to continuously optimize and adapt.

[0015] A further improvement of this invention is that the AI ​​algorithm processing module records each partial discharge pulse into a phase distribution spectrum according to its AC phase and discharge signal amplitude to form a partial discharge phase-resolved spectrum, which serves as one of the feature inputs for acoustic and electrical signal fusion to identify insulation discharge faults.

[0016] A further improvement of this invention is that when the AI ​​algorithm processing module receives a trend of continuous decline in the health score of the power module, it enters a preventive fault-tolerant control mode, automatically adjusts the operating parameters of the module to reduce its thermal or electrical stress, and dispatches maintenance personnel to carry out maintenance, thereby preventing the fault from worsening further.

[0017] Compared with the prior art, the present invention has at least the following beneficial technical effects: This invention proposes a non-contact sensing structure based on the fusion of acoustic, optical, and electrical signals, which can effectively improve the fault detection range and resolution of the MMC submodule and adapt to the strong electromagnetic environment of the converter valve. A diagnostic algorithm based on the combination of Transformer and PCNN can handle nonlinear, strongly coupled data structures and maintain high accuracy and stability even under conditions of scarce early fault samples. The health assessment mechanism introduces multi-physical parameter calculation and degradation function modeling, supporting digital expression of module-level health status and quantifying failure risk. A diagnostic deployment method oriented towards an edge processing architecture is proposed, with the algorithm running in real time on the local side, achieving millisecond-level response time and significantly shortening the fault closed-loop control time. A "sensing-analysis-diagnosis-control" system is constructed. The integrated closed-loop framework of "prediction" enhances the system's autonomous perception and operation and maintenance capabilities, reducing reliance on manual inspections; the design incorporates a learning structure with embedded physical mechanisms to improve the credibility and engineering deployability of AI models, adapting to high security standards in the environment; it supports rapid iteration and upgrades of edge-side models, and combines cloud-based expert systems to achieve algorithm generalization and knowledge transfer, effectively supporting the reuse of common fault knowledge across multiple stations; the end-to-end non-intrusive architecture avoids system access interference, does not affect converter station operation, and facilitates modular integration and widespread application; it introduces advanced formula modules such as residual attention mechanisms, multi-task loss functions, and window energy functions, enabling the model to possess expressiveness, controllability, and interpretability, promoting stable implementation in industrial scenarios.

[0018] In summary, this invention provides a forward-looking solution for wind power flexible DC systems with multimodal fusion, intelligent identification, and fault-tolerant control capabilities, and has significant engineering application value and promising prospects for promotion. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific 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 from these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the architecture of the wind power generation MMC fault detection system of the present invention; Figure 2 This is a flowchart of the fault diagnosis method of the present invention; Figure 3 This invention relates to a multimodal feature fusion strategy and fusion mechanism. Figure 4 This is a structural diagram of the wind power generation MMC module fault handling system described in this invention; Figure 5 This is a flowchart of the electro-acoustic-optical fusion AI algorithm processing module described in this invention. Detailed Implementation

[0021] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0022] In the description of this invention, it should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] The accompanying drawings illustrate various structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0026] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] Example 1 The fault detection system based on acoustic-optical-electrical fusion in the wind power MMC system provided by this invention includes an acoustic monitoring unit, an optical monitoring unit, an electrical acquisition unit, and an AI algorithm processing module; The acoustic monitoring unit is used to collect acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit is used to collect the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit is used to acquire the voltage, current and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module is used to perform fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and to perform control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

[0028] In this embodiment, the AI ​​algorithm processing module fuses and analyzes multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, outputting the data in two paths: one path directly enters the "sudden fault" detection module to identify short-term high-frequency anomalies; the other path enters the "intelligent feature analysis" module as input for deep feature extraction. This bidirectional path design reflects the dual role of data layer fusion: rapidly detecting early abnormal signals and achieving immediate alarms; and providing high-fidelity raw input for subsequent "feature layer information fusion" and "decision layer fusion," thereby improving overall diagnostic accuracy.

[0029] To improve the robustness and efficiency of data layer fusion, a multi-modal time window synchronization mechanism is adopted: multi-sensor sampling is controlled by a master clock signal (e.g., FPGA triggering); adaptive weighted fusion is used: the weights of each modality are dynamically allocated based on the signal-to-noise ratio.

[0030] Feature initial screening mechanism: Before fusion, low-relevance dimensions are screened out using principal component analysis (PCA) or mutual information to reduce computational burden. This layer of fusion provides a unified input basis for subsequent "feature layer information fusion" and "decision layer information fusion". When a sudden anomaly is detected, it can quickly enter the "fault detection - fault reproduction - fault isolation" loop through the feedback path, realizing closed-loop intelligent diagnosis from early detection to decision control.

[0031] In this embodiment, the acoustic monitoring unit includes one or more acoustic emission sensors (AE sensors) and ultrasonic sensors. The sensors are installed on or near the power module package surface of the MMC converter valve and are used to detect acoustic signals generated by cracks, partial discharges, and vibrations of cooling components inside the power device.

[0032] In this embodiment, the optical monitoring unit includes an infrared thermal imager, an ultraviolet light sensor, and a fiber Bragg grating (FBG) sensor. The infrared thermal imager is used to monitor the surface temperature distribution of the IGBT chip and power connectors in the MMC submodule to detect local overheating faults. The ultraviolet light sensor is used to detect partial discharge ultraviolet radiation on the surface of the insulator in the MMC valve hall. The fiber Bragg grating (FBG) sensor is arranged at key locations in the submodule or busbar to measure temperature changes and realize distributed temperature monitoring.

[0033] In this embodiment, the electrical acquisition unit includes a voltage sensor, a current sensor, and an ultra-high frequency partial discharge detection antenna; wherein the voltage sensor and the current sensor are connected to the MMC submodule for real-time acquisition of the submodule capacitor voltage and bridge arm current signals, and the ultra-high frequency partial discharge detection antenna is installed in the converter valve chamber for capturing ultra-high frequency electromagnetic pulse signals generated by partial discharge.

[0034] Example 2 The fault detection method based on acoustic-optical-electrical fusion in the wind power MMC system provided by this invention includes: The acoustic monitoring unit collects acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit collects the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit acquires the voltage, current, and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and performs control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

[0035] In this embodiment, the method specifically includes: First, raw information from three dimensions—electrical, acoustic, and optical—is collected through acoustic monitoring units, optical monitoring units, and electrical acquisition units. Electrical signals reflect changes in voltage, current, and electromagnetic discharge of the power module; acoustic signals record mechanical vibrations and acoustic emission events; and optical signals are monitored for temperature distribution and insulation status using infrared, ultraviolet, and fiber optic sensors. The second step involves time synchronization and data alignment, using a unified clock and timestamp mechanism to ensure that all modal data correspond to the same time reference. Subsequently, the modal preprocessing stage involves filtering, denoising, and standardizing the raw signals. Electrical signals are filtered to remove power frequency interference using bandpass or notch filtering; acoustic signals undergo envelope extraction to enhance weak features; and optical signals are calibrated using radiometric calibration and image processing. The system enhances the identification of local hotspots and discharge areas. After processing, the data is converted into quantifiable feature vectors during the feature extraction stage: energy, frequency, and waveform abrupt change indicators are extracted for electrical signals; pulse counts and energy distribution are extracted for acoustic signals; and maximum temperature, hotspot area, and light intensity variation trends are extracted for optical signals. After standardization and alignment, the features of different modes are uniformly mapped to a shared embedding space, ensuring that each mode expresses similar meanings in the same dimension. Next, the core cross-modal fusion stage begins. The AI ​​algorithm processing module automatically learns the correlation and weights between electrical, acoustic, and optical signals through an attention mechanism, enabling the AI ​​model to adaptively strengthen key modal information according to different fault types, obtaining a fused comprehensive feature representation. The fusion result is input to the fault identification module, which determines whether the current equipment is abnormal, identifies the fault type and specific location (e.g., IGBT open circuit, short circuit, capacitor breakdown, or partial discharge), and outputs a health score to quantify the fault severity. Finally, the closed-loop and self-learning stage feeds the diagnostic results back to the control system, enabling fault isolation, derated operation, or alarms. Simultaneously, new data is used to update the AI ​​model, allowing for continuous algorithm optimization and adaptation. The entire process realizes a closed loop from "multi-source perception - feature fusion - intelligent diagnosis - self-learning optimization", providing core technical support for high-precision fault detection and intelligent operation and maintenance of wind power MMC systems.

[0036] This embodiment demonstrates how a typical multimodal intelligent fault diagnosis system can achieve cross-modal information fusion and decision-making in wind power MMC equipment. First, three types of signals are simultaneously acquired: electrical signals (including voltage, current, and UHF partial discharge characteristics), acoustic signals (vibration and acoustic emission information acquired by AE sensors or ultrasonic transducers), and optical signals (infrared thermography, ultraviolet radiation, and FBG fiber optic sensing data). Subsequently, these signals are synchronized through unified time triggering and sampling frequency alignment to ensure that the data from different modes are synchronized under the same time reference, laying the foundation for subsequent feature mapping and fusion.

[0037] In the preprocessing stage, each modal signal undergoes denoising and calibration: electrical signals are bandpass or notch filtered to suppress power frequency interference; acoustic signals undergo pre-amplification, bandpass, and Hilbert transform to extract the envelope; and optical signals are radiometrically calibrated and image-cropped to highlight the target region. Next, key features are extracted for each modality: short-time Fourier energy, wavelet features, harmonic components, and PRPD spectral parameters are extracted for electrical signals; pulse counts, energy spectrum, and MFCC coefficients are extracted for acoustic signals; and maximum temperature rise, temperature gradient, hotspot area, and UV scintillation frequency are extracted for optical signals. The extracted features are then normalized, missing data is filled in, and dimensionally aligned to generate a unified multimodal feature matrix.

[0038] Subsequently, features from different modalities are fed into dedicated encoding networks: electrical signals are processed by 1D-CNN or Transformer-Encoder to extract temporal features, acoustic signals are processed by 1D-CNN / LSTM to capture short-term and long-term dependencies, and optical signals are processed by lightweight 2D-CNN or backbone networks to extract spatial features. The resulting embedding vectors are then input into a cross-modal fusion layer, which is the core of the entire process. This layer utilizes a self-attention mechanism to achieve Transformer-level fusion, learning the correlation between modalities through cross-attention, and automatically adjusting the weights of different signals in the weighted concatenation and residual structure, so that the model focuses on the modalities and time segments most relevant to the fault. If necessary, physical constraints (such as thermo-electrical consistency) can be introduced to enhance the interpretability of the model.

[0039] The fused features are fed into the classification and localization module. The model outputs the specific fault type (such as IGBT open circuit, short circuit, capacitor breakdown, partial discharge, etc.) and its corresponding location (submodule number or bridge arm position), and simultaneously generates confidence or health scores to assess the equipment's operating status. When an anomaly is detected, the results are transmitted back to the edge or cloud database for event logging and knowledge accumulation. Continuous data and label writing supports model retraining and adaptive threshold updates, thus constructing a closed-loop process of "multi-source acquisition - fusion analysis - fault identification - feedback learning." Overall, this algorithm module, through deep learning feature encoding and cross-modal fusion, enables the coordinated expression of electrical, acoustic, and optical information in time and space, significantly improving the fault identification accuracy, localization capability, and long-term stability of the wind power MMC system.

[0040] Example 3 Figure 1This is a functional structure diagram of the fault detection system. It includes five parts: an acoustic monitoring subsystem, an optical monitoring subsystem, an electrical quantity acquisition subsystem, an edge intelligent processing unit, and a diagnostic control interface module. The acoustic monitoring subsystem uses a multi-channel acoustic emission and ultrasonic sensor array deployed on the surface of the MMC converter valve submodule housing to identify high-frequency vibration signals generated by bonding wire detachment, module packaging defects, or localized arc discharge. Each sensor channel is equipped with a preamplifier and bandpass filter circuit to improve the signal-to-noise ratio and directional resolution. The optical monitoring subsystem uses a multi-point temperature and vibration sensing unit based on a fiber Bragg grating (FBG). The FBG sensor is small and can be embedded in the submodule busbar, the thermally sensitive area of ​​the IGBT chip, and the top of the capacitor to sense temperature changes in real time. The system uses a demodulator to convert wavelength signals and utilizes the center wavelength drift. With temperature or strain Linear relationship:

[0041] in, and To enhance sensitivity, temperature and vibration are decoupled into dual channels. The FBG sensor also boasts advantages such as electromagnetic interference resistance and remote fiber optic transmission, making it particularly suitable for embedded deployment within high-voltage, high-interference power systems. The electrical quantity acquisition subsystem is equipped with a high-frequency broadband current transformer and a voltage isolation amplifier module, combined with an ultra-high frequency (UHF) partial discharge detection antenna to capture high-frequency pulses generated by module capacitor breakdown, insulation aging, etc. The electrical quantity signal is filtered and gain adjusted by the sampling and conditioning module before being sent to a high-speed AD chip for digitization and input to the edge processing platform. As shown in the figure, the system supports flexible deployment at key potential nodes such as generators, circuit breakers, transformers, and transmission line (TL) buses, unconstrained by potential, comprehensively improving the reliability, spatial adaptability, and parameter sensing capabilities of equipment monitoring.

[0042] Figure 2This is a flowchart of the fault detection method based on acoustic-optical-optical fusion in the wind power MMC system described in this invention. Its structural logic includes five main stages: initialization, signal acquisition, multimodal fusion processing, diagnostic judgment, and maintenance response. The functions of each stage are as follows: The "Start" node indicates the initiation of the system fault detection process, entering the initialization stage; the "Initialization" module is used to complete the power-on, self-test, and communication initialization of various acoustic, optical, and electrical sensors, and simultaneously set the synchronous sampling frequency of the edge processing unit and the sampling window size of each channel to ensure consistency of modal collaborative acquisition; the "Detect Electrical Signal," "Detect Optical Signal," and "Detect Acoustic Signal" modules execute in parallel, respectively calling the voltage / current / EMI sub-channel, infrared / ultraviolet / FBG optical channel, and acoustic emission / ultrasonic acoustic channel to complete real-time data acquisition; the "AD Sampling" unit performs high-precision analog-to-digital conversion on the analog signal, outputting a standardized time-series digital signal stream; the "Electroacoustic-optical fusion AI algorithm processing module" is the core of this invention. The system's core analysis module integrates preprocessing, modal alignment, feature extraction, Transformer cross-modal fusion, and physical prior enhancement models, uniformly outputting two key results: fault identification results and equipment health scores. The process then enters a dual-path decision structure: if the "fault identification" module detects an anomaly, it proceeds to the decision node "whether it exceeds the normal value"; if the result is "Y (yes)," it is determined to be a major fault, and the maintenance process begins; otherwise, the system continues running and returns to the upper layer for further monitoring. If the "health score" module's output value is lower than a set threshold, it indicates a potential degradation risk to the module, and the system maintenance and inspection process also begins. The "system maintenance and repair" module triggers fault-tolerant control mechanisms (such as module bypass, current redistribution, derating operation, etc.), simultaneously reporting diagnostic results to the station control or cloud system and recording abnormal data for subsequent model training and trend analysis. The "end" node indicates the completion of this round of diagnostic loop, and the system enters the next round of detection cycle.

[0043] Figure 3 This invention relates to a multi-sensor information fusion-based transmission system fault diagnosis method and system. In the diagram, the "data layer information fusion" module is located at the very front, immediately following "sensor 1…N," and is in the raw signal convergence stage. This module directly receives synchronously sampled data from multiple sources of sensors, including electrical, acoustic, and optical sensors, and completes the fusion of raw signals before entering the "intelligent feature analysis" stage. The function of this layer is to: align the raw data from different types of sensors (current, voltage, acoustic emission, spectrum, temperature, etc.) in the dimensions of time, space, and sampling rate; construct a multimodal joint input matrix to provide a unified data foundation for subsequent feature and decision layers; and achieve rapid response and early warning for "sudden faults" such as short-term arcing and insulation breakdown.

[0044] At the data layer, early fusion is often achieved through the following methods: time synchronization splicing: timestamp alignment of multi-source time series, splicing of each modality data into a unified input matrix according to time windows.

[0045]

[0046] Channel stacking fusion: For image or 2D waveform data, the outputs of different sensors are stacked into a multi-channel input tensor, for example, combining infrared, ultraviolet, and spectrogram images into a three-channel input. Multi-sampling rate normalization: Interpolation or downsampling techniques enhance the comparability of sampled data at different frequencies in the time domain. Noise suppression and normalization: Filtering (such as Kalman filtering and wavelet denoising) and Z-score normalization are applied before fusion to improve the robustness of the fused signal. These operations ensure that the input obtained from data layer fusion contains the richest underlying information, helping subsequent modules discover cross-modal detail correlations during the learning phase.

[0047] Figure 4 This is a structural diagram of the wind power generation MMC module fault handling system described in this invention. The system first collects raw information from three dimensions: electrical, acoustic, and optical, using multiple sensors. Electrical signals reflect changes in the power module's voltage, current, and electromagnetic discharge; acoustic signals record mechanical vibrations and acoustic emission events; and optical signals monitor temperature distribution and insulation status using infrared, ultraviolet, and fiber optic sensors. Because the sampling frequencies and trigger times of different sensors are inconsistent, the system performs time synchronization and data alignment in the second step. Through a unified clock and timestamp mechanism, the data from each mode correspond under the same time reference, thereby ensuring that subsequent algorithms can accurately identify cross-modal correlations.

[0048] Figure 5 This is a flowchart of the electro-acoustic-optical fusion AI algorithm processing module described in this invention, demonstrating how a typical multimodal intelligent fault diagnosis system achieves cross-modal information fusion and decision-making in wind power MMC equipment. The system first simultaneously acquires three types of signals: electrical signals (including voltage, current, and UHF partial discharge characteristics), acoustic signals (vibration and acoustic emission information acquired by AE sensors or ultrasonic transducers), and optical signals (infrared thermography, ultraviolet radiation, and FBG fiber optic sensing data). Subsequently, these signals undergo unified time triggering and sampling frequency alignment to ensure that different modal data are synchronized under the same time reference.

[0049] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0050] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. 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 based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A fault detection system based on acoustic-optical-electrical fusion in a wind power MMC system, characterized in that, It includes an acoustic monitoring unit, an optical monitoring unit, an electrical data acquisition unit, and an AI algorithm processing module; The acoustic monitoring unit is used to collect acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit is used to collect the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit is used to acquire the voltage, current and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module is used to perform fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and to perform control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

2. The fault detection system based on acoustic-optical-electric fusion in the wind power MMC system according to claim 1, characterized in that, The AI ​​algorithm processing module includes a "sudden fault" detection module and an "intelligent feature analysis" module. It performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and outputs the data in two ways: one way goes directly to the "sudden fault" detection module to identify short-term high-frequency anomalies; the other way goes to the "intelligent feature analysis" module as input for deep feature extraction.

3. The fault detection system based on acoustic-optical-electric fusion in the wind power MMC system according to claim 1, characterized in that, The acoustic monitoring unit includes one or more acoustic emission sensors and ultrasonic sensors, which are installed on or near the power module package surface of the MMC converter valve to detect acoustic signals generated by cracks, partial discharges, and vibrations of cooling components inside the power device.

4. The fault detection system based on acoustic-optical-electric fusion in the wind power MMC system according to claim 1, characterized in that, The optical monitoring unit includes an infrared thermal imager, an ultraviolet light sensor, and a fiber Bragg grating sensor. The infrared thermal imager is used to monitor the surface temperature distribution of the IGBT chips and power connectors in the MMC submodule to detect local overheating faults. The ultraviolet light sensor is used to detect the partial discharge ultraviolet radiation on the surface of the insulator in the MMC valve hall. The fiber Bragg grating sensor is arranged at key locations in the submodule or busbar to measure temperature changes and realize distributed temperature monitoring.

5. The fault detection system based on acoustic-optical-electric fusion in the wind power MMC system according to claim 1, characterized in that, The electrical acquisition unit includes a voltage sensor, a current sensor, and an ultra-high frequency partial discharge detection antenna; The voltage and current sensors are connected to the MMC submodule to collect the capacitor voltage and bridge arm current signals of the submodule in real time. The ultra-high frequency partial discharge detection antenna is installed in the converter valve hall to capture the ultra-high frequency electromagnetic pulse signal generated by partial discharge.

6. A fault detection method based on acoustic-optical-electrical fusion in a wind power MMC system, characterized in that, include: The acoustic monitoring unit collects acoustic emission signals and ultrasonic signals from the power module inside the MMC converter valve; The optical monitoring unit collects the infrared temperature field and insulation discharge ultraviolet radiation signal of the power module; The electrical acquisition unit acquires the voltage, current, and partial discharge electromagnetic signals of the power module; The AI ​​algorithm processing module performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and performs control on the MMC system based on the diagnostic results of the AI ​​algorithm processing module.

7. The fault detection method based on acoustic-optical-electric fusion in the wind power MMC system according to claim 6, characterized in that, The method specifically includes: First, raw information from three dimensions—electrical, acoustic, and optical—is collected through acoustic monitoring, optical monitoring, and electrical acquisition units. Electrical signals reflect changes in voltage, current, and electromagnetic discharge of the power module; acoustic signals record mechanical vibrations and acoustic emission events; and optical signals monitor temperature distribution and insulation status using infrared, ultraviolet, and fiber optic sensors. The second step involves time synchronization and data alignment, using a unified clock and timestamp mechanism to ensure that all modal data correspond to the same time reference. Following this, the modal preprocessing stage filters, denoises, and standardizes the raw signals. Electrical signals are filtered to remove power frequency interference using bandpass or notch filtering; acoustic signals undergo envelope extraction to enhance weak features; and optical signals are enhanced through radiometric calibration and image enhancement to highlight local hotspots and discharge areas. The processed data is then converted into quantifiable feature vectors during the feature extraction stage: energy, frequency, and waveform are extracted from electrical signals. The system extracts mutation indicators, including pulse counts and energy distribution from acoustic signals, and maximum temperature, hotspot area, and light intensity trends from optical signals. After standardization and alignment, the features of different modalities are uniformly mapped to a shared embedding space, ensuring that each modality expresses similar meanings on the same dimension. Next, the core cross-modal fusion stage begins. The AI ​​algorithm processing module automatically learns the correlations and weights between electrical, acoustic, and optical signals through an attention mechanism. This allows the AI ​​algorithm processing module to adaptively strengthen key modal information according to different fault types, obtaining a fused comprehensive feature representation. The system then determines whether the current equipment is abnormal, identifies the fault type and specific location, and outputs a health score to quantify the fault severity. Finally, the closed-loop and self-learning stage feeds the diagnostic results back to the control system, enabling fault isolation, derated operation, or alarms. Simultaneously, new data is used to update the AI ​​model, allowing the algorithm to continuously optimize and adapt.

8. The fault detection method based on acoustic-optical-electric fusion in the wind power MMC system according to claim 7, characterized in that, The AI ​​algorithm processing module records each partial discharge pulse into a phase distribution spectrum according to its AC phase and discharge signal amplitude to form a partial discharge phase-resolved spectrum, which serves as one of the feature inputs for acoustic and electrical signal fusion to identify insulation discharge faults.

9. The fault detection method based on acoustic-optical-electric fusion in the wind power MMC system according to claim 7, characterized in that, When the AI ​​algorithm processing module receives a trend of a continuous decline in the health score of the power module, it enters a preventive fault-tolerant control mode, automatically adjusts the operating parameters of the module to reduce its thermal or electrical stress, and dispatches maintenance personnel to carry out repairs, thereby preventing the fault from worsening further.

10. The fault detection method based on acoustic-optical-electric fusion in the wind power MMC system according to claim 7, characterized in that, The AI ​​algorithm processing module includes a "sudden fault" detection module and an "intelligent feature analysis" module. It performs fusion analysis on multimodal data from the acoustic monitoring unit, optical monitoring unit, and electrical acquisition unit to identify faults, and outputs the data in two ways: one way goes directly to the "sudden fault" detection module to identify short-term high-frequency anomalies; the other way goes to the "intelligent feature analysis" module as input for deep feature extraction.

Images