Cross-device power adapter fault diagnosis method and system

Abstract

The invention relates to the technical field of fault diagnosis, and discloses a cross-device power adapter fault diagnosis method and system, and the method comprises the steps: collecting an operation data set of multiple types of devices; extracting a universal feature vector; calculating a deviation value between the general feature vector and a preset normal operation vector, and if the deviation value exceeds a preset deviation tolerance threshold value, marking the general feature vector as an abnormal vector and performing classification to obtain a fault classification label; performing cross-device unified correction on the fault feature vectors matched with the tags, and extracting fault distribution features; in combination with the real-time feature vector, standardized fault modes are extracted and integrated, and an initial diagnosis standard is obtained; performing matching analysis on the real-time operation parameters and an initial diagnosis standard, and obtaining an optimized diagnosis standard according to an analysis result; and monitoring the operation state of each device and performing diagnosis by using the optimized diagnosis standard to obtain a final fault diagnosis result. The method can unify the feature data of different devices under the same type of faults, and flexibly deals with the difference between the devices.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and system for diagnosing power adapter faults across devices. Background Technology

[0002] In modern industry and daily life, power adapters serve as the core energy supply for various electronic devices, and their stable operation is crucial to the safety and efficiency of these devices. The MCU (Microcontroller Unit), as the core control and data processing unit of the power adapter, determines the accurate monitoring of the adapter's operating status and its ability to execute commands. Therefore, fault diagnosis technology for power adapters not only needs to cover the identification of anomalies in key parameters such as voltage and current, but also needs to combine the operating characteristics of the MCU to achieve concrete and precise fault location of the adapter, which directly affects the device's lifespan and user experience.

[0003] In current mainstream methods, it is usually necessary to rely on historical data or fixed rules for specific devices. In practical applications, historical data of a single model of device is collected to train the model, or fixed diagnostic rules or thresholds are formulated according to the physical principles and operating experience of a single model of device, and the data of the device in operation is judged based on these rules or thresholds.

[0004] When dealing with cross-device applications, this approach ignores the impact of differences in the control logic and data output format of MCUs in different adapter models on fault characteristics. It lacks the ability to flexibly respond to differences between devices. Therefore, in actual use, strategies need to be adjusted frequently, which increases time consumption and results in low coordination efficiency. Summary of the Invention

[0005] This invention provides a cross-device power adapter fault diagnosis method and system to unify the characteristic data of different devices under the same type of fault in a multi-device environment, and to flexibly deal with the differences between devices.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a cross-device power adapter fault diagnosis method, comprising: The system collects operating data and historical operating records from multiple equipment models, merges the operating data and historical operating records, and corrects them to obtain an operating dataset; wherein the operating data includes: voltage data, current data, and temperature data; Extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results, and perform vectorization processing to obtain a general feature vector. Calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, mark the general feature vector as an abnormal vector and use support vector machine for classification to obtain a fault classification label. Obtain the fault feature vector that matches the fault classification label, perform cross-device unified correction on the fault feature vector, and extract the fault distribution features from the corrected vector; Based on the fault distribution characteristics, combined with the real-time feature vectors acquired in real time, standardized fault modes are extracted and integrated to obtain initial diagnostic criteria applicable to cross-device diagnosis. Obtain real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria; Monitor the operating status of each device and use the optimized diagnostic criteria to perform diagnosis to obtain the final fault diagnosis result.

[0007] Secondly, embodiments of the present invention also provide a cross-device power adapter fault diagnosis system, comprising: The data acquisition and preprocessing module is used to collect operating data and historical operating records of multiple equipment models, merge the operating data and the historical operating records and correct them to obtain an operating dataset; wherein, the operating data includes: voltage data, current data and temperature data; The feature extraction module is used to extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results, and perform vectorization processing to obtain a general feature vector. The fault classification module is used to calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an abnormal vector and classified using a support vector machine to obtain a fault classification label. A cross-device unified module is used to obtain fault feature vectors that match the fault classification labels, perform cross-device unified correction on the fault feature vectors, and extract fault distribution features from the corrected vectors. The preliminary diagnosis module is used to extract and integrate standardized fault modes based on the fault distribution characteristics and real-time feature vectors acquired in real time, so as to obtain an initial diagnostic standard applicable to cross-device diagnosis. The diagnostic criteria optimization module is used to acquire real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria. The final diagnostic module is used to monitor the operating status of each device and perform diagnosis using the optimized diagnostic criteria to obtain the final fault diagnosis result.

[0008] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: In this invention, principal component analysis is used to extract a common feature set across devices, solving the problem of inconsistent diagnostic standards caused by differences in parameter dimensions and numerical distributions among different device models. By using a transformation matrix to correct for differences in feature scales between devices, the same fault mode across different devices is mapped to a unified feature space. This enables a single diagnostic system to adapt to multiple device models, eliminating the need to develop separate diagnostic models for individual devices and significantly reducing system deployment and maintenance costs in multi-device scenarios. A dynamic standard optimization mechanism combined with gradient descent adaptive step size adjustment corrects for deviations between standard parameters and device operating characteristics in real time, ensuring that diagnostic results adapt to changes in device operating status and avoiding missed diagnoses due to rigid standards. Attached Figure Description

[0009] Figure 1 This is a schematic flowchart of the power adapter fault diagnosis method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the support vector machine fault classification hyperplane provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the power adapter fault diagnosis system provided in an embodiment of the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Reference Figure 1 The first embodiment of the present invention provides a power adapter fault diagnosis method, including steps S11 to S17, as follows: S11, collect operating data and historical operating records of multiple equipment models, merge the operating data and the historical operating records and make corrections to obtain an operating dataset; wherein, the operating data includes: voltage data, current data and temperature data; S12, extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results and perform vectorization processing to obtain a general feature vector; S13, calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, mark the general feature vector as an abnormal vector and use support vector machine for classification to obtain a fault classification label. S14, obtain the fault feature vector that matches the fault classification label, perform cross-device unified correction on the fault feature vector, and extract the fault distribution features from the corrected vector; S15. Based on the fault distribution characteristics and combined with the real-time feature vectors acquired in real time, standardized fault modes are extracted and integrated to obtain initial diagnostic criteria applicable to cross-device diagnosis. S16, Obtain real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, optimize the initial diagnostic criteria based on the analysis results, and obtain optimized diagnostic criteria; S17. Monitor the operating status of each device and use the optimized diagnostic criteria to perform diagnosis to obtain the final fault diagnosis result.

[0012] In step S11, it is necessary to collect operating data and historical operating records of multiple device models, merge the operating data and the historical operating records, and correct them to obtain an operating dataset, including: Based on a pre-established equipment identification mapping table, voltage, current, and temperature data are collected from multiple equipment models, and the equipment model identification is recorded to obtain the first operating data; Extract the historical operation records corresponding to the device model identifier from the pre-established historical record table, and merge the historical operation records with the first operation data according to the time sequence to obtain the second operation data; If any value in the second running data exceeds the preset running data threshold, it is corrected by linear interpolation to obtain the third running data. If any value in the second running data does not exceed the preset running data threshold, the second running data is directly used as the third running data. The third running data is then converted to a unified format to obtain the running dataset.

[0013] First, it's important to clarify that the device identification mapping table is a pre-established table that records the correspondence between device models, communication protocols, and data address mappings. For example, one record might show an ABB frequency converter with model number ACS880, using the Modbus communication protocol, and data address mappings of 40001-40010. When collecting voltage, current, and temperature data from the device based on this mapping table, a running data record table can be created first. The initial content of this record table only includes the device model and its corresponding identifier, such as the ABB frequency converter model ACS880, corresponding to the identifier VFD001 (this operation converts ambiguous device model numbers into more specific information; VFD001 indicates the first VFD device). Then, based on the device identification mapping table, the corresponding data address mapping is queried, and data collection begins. It's worth noting that the collected voltage data includes parameters such as the effective value of three-phase voltage and voltage imbalance; current data includes indicators such as phase current, zero-sequence current, and power factor; and temperature data includes the real-time temperature of key internal components such as the power module, radiator, and motor windings. After collecting this data, the completeness of the data collection can be checked against the communication protocol recorded in the device identifier mapping table. The collected data is then entered into the data record section of the corresponding device in the operation data record table to obtain the first set of operation data.

[0014] When acquiring the second operational data, the historical operational records of the same device can be queried from the database based on the device identifier. Then, the historical operational records and the first operational data are merged according to the time series. For example, the first operational data collected by a device at the current time 18:00 is taken as a data point. The historical operational records are queried backward to 18:00 30 days ago. From that time, the historical operational records recorded operational data once every hour. Before the current time 18:00, 24×30-1=719 data points were recorded. These 719 data points are merged with the data points at the current time in chronological order to obtain a time series data record containing 720 data points, which is the second operational data.

[0015] It should be noted that the second operation record may contain data records that are clearly illogical due to abnormal data transmission or other reasons. For these data records, outlier removal can be performed using the statistical 3σ criterion combined with a preset operating data range for normal equipment operation. After removing outliers, blank values ​​can be filled by linear interpolation between the two normal data points before and after the blank. For example, in the voltage data record, there are data records such as 385V, 450V, and 390V. The 450V value exceeds the normal operating voltage range of 320V-420V and is therefore identified as an outlier. After removing this value, linear interpolation is performed based on 385V and 390V to obtain the corrected record of 385V, 387.5V, and 390V. A similar correction operation is performed on all data in the second operation data to obtain the third operation data.

[0016] Finally, the third set of running data undergoes a standardized format processing to ensure consistent data format for subsequent use, eliminating the need for additional data processing. For example, the data is stored in IEEE 754 floating-point format, the timestamp uses the ISO 8601 standard, and the device identifier follows the IEC 61850 naming convention. During the format conversion, the original integer current value 45 is converted to a floating-point format of 45.00, and the timestamp is converted from the Unix timestamp 1705298425 to the format 2024-01-15T14:30:25Z, ultimately yielding the running dataset.

[0017] In step S12, it is necessary to extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, and generate a general feature set based on the analysis results, and then perform vectorization processing to obtain a general feature vector, including: The voltage, current, and temperature data in the aforementioned dataset are subjected to dimensionality reduction transformation using principal component analysis to obtain the core parameter combination. Calculate the correlation degree of each component in the core parameter combination; If the correlation value is greater than the preset correlation threshold, it is included in the general feature set; if the correlation value is less than the preset correlation threshold, it is not included in the general feature set. For each feature component of the general feature set, a corresponding general feature weight coefficient is assigned to obtain a weighted feature sequence; wherein, the general feature weight coefficient is normalized. The weighted feature sequences are combined and encapsulated according to a predefined vector format to obtain a general feature vector.

[0018] It should be noted that when performing principal component analysis on the running dataset, the parameters are first standardized using Z-scores to construct the original data matrix. Then, the covariance of the original data matrix is ​​calculated to construct the covariance matrix. For example, the covariance of the voltage and current parameters of the aforementioned ACS880 frequency converter is calculated to be 0.85, the covariance of the current and temperature parameters is 0.72, and the covariance of the voltage and temperature parameters is 0.63, yielding the corresponding covariance matrices. Eigenvalue decomposition is then performed on the covariance matrix to obtain the corresponding eigenvalues ​​and eigenvectors. The ratio of a specific eigenvalue to the total eigenvalues ​​is used as the contribution rate. The cumulative contribution rate of the first k principal components is calculated from the highest contribution rate. If the cumulative contribution rate is ≥90%, the first k principal components are selected as the core parameter combination after dimensionality reduction. For example, if the extracted contribution rates are 68.5%, 23.2% and 8.3% respectively, since the cumulative contribution rate of the first two principal components is 91.7%≥90%, the first two principal components are selected as the core parameter combination after dimensionality reduction.

[0019] Cross-device correlation assessment involves calculating the correlation coefficient between a principal component (all are first principal components or all are second principal components, etc.) of the core parameter combinations of different devices. If the correlation coefficient exceeds a preset correlation threshold, it indicates a significant correlation in the operating states of the devices; that is, the same parameter exhibits similar trends when the same fault occurs in different models of devices. Based on this correlation analysis, strongly correlated features are included in a general feature set. It should be noted that the Pearson correlation coefficient is used for cross-device correlation assessment, with 0.75 selected as the correlation threshold (this threshold is a commonly used engineering constraint; those skilled in the art will understand that this threshold can be dynamically adjusted according to actual engineering needs, with a preferred range of 0.6-0.85). Exceeding this threshold indicates a strong correlation. For example, calculations revealed that the correlation coefficient between the first principal component data of the ACS880 frequency converter and the SINAMICS G120 motor controller was 0.78, exceeding the correlation threshold; the correlation coefficient between the temperature control device and the frequency converter was 0.82, also exceeding the correlation threshold; finally, a general feature set including power load characteristics, thermal balance characteristics, and electrical stability characteristics was determined.

[0020] When assigning weights to each feature component of the general feature set, the feature weight allocation rule quantifies the importance of each feature to the device state representation, assigning a corresponding normalized feature weight coefficient. For example, the power load feature is assigned a weight coefficient of 0.45 due to its direct impact on device operating state, the thermal balance feature receives a weight coefficient of 0.35 considering the critical role of temperature in device lifespan, and the electrical stability feature is assigned a weight coefficient of 0.20. The features in the general feature set are weighted and fused with their corresponding feature weight coefficients to form a weighted feature sequence. This sequence is then combined and encapsulated according to the IEEE standard format, with vector length standardized and data type unified, resulting in a general feature vector that eliminates the influence of differences in feature dimensions. For example, multiplying the power load feature value of 12.8, the thermal balance feature value of 8.6, and the electrical stability feature value of 5.2 by their corresponding weight coefficients yields weighted feature components of 5.76, 3.01, and 1.04, respectively. These feature components are arranged according to the predefined vector format [5.76, 3.01, 1.04]. The Euclidean norm of the vector is calculated to be 6.58. Dividing each component by this norm value yields the unit vector [0.875, 0.457, 0.158], thus obtaining the general feature vector.

[0021] In step S13, the deviation value between the general feature vector and the preset normal operation vector needs to be calculated. If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an anomaly vector and classified using a support vector machine to obtain a fault classification label, including: Calculate the deviation between the general feature vector and the preset normal operation vector; If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an anomaly vector. The anomaly vectors are classified using a support vector machine to obtain fault classification labels.

[0022] It should be noted that the normal operation vector is obtained by collecting voltage, current, and temperature data of each device under fault-free and rated operating conditions. The sample size must meet statistical significance requirements (e.g., 1000+ normal samples per type of device, covering different working periods and load intensities). After processing the data according to S11 and S12, the 95th percentile of each component is calculated. The 99th percentile of each component is used as the deviation tolerance threshold. Using a deviation quantization function, the deviation value between the general feature vector and each component of the normal operation vector is calculated. If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an abnormal vector. For example, the normal operation vector of the ACS880 frequency converter is calculated to be [0.820, 0.430, 0.180] and the deviation tolerance threshold is 0.050. The deviation values ​​of each component of the general feature vector [0.875, 0.457, 0.158] from the normal operation vector are calculated to be 0.055, 0.027 and 0.022 respectively. It is found that the deviation of the first component 0.055 exceeds the deviation tolerance threshold. The vector is immediately marked as an abnormal vector and the subsequent analysis process is triggered.

[0023] It should be noted that a pre-trained Support Vector Machine (SVM) model is used for classification. For example, the hyperparameter settings of this model are derived based on the practicality and data characteristics of the fault diagnosis scenario. The core hyperparameter γ (RBF kernel width) is set to 0.5. This value was verified and determined using a grid search method. After testing candidate values ​​such as γ=0.1, 0.3, 0.5, 0.7, and 1.0, it was found that γ=0.5 resulted in the highest cross-validation accuracy for outlier vectors. This balances local fitting and global generalization ability, avoiding overfitting or underfitting in small sample scenarios, while accurately distinguishing similar abnormal patterns such as electrical overload and thermal imbalance. The classification confidence threshold is set to 0.80, considering the safety requirements of power adapter fault diagnosis, to ensure the reliability of category determination and reduce the risk of misdiagnosis. The implicit hyperparameter penalty coefficient C is set to 1.0, balancing the maximization of the classification margin with the penalty for misclassified samples, enabling the model to maintain good robustness while ensuring discriminative power.

[0024] The model training process revolves around the objective of "maximizing the classification margin." First, 1500 standardized anomaly vectors are pre-collected and labeled to identify their corresponding real-world fault types (e.g., electrical overload, thermal imbalance), thus constructing a training dataset. Then, the similarity matrix between vectors is calculated using an RBF kernel function with γ=0.5, mapping the original features to a high-dimensional space. Next, the SMO (Sequence Minimum Optimization) algorithm is used to solve a convex quadratic programming problem, decomposing the multi-sample optimization into sub-problems for iterative solving, gradually determining the normal vector and bias parameters of the classification hyperplane, referring to... Figure 2The optimal hyperplane that maximizes the classification margin between the two types of anomaly vectors is found through multiple rounds of iteration. Finally, the model is optimized by leave-one-out cross-validation to ensure that the trained model can accurately adapt to the cross-device anomaly classification requirements.

[0025] After model training is complete, a reasoning process of "directed distance calculation - probability transformation - label determination" is used for newly input anomaly vectors. The directed distance from the new vector to the hyperplane is calculated using the normal vector and bias parameters of the optimal classification hyperplane. The sign of the distance represents the direction of class assignment, and the absolute value reflects the certainty of the assignment. The directed distance is then mapped to the [0,1] interval using the sigmoid function to obtain the class probability. The transformation logic of this function is (d is the directed distance). When the probability value exceeds the preset classification confidence level of 0.80, the corresponding class label (e.g., "TYPE_A_ELECTRICAL_OVERLOAD") is output.

[0026] In step S14, it is necessary to obtain the fault feature vector that matches the fault classification label, perform cross-device unified correction on the fault feature vector, and extract the fault distribution features from the corrected vector.

[0027] The process includes performing cross-device unified correction on the fault feature vectors and extracting fault distribution features from the corrected vectors, including: Obtain the hardware configuration parameter matrix of each device. If the feature dimensions of different devices do not match, zero-padding is performed on the low-dimensional feature vectors to obtain a device feature matrix with a unified dimension. Based on the device feature matrix, the feature offset between different models of devices is calculated, and the parameters of the linear transformation matrix are determined. If the feature offset exceeds the preset alignment accuracy threshold, iterative optimization is performed until the feature offset meets the requirements, and the transformation matrix for cross-device feature alignment is obtained. Based on the transformation matrix, the fault feature vector is subjected to spatial mapping transformation, and the transformed distribution density is calculated to obtain the aligned feature distribution density map. Based on the feature distribution density map, the anomaly degree value of each feature vector relative to its neighboring points is calculated to obtain the fault distribution characteristics.

[0028] It should be noted that obtaining the fault feature vector can be achieved through a pre-established fault mode mapping table, which is a pre-established identifier table that records the correspondence between category labels and specific fault type descriptions. For example, one record records the label "TYPE_A_ELECTRICAL_OVERLOAD," corresponding to the fault type description "abnormal fluctuations in electrical parameters caused by overload." The rule base is a pre-established identifier table that records the correspondence between category labels and typical descriptions of specific fault features. For example, one record is the label "TYPE_A_ELECTRICAL_OVERLOAD," corresponding to typical fault feature descriptions such as voltage fluctuation exceeding 8% of the rated value, peak current duration exceeding 200 milliseconds, and temperature rise rate exceeding 3 degrees Celsius per minute. By associating the category label with the fault mode mapping table and the rule base, a typical fault feature description is obtained. Key information is extracted from this description, such as 8% of the voltage fluctuation exceeding 8% of the rated value. The extracted key information is concatenated into a vector and Z-score standardized to obtain the fault feature vector.

[0029] Subsequently, a dimension alignment operation is performed. It should be noted that the hardware configuration parameter matrix is ​​a matrix composed of feature vectors from different devices of the same dimension. The device feature vectors are obtained by parsing the device model identifier. If the dimensions of the feature vectors from different devices do not match, zero-padding is performed on the lower-dimensional feature vectors to obtain a device feature matrix of uniform dimension. For example, the hardware configuration parameters of the ABB ACS880 frequency converter and the Siemens SINAMICS G120 controller are parsed by the device model identifier, resulting in a 5-dimensional feature vector [0.875, 0.457, 0.158, 0.692, 0.334] output by the ACS880 device, while the G120 device only outputs a 3-dimensional feature vector [0.821, 0.445, 0.167]. Zero-padding is performed on the lower-dimensional vector of the G120 device using a feature dimension expansion function to generate a uniform 5-dimensional feature vector [0.821, 0.445, 0.167, 0.000, 0.000], ensuring that all device feature vectors have the same data structure.

[0030] A linear transformation matrix of the same dimension is set according to the equipment feature matrix to correct the systematic shift of feature vectors of different equipment under the same fault type. Specifically, the linear transformation matrix is ​​a diagonal matrix that only corrects scale differences without changing the physical meaning. The parameters of the linear transformation matrix are determined by the least squares method, using the feature vector of the high-dimensional equipment under the same fault type as the target vector and the feature vector of the low-dimensional equipment as the vector to be corrected. For example, for the equipment feature matrix with feature vectors of 20 ACS880 equipment and 15 G120 equipment, there are 10 ACS880 equipment with electrical overload fault type, one of which has a feature vector of [0.891, 0.445, 0.142, 0.678, 0.298], while there are 5 G120 equipment, one of which has a feature vector of [0.856, 0.421, 0.135, 0.000, 0.000]. The dimension of the linear transformation matrix is ​​determined to be 5. Taking the first dimension as an example, the first dimension of the 10 feature vectors corresponding to the ACS880 device is extracted. The same applies to G120. A linear model is established as the first dimension vector of ACS880 ≈ the first dimension vector of G120 x the parameters of the corresponding first dimension of the linear transformation matrix. The goal is to minimize the sum of squared errors of this model. The parameters of the corresponding first dimension of the linear transformation matrix are solved. The same applies to other dimensions. Finally, the diagonal parameters of the linear transformation matrix are obtained as 1.041, 1.057, 1.052, 1.000, and 1.000.

[0031] An alignment accuracy threshold of 0.08 is set as the acceptable feature vector offset difference. The obtained linear transformation matrix is ​​used to multiply the vectors to be corrected of all G120 devices. The Euclidean distance between the corrected vector and the target vector is calculated as the offset. If the offset of each component is lower than the alignment accuracy threshold, no further iteration is needed, and the transformation matrix for cross-device feature alignment is directly obtained. If the offset of any component is higher than the alignment accuracy threshold, the least squares method is re-applied, the corresponding scaling factor is adjusted, and iterative optimization is performed until the offset meets the requirements, thus obtaining the transformation matrix for cross-device feature alignment. For example, after three rounds of iterative calculation, the corrected feature vector of the G120 device is [0.891, 0.445, 0.142, 0.000, 0.000], which achieves high-precision alignment with the corresponding feature vector of the ACS880 device [0.891, 0.445, 0.142, 0.678, 0.298] in the first three dimensions, and the offset is reduced to below 0.003, which is lower than the alignment accuracy threshold.

[0032] For each fault type, a transformation matrix is ​​used to multiply the corresponding equipment's fault feature vector, resulting in an aligned feature matrix that has undergone high-precision dimensional alignment for different fault types and equipment. This maps the fault feature vectors of different fault types and equipment to a unified space. The Euclidean distance between the transformed feature vectors is calculated, and the density centers formed by different fault types are identified, resulting in an aligned feature distribution density map. For example, after mapping the 35 feature vectors of the equipment feature matrix to a unified space according to fault type, it is found that electrical overload fault feature points form a high-density clustering region near coordinates [0.89, 0.44, 0.14], i.e., a density center with a cluster radius of approximately 0.12. Meanwhile, thermal imbalance fault feature points are distributed around coordinates [0.76, 0.52, 0.23], forming another density center.

[0033] It should be noted that for a feature vector to be anomaly identified, the average distance between the feature vector mapped to a unified space and its five nearest neighboring points (the number of neighboring points can be set according to the data distribution density, usually an integer between 5 and 10) is calculated. The ratio of this distance to the average distance of these five neighboring points, i.e., the density ratio, is used to quantify the degree of anomaly of the feature vector. An anomaly threshold of 1.5 is set. Exceeding the anomaly threshold indicates that the fault corresponding to the feature vector is a rare type among the known equipment fault groups. For example, when processing the fault feature vector [0.923, 0.467, 0.128, 0.000, 0.000] of a Schneider ATV320 frequency converter, the average distance between this point and its five nearest neighboring points is calculated to be 0.087, while the average distance between the neighboring points is 0.045. Based on the density ratio, the anomaly score of this feature point is 1.93, exceeding the normal range of the anomaly threshold of 1.5, indicating that this fault mode is a rare type among the current equipment group.

[0034] The final result is a fault distribution feature that includes spatial clustering features of fault types (coordinates of cluster centers and cluster radius and density) and distribution and quantification features of outliers (coordinates of outliers and their attribution and anomaly severity score).

[0035] In step S15, based on the fault distribution characteristics and combined with the real-time feature vectors acquired in real time, standardized fault modes need to be extracted and integrated to obtain initial diagnostic criteria applicable to cross-device diagnosis, including: Based on the fault distribution characteristics, similar fault patterns are classified and merged by clustering to obtain a limited number of fault patterns. The feature parameters of the fault patterns are then normalized to obtain a standardized fault pattern library. Obtain real-time operating parameters of the device during operation, and obtain time window segments based on the standardized fault mode library and the real-time operating parameters, and extract the corresponding real-time feature vectors from them; The similarity between the standardized fault mode library and the real-time feature vector is calculated and weighted to obtain a matching degree matrix; The weighted features of each fault mode in the matching degree matrix are summed to obtain an initial diagnostic standard applicable to cross-device diagnosis.

[0036] It should be noted that, for the fault feature vectors in the fault distribution characteristics, the K-means clustering algorithm is used to classify and merge similar fault modes. The initial number of clusters is set to 3 (corresponding to 3 potential fault modes). Three cluster centers are manually or randomly selected, and a similarity threshold of 0.075 is set to determine whether the cluster centers are too close, triggering mode merging. The convergence condition is that the change in the cluster center position is less than the preset threshold of 0.001, or the number of iterations reaches the maximum of 10 rounds. The Euclidean distance between each fault feature vector and the three initial cluster centers is calculated as the similarity between the sample and the center; the smaller the distance, the more similar the samples. Based on the highest similarity value, the fault feature vector is assigned to the corresponding cluster, forming 3 initial clusters. The mean of the feature vectors within each cluster is calculated, and this mean is used as the new cluster center. The Euclidean distance between all samples and the new cluster center is recalculated, and the samples are reassigned according to the "nearest principle." This process of "updating centers → reassigning samples" is repeated iteratively until the convergence condition is met. Next, the Euclidean distances between each pair of the three final cluster centers are calculated and compared with a similarity threshold. When the Euclidean distance is less than the similarity threshold, the corresponding fault type differences are considered minor, and the two clusters are merged into one. The mean of the merged cluster is then recalculated as the new cluster center. This process ultimately forms a few core fault patterns, eliminating interference from "different manifestations of the same fault" (such as classifying bearing wear characteristics from different devices into one category) and avoiding excessive subdivision of fault patterns by merging similar clusters (reducing diagnostic complexity). This provides a clear categorization basis for subsequent extraction of standardized fault patterns and construction of cross-device diagnostic standards.

[0037] For example, for 45 cross-device aligned fault feature vectors, three initial cluster center coordinates were selected: [0.82, 0.41, 0.15], [0.76, 0.53, 0.28], and [0.91, 0.38, 0.12]. The Euclidean distance between each of the 45 fault feature vectors and the three initial cluster centers was calculated, resulting in cluster 1 containing 14 samples, cluster 2 containing 15 samples, and cluster 3 containing 16 samples. Iterative calculations were then performed. After eight rounds of iterative calculations, it was found that the positions of the three cluster centers were stable, with the change being less than the preset threshold of 0.001, thus reaching the convergence condition. One of the clusters represents motor bearing wear faults, with the cluster center located at [0.834, 0.425, 0.163], containing 16 feature sample points. The calculated Euclidean distance between adjacent cluster centers was 0.067, which is less than the similarity threshold of 0.075. Therefore, the clusters were merged, unifying two fault types with minimal differences into the motor bearing wear fault type.

[0038] It should be noted that for all feature vectors of the same fault type after merging, Z-score normalization is applied dimension-by-dimensionally to form a standardized feature vector group for that fault type. Standardizing all merged fault types yields a standardized fault pattern library, where each pattern includes a standardized feature vector, a confidence interval range, and a description of typical fault symptoms. The confidence interval range is calculated based on the standardized feature vector group, using a 95% confidence interval for each dimension, clearly defining the normal fluctuation range of the fault pattern. The description of typical fault symptoms is obtained by combining the original fault data and annotating the typical manifestations of the pattern. Standardization eliminates feature scale bias caused by differences in equipment models and allows for direct comparison of fault type feature parameters, laying the foundation for subsequent calculations of the matching degree between real-time data and standard patterns.

[0039] For example, all feature vectors for the winding overheating fault type are calculated, resulting in a mean of [0.756, 0.489, 0.234] and a standard deviation of [0.089, 0.067, 0.045]. One feature vector, [0.845, 0.556, 0.279], is standardized using Z-scores to obtain a standardized feature vector of [1.0, 1.0, 1.0]. All standardized feature vectors for the winding overheating fault mode are then aggregated to form a standardized feature vector group for that mode. Based on this standardized feature vector group, a 95% confidence interval is calculated for each dimension (calculated by multiplying ±1.96 by the standard deviation; since the standard deviation is 1 after standardization, the confidence interval is [-1.96, 1.96]). Combined with the original fault data, the typical characteristics of the winding overheating fault type are: a standardized second dimension value > 1.5, and temperature-related features consistently exceeding the confidence interval. Following the above process, Z-score standardization is performed on six other fault modes, including electrical overload and thermal imbalance, and finally the information of the seven fault modes is integrated to form a standardized fault mode library.

[0040] When acquiring real-time feature vectors, the time-series data of various parameters during equipment operation are first collected in real time. The collected parameters include, but are not limited to, current, voltage, and temperature, as the vector dimensions in the previously established standardized fault mode library are based on these three data types. In actual engineering, the vector dimensions in the standardized fault mode library can be increased according to actual control requirements to achieve higher precision control. Correspondingly, the parameters collected here also need to contain operating parameters of the corresponding dimensions. The time window length is set to 30 seconds, the sliding step size to 10 seconds, and the overlap between adjacent windows is 20 seconds to avoid feature distortion caused by data breaks. The sampling density is 150 sampling points per window, corresponding to the three parameter types: current, voltage, and temperature, with 50 sampling points for each parameter type, ensuring the reliability of statistical features. The 150 sampling points in each window are split according to parameter type, resulting in three sets of data: 50 current sampling points, 50 voltage sampling points, and 50 temperature sampling points. The mean of each set of data is calculated as the core statistic and combined in a fixed order of current, voltage, and temperature to form a standardized 3D feature vector, completing the feature extraction for a single window. Feature extraction was performed on each of the 16 window segments, resulting in 16 3D real-time feature vectors, which formed a real-time feature vector group corresponding to the standardized fault mode library. This approach ensured both the continuity and integrity of the data, while also achieving dimensionality reduction of the high-dimensional original data through statistical analysis, thus adapting to the needs of subsequent fault mode matching.

[0041] For example, when processing the real-time data stream of a Rockwell PowerFlex 525 frequency converter, the algorithm extracts the monitoring data from seconds 0 to 30 as the first window segment, containing 150 sampling points within that time period. Each time, the window is moved forward 10 seconds along the time axis, generating a new window—the second window is for seconds 10-40, the third for seconds 20-50, and so on. Generation stops when the window ends after 180 seconds, ultimately extracting 16 window segments precisely from the 180 seconds of data (calculation logic: (180-30)÷10+1=16, ensuring that the beginning and end data are covered). Each window contains 150 sampling points of current, voltage, and temperature data (50 sampling points for each parameter type). After splitting by parameter type, the mean values ​​of current, voltage, and temperature data were calculated to be 45.2A, 382.5V, and 62.3℃, respectively, forming a standardized 3D feature vector [45.2, 382.5, 62.3]. This vector was then standardized using the Z-socre standard used in constructing the standardized fault mode library to obtain a real-time feature vector [0.823, 0.441, 0.178], completing the feature extraction for a single window. The real-time feature vector group was obtained by calculating the values ​​for each of the 16 window segments.

[0042] Subsequently, the similarity between the fault modes in the standardized fault mode library and the real-time feature vectors was measured using cosine similarity. When the real-time feature vector [0.823, 0.441, 0.178] was compared with the bearing fault modes [-0.234, 0.567, -0.891] in the standardized fault mode library, the cosine similarity result was 0.742. The calculations for each of the seven fault modes were completed one by one, resulting in the corresponding set of similarity values. Then, weights were assigned to the standard fault modes based on the historical frequency and severity of the faults. Specifically, the occurrence frequency of each fault mode within a preset time window (e.g., the past year) was statistically analyzed, and its proportion to the total number of faults was calculated. The higher the proportion, the larger the weight coefficient. The weight coefficients of all fault modes fell within the range of [0.3, 1.2], with the lowest weight not lower than 0.3 (ensuring that minor faults still have some influence) and the highest weight not exceeding 1.2 (avoiding a single fault having an excessively high weight that dominates the diagnostic results). For example, the historical occurrence counts of the seven fault modes were obtained as 32, 28, 22, 15, 12, 8, and 5 times, respectively, for a total of 122 occurrences. The frequency proportion of bearing faults (32 occurrences) was 32 / 122≈0.262, corresponding to a weight of 1.2; the frequency proportion of winding faults (28 occurrences) was 28 / 122≈0.229, corresponding to a weight of 1.0. Finally, the set of weight coefficients [1.2, 1.0, 0.9, 0.7, 0.6, 0.4, 0.3] was obtained. The cosine similarity of each fault mode was multiplied by the corresponding weight coefficient to obtain the weighted matching score, which was then arranged in order to construct a single-row matching matrix.

[0043] The weighted matching degree values ​​of the seven fault modes in the matching degree matrix are directly added together to obtain a comprehensive diagnostic score. A warning threshold of 2.5 is set (this threshold is calculated based on historical data of the equipment operating normally; 100 sets of operating data from different devices under normal conditions are taken, and the comprehensive diagnostic score is calculated, with the average value increased by 15% as the threshold). When the comprehensive diagnostic score exceeds the warning threshold, the corresponding equipment is deemed to have a potential fault risk; if it does not exceed the threshold, the current operating status is determined to be normal, thus providing a unified and quantitative fault diagnosis reference for multiple equipment models. For example, the weighted matching degree for bearing faults is 0.742 × 1.2 = 0.891, and the weighted matching degree for winding faults is 0.756 × 1.0 = 0.756. The weighted matching scores for the seven fault modes were obtained [0.891, 0.756, 0.623, 0.445, 0.334, 0.223, 0.112]. These scores were then summed to obtain a comprehensive diagnostic score of 3.384. This score exceeds the warning threshold of 2.5, indicating a potential fault risk in the equipment.

[0044] In step S16, it is necessary to perform a matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria, including: Calculate the degree of matching between the real-time operating parameters and the initial diagnostic criteria; When the matching degree value is lower than the preset matching degree threshold, the deviation value of each component is calculated and combined to obtain the initial parameter deviation vector. For the initial parameter deviation vector, a gradient descent optimization strategy is used to optimize the step size. An error feedback mechanism is used to monitor the deviation change trend and obtain the optimal parameter update step size value. Specifically, when the deviation is decreasing, the current step size is maintained; when the deviation oscillates, the adjustment step size is reduced. For the key parameters in the initial diagnostic criteria, set corresponding update weight coefficients, update the step size value according to the optimal parameters, and iteratively update the initial diagnostic criteria. When the matching degree value between the updated diagnostic criteria and the real-time running parameters exceeds the matching degree threshold, the optimized diagnostic criteria are obtained.

[0045] It should be noted that the matching degree value adopts a multi-dimensional evaluation mechanism, calculating the cosine similarity, Euclidean distance, and parameter conformity between the real-time feature vector of the equipment and the feature vectors of each fault type in the standardized fault mode library. Parameter conformity is calculated by comparing the real-time operating parameters with the typical fault parameter range of the standard mode, and statistically analyzing the percentage of conforming items. A higher conformity percentage indicates a better match between the actual operating state and the characteristics of the fault mode. For example, taking bearing faults as an example, the typical parameter range is "current fluctuation > 5%, temperature > 65℃, vibration frequency > 200Hz". If two of these conditions are met in the real-time data, the parameter conformity is 2 / 3 ≈ 0.667. The Euclidean distance is converted to similarity to ensure that the results of all three dimensions are within the range [0,1]. For example, the maximum distance is set to 5, and the similarity corresponding to a distance of 1.551 is 1 - 1.551 / 5 ≈ 0.689. Based on the importance of each dimension to fault identification, dimension weights are assigned: cosine similarity (0.4), Euclidean distance (0.3), and parameter conformity (0.3). The calculation results of the three dimensions are then fused according to their weights to obtain the final matching degree. The above steps are repeated to calculate the matching degree between the real-time feature vector and each fault mode. A matching degree adjustment threshold of 0.750 is set. If the matching degree value is lower than the matching degree adjustment threshold, it is determined that the current standard and real-time parameters are not sufficiently compatible, and the parameter adjustment mechanism is automatically triggered; if the matching degree value is higher than the matching degree adjustment threshold, the parameters are not adjusted. For example, the matching degree between the real-time feature vector of a PowerFlex 525 inverter and the standardized bearing fault mode is 0.623, which is lower than the matching degree threshold of 0.750, thus automatically triggering the parameter adjustment mechanism.

[0046] It should be noted that the parameter adjustment mechanism first performs a difference operation on the corresponding dimensions of the real-time feature vector and a standard vector of a standard fault mode used in the initial diagnostic criteria to obtain a deviation vector, thus defining the adjustment direction. The positive and negative signs represent the deviation direction, and the absolute value represents the deviation magnitude. For example, the deviation vector between the real-time feature vector [0.789, 0.456, 0.201] and the standard vector [0.834, 0.425, 0.163] is [-0.045, 0.031, 0.038]. Based on this, it is determined that the standard parameters need to be adjusted by 0.045 units in the negative direction and 0.031 and 0.038 units in the positive direction, respectively. An adaptive step size optimization adjustment step size based on a gradient descent strategy is adopted. An initial step size of 0.01 is set, and the first parameter correction is performed at a step size of 0.01. The deviation change is monitored; if the deviation shows a decreasing trend, the current step size is maintained; if the deviation oscillates, the current step size is reduced. Multiple adjustments are made to obtain the optimal parameter update step size. For example, the first parameter correction is performed with a step size of 0.01, and the deviation changes from 0.127 to 0.089. Since the deviation shows a decreasing trend, the current step size is maintained. After the third round of adjustment, if the deviation rebounds from 0.067 to 0.078 (oscillation occurs), the error feedback mechanism automatically reduces the step size to 0.005 to avoid the parameter adjustment from exceeding the optimal range. After 5 rounds of iterative testing, the optimal parameter update step size of 0.007 is found to balance the adjustment efficiency and stability.

[0047] The impact of various parameters on fault identification accuracy was statistically analyzed using the controlled variable method. If removing a parameter resulted in a decrease in diagnostic accuracy ≥15%, it was classified as a high-contribution parameter; a decrease of 5%-15% was considered a medium-contribution parameter; and a decrease of <5% was considered a low-contribution parameter. For example, removing the current amplitude parameter reduced the bearing fault diagnosis accuracy from 89% to 67% (a decrease of 22%), classifying it as a high-contribution parameter; removing the temperature change rate parameter reduced the accuracy from 89% to 78% (a decrease of 11%), classifying it as a medium-contribution parameter. The correlation between parameter fluctuations and fault misjudgment rates was calculated—for every 10% fluctuation in a parameter, an increase in misjudgment rate ≥8% was considered a high-sensitivity parameter; an increase of 3%-8% was considered a medium-sensitivity parameter; and a decrease of <3% was considered a low-sensitivity parameter. For example, a 10% fluctuation in current amplitude resulted in a 12% increase in misjudgment rate (high sensitivity); a 10% fluctuation in temperature change rate resulted in a 6% increase in misjudgment rate (medium sensitivity). Different parameter update weights were assigned based on the importance and sensitivity of different standard parameters; higher weights resulted in higher adjustment priorities. The weight range for parameters with high contribution and high sensitivity is 1.3 to 1.5; the weight range for parameters with medium contribution and medium sensitivity is 1.0 to 1.2; and the weight range for parameters with low contribution and low sensitivity is 0.7 to 0.8. For example, the current amplitude is 1.5, the temperature change rate is 1.2, and the vibration frequency is 0.8. The standard parameters are iteratively adjusted according to the optimal step size, weight ratio, and deviation value vector. Multiple rounds of iterative optimization are performed, with each round fine-tuning the parameters according to the weight ratio based on the deviation feedback from the previous round, gradually approaching the optimal standard. A convergence threshold of 0.003 is set; when the parameter change is below the convergence threshold for three consecutive iterations, the system is considered stable, yielding the final optimization diagnostic standard. For example, the first iteration of the current standard is 0.834 + (-0.045) × (0.007 / 0.01) × 1.5 ≈ 0.827. The temperature and vibration standards are treated similarly, resulting in a temporary set of standard parameters [0.827, 0.432, 0.168]. The parameter changes are continuously recorded across multiple iterations; if three consecutive changes are 0.002, 0.001, and 0.002 (all less than 0.003), the optimization process is considered to have reached convergence. The final optimized standard vector [0.825, 0.434, 0.170] is obtained. At this point, the matching degree between the real-time parameters and the standard is improved to 0.856, which meets the threshold requirement. The standard vectors of the fault modes used in all the initial diagnostic standards are optimized to obtain the optimized diagnostic standard.

[0048] In step S17, it is necessary to monitor the operating status of each device and use the optimized diagnostic criteria to perform diagnosis, thereby obtaining the final fault diagnosis result, including: Obtain the diagnostic operating parameters of multiple equipment models; Calculate the degree of matching between the operating parameters to be diagnosed and the optimized diagnostic criteria; Obtain historical operating data corresponding to the equipment model, calculate the correlation between the historical operating data and the operating parameter to be diagnosed, and obtain the historical data reference weight coefficient; The matching degree value and the historical data reference weight coefficient are weighted and fused to obtain a comprehensive diagnostic index; When the comprehensive diagnostic index exceeds the preset fault judgment threshold, the confidence level of each type of fault is calculated. When the highest confidence level is greater than the preset confidence threshold and the difference between the highest and second highest confidence levels exceeds the preset discrimination threshold, the corresponding fault type is taken as the final fault diagnosis result.

[0049] It should be noted that the preset supported model list includes the identification codes of all processable devices. A unified interface compatible with protocols such as Modbus and Profinet is used to extract key operating parameters, which are then converted according to the format specified in S11 to obtain the operating parameters to be diagnosed. For example, the preset supported model list includes the identification codes of mainstream frequency converter models such as PowerFlex 525, ABB ACS580, and Schneider ATV320, such as device number VFD-G120-005. When a Siemens G120 frequency converter is detected connected, the system automatically identifies its device identification code and extracts key operating parameters such as the effective current value of 2.34 amps, output frequency of 48.7 Hz, and bus voltage of 385 volts. These parameters are converted to IEEE 754 floating-point format, and the timestamps are standardized to ISO 8601.

[0050] Calculate the Euclidean distance between the operating parameters to be diagnosed and the diagnostic criteria optimized by S106, normalize the distance value to the [0,1] interval, and use it as the initial matching degree. The smaller the distance, the higher the matching degree.

[0051] Based on the equipment model index, historical operating records under the same conditions within the past 30 days are extracted from the database, including normal parameter baseline values ​​and abnormal feature patterns. Using the S15 sliding window setting, several historical window segments are generated according to the formula: window number = (T-30)÷10+1, where T is the total duration of historical data under the same conditions within the past 30 days. For each historical window, data is split by current, frequency, and voltage, and the mean and trend slope of each parameter are calculated to generate a historical window feature vector. The S15 method is used to generate real-time window feature vectors (with the added calculation of trend slope). All historical window feature vectors are classified into normal and abnormal categories, forming a normal historical pattern set and an abnormal historical pattern set. For each real-time window feature vector, the Pearson correlation coefficient is used to calculate the trend similarity with all feature vectors in the historical abnormal pattern set. The historical reference correlation threshold is set to 0.7. For a single real-time window, the mean correlation coefficient between it and all historical abnormal windows is calculated. If the mean is ≥0.7, the real-time window is considered strongly correlated with the historical abnormal pattern. The proportion of windows strongly correlated with historical anomaly patterns across all real-time windows is calculated to obtain the historical data reference weight. For example, the current trend slope of a real-time window is 0.03 A / s, and the current trend slope of a historical bearing failure window is 0.028 A / s; their correlation coefficient is 0.92, indicating a high degree of trend agreement. Of the 16 real-time windows, 11 have an average correlation ≥ 0.7 with the historical anomaly pattern. Therefore, the proportion of correlated windows is approximately 11 / 16 ≈ 0.69, meaning the historical data reference weight is 0.69.

[0052] Following the principle of prioritizing real-time data and supplementing with historical data, a weighting ratio of 7:3 (70% initial matching degree, 30% historical weight) is set. The initial matching degree and historical data reference weights are weighted and fused according to the weighting ratio to obtain a comprehensive diagnostic index. A fault determination threshold of 0.7 is set. This threshold is obtained through statistical analysis of a large number of historical fault records. Specifically, the comprehensive diagnostic index of a large number of historical fault records is calculated, and the 95th percentile is used to determine it. If the comprehensive diagnostic index exceeds the fault determination threshold, it indicates that the equipment corresponding to the data has failed and further analysis is required. For example, with an initial matching degree of 0.742 and a historical data reference weight of 0.69, the resulting comprehensive diagnostic index is 0.726, which is greater than the fault determination threshold. The comprehensive diagnostic index, real-time parameter vector, and historical abnormal pattern features are used as inputs and fed into a Bayesian classifier. The classifier loads preset fault prior probabilities (e.g., 32% historical occurrence probability of bearing failure, 28% probability of insulation aging). The posterior probabilities of various faults are calculated to obtain the confidence distribution. For example, the calculated confidence levels are 0.78 for bearing failure, 0.65 for insulation aging failure, and 0.43 for overload failure. A confidence threshold of 0.75 and a discrimination threshold of 0.10 are set. When the highest confidence value is greater than the preset confidence threshold, and the difference between it and the second-highest confidence value exceeds the preset discrimination threshold, the corresponding failure type is taken as the final fault diagnosis result, and the confidence level is marked as high confidence. For example, when the highest confidence value of bearing failure (0.78) is greater than the confidence threshold, and the difference between it and the second-highest confidence value (0.65) (0.13) exceeds the discrimination threshold, the reliability of the diagnosis result is confirmed, and the final diagnosis result is output as bearing failure.

[0053] In summary, this invention discloses a cross-device power adapter fault diagnosis method. By extracting a common feature set across devices through principal component analysis, it solves the problem of inconsistent diagnostic standards caused by differences in parameter dimensions and numerical distributions among different device models. By using a transformation matrix to correct the differences in feature scales between devices, the same fault mode across different devices is mapped to a unified feature space, enabling a single diagnostic system to adapt to multiple device models. This eliminates the need to develop separate diagnostic models for individual devices, significantly reducing system deployment and maintenance costs in multi-device scenarios. Through a dynamic standard optimization mechanism combined with gradient descent adaptive step size adjustment, the deviation between standard parameters and device operating characteristics is corrected in real time, ensuring that diagnostic results adapt to changes in device operating status and avoiding missed diagnoses due to rigid standards.

[0054] Reference Figure 3 This invention also provides a cross-device power adapter fault diagnosis system, capable of implementing all processes of the above-described power adapter fault diagnosis method, including: The data acquisition and preprocessing module is used to collect operating data and historical operating records of multiple equipment models, merge the operating data and the historical operating records and correct them to obtain an operating dataset; wherein, the operating data includes: voltage data, current data and temperature data; The feature extraction module is used to extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results, and perform vectorization processing to obtain a general feature vector. The fault classification module is used to calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an abnormal vector and classified using a support vector machine to obtain a fault classification label. A cross-device unified module is used to obtain fault feature vectors that match the fault classification labels, perform cross-device unified correction on the fault feature vectors, and extract fault distribution features from the corrected vectors. The preliminary diagnosis module is used to extract and integrate standardized fault modes based on the fault distribution characteristics and real-time feature vectors acquired in real time, so as to obtain an initial diagnostic standard applicable to cross-device diagnosis. The diagnostic criteria optimization module is used to acquire real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria. The final diagnostic module is used to monitor the operating status of each device and perform diagnosis using the optimized diagnostic criteria to obtain the final fault diagnosis result.

[0055] It should be noted that the cross-device power adapter fault diagnosis system provided in this embodiment of the invention is used to execute all the process steps of the cross-device power adapter fault diagnosis method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0056] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0057] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for diagnosing power adapter faults across devices, characterized in that, include: The system collects operating data and historical operating records from multiple equipment models, merges the operating data and historical operating records, and corrects them to obtain an operating dataset; wherein the operating data includes: voltage data, current data, and temperature data; Extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results, and perform vectorization processing to obtain a general feature vector. Calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, mark the general feature vector as an abnormal vector and use support vector machine for classification to obtain a fault classification label. Obtain the fault feature vector that matches the fault classification label, perform cross-device unified correction on the fault feature vector, and extract the fault distribution features from the corrected vector; Based on the fault distribution characteristics, combined with the real-time feature vectors acquired in real time, standardized fault modes are extracted and integrated to obtain initial diagnostic criteria applicable to cross-device diagnosis. Obtain real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria; Monitor the operating status of each device and use the optimized diagnostic criteria to perform diagnosis to obtain the final fault diagnosis result.

2. The power adapter fault diagnosis method according to claim 1, characterized in that, The process involves collecting operational data and historical operational records from multiple device models, merging the operational data with the historical operational records, and making corrections to obtain an operational dataset, including: Based on a pre-established equipment identification mapping table, voltage, current, and temperature data are collected from multiple equipment models, and the equipment model identification is recorded to obtain the first operating data; Extract the historical operation records corresponding to the device model identifier from the pre-established historical record table, and merge the historical operation records with the first operation data according to the time sequence to obtain the second operation data; If any value in the second running data exceeds the preset running data threshold, it is corrected by linear interpolation to obtain the third running data. If any value in the second running data does not exceed the preset running data threshold, the second running data is directly used as the third running data. The third running data is then converted to a unified format to obtain the running dataset.

3. The power adapter fault diagnosis method according to claim 1, characterized in that, The process involves extracting core parameter combinations from the runtime dataset, performing cross-device correlation analysis on these core parameter combinations, generating a general feature set based on the analysis results, and vectorizing it to obtain a general feature vector, including: The voltage, current, and temperature data in the aforementioned dataset are subjected to dimensionality reduction transformation using principal component analysis to obtain the core parameter combination. Calculate the correlation degree of each component in the core parameter combination; If the correlation value is greater than the preset correlation threshold, it is included in the general feature set; if the correlation value is less than the preset correlation threshold, it is not included in the general feature set. For each feature component of the general feature set, a corresponding general feature weight coefficient is assigned to obtain a weighted feature sequence; wherein, the general feature weight coefficient is normalized. The weighted feature sequences are combined and encapsulated according to a predefined vector format to obtain a general feature vector.

4. The power adapter fault diagnosis method according to claim 1, characterized in that, The step involves calculating the deviation between the general feature vector and a preset normal operation vector. If the deviation exceeds a preset deviation tolerance threshold, the general feature vector is marked as an anomaly vector and classified using a support vector machine to obtain a fault classification label, including: Calculate the deviation between the general feature vector and the preset normal operation vector; If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an anomaly vector. The anomaly vectors are classified using a support vector machine to obtain fault classification labels.

5. The power adapter fault diagnosis method according to claim 1, characterized in that, The step of performing cross-device unified correction on the fault feature vector and extracting fault distribution features from the corrected vector includes: Obtain the hardware configuration parameter matrix of each device. If the feature dimensions of different devices do not match, zero-padding is performed on the low-dimensional feature vectors to obtain a device feature matrix with a unified dimension. Based on the device feature matrix, the feature offset between different models of devices is calculated, and the parameters of the linear transformation matrix are determined. If the feature offset exceeds the preset alignment accuracy threshold, iterative optimization is performed until the feature offset meets the requirements, and the transformation matrix for cross-device feature alignment is obtained. Based on the transformation matrix, the fault feature vector is subjected to spatial mapping transformation, and the transformed distribution density is calculated to obtain the aligned feature distribution density map. Based on the feature distribution density map, the anomaly degree value of each feature vector relative to its neighboring points is calculated to obtain the fault distribution characteristics.

6. The power adapter fault diagnosis method according to claim 1, characterized in that, Based on the fault distribution characteristics and combined with the real-time feature vectors acquired in real time, standardized fault modes are extracted and integrated to obtain initial diagnostic criteria applicable to cross-device diagnosis, including: Based on the fault distribution characteristics, similar fault patterns are classified and merged by clustering to obtain a limited number of fault patterns. The feature parameters of the fault patterns are then normalized to obtain a standardized fault pattern library. Obtain real-time operating parameters of the device during operation, and obtain time window segments based on the standardized fault mode library and the real-time operating parameters, and extract the corresponding real-time feature vectors from them; The similarity between the standardized fault mode library and the real-time feature vector is calculated and weighted to obtain a matching degree matrix; The weighted features of each fault mode in the matching degree matrix are summed to obtain an initial diagnostic standard applicable to cross-device diagnosis.

7. The power adapter fault diagnosis method according to claim 1, characterized in that, The process of matching and analyzing the real-time operating parameters with the initial diagnostic criteria, and optimizing the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria, includes: Calculate the degree of matching between the real-time operating parameters and the initial diagnostic criteria; When the matching degree value is lower than the preset matching degree threshold, the deviation value of each component is calculated and combined to obtain the initial parameter deviation vector. For the initial parameter deviation vector, a gradient descent optimization strategy is used to optimize the step size. An error feedback mechanism is used to monitor the deviation change trend and obtain the optimal parameter update step size value. Specifically, when the deviation is decreasing, the current step size is maintained; when the deviation oscillates, the adjustment step size is reduced. For the key parameters in the initial diagnostic criteria, set corresponding update weight coefficients, update the step size value according to the optimal parameters, and iteratively update the initial diagnostic criteria. When the matching degree value between the updated diagnostic criteria and the real-time running parameters exceeds the matching degree threshold, the optimized diagnostic criteria are obtained.

8. The power adapter fault diagnosis method according to claim 1, characterized in that, The monitoring of the operating status of each device and the use of the optimized diagnostic criteria to perform diagnosis yield the final fault diagnosis result, including: Obtain the diagnostic operating parameters of multiple equipment models; Calculate the degree of matching between the operating parameters to be diagnosed and the optimized diagnostic criteria; Obtain historical operating data corresponding to the equipment model, calculate the correlation between the historical operating data and the operating parameter to be diagnosed, and obtain the historical data reference weight coefficient; The matching degree value and the historical data reference weight coefficient are weighted and fused to obtain a comprehensive diagnostic index; When the comprehensive diagnostic index exceeds the preset fault judgment threshold, the confidence level of each type of fault is calculated. When the highest confidence level is greater than the preset confidence threshold and the difference between the highest and second highest confidence levels exceeds the preset discrimination threshold, the corresponding fault type is taken as the final fault diagnosis result.

9. A cross-device power adapter fault diagnosis system, characterized in that, include: The data acquisition and preprocessing module is used to collect operating data and historical operating records of multiple equipment models, merge the operating data and the historical operating records and correct them to obtain an operating dataset; wherein, the operating data includes: voltage data, current data and temperature data; The feature extraction module is used to extract core parameter combinations from the running dataset, perform cross-device correlation analysis on the core parameter combinations, generate a general feature set based on the analysis results, and perform vectorization processing to obtain a general feature vector. The fault classification module is used to calculate the deviation value between the general feature vector and the preset normal operation vector. If the deviation value exceeds the preset deviation tolerance threshold, the general feature vector is marked as an abnormal vector and classified using a support vector machine to obtain a fault classification label. A cross-device unified module is used to obtain fault feature vectors that match the fault classification labels, perform cross-device unified correction on the fault feature vectors, and extract fault distribution features from the corrected vectors. The preliminary diagnosis module is used to extract and integrate standardized fault modes based on the fault distribution characteristics and real-time feature vectors acquired in real time, so as to obtain an initial diagnostic standard applicable to cross-device diagnosis. The diagnostic criteria optimization module is used to acquire real-time operating parameters, perform matching analysis between the real-time operating parameters and the initial diagnostic criteria, and optimize the initial diagnostic criteria based on the analysis results to obtain optimized diagnostic criteria. The final diagnostic module is used to monitor the operating status of each device and perform diagnosis using the optimized diagnostic criteria to obtain the final fault diagnosis result.

Images