Wireless charging equipment fault detection method and system based on artificial intelligence

Abstract

The invention discloses a wireless charging equipment fault detection method and system based on artificial intelligence, and the method comprises the steps: analyzing a spatial mode of thermal imaging data through a 3D convolutional layer in a first path based on a DP-CNN dual-path convolutional neural network, and analyzing a time sequence mode of electromagnetic field data through an LSTM network in a second path, constructing a dual-path feature fusion network model through the contribution degree of attention mechanism weighted features; optimizing hyper-parameters of the dual-path feature fusion network model by using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; and inputting the equipment feature data into the target dual-path feature fusion network model, outputting a fault probability index, and generating a diagnosis report according to the fault probability index and a preset confidence threshold. According to the invention, fine features of various faults can be accurately captured, the accuracy of fault detection is greatly improved, and the occurrence of leak detection and false detection is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless charging equipment, and in particular to a method and system for detecting faults in wireless charging equipment based on artificial intelligence. Background Art

[0002] Failures in in-vehicle and portable wireless charging devices are becoming increasingly common. Existing fault detection technologies rely primarily on sensors collecting basic parameters and determining anomalies using preset thresholds. Traditional threshold methods are poorly adaptable to complex operating conditions and are prone to false positives or omissions, limiting the accuracy of wireless charging device detection. Fault identification relies on offline analysis, which cannot meet real-time requirements, resulting in slow response times and low detection efficiency. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a wireless charging device fault detection method and system based on artificial intelligence.

[0004] The technical solution of the present invention to achieve the above object is that, further, in the above-mentioned wireless charging device fault detection method based on artificial intelligence, the wireless charging device fault detection method includes the following steps: Collect data from multiple charging devices in real time, align the time series of the data using the DTW dynamic time warping algorithm, and extract electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data; Based on the DP-CNN dual-path convolutional neural network, the first path uses 3D convolutional layers to analyze the spatial patterns of thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series patterns of electromagnetic field data. The contribution of features is weighted through the attention mechanism to construct a dual-path feature fusion network model. Optimizing the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; The device feature data is input into the target dual-path feature fusion network model, a fault probability index is output, and a diagnosis report is generated according to the fault probability index and a preset confidence threshold.

[0005] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the real-time collection of multi-source charging device data, the time series alignment of the multi-source charging device data using the DTW dynamic time warping algorithm, and the extraction of electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data include: Real-time collection of electromagnetic field distribution data, thermal imaging data, and power transmission efficiency data of wireless charging devices to obtain multi-source charging device data; The power transmission efficiency data with the highest sampling frequency was selected as the reference sequence. The Euclidean distance between the electromagnetic field distribution data and thermal imaging data and each data point in the reference sequence was calculated to construct a distance matrix. Through the dynamic programming method, the path with the minimum cumulative distance is found in the distance matrix. The minimum path is the optimal matching path, and the aligned data sequence is obtained.

[0006] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the real-time collection of multi-source charging device data, the time series alignment of the multi-source charging device data using the DTW dynamic time warping algorithm, and the extraction of electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data include: By calculating the deviation between the actual electromagnetic field distribution in the aligned data sequence and the normal state, the sum of the absolute values ​​of the deviation values ​​is used as the electromagnetic field distortion characteristic value; Calculate the temperature difference between each pixel in the thermal imaging data and its surrounding adjacent pixels, take the absolute value and average it as the thermal distribution gradient value of the pixel, perform statistics on the gradient values ​​of the thermal imaging area, and extract the thermal distribution gradient characteristics; The standard deviation of the power transmission efficiency data within the time window is calculated, and the coefficient of variation is the ratio of the standard deviation to the mean value, thus obtaining the power fluctuation characteristics.

[0007] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the DP-CNN dual-path convolutional neural network is based on a first path that uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data, and a second path that uses an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data. The contribution of features is weighted through an attention mechanism to construct a dual-path feature fusion network model, including: The 3D convolutional layer is used to convert the time-aligned thermal imaging data into a 3D cube structure. Each cube contains 16 consecutive frames of thermal images, preserving the temporal continuity and spatial distribution characteristics of temperature changes. The continuous electromagnetic field data is divided into subsequences according to fixed time windows through the LSTM long short-term memory network. Each subsequence contains 200 data points to obtain time features.

[0008] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the DP-CNN dual-path convolutional neural network is based on a first path that uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data, and a second path that uses an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data. The contribution of features is weighted through an attention mechanism to construct a dual-path feature fusion network model, including: The one-dimensional vector output by the 3D convolutional layer and the vector output by the LSTM network are used as the input of the attention mechanism respectively, and the attention weight of each feature is calculated; The features of the two paths are weighted and summed according to the attention weight to obtain a fused feature vector; the fused feature vector is passed through a fully connected layer to output the fault probability index.

[0009] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the Bayesian optimization algorithm is used to optimize the hyperparameters of the dual-path feature fusion network model to obtain the target dual-path feature fusion network model, including: A set of hyperparameter combinations is randomly generated as the initial sample points. The dual-path feature fusion network model is trained using this hyperparameter combination, and the loss function value of the model is calculated by cross-validation as the evaluation indicator. A Gaussian process surrogate model is constructed based on the existing sample points and corresponding evaluation indicators, and the expected improvement criterion is used according to the surrogate model to select the hyperparameter combination of the optimized evaluation indicators.

[0010] Furthermore, in the above-mentioned artificial intelligence-based wireless charging device fault detection method, the device feature data is input into the target dual-path feature fusion network model, a fault probability index is output, and a diagnostic report is generated based on the fault probability index and a preset confidence threshold, including: If the fault probability index is greater than the preset confidence threshold, the fault type is determined based on the range of the fault probability index, and the fault type, fault probability index and fault location are described in the diagnostic report.

[0011] Furthermore, in the artificial intelligence-based wireless charging device fault detection system, the wireless charging device fault detection system includes the following modules: A feature data extraction module is used to collect multi-source charging device data in real time, align the time series of the multi-source charging device data using the DTW dynamic time warping algorithm, and extract electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data; A fusion model building module is used to build a dual-path convolutional neural network based on the DP-CNN. The first path uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data. The attention mechanism is used to weight the contribution of features to build a dual-path feature fusion network model. A fusion model optimization module is used to optimize the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; The device fault diagnosis module is used to input the device feature data into the target dual-path feature fusion network model, output a fault probability index, and generate a diagnosis report based on the fault probability index and a preset confidence threshold.

[0012] Furthermore, in the artificial intelligence-based wireless charging device fault detection system, the fusion model optimization module includes the following submodules: The generation submodule is used to randomly generate a set of hyperparameter combinations as initial sample points, use the hyperparameter combination to train the dual-path feature fusion network model, and use cross-validation to calculate the model's loss function value as an evaluation indicator; The selection submodule is used to construct a Gaussian process proxy model based on the existing sample points and corresponding evaluation indicators, and select the hyperparameter combination of the optimization evaluation indicators according to the expected improvement criterion of the proxy model.

[0013] Furthermore, in the artificial intelligence-based wireless charging device fault detection system, the device fault diagnosis module includes the following submodules: The diagnosis submodule is used to determine the fault type according to the range of the fault probability index if the fault probability index is greater than a preset confidence threshold, and to describe the fault type, fault probability index and fault location in the diagnosis report.

[0014] Its beneficial effects are: 1. It can accurately capture the subtle characteristics of various types of faults, greatly improving the accuracy of fault detection and effectively reducing the occurrence of missed detections and false detections. 2. It can quickly complete data time series alignment, feature extraction, and fault probability index calculation. Compared with traditional manual detection or single indicator detection methods, it shortens the time for fault detection, can promptly detect potential equipment faults, and gain valuable time for equipment maintenance. 3. It can effectively reduce equipment maintenance costs. Through accurate and timely fault detection, it can avoid the expansion of equipment damage caused by untimely fault detection, reduce unnecessary maintenance costs and equipment replacement costs. At the same time, the automated detection process reduces dependence on manual labor, reduces the labor and time costs of manual detection, and improves the economic efficiency of equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0016] Figure 1 Schematic diagram of a first embodiment of a method for detecting faults of a wireless charging device based on artificial intelligence in an embodiment of the present invention; Figure 2Schematic diagram of a second embodiment of a method for detecting faults of a wireless charging device based on artificial intelligence in an embodiment of the present invention; Figure 3 Schematic diagram of a first embodiment of an artificial intelligence-based wireless charging device fault detection system in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] It will be understood by those skilled in the art that, unless otherwise stated, the singular forms "a", "an", and "the" used herein may also include the plural forms. It should be further understood that the terms "include" used in the specification of the present invention refer to the presence of features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0019] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 As shown, a wireless charging device fault detection method based on artificial intelligence includes the following steps: Step 101: Real-time data collection from multiple charging devices is performed. The time series of the data is aligned using the DTW dynamic time warping algorithm. Electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics are extracted to obtain device feature data. Specifically, in this embodiment, electromagnetic field distribution data, thermal imaging data, and power transmission efficiency data of wireless charging devices are collected in real time to obtain multi-source charging device data; The power transmission efficiency data with the highest sampling frequency was selected as the reference sequence. The Euclidean distance between the electromagnetic field distribution data and thermal imaging data and each data point in the reference sequence was calculated to construct a distance matrix. Through the dynamic programming method, the path with the minimum cumulative distance is found in the distance matrix. The minimum path is the optimal matching path, and the aligned data sequence is obtained.

[0020] By calculating the deviation between the actual electromagnetic field distribution in the aligned data sequence and the normal state, the sum of the absolute values ​​of the deviation values ​​is used as the electromagnetic field distortion characteristic value; Calculate the temperature difference between each pixel in the thermal imaging data and its surrounding adjacent pixels, take the absolute value and average it as the thermal distribution gradient value of the pixel, perform statistics on the gradient values ​​of the thermal imaging area, and extract the thermal distribution gradient characteristics; The standard deviation of the power transmission efficiency data within the time window is calculated, and the coefficient of variation is the ratio of the standard deviation to the mean value, thus obtaining the power fluctuation characteristics.

[0021] Specifically, 1. Real-time collection of data from multiple charging devices During the real-time collection phase, a variety of high-precision equipment must be used to ensure that the data obtained is accurate and comprehensive.

[0022] Electromagnetic field distribution data collection: A high-precision electromagnetic field sensor array consisting of 32 sensors is distributed within a circular area with a radius of 1.5 meters centered on the wireless charging device. The array collects electromagnetic field distribution data around the device every 0.1 seconds. The sensors have a measurement range of 0-100μT and an accuracy of ±0.1μT, accurately covering the device's primary electromagnetic field radiation area during operation.

[0023] Thermal imaging data acquisition: Utilizing a high-resolution infrared thermal imager with a resolution of 640 x 512 pixels and a thermal sensitivity of up to 0.05°C, thermal imaging data is acquired at 30 frames per second. This capture covers the wireless charging device's transmitter, receiver, and key circuit components, ensuring detailed temperature changes across all device components are captured.

[0024] Power transmission efficiency data collection: A professional power meter monitors power transmission efficiency data in real time. The power meter has a measurement range of 0-500W, an accuracy of ±0.5%, and a sampling interval of 0.05 seconds. The power meter is connected to the input and output of the wireless charging device to ensure data continuity and accuracy, and can reflect efficiency changes during power transmission in real time.

[0025] 2. Data timing alignment and feature extraction.

[0026] (1) Data timing alignment.

[0027] In view of the different sampling frequencies and time series lengths of multi-source charging equipment data, the DTW dynamic time warping algorithm is used to achieve data time series alignment. The specific steps are as follows: Determine the reference sequence: Select the power transmission efficiency data with the highest sampling frequency as the reference sequence.

[0028] Calculate the distance matrix: For the electromagnetic field distribution data and thermal imaging data, calculate the Euclidean distance between them and each data point in the reference sequence and construct a distance matrix.

[0029] Finding the optimal path: Using dynamic programming methods, we search for the path with the smallest cumulative distance in the distance matrix. This path is the optimal matching path, which enables alignment of different data sequences.

[0030] Parameter setting: The window size in the DTW algorithm is set to 10% of the reference sequence length to balance computational efficiency and alignment accuracy.

[0031] (2) Feature extraction.

[0032] Electromagnetic field distortion feature extraction: Establish a normal database: Collect electromagnetic field distribution data of the equipment under different workloads (low power, medium power, high power) in a fault-free state, and build a normal electromagnetic field feature library as a benchmark for judging distortion.

[0033] Calculating the degree of distortion: The real-time collected and aligned electromagnetic field data is compared point by point with the electromagnetic field data of the corresponding operating conditions in the normal database. The area and magnitude of deviation from the normal range are calculated. For example, if the electromagnetic field intensity in a certain area is more than 30% higher than the normal state and lasts for more than 5 seconds, it is marked as a significantly distorted area.

[0034] Feature quantification: Integrate indicators such as the total area of ​​the distortion region, the maximum deviation amplitude, and the average deviation amplitude into a comprehensive electromagnetic field distortion characteristic value. The larger the value, the more obvious the electromagnetic field anomaly.

[0035] Thermal distribution gradient feature extraction: Temperature zone division: Thermal imaging data is divided into multiple key areas according to the device structure, such as the transmitting coil area, receiving coil area, control circuit area, etc., and the temperature changes in each area are analyzed separately.

[0036] Gradient calculation: For each area, the temperature difference between adjacent pixels is calculated to reflect the speed of temperature change. For example, in the transmitting coil area, if the temperature difference between two adjacent points exceeds 2°C, it indicates that there is a significant temperature gradient in the area.

[0037] Feature integration: Count the maximum temperature gradient, average temperature gradient, and number of pixels with dramatic gradient changes in each area. This data is aggregated into thermal distribution gradient features to determine whether the device has local overheating or heat dissipation anomalies.

[0038] Power fluctuation feature extraction: Fluctuation interval division: Using 1 minute as a time window, the power transmission efficiency data is segmented and analyzed to capture the power changes in different time periods.

[0039] Fluctuation indicator calculation: Calculate the difference between the maximum and minimum power data values ​​within each time window (i.e., the fluctuation amplitude), the trend of power values ​​changing for more than three consecutive times (continuous increase, continuous decrease, or fluctuating highs and lows), and the number of times and duration that the power value deviates from the average efficiency.

[0040] Feature summary: The fluctuation amplitude, change trend stability, and other indicators in each time window are combined into power fluctuation characteristics. If the fluctuation amplitude exceeds the normal range (±5%) and the number of consecutive fluctuations is high, it indicates that there may be abnormalities in power transmission.

[0041] Step 102: Based on the DP-CNN dual-path convolutional neural network, the first path uses a 3D convolutional layer to analyze the spatial pattern of the thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series pattern of the electromagnetic field data. The contribution of the features is weighted through the attention mechanism to construct a dual-path feature fusion network model. Specifically, in this embodiment, a 3D convolution layer is used to convert the time-aligned thermal imaging data into a three-dimensional cube structure. Each cube contains 16 consecutive frames of thermal images, preserving the temporal continuity and spatial distribution characteristics of temperature changes. The continuous electromagnetic field data is divided into subsequences according to fixed time windows through the LSTM long short-term memory network. Each subsequence contains 200 data points to obtain time features.

[0042] The one-dimensional vector output by the 3D convolutional layer and the vector output by the LSTM network are used as the input of the attention mechanism respectively, and the attention weight of each feature is calculated; The features of the two paths are weighted and summed according to the attention weight to obtain a fused feature vector; the fused feature vector is passed through a fully connected layer to output the fault probability index.

[0043] Specifically, The first path: 3D convolution layer analyzes the spatial pattern of thermal imaging data. The core function of the 3D convolution layer is to capture the spatial correlation and dynamic changes of temperature distribution from the time series data of thermal imaging. The specific process is as follows: Data input processing: The time-aligned thermal imaging data is converted into a three-dimensional cube structure. Each cube contains 16 consecutive frames of thermal images, and the size of each frame is 64×64 pixels, so as to preserve the temporal continuity and spatial distribution characteristics of temperature changes.

[0044] Multi-layer feature extraction: The first convolution layer uses 16 3×3×3 convolution kernels to focus on capturing the temperature gradient changes in local areas, such as the hot spot diffusion trend near the coil; After 2×2×2 maximum pooling, redundant information is filtered out and key temperature features are retained; The second convolution layer uses 32 convolution kernels of the same size to further explore temperature correlation patterns between different regions, such as the temperature conduction relationship between the heat sink and the core components; Finally, global average pooling is used to compress the three-dimensional features into a fixed-length vector, highlighting abnormal spatial patterns in thermal distribution.

[0045] Second path: LSTM network analyzes the time series pattern of electromagnetic field data.

[0046] The LSTM network focuses on capturing the dynamic changes of electromagnetic field data over time and is particularly good at identifying abnormal patterns hidden in long-term sequences: Sequence segmentation processing: Split the continuous electromagnetic field data into subsequences according to a fixed time window (10 seconds). Each subsequence contains 200 data points to ensure that the typical fault evolution cycle is fully covered. Memory unit design: The first LSTM layer has 64 memory cells, which are responsible for extracting high-frequency fluctuation features, such as sudden changes in the electromagnetic field when the device is started; The second LSTM layer has 32 memory units, focusing on capturing low-frequency trend features, such as the attenuation law of the electromagnetic field after long-term operation; Automatically filter irrelevant noise through the forget gate mechanism, focusing on retaining key change patterns related to faults; Feature output: converted into a feature vector that matches the dimension of the 3D convolutional layer through the fully connected layer, which is convenient for subsequent fusion processing.

[0047] Attention mechanism and feature fusion.

[0048] The attention mechanism can dynamically adjust the contribution weights of different features, allowing the model to focus more on information that plays a key role in fault diagnosis: Weight calculation logic: Each feature is assigned a learnable weight parameter, which is automatically adjusted by comparing the strength of the feature's association with historical fault samples. For example, when localized high temperatures appear in thermal imaging, the weight of the corresponding feature is significantly increased. Fusion strategy: A combination of weighted summation and feature concatenation is used to preserve the unique features of a single path while strengthening cross-path correlation information. For example, the features of "high temperature in the coil area" and "electromagnetic field distortion at the corresponding position" are correlated and enhanced. Output layer optimization: The fused features are deeply processed through a multi-layer perceptron, and the normalized fault probability index is finally output to ensure that the judgment criteria for different types of faults are consistent.

[0049] Step 103: Optimize the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; Specifically, in this embodiment, a set of hyperparameter combinations is randomly generated as initial sample points, the dual-path feature fusion network model is trained using the hyperparameter combination, and the loss function value of the model is calculated by cross-validation as an evaluation indicator; A Gaussian process surrogate model is constructed based on the existing sample points and corresponding evaluation indicators, and the expected improvement criterion is used according to the surrogate model to select the hyperparameter combination of the optimized evaluation indicators.

[0050] Specifically, 1. Parameter determination, Determine the optimization goal: Use the fault detection accuracy of the dual-path feature fusion network model as the core optimization goal, while also taking into account the model's training efficiency to avoid excessive pursuit of accuracy that leads to overly complex models and prolonged training time.

[0051] Defining hyperparameter ranges: The number of convolution kernels in the 3D convolution layer is set to 12-20 in the first layer and 28-36 in the second layer. This range ensures the feature extraction capability without increasing the computational burden due to an excessive number of kernels.

[0052] The number of hidden units in the LSTM network is between 56 and 72 in the first layer and between 28 and 36 in the second layer, which can adapt to electromagnetic field time series patterns of different complexities.

[0053] Dropout rate: range is 0.1-0.3, in order to find a balance between preventing overfitting and retaining effective features.

[0054] Learning rate: between 0.0005 and 0.005, so that the model can converge quickly during training while avoiding convergence to the local optimal solution.

[0055] Batch size: Optional values ​​are 24, 40, and 56. These values ​​are suitable for the data scale of the model and can balance training speed and memory usage.

[0056] 2. The specific process of Bayesian optimization.

[0057] Initial sample selection: Randomly select 10 different hyperparameter combinations within the defined hyperparameter range as the initial samples. These samples are then fed into the dual-path feature fusion network model for training. The model fault detection accuracy and training time corresponding to each hyperparameter combination are recorded.

[0058] Build a surrogate model: Based on the initial sample's hyperparameter combinations and their corresponding model performance metrics (accuracy and training time), a Gaussian process surrogate model is constructed. This model simulates the relationship between hyperparameters and model performance and can predict the performance of untried hyperparameter combinations.

[0059] Select the next set of hyperparameters: Based on the surrogate model, an expected improvement strategy is used to select the next set of hyperparameters. This strategy considers the potential performance improvement of each hyperparameter combination and the uncertainty of the prediction, prioritizing those hyperparameter combinations that are likely to significantly improve model performance.

[0060] Model training and updating the proxy model: The selected hyperparameter combination is input into the model for training to obtain new performance indicators. This set of hyperparameters and performance indicators are added to the sample set and the proxy model is updated.

[0061] Repeat the process of selecting hyperparameters, training the model, and updating the surrogate model until 50 iterations have been completed. At this point, the surrogate model should have accurately simulated the relationship between hyperparameters and model performance.

[0062] 3. Determine the optimal hyperparameter combination.

[0063] After all iterations are complete, the optimal hyperparameter combination with the highest fault detection accuracy and acceptable training time is selected from all sample hyperparameter combinations. This optimal hyperparameter combination is then applied to the dual-path feature fusion network model to obtain the target dual-path feature fusion network model. This model achieves high fault detection accuracy while also achieving high training and runtime efficiency, better meeting practical application requirements.

[0064] Step 104: Input the device feature data into the target dual-path feature fusion network model, output the fault probability index, and generate a diagnosis report based on the fault probability index and a preset confidence threshold.

[0065] Specifically, in this embodiment, if the fault probability index is greater than a preset confidence threshold, the fault type is determined according to the range of the fault probability index, and the fault type, fault probability index and fault location are described in the diagnosis report.

[0066] Specifically, 1. Preprocessing of device feature data, Before inputting the device feature data into the target dual-path feature fusion network model, a series of preprocessing operations are required to ensure the validity and consistency of the data.

[0067] Data cleaning: Check feature data for outliers, such as values ​​significantly outside the reasonable range due to sensor failure. These outliers are corrected using interpolation. This involves estimating a reasonable value based on the normal data before and after the data point to prevent the outlier from interfering with the model output.

[0068] Data standardization: Because different types of feature data (electromagnetic field distortion, thermal distribution gradient, and power fluctuation) have different dimensions and numerical ranges, they need to be uniformly converted to the same numerical range (0-1). By calculating the maximum and minimum values ​​of each feature data and normalizing it according to the formula (data - minimum) / (maximum - minimum), the model can treat different features more fairly, improving model stability and accuracy.

[0069] 2. The target model outputs the failure probability index.

[0070] The preprocessed device feature data is input into the target dual-path feature fusion network model. The model analyzes and processes the data based on its internal computational logic, ultimately outputting a failure probability index. This index is a value between 0 and 1, with values ​​closer to 1 indicating a greater likelihood of device failure; values ​​closer to 0 indicate a greater likelihood of normal operation. The model's output comprehensively considers the spatial pattern characteristics of the thermal imaging data and the time series pattern characteristics of the electromagnetic field data, as well as their weighted fusion features via an attention mechanism, to produce a comprehensive and objective assessment of the failure probability.

[0071] 3. Generate a diagnostic report based on the failure probability index and confidence threshold.

[0072] Fault diagnosis: Compare the fault probability index output by the model with the preset confidence threshold. If the fault probability index is greater than the confidence threshold, the device is judged to be faulty; if it is less than or equal to the confidence threshold, the device is judged to be operating normally. It should be noted that for wireless charging devices with different application scenarios and importance, the confidence threshold can be flexibly adjusted according to actual needs. For example, in the wireless charging scenario of medical equipment with extremely high safety requirements, the confidence threshold can be appropriately lowered to more sensitively detect potential faults; while in some ordinary consumer electronic device charging scenarios, the threshold can be appropriately increased to reduce false alarms.

[0073] Diagnostic Report Content: The diagnostic report should include basic device information (device number, model, system, etc.), detection time, failure probability index, comparison results with the confidence threshold, and the final conclusion. If a device fault is determined, the likely fault type and location must be inferred based on the magnitude of the failure probability index and historical fault data. For example, when the failure probability index is high and the thermal distribution gradient is abnormally pronounced, the device's cooling system may be at fault; when electromagnetic field distortion is prominent, the coil may be at fault. The report also includes recommended measures, such as recommended maintenance items and their priority levels for the inferred fault type, providing clear and practical guidance for equipment maintenance personnel.

[0074] Its beneficial effects are: 1. It can accurately capture the subtle characteristics of various types of faults, greatly improving the accuracy of fault detection and effectively reducing the occurrence of missed detections and false detections. 2. It can quickly complete data time series alignment, feature extraction, and fault probability index calculation. Compared with traditional manual detection or single indicator detection methods, it shortens the time for fault detection, can promptly detect potential equipment faults, and gain valuable time for equipment maintenance. 3. It can effectively reduce equipment maintenance costs. Through accurate and timely fault detection, it can avoid the expansion of equipment damage caused by untimely fault detection, reduce unnecessary maintenance costs and equipment replacement costs. At the same time, the automated detection process reduces dependence on manual labor, reduces the labor and time costs of manual detection, and improves the economic efficiency of equipment maintenance.

[0075] See also Figure 2 In the artificial intelligence-based wireless charging device fault detection method, multi-source charging device data is collected in real time. The DTW dynamic time warping algorithm is used to align the time series of the multi-source charging device data, and the electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics are extracted to obtain device feature data. The following steps are included: Step 201: Calculate the deviation between the actual electromagnetic field distribution in the aligned data sequence and the normal state, and use the sum of the absolute values ​​of the deviations as the electromagnetic field distortion characteristic value; Step 202: Calculate the temperature difference between each pixel in the thermal imaging data and its surrounding adjacent pixels, take the absolute value and average it as the thermal distribution gradient value of the pixel, perform statistics on the gradient values ​​of the thermal imaging area, and extract the thermal distribution gradient characteristics; Step 203: Calculate the standard deviation of the power transmission efficiency data within the time window. The coefficient of variation is the ratio of the standard deviation to the average value, and obtain the power fluctuation characteristics.

[0076] The above is an introduction to the embodiment of the wireless charging device fault detection method based on artificial intelligence of the present invention. Figure 3 ,In the wireless charging equipment fault detection system based on artificial intelligence, the wireless charging equipment fault detection system includes the following modules: The feature data extraction module is used to collect data from multiple charging devices in real time. It uses the DTW dynamic time warping algorithm to align the time series of the data and extract electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data. A fusion model building module is used to build a dual-path convolutional neural network based on the DP-CNN. The first path uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data. The attention mechanism is used to weight the contribution of features to build a dual-path feature fusion network model. The fusion model optimization module is used to optimize the hyperparameters of the dual-path feature fusion network model using the Bayesian optimization algorithm to obtain the target dual-path feature fusion network model; The equipment fault diagnosis module is used to input equipment feature data into the target dual-path feature fusion network model, output the fault probability index, and generate a diagnosis report based on the fault probability index and a preset confidence threshold.

[0077] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A wireless charging device fault detection method based on artificial intelligence, characterized in that: The wireless charging device fault detection method comprises the following steps: Collect data from multiple charging devices in real time, align the time series of the data using the DTW dynamic time warping algorithm, and extract electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data; Based on the DP-CNN dual-path convolutional neural network, the first path uses 3D convolutional layers to analyze the spatial patterns of thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series patterns of electromagnetic field data. The contribution of features is weighted through the attention mechanism to construct a dual-path feature fusion network model. Optimizing the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; The device feature data is input into the target dual-path feature fusion network model, a fault probability index is output, and a diagnosis report is generated according to the fault probability index and a preset confidence threshold.

2. The artificial intelligence-based wireless charging device fault detection method according to claim 1, characterized in that: The real-time collection of multi-source charging device data, alignment of the time series of the multi-source charging device data using the DTW dynamic time warping algorithm, and extraction of electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data include: Real-time collection of electromagnetic field distribution data, thermal imaging data, and power transmission efficiency data of wireless charging devices to obtain multi-source charging device data; The power transmission efficiency data with the highest sampling frequency was selected as the reference sequence. The Euclidean distance between the electromagnetic field distribution data and thermal imaging data and each data point in the reference sequence was calculated to construct a distance matrix. Through the dynamic programming method, the path with the minimum cumulative distance is found in the distance matrix. The minimum path is the optimal matching path, and the aligned data sequence is obtained.

3. The artificial intelligence-based wireless charging device fault detection method according to claim 1, characterized in that: The real-time collection of multi-source charging device data, alignment of the time series of the multi-source charging device data using the DTW dynamic time warping algorithm, and extraction of electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data include: By calculating the deviation between the actual electromagnetic field distribution in the aligned data sequence and the normal state, the sum of the absolute values ​​of the deviation values ​​is used as the electromagnetic field distortion characteristic value; Calculate the temperature difference between each pixel in the thermal imaging data and its surrounding adjacent pixels, take the absolute value and average it as the thermal distribution gradient value of the pixel, perform statistics on the gradient values ​​of the thermal imaging area, and extract the thermal distribution gradient characteristics; The standard deviation of the power transmission efficiency data within the time window is calculated, and the coefficient of variation is the ratio of the standard deviation to the mean value, thus obtaining the power fluctuation characteristics.

4. The artificial intelligence-based wireless charging device fault detection method according to claim 1, wherein: The DP-CNN dual-path convolutional neural network uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data in the first path, and an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data in the second path. The attention mechanism is used to weight the contribution of features to construct a dual-path feature fusion network model, including: The 3D convolutional layer is used to convert the time-aligned thermal imaging data into a 3D cube structure. Each cube contains 16 consecutive frames of thermal images, preserving the temporal continuity and spatial distribution characteristics of temperature changes. The continuous electromagnetic field data is divided into subsequences according to fixed time windows through the LSTM long short-term memory network. Each subsequence contains 200 data points to obtain time features.

5. The artificial intelligence-based wireless charging device fault detection method according to claim 1, characterized in that: The DP-CNN dual-path convolutional neural network uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data in the first path, and an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data in the second path. The attention mechanism is used to weight the contribution of features to construct a dual-path feature fusion network model, including: The one-dimensional vector output by the 3D convolutional layer and the vector output by the LSTM network are used as the input of the attention mechanism respectively, and the attention weight of each feature is calculated; The features of the two paths are weighted and summed according to the attention weight to obtain a fused feature vector; the fused feature vector is passed through a fully connected layer to output the fault probability index.

6. The artificial intelligence-based wireless charging device fault detection method according to claim 1, characterized in that: The method of optimizing the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model includes: A set of hyperparameter combinations is randomly generated as the initial sample points. The dual-path feature fusion network model is trained using this hyperparameter combination, and the loss function value of the model is calculated by cross-validation as the evaluation indicator. A Gaussian process surrogate model is constructed based on the existing sample points and corresponding evaluation indicators, and the expected improvement criterion is used according to the surrogate model to select the hyperparameter combination of the optimized evaluation indicators.

7. The artificial intelligence-based wireless charging device fault detection method according to claim 1, wherein: The step of inputting the device feature data into the target dual-path feature fusion network model, outputting a fault probability index, and generating a diagnostic report based on the fault probability index and a preset confidence threshold comprises: If the fault probability index is greater than the preset confidence threshold, the fault type is determined based on the range of the fault probability index, and the fault type, fault probability index and fault location are described in the diagnostic report.

8. The wireless charging equipment fault detection system based on artificial intelligence is characterized by: The wireless charging device fault detection system includes the following modules: A feature data extraction module is used to collect multi-source charging device data in real time, align the time series of the multi-source charging device data using the DTW dynamic time warping algorithm, and extract electromagnetic field distortion characteristics, thermal distribution gradient characteristics, and power fluctuation characteristics to obtain device feature data; A fusion model building module is used to build a dual-path convolutional neural network based on the DP-CNN. The first path uses a 3D convolutional layer to analyze the spatial pattern of thermal imaging data, and the second path uses an LSTM long short-term memory network to analyze the time series pattern of electromagnetic field data. The attention mechanism is used to weight the contribution of features to build a dual-path feature fusion network model. A fusion model optimization module is used to optimize the hyperparameters of the dual-path feature fusion network model using a Bayesian optimization algorithm to obtain a target dual-path feature fusion network model; The device fault diagnosis module is used to input the device feature data into the target dual-path feature fusion network model, output a fault probability index, and generate a diagnosis report based on the fault probability index and a preset confidence threshold.

9. The artificial intelligence-based wireless charging device fault detection system according to claim 8, characterized in that: The fusion model optimization module includes the following submodules: The generation submodule is used to randomly generate a set of hyperparameter combinations as initial sample points, use the hyperparameter combination to train the dual-path feature fusion network model, and use cross-validation to calculate the model's loss function value as an evaluation indicator; The selection submodule is used to construct a Gaussian process proxy model based on the existing sample points and corresponding evaluation indicators, and select the hyperparameter combination of the optimization evaluation indicators according to the expected improvement criterion of the proxy model.

10. The artificial intelligence-based wireless charging device fault detection system according to claim 8, characterized in that: The equipment fault diagnosis module includes the following submodules: The diagnosis submodule is used to determine the fault type according to the range of the fault probability index if the fault probability index is greater than a preset confidence threshold, and to describe the fault type, fault probability index and fault location in the diagnosis report.

Images