Electromagnetic coil diagnostic method, apparatus, device, medium, and program product

Abstract

The application provides an electromagnetic coil diagnosis method, device, equipment, medium and program product, and particularly relates to the technical field of electromagnetic coil health diagnosis. The method comprises the following steps: acquiring multi-source heterogeneous monitoring data of a target electromagnetic coil, wherein the multi-source heterogeneous monitoring data comprises static scalar data reflecting inherent characteristics of the coil and dynamic waveform data representing the running state of the coil; converting the static scalar data into a Gaussian distribution heat map and converting the dynamic waveform data into a kernel density estimation heat map; performing multi-dimensional feature joint analysis based on the Gaussian distribution heat map and the kernel density estimation heat map in combination with a preset intelligent diagnosis model to obtain a health state detection result for the target electromagnetic coil. The method is used to improve the accuracy and robustness of the health state diagnosis of the electromagnetic coil.

Description

Technical Field

[0001] This application relates to the field of electromagnetic coil health diagnosis technology, and in particular to an electromagnetic coil diagnosis method, apparatus, equipment, medium and program product. Background Technology

[0002] Electromagnetic coils, as core components in power systems, industrial automation, and transportation equipment, are widely used in transformers, relays, solenoid valves, motors, and other devices. Their health status directly affects the stability of power systems, the operating efficiency of industrial equipment, and the safety of transportation equipment. For example, in power systems, insulation aging or short circuits in electromagnetic coils can lead to transformer failures and large-scale power outages; in industrial automation, coil performance degradation can cause production line shutdowns, affecting production efficiency; and in rail transit, coil failure can lead to signal control malfunctions, threatening passenger safety. However, existing health diagnosis methods suffer from poor multi-source data fusion capabilities and insufficient feature extraction, making it difficult to achieve high-precision, real-time fault warnings under complex operating conditions. Summary of the Invention

[0003] This application provides methods, apparatus, devices, media, and program products for diagnosing electromagnetic coils, in order to improve the accuracy and robustness of diagnosing the health status of electromagnetic coils.

[0004] In a first aspect, embodiments of this application provide an electromagnetic coil diagnostic method, including:

[0005] Acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state.

[0006] Convert static scalar data into Gaussian distribution heatmaps and dynamic waveform data into kernel density estimation heatmaps;

[0007] Based on Gaussian distribution heatmaps and kernel density estimation heatmaps, and combined with a pre-set intelligent diagnostic model, multi-dimensional feature joint analysis is performed to obtain the health status detection results for the target electromagnetic coil.

[0008] Secondly, embodiments of this application provide an electromagnetic coil diagnostic device, comprising:

[0009] The acquisition module is used to acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state.

[0010] The conversion module is used to convert static scalar data into Gaussian distribution heatmaps and dynamic waveform data into kernel density estimation heatmaps.

[0011] The analysis module is used to perform multi-dimensional feature joint analysis based on Gaussian distribution heatmap and kernel density estimation heatmap, combined with a preset intelligent diagnostic model, to obtain the health status detection results for the target electromagnetic coil.

[0012] Thirdly, embodiments of this application provide an electromagnetic coil diagnostic device, including: a memory and a processor;

[0013] The memory stores the instructions that the computer executes;

[0014] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0017] The electromagnetic coil diagnostic method, apparatus, device, medium, and program products provided in this application address the differences in data format between static scalar data and dynamic waveform data by employing Gaussian distribution heatmaps and kernel density estimation heatmaps for unified graphical representation. This eliminates modal barriers between multi-source heterogeneous data and achieves compatible expression of heterogeneous data within the same feature space. The Gaussian distribution heatmap of static scalar data preserves the probability distribution characteristics of the coil's inherent properties, while the kernel density estimation heatmap of dynamic waveform data depicts the time-frequency evolution of the operating state. The fusion of these two heatmaps forms a highly complementary and information-density multi-channel input, enhancing the diagnostic model's comprehensive perception of the coil's health status and improving the accuracy and robustness of electromagnetic coil health status diagnosis. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] Figure 1 Flowchart of the electromagnetic coil diagnostic method provided in this application Figure 1 ;

[0020] Figure 2 Flowchart of the electromagnetic coil diagnostic method provided in this application Figure 2 ;

[0021] Figure 3 A schematic diagram of the overall process of the electromagnetic coil diagnostic method provided in this application;

[0022] Figure 4 A schematic diagram of the electromagnetic coil diagnostic device provided in this application;

[0023] Figure 5 This is a schematic diagram of the electromagnetic coil diagnostic device provided in this application.

[0024] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0026] Currently, the health status diagnosis of electromagnetic coils mainly relies on monitoring and analyzing single-type sensor signals. Common signal types include physical quantities such as current, temperature, voltage, and coil travel. Traditional diagnostic methods can be categorized as follows:

[0027] 1. Threshold-based alarm method

[0028] This method sets safety thresholds for key parameters such as coil current, temperature, or resistance, triggering an alarm when the monitored value exceeds the set threshold. While simple to implement and highly real-time, this method is extremely sensitive to environmental noise and parameter fluctuations, easily leading to false alarms or missed alarms, and is ill-suited to adapting to dynamic changes under complex operating conditions.

[0029] 2. Signal processing methods based on time-frequency analysis

[0030] This method utilizes signal analysis techniques such as Fourier Transform (FFT) and Wavelet Transform (WT) to extract time-frequency domain features from coil current or voltage waveforms to identify abnormal states such as insulation aging, coil short circuits, or core faults. While this method can reveal some non-stationary features, its ability to model multi-dimensional signal correlations under complex operating conditions is limited, and it requires high signal quality.

[0031] 3. Classification and recognition methods based on traditional machine learning

[0032] With the improvement of sensor sampling accuracy, some studies have adopted algorithms such as Support Vector Machine (SVM), Decision Tree (DT), and K-Nearest Neighbors (KNN) to classify and model the extracted time-domain and frequency-domain features in order to achieve automatic identification of different fault modes. The above models perform well in small-scale and specific working conditions, but they require manually designed feature engineering, have poor model transferability, and limited generalization ability.

[0033] 4. Deep Learning-Based Automatic Feature Extraction and Intelligent Recognition Methods

[0034] In recent years, deep learning architectures such as Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and Autoencoders (AE) have been introduced into the field of electromagnetic coil health monitoring for automatic feature extraction and state recognition of various types of data, including time-series signals, infrared images, and vibration spectra. Compared to traditional methods, deep learning models have strong nonlinear mapping and feature abstraction capabilities, but they typically rely on a large number of labeled samples and are highly sensitive to model structure and training parameters, resulting in high deployment costs.

[0035] In summary, the existing technical solutions have the following technical defects:

[0036] 1. Insufficient ability to fuse multi-source heterogeneous data

[0037] Existing methods typically construct diagnostic models for single signal types, such as judging the state based solely on current waveforms or temperature changes. This makes it difficult to simultaneously integrate heterogeneous monitoring data from multiple sources, including temperature, current, voltage, and travel. Different types of signals differ significantly in time scale, noise characteristics, and physical meaning. Without a unified framework for heterogeneous data fusion, key diagnostic information will be underutilized, reducing the overall robustness and accuracy of the diagnosis.

[0038] 2. Limited ability to express dynamic waveform features

[0039] For the dynamic response characteristics of complex electromagnetic systems, one-dimensional time-series signals are difficult to fully express their spatiotemporal distribution features in traditional convolutional or recursive networks. Especially under non-stationary and nonlinear dynamic conditions, single-dimensional convolutional operations cannot simultaneously capture the global evolution trend and local transient changes of the signal, resulting in feature extraction results lacking physical interpretability and fault mode discrimination.

[0040] 3. Weak ability to process static scalar data

[0041] The operating state of an electromagnetic coil is affected not only by dynamic signals but also by static scalar parameters such as temperature rise, insulation resistance, and leakage current. However, mainstream deep learning models (such as CNN and LSTM) are mostly designed for time-series or image data, and their ability to fuse such static numerical features is relatively weak. Existing research usually requires the introduction of additional feature concatenation, attention mechanisms, or multimodal fusion structures, resulting in high algorithm complexity, large computational overhead, and reduced real-time performance.

[0042] 4. Insufficient generalization ability of the model across working conditions

[0043] In practical industrial applications, the operating environment of electromagnetic coils is complex and variable, with significant differences in power supply frequency, sampling frequency, and noise levels under different operating conditions. Most existing models are only trained and validated under specific experimental conditions, lacking the ability to adapt across devices and operating conditions. When external interference increases or system parameters drift, the diagnostic performance of the models often deteriorates significantly, making it difficult to meet the reliability requirements for long-term stable operation in industrial scenarios.

[0044] Therefore, this application provides an electromagnetic coil diagnostic method. Addressing the significant differences in data format, dimensionality, and physical meaning among different types of monitoring data such as temperature, current, and voltage, this method unifies multi-source heterogeneous data into a standardized heatmap representation, achieving compatible expression and alignment of heterogeneous data within the same visual feature space. To address the significant differences in data form, temporal characteristics, and information dimensions between dynamic waveform data and static scalar parameters, this method constructs kernel density estimation heatmaps and Gaussian distribution heatmaps separately, and then stacks and fuses them at the channel level. This effectively combines dynamic temporal features with static state features, improving the completeness and complementarity of diagnostic information. To address the gradient vanishing, insufficient feature extraction, and unstable training issues that traditional deep learning models often encounter when processing electromagnetic coil monitoring data, this method introduces a deep residual learning mechanism using residual networks. A skip connection structure is constructed to optimize the gradient backpropagation path, achieving efficient extraction and robust learning of deep discriminative features from multi-channel heatmaps. To address the problem that existing methods can only identify faults after they have become significant and are difficult to detect early signs of coil aging, this paper proposes a multi-dimensional feature joint analysis and a quantitative health scoring mechanism to achieve sensitive perception and early warning of latent coil faults and early performance degradation.

[0045] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0046] Figure 1 Flowchart of the electromagnetic coil diagnostic method provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0047] S101. Acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state.

[0048] In this step, multi-dimensional monitoring data of the target electromagnetic coil is collected. Static scalar data is used to describe the steady-state physical properties of the coil, while dynamic waveform data is used to record the transient operation process of the coil. Together, they form the data basis for health status assessment.

[0049] S102. Convert static scalar data into Gaussian distribution heatmaps and dynamic waveform data into kernel density estimation heatmaps.

[0050] In this step, heterogeneous monitoring data are uniformly represented graphically: for static scalar data, a probability distribution heatmap centered on the measured value is generated using a Gaussian function; for dynamic waveform data, a time-frequency domain heatmap reflecting the signal density distribution is generated using kernel density estimation, thereby achieving compatible representation of multi-source data in the same feature space.

[0051] S103. Based on the Gaussian distribution heatmap and kernel density estimation heatmap, and combined with the preset intelligent diagnostic model, multi-dimensional feature joint analysis is performed to obtain the health status detection results for the target electromagnetic coil.

[0052] In this step, the Gaussian distribution heatmap and kernel density estimation heatmap are input into the preset intelligent diagnostic model. Through deep network, static and dynamic features are jointly extracted and fused for analysis, and the quantitative assessment results of the coil health status are output, realizing a complete mapping from data representation to diagnostic decision.

[0053] The electromagnetic coil diagnostic method provided in this application addresses the differences in data form between static scalar data and dynamic waveform data by employing Gaussian distribution heatmaps and kernel density estimation heatmaps for unified graphical representation. This eliminates the modal barriers between multi-source heterogeneous data and achieves compatible representation of heterogeneous data within the same feature space. The Gaussian distribution heatmap of static scalar data preserves the probability distribution characteristics of the coil's inherent properties, while the kernel density estimation heatmap of dynamic waveform data depicts the time-frequency evolution of the operating state. The fusion of these two heatmaps forms a highly complementary and information-density multi-channel input, enhancing the diagnostic model's comprehensive perception of the coil's health status and improving the accuracy and robustness of electromagnetic coil health status diagnosis.

[0054] Figure 2 Flowchart of the electromagnetic coil diagnostic method provided in this application Figure 2 ,like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the electromagnetic coil diagnostic method is described in detail, which includes:

[0055] S201. Acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state.

[0056] In this step, raw monitoring data of the target electromagnetic coil is collected, including at least one of temperature rise and time-voltage waveform. The raw monitoring data is then normalized to eliminate the influence of differences in the dimensions of different physical quantities on subsequent analysis, resulting in normalized multi-source heterogeneous monitoring data.

[0057] S202. Convert static scalar data into Gaussian distribution heatmaps and dynamic waveform data into kernel density estimation heatmaps.

[0058] In one possible implementation, converting static scalar data into a Gaussian distribution heatmap may specifically include the following steps:

[0059] Based on the numerical distribution characteristics of static scalar data, the coordinates of the center point and the distribution amplitude of the two-dimensional network are determined.

[0060] Based on the center point coordinates and distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated using the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix.

[0061] A Gaussian distribution heatmap is generated by color encoding a two-dimensional Gaussian distribution matrix using a pseudo-color mapping function.

[0062] In this embodiment, the scalar measurement value μ (e.g., a temperature rise of 65K) is mapped to the center of a two-dimensional grid, serving as the coordinates of the center point of the Gaussian distribution. Based on a defined standard deviation σ, the intensity value of each pixel (i, j) on an N×N (e.g., 64×64) two-dimensional grid G ​​is calculated using the following formula. :

[0063] ;

[0064] Where A is the distribution amplitude, which is usually normalized to 1.

[0065] The two-dimensional Gaussian distribution matrix is ​​color-coded using pseudo-color mapping functions such as the Viridis color map or the Inferno color map, generating a colored Gaussian distribution heatmap with a continuous color gradient centered on the scalar measurement value μ. The color coding process can be represented as follows:

[0066] .

[0067] In one possible implementation, the intensity value of each pixel in the two-dimensional network is calculated using a Gaussian distribution function based on the center point coordinates and the distribution amplitude, thereby generating a two-dimensional Gaussian distribution matrix. This may specifically include the following steps:

[0068] Calculate the first statistical property of the static scalar data, and adjust the standard deviation of the Gaussian distribution based on the first statistical property;

[0069] Based on the adjusted standard deviation, center point coordinates and distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated by the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix.

[0070] The first statistical characteristic includes at least one of variance, maxima, and skewness.

[0071] The first statistical characteristic refers to the statistical quantities that characterize the dispersion and distribution pattern of static scalar data, including variance (reflecting the degree of data fluctuation), kurtosis (reflecting the sharpness of the data distribution), and skewness (reflecting the degree of asymmetry in the data distribution). When the variance or kurtosis of the input data increases, it indicates that the data dispersion has increased or there are more extreme values. In this case, the standard deviation σ should be increased accordingly to expand the perception range of the Gaussian distribution and avoid over-focusing on local areas. Conversely, when the variance or kurtosis decreases, it indicates that the data distribution is concentrated. In this case, the standard deviation σ should be decreased to enhance the local detail resolution capability of the heatmap and improve the sensitivity to subtle changes. By introducing a data-driven dynamic parameter adjustment strategy, the generation process of the Gaussian distribution heatmap can adaptively match the statistical characteristics of the input data, thereby optimizing the flexibility and discriminativeness of feature representation and providing higher-quality input for feature extraction in subsequent deep models.

[0072] In one possible implementation, converting dynamic waveform data into a kernel density estimation heatmap may specifically include the following steps:

[0073] The dynamic waveform data is normalized to form a normalized waveform curve;

[0074] On a two-dimensional plane consisting of time and amplitude dimensions, the probability density value of the normalized waveform curve in each pixel region is calculated based on the kernel density estimation method to generate a density matrix.

[0075] A kernel density estimation heatmap is generated by color encoding the density matrix using a pseudo-color mapping function.

[0076] In this embodiment, time series data such as stroke-current waveform, coil current-time waveform, and impulse voltage waveform are used. First, normalize the waveform so that its time axis t is mapped to the range [0,T] and its amplitude axis A is mapped to the range [0,1], thus obtaining the normalized waveform curve. Subsequently, the normalized waveform curve is plotted on a two-dimensional plane P consisting of time (horizontal axis) and amplitude (vertical axis). Treating the curve as a continuous path, the probability density value of each pixel region (i, j) is calculated using the kernel density estimation method, generating a density matrix D. The density values ​​of each element in density matrix D are... Calculate using the following formula:

[0077] ;

[0078] Where K(·) is the kernel function (such as a Gaussian kernel), h is the bandwidth parameter used to control the smoothness of density diffusion, and dist(·) is the distance from pixel coordinates (i, j) to waveform data points. A function of distance, This represents the total number of data points.

[0079] The density matrix D is color-coded using a pseudo-color mapping function to generate the final color kernel density estimation heatmap. The color coding process can be represented as follows:

[0080] .

[0081] In one possible implementation, on a two-dimensional plane consisting of a time dimension and an amplitude dimension, the probability density value of the normalized waveform curve in each pixel region is calculated based on the kernel density estimation method to generate a density matrix. Specifically, this may include the following steps:

[0082] Calculate the second statistical property of the dynamic waveform data, and adjust the bandwidth parameter of the kernel density estimation based on the second statistical property;

[0083] Based on the adjusted bandwidth parameters, the probability density value of the normalized waveform curve in each pixel region is calculated using the kernel density estimation method on a two-dimensional plane consisting of time and amplitude dimensions, and a density matrix is ​​generated.

[0084] The second statistical characteristic includes at least one of variance, maxima, and skewness.

[0085] The second statistical characteristic refers to the statistical quantities that characterize the dispersion, sharpness, and asymmetry of dynamic waveform data, including variance (reflecting the degree of waveform amplitude fluctuation), kurtosis, and skewness (reflecting the degree of asymmetry of the waveform distribution relative to the mean). When the variance or kurtosis of the input waveform increases, it indicates that the signal fluctuates violently or there are many transient impacts. In this case, the bandwidth parameter h is increased accordingly to expand the smoothing range of the kernel function, reduce noise sensitivity, and retain the overall trend characteristics. Conversely, when the variance or kurtosis decreases, it indicates that the signal is stable and rich in details. In this case, the bandwidth parameter h is decreased to improve the resolution of kernel density estimation and enhance the ability to capture subtle local changes. By introducing a data-driven dynamic parameter adjustment strategy, the generation process of the kernel density estimation heatmap can adaptively match the statistical characteristics of the dynamic waveform data, optimize the flexibility and discriminativeness of time-frequency feature representation, provide higher quality input for feature extraction of subsequent deep models, and thus improve the detection capability of early faults and transient anomalies in electromagnetic coils.

[0086] S203. In the preset intelligent diagnostic model, Gaussian distribution heatmap and kernel density estimation heatmap are stacked and fused in the channel dimension to construct a multi-channel input tensor.

[0087] This step achieves spatial alignment and channel fusion of heterogeneous data. The Gaussian distribution heatmap representing the inherent characteristics of the coil and the kernel density estimation heatmap representing the operating state are stacked as two independent channels to form a unified input representation that combines static attributes and dynamic behavior information, laying the data foundation for subsequent joint feature extraction.

[0088] S204. Input the multi-channel input tensor into the residual network, and use the deep residual learning mechanism of the residual network to jointly extract static scalar features and temporal dynamic features to obtain a multi-dimensional fused feature vector.

[0089] This step utilizes the deep residual learning mechanism of residual networks to perform end-to-end feature extraction and abstract mapping on the dual-channel input: local spatial features are extracted using convolutional layers, and the gradient vanishing problem is alleviated through skip connection structures, achieving deep fusion and high-level semantic representation of static scalar features and temporal dynamic features, and outputting a multi-dimensional fused feature vector with strong discriminativeness.

[0090] S205. Classify and evaluate the health status of electromagnetic coils based on multi-dimensional fusion feature vectors, and generate health status detection results.

[0091] This step maps multi-dimensional fused feature vectors to the health status decision space, and realizes diagnostic decisions through classification and scoring mechanisms.

[0092] Before using a pre-defined intelligent diagnostic model for diagnostic decisions, a training dataset needs to be constructed to train the model. Specifically, multiple generated experimental heatmaps can be stacked along the channel dimension to form a multi-channel tensor sample. The process of constructing the multi-channel tensor sample can be represented as follows:

[0093] ;

[0094] This represents the k-th multichannel tensor sample; This is a function for stacking operations along the channel dimension; This represents the M different types of heatmaps corresponding to the k-th sample.

[0095] Multiple experimental heatmaps are included, comprising at least two of the following: temperature rise sinusoidal distribution heatmap, stroke-current kernel density estimation heatmap, and current-time kernel density estimation heatmap. Each multi-channel tensor sample corresponds to a state label, which characterizes the normal state, mild aging state, severe aging state, or failure state of the electromagnetic coil. The set of state label values ​​can be represented as follows:

[0096] ;

[0097] The state label corresponding to the k-th multichannel tensor sample is represented by C, which is the true classification label; C represents the total number of health status categories, which is the number of categories in the classification task.

[0098] A residual network is adopted as the core architecture of the model. The residual network contains multiple residual blocks. Each residual block achieves the fusion of identity mapping and convolution transformation through skip connections. Its basic computational unit is represented as follows:

[0099] ;

[0100] Where x is the residual block input and y is the residual block output; Let represent the residual mapping function, and represent the feature obtained by performing a nonlinear transformation on the input x; where is the set of weight parameters for the convolutional layer, where i represents the layer index.

[0101] The cross-entropy loss function is used to measure the difference between the predicted output and the true label. The gradient is calculated using the backpropagation algorithm, and the Adam optimizer is used for end-to-end training to optimize the network parameters to minimize the loss function value, resulting in the trained preset intelligent diagnostic model. The loss function formula can be shown below:

[0102] ;

[0103] in, The value represents the loss function, indicating the degree of difference between the model's predicted output and the true label; N is the total number of samples. This indicates summing over samples 1 to N; This indicates summing over categories 1 through C; This represents the true label of the k-th sample in the i-th category; This represents the predicted probability of the k-th sample in the i-th category.

[0104] In one possible implementation, the health status of electromagnetic coils is classified and assessed based on multi-dimensional fused feature vectors to generate health status detection results, which may specifically include the following steps:

[0105] The multi-dimensional fused feature vector is input into the fully connected layer, and the output is a multi-class probability distribution vector. Each element in the multi-class probability distribution vector corresponds to the confidence probability of different health status levels of the coil.

[0106] Based on the preset scoring weighting formula, the multi-class probability distribution vectors are weighted and summed to obtain the comprehensive score of coil health status;

[0107] Based on the comparison between the comprehensive health status score and the preset threshold range, the health status level of the target electromagnetic coil is determined to generate a health status detection result.

[0108] Specifically, the fully connected layer contains four neurons, corresponding to four health state levels. A four-class probability distribution vector [P1, P2, P3, P4] is generated using the Softmax activation function, where P1, P2, P3, and P4 represent the confidence probabilities of a healthy state, a mildly aged state, a severely aged state, and a failed state, respectively, and satisfy P1+P2+P3+P4=1. For example, when the generated four-class probability distribution vector is [P1, P2, P3, P4]=[0.80, 0.09, 0.06, 0.05], it indicates that the coil is in a healthy state with a confidence level of 80%.

[0109] The weighted scoring formula can be expressed as:

[0110] ;

[0111] in, , , , These are the weighting coefficients corresponding to each health status level.

[0112] With weighting coefficients =0.99、 =0.70、 =0.50、 Taking = 0.01 as an example, when the four-class probability distribution vector is [0.80, 0.09, 0.06, 0.05]:

[0113] ;

[0114] According to the comparison result of the comprehensive health score Score and the preset threshold interval, the health status level is determined:

[0115] When 0.70 < Score < 0.99, it is determined as a healthy state, the coil is in the best working state, and all performance parameters are better than the standard requirements;

[0116] When 0.50 < Score ≤ 0.70, it is determined as a mild aging state. Early signs of performance decay begin to appear in the coil, but it can still meet the basic operation requirements. This stage is the critical window period for predictive maintenance;

[0117] When 0.10 < Score ≤ 0.50, it is determined as a severe aging state. The coil performance has severely declined and is close to or has reached the operation limit, and there is a risk of failure in the short term;

[0118] When 0.01 ≤ Score ≤ 0.10, it is determined as a failure state. The coil has lost its main function, cannot operate normally, or there are safety hazards.

[0119] Refer to Figure 3 As shown, it is the overall flow schematic diagram of the electromagnetic coil diagnosis method provided by this application. Multi-source heterogeneous monitoring data is used as the system input, including static scalar data (such as temperature rise, insulation resistance, leakage current, etc.) reflecting the inherent characteristics of the coil and dynamic waveform data (such as stroke-current waveform, current-time waveform, impulse voltage waveform, etc.) characterizing its operating state. The static scalar data is processed by the Gaussian distribution function to generate a Gaussian distribution heat map; the dynamic waveform data is normalized and kernel density estimated to generate a kernel density estimation heat map. This layer realizes the unified graphical representation of heterogeneous data and eliminates the modality barrier. The Gaussian distribution heat map and the kernel density estimation heat map are stacked in the channel dimension to construct a multi-channel input tensor; the multi-channel input tensor is input into the residual network, and the deep residual learning mechanism is used to jointly extract the static scalar features and time-domain dynamic features to obtain a multi-dimensional fusion feature vector. Based on the multi-dimensional fusion feature vector, health status classification and evaluation are carried out: the multi-class probability distribution vector is output through the fully connected layer and the Softmax activation function, the comprehensive health score is calculated through the scoring weighted formula, and finally the health status level is determined according to the preset threshold interval to generate the health status detection result.

[0120] The electromagnetic coil diagnostic method provided in this application uses Gaussian distribution heatmaps and kernel density estimation heatmaps to provide unified visual representations of static scalar data and dynamic waveform data, respectively. This transforms multimodal heterogeneous signals into standardized visual features, achieving compatible expression and deep fusion of heterogeneous data within the same feature space, thus overcoming the bottleneck of limited information dimensions from a single signal source. The heatmap representation not only preserves the numerical information of the original data but also enhances the visual discriminative power of features through color gradients and spatial distribution, providing a more discriminative input representation for deep learning models and facilitating the extraction of deep, highly abstract fault-sensitive features. Employing a residual network as the core architecture, the method mitigates the gradient vanishing problem in deep networks through a skip connection mechanism, enabling effective learning of complex feature distributions and improving the model's generalization ability and diagnostic stability under noise interference and changing operating conditions. From multi-source data acquisition, heatmap generation, deep feature extraction to health status determination, the entire process is automated, eliminating the need for manual feature engineering design, significantly improving diagnostic efficiency and result consistency, and reducing engineering deployment costs. The generated heatmap presents the signal feature distribution in a visual form, enabling technicians to intuitively understand the basis of the model's decision-making, enhancing the credibility and acceptability of the intelligent diagnostic system, and facilitating fault tracing and maintenance decisions in engineering practice.

[0121] The electromagnetic coil diagnostic method provided in this application addresses the differences in data form between static scalar data and dynamic waveform data by employing Gaussian distribution heatmaps and kernel density estimation heatmaps for unified graphical representation. This eliminates the modal barriers between multi-source heterogeneous data and achieves compatible representation of heterogeneous data within the same feature space. The Gaussian distribution heatmap of static scalar data preserves the probability distribution characteristics of the coil's inherent properties, while the kernel density estimation heatmap of dynamic waveform data depicts the time-frequency evolution of the operating state. The fusion of these two heatmaps forms a highly complementary and information-density multi-channel input, enhancing the diagnostic model's comprehensive perception of the coil's health status and improving the accuracy and robustness of electromagnetic coil health status diagnosis.

[0122] Figure 4 This is a schematic diagram of the electromagnetic coil diagnostic device provided in this application, as shown below. Figure 4 As shown, the electromagnetic coil diagnostic device 40 provided in this embodiment includes:

[0123] The acquisition module 401 is used to acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state.

[0124] The conversion module 402 is used to convert static scalar data into Gaussian distribution heatmaps and dynamic waveform data into kernel density estimation heatmaps.

[0125] Analysis module 403 is used to perform multi-dimensional feature joint analysis based on Gaussian distribution heatmap and kernel density estimation heatmap, combined with a preset intelligent diagnostic model, to obtain the health status detection results for the target electromagnetic coil.

[0126] In one possible implementation, the analysis module is specifically used for:

[0127] In the preset intelligent diagnostic model

[0128] The Gaussian distribution heatmap and the kernel density estimation heatmap are stacked and fused along the channel dimension to construct a multi-channel input tensor;

[0129] The multi-channel input tensor is fed into the residual network, and the deep residual learning mechanism of the residual network is used to jointly extract static scalar features and temporal dynamic features to obtain a multi-dimensional fused feature vector.

[0130] The health status of electromagnetic coils is classified and evaluated based on multi-dimensional fusion feature vectors, and health status detection results are generated.

[0131] In one possible implementation, the analysis module is specifically used for:

[0132] The multi-dimensional fused feature vector is input into the fully connected layer, and the output is a multi-class probability distribution vector. Each element in the multi-class probability distribution vector corresponds to the confidence probability of different health status levels of the coil.

[0133] Based on the preset scoring weighting formula, the multi-class probability distribution vectors are weighted and summed to obtain the comprehensive score of coil health status;

[0134] Based on the comparison between the comprehensive health status score and the preset threshold range, the health status level of the target electromagnetic coil is determined to generate a health status detection result.

[0135] In one possible implementation, the conversion module is specifically used for:

[0136] Based on the numerical distribution characteristics of static scalar data, the coordinates of the center point and the distribution amplitude of the two-dimensional network are determined.

[0137] Based on the center point coordinates and distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated using the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix.

[0138] A Gaussian distribution heatmap is generated by color encoding a two-dimensional Gaussian distribution matrix using a pseudo-color mapping function.

[0139] In one possible implementation, the conversion module is specifically used for:

[0140] The dynamic waveform data is normalized to form a normalized waveform curve;

[0141] On a two-dimensional plane consisting of time and amplitude dimensions, the probability density value of the normalized waveform curve in each pixel region is calculated based on the kernel density estimation method to generate a density matrix.

[0142] A kernel density estimation heatmap is generated by color encoding the density matrix using a pseudo-color mapping function.

[0143] In one possible implementation, the conversion module is specifically used for:

[0144] Calculate the first statistical property of the static scalar data, and adjust the standard deviation of the Gaussian distribution based on the first statistical property;

[0145] Based on the adjusted standard deviation, center point coordinates and distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated by the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix.

[0146] The first statistical characteristic includes at least one of variance, maxima, and skewness.

[0147] In one possible implementation, the conversion module is specifically used for:

[0148] Calculate the second statistical property of the dynamic waveform data, and adjust the bandwidth parameter of the kernel density estimation based on the second statistical property;

[0149] Based on the adjusted bandwidth parameters, the probability density value of the normalized waveform curve in each pixel region is calculated using the kernel density estimation method on a two-dimensional plane consisting of time and amplitude dimensions, and a density matrix is ​​generated.

[0150] The second statistical characteristic includes at least one of variance, maxima, and skewness.

[0151] The electromagnetic coil diagnostic device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0152] Figure 5 This is a schematic diagram of the electromagnetic coil diagnostic device provided in this application. Figure 5 As shown, the electromagnetic coil diagnostic device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 further includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus.

[0153] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0154] The specific implementation process of processor 501 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0155] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0156] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0157] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0158] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0159] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0160] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0161] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0162] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0163] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0164] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0165] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0166] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0167] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for diagnosing electromagnetic coils, characterized in that, include: Acquire multi-source heterogeneous monitoring data of the target electromagnetic coil, wherein the multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state; The static scalar data is converted into a Gaussian distribution heatmap, and the dynamic waveform data is converted into a kernel density estimation heatmap. Based on the Gaussian distribution heatmap and the kernel density estimation heatmap, a multi-dimensional feature joint analysis is performed using a preset intelligent diagnostic model to obtain the health status detection results for the target electromagnetic coil.

2. The method according to claim 1, characterized in that, The method involves combining the Gaussian distribution heatmap and the kernel density estimation heatmap with a preset intelligent diagnostic model to perform multi-dimensional feature joint analysis, thereby obtaining the health status detection results for the target electromagnetic coil, including: In the preset intelligent diagnostic model, The Gaussian distribution heatmap and the kernel density estimation heatmap are stacked and fused along the channel dimension to construct a multi-channel input tensor; The multi-channel input tensor is input into the residual network, and the deep residual learning mechanism of the residual network is used to jointly extract static scalar features and temporal dynamic features to obtain a multi-dimensional fused feature vector. The health status of electromagnetic coils is classified and evaluated based on the multi-dimensional fused feature vector, and the health status detection results are generated.

3. The method according to claim 2, characterized in that, The process of classifying and evaluating the health status of electromagnetic coils based on the multi-dimensional fused feature vectors, and generating the health status detection results, includes: The multi-dimensional fused feature vector is input into the fully connected layer, and a multi-class probability distribution vector is output. Each element in the multi-class probability distribution vector corresponds to the confidence probability of different health status levels of the coil. Based on a preset scoring weighting formula, the multi-class probability distribution vector is weighted and summed to obtain a comprehensive score of coil health status. Based on the comparison between the comprehensive health status score and the preset threshold range, the health status level of the target electromagnetic coil is determined to generate the health status detection result.

4. The method according to any one of claims 1-3, characterized in that, The step of converting the static scalar data into a Gaussian distribution heatmap includes: Based on the numerical distribution characteristics of the static scalar data, the coordinates of the center point and the distribution amplitude of the two-dimensional network are determined. Based on the center point coordinates and distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated using the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix. The Gaussian distribution heatmap is generated by color encoding the two-dimensional Gaussian distribution matrix using a pseudo-color mapping function.

5. The method according to any one of claims 1-3, characterized in that, The step of converting the dynamic waveform data into a kernel density estimation heatmap includes: The dynamic waveform data is normalized to form a normalized waveform curve; On a two-dimensional plane consisting of time and amplitude dimensions, the probability density value of the normalized waveform curve in each pixel region is calculated based on the kernel density estimation method to generate a density matrix. The density matrix is ​​color-coded using a pseudo-color mapping function to generate the kernel density estimation heatmap.

6. The method according to claim 4, characterized in that, The step of calculating the intensity value of each pixel in the two-dimensional network based on the center point coordinates and distribution amplitude using a Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix includes: Calculate the first statistical characteristic of the static scalar data, and adjust the standard deviation of the Gaussian distribution based on the first statistical characteristic; Based on the adjusted standard deviation, the coordinates of the center point, and the distribution amplitude, the intensity value of each pixel in the two-dimensional network is calculated using the Gaussian distribution function to generate a two-dimensional Gaussian distribution matrix. The first statistical characteristic includes at least one of variance, maxima, and skewness.

7. The method according to claim 5, characterized in that, The process of calculating the probability density value of the normalized waveform curve in each pixel region on a two-dimensional plane consisting of time and amplitude dimensions, and generating a density matrix based on the kernel density estimation method, includes: Calculate the second statistical characteristic of the dynamic waveform data, and adjust the bandwidth parameter of the kernel density estimation based on the second statistical characteristic; Based on the adjusted bandwidth parameters, on a two-dimensional plane consisting of time and amplitude dimensions, the probability density value of the normalized waveform curve in each pixel region is calculated using the kernel density estimation method to generate a density matrix. The second statistical characteristic includes at least one of variance, maxima, and skewness.

8. An electromagnetic coil diagnostic device, characterized in that, include: The acquisition module is used to acquire multi-source heterogeneous monitoring data of the target electromagnetic coil. The multi-source heterogeneous monitoring data includes static scalar data reflecting the inherent characteristics of the coil and dynamic waveform data characterizing its operating state. The conversion module is used to convert the static scalar data into a Gaussian distribution heatmap and the dynamic waveform data into a kernel density estimation heatmap. The analysis module is used to perform multi-dimensional feature joint analysis based on the Gaussian distribution heatmap and the kernel density estimation heatmap, combined with a preset intelligent diagnostic model, to obtain the health status detection results for the target electromagnetic coil.

9. An electromagnetic coil diagnostic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium or computer program product, characterized in that, A computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as claimed in any one of claims 1-7; or, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images