Sensor-based explosion-proof junction box fault diagnosis method and system

Abstract

The invention discloses an explosion-proof junction box fault diagnosis method and system based on a sensor, which are used in the field of fault diagnosis, and the method comprises the following steps: obtaining historical multi-source sensor data reflecting the state of an explosion-proof junction box; extracting a time sequence sample of the preprocessed historical multi-source sensor data by using a sliding time window method; multi-domain features reflecting fault types are extracted from the time sequence samples, and a fault diagnosis model is constructed based on the feature combination with the optimal diagnosis performance; performing feature extraction on the multi-source sensor data acquired in real time, and inputting the extracted features into a fault diagnosis model for fault diagnosis; and the operation state and the diagnosis result of the explosion-proof junction box are displayed in real time. According to the method, the subtle change of the fault signal is captured from multiple angles, the Pearson correlation coefficient is used for screening the features significantly related to the fault type, and the dimension reduction processing is carried out through principal component analysis, so that the operation efficiency and accuracy of the model are improved.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a sensor-based explosion-proof junction box fault diagnosis method and system. Background Art

[0002] Explosion-proof junction boxes are specially designed equipment to ensure the safety of electrical connections used in flammable and explosive environments; they are mainly used in industries such as petroleum, chemical, and mining, and can effectively prevent explosions and fires caused by electrical sparks or high temperatures; through their special structural design and material selection, explosion-proof junction boxes can limit the sparks or heat generated by electrical equipment to a sealed casing, avoiding contact with external flammable gases or dust, thereby protecting the safety of equipment and personnel.

[0003] In practical applications, the design and manufacture of explosion-proof junction boxes must comply with strict safety standards and certification requirements to ensure their safety and reliability under extreme conditions; by using the fault diagnosis and monitoring system described in the present invention, the operational safety and maintenance efficiency of the explosion-proof junction box can be further improved, and real-time monitoring of its status can be achieved, potential fault problems can be discovered and handled in a timely manner, and the stability and safety of equipment operation can be ensured; many sensors, such as temperature, pressure, and humidity sensors, will convert the detected physical quantities into voltage or current signals, and the required physical parameter values ​​can be obtained by measuring the voltage or current changes.

[0004] In traditional methods, data quality is often affected by outliers, invalid values, and missing values. At the same time, the inconsistency of sampling frequency and timestamps between different sensors will also reduce the accuracy of data analysis. Due to the failure to effectively extract and select features, the data processing efficiency in traditional methods is low, and the accuracy of the model is limited by the interference of redundant and irrelevant features. There is a lack of effective real-time data collection and processing mechanisms, making it difficult to achieve dynamic monitoring of equipment status and timely reflect the real-time status of equipment. In addition, the response to fault prevention and response is not timely enough. There is a lack of intuitive user interface to display the equipment operating status and fault diagnosis results. There is also a lack of alarm information and maintenance suggestions automatically generated based on the diagnosis results, which affects the efficiency of equipment management and maintenance.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In order to overcome the above problems, the present invention aims to propose a sensor-based explosion-proof junction box fault diagnosis method and system, with the aim of solving the problem that in traditional methods, data quality is often affected by outliers, invalid values ​​and missing values. At the same time, the inconsistency of sampling frequency and timestamp between different sensors will also lead to reduced accuracy of data analysis. Due to the failure to effectively extract and select features, the data processing efficiency in traditional methods is low, and the accuracy of the model is limited by the interference of redundant and irrelevant features. There is a lack of effective real-time data acquisition and processing mechanism, making it difficult to realize dynamic monitoring of the equipment status and timely reflect the real-time status of the equipment.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a sensor-based explosion-proof junction box fault diagnosis method is provided, and the explosion-proof junction box fault diagnosis method comprises the following steps:

[0009] S1, obtaining historical multi-source sensor data reflecting the status of the explosion-proof junction box and performing preprocessing;

[0010] S2, extracting time series samples of preprocessed historical multi-source sensor data using the sliding time window method;

[0011] S3. Extract multi-domain features reflecting the fault type from the time series samples, use the Pearson correlation coefficient and principal component analysis method to remove redundant and irrelevant features, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance;

[0012] S4, extracting features from multi-source sensor data collected in real time, and inputting the extracted features into a fault diagnosis model for fault diagnosis;

[0013] S5. Build a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnostic results.

[0014] Optionally, extracting the time series samples of the preprocessed historical multi-source sensor data using the sliding time window method includes the following steps:

[0015] S21, detecting and removing abnormal values ​​and invalid values ​​in historical multi-source sensor data, and deleting missing values ​​in multi-source sensor data;

[0016] S22, time synchronization of historical multi-source sensor data to eliminate the impact of inconsistent sampling frequencies and timestamps of different sensors;

[0017] S23, normalizing the historical multi-source sensor data, setting a time window of fixed length, sliding on the time axis with a fixed step length, extracting the historical multi-source sensor data within the window, and forming a time series sample;

[0018] S24. Use oversampling method to balance the number of samples of different fault categories.

[0019] Optionally, using an oversampling method to balance the number of samples of different fault categories includes the following steps:

[0020] S241, for a fault category whose sample number is less than a preset reference category, randomly select a sample as a start;

[0021] S242, generating a new time series sample for the currently selected sample, and randomly selecting a time point in the time series of the time series sample;

[0022] S243, randomly selecting a time period before and after the selected time point, and cutting a data segment of the time period;

[0023] S244, inserting the cut data segment into a random position of the time series sample to form a new time series sample;

[0024] S245. Repeat steps S222-S244 until the number of samples in the category is the same as the number of reference categories.

[0025] Optionally, extracting multi-domain features reflecting the fault type from the time series samples, removing redundant and irrelevant features by using the Pearson correlation coefficient and principal component analysis method, obtaining a feature combination with the best diagnostic performance, and constructing a fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps:

[0026] S31, extracting multi-domain features including time domain features, frequency domain features and time-frequency domain features from the time series samples;

[0027] S32, using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable, screening out a feature subset associated with the preset target variable;

[0028] S33, normalizing the feature subset;

[0029] S34, using principal component analysis to perform dimensionality reduction processing on the normalized feature subset;

[0030] S35. Use the random forest model to evaluate the diagnostic performance of the feature combination on the feature subset after dimensionality reduction, determine the feature combination with the best diagnostic performance based on the evaluation results, and build the final fault diagnosis model based on the feature combination with the best diagnostic performance.

[0031] Optionally, using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable, screening out a feature subset associated with the preset target variable comprises the following steps:

[0032] S321. For each feature, calculate the Pearson correlation coefficient between each feature and the target variable;

[0033] S322. Based on the calculated Pearson correlation coefficient, a threshold is set, and features associated with the target variable are screened according to the set threshold;

[0034] Among them, the formula for calculating the Pearson correlation coefficient between each feature and the target variable is:

[0035]

[0036] In the formula, r represents the Pearson correlation coefficient between each feature and the target variable;

[0037] m a represents the observed value of the eigenvalue;

[0038] n a represents the observed value of the target variable;

[0039] represents the average value of the eigenvalues;

[0040] represents the mean value of the target variable.

[0041] Optionally, performing dimensionality reduction processing on the normalized feature subset using principal component analysis includes the following steps:

[0042] S341, calculating the covariance matrix of the normalized feature subset, and calculating all eigenvalues ​​and corresponding eigenvectors based on the covariance matrix;

[0043] S343, according to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components;

[0044] S344, constructing a projection matrix using the eigenvector;

[0045] S345, using a projection matrix to transform the normalized data into a new feature space composed of principal components, to obtain a feature subset after dimensionality reduction;

[0046] Among them, the formula for calculating the covariance matrix of the normalized feature subset is:

[0047]

[0048] In the formula, C ijrepresents the covariance between feature i and feature j;

[0049] n represents the number of samples in the feature subset;

[0050] k represents the sum index;

[0051] X ki Represents the value of the i-th feature of the k-th sample;

[0052] X kj Represents the value of the jth feature of the kth sample.

[0053] Optionally, using a random forest model, evaluating the diagnostic performance of a feature combination on a feature subset after dimensionality reduction, determining a feature combination with the best diagnostic performance according to the evaluation results, and constructing a final fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps:

[0054] S351, initializing the random forest model and setting the parameters of the random forest model;

[0055] S352, using the training set to train the random forest model, during the training process, constructing multiple decision trees by randomly selecting features and data samples;

[0056] S353. In multiple decision trees, the feature importance evaluation function of the random forest model is used to evaluate the contribution score of each feature to the model prediction performance;

[0057] S354. Filter the features that contribute most to model performance according to the contribution score;

[0058] S355, Use the test set to evaluate the diagnostic performance of the random forest model;

[0059] S356, based on the contribution score of the performance evaluation results of the random forest model, screen the best feature combination;

[0060] S357. Use the determined optimal feature combination and the adjusted model parameters to retrain the random forest model and construct the final fault diagnosis model.

[0061] Optionally, in multiple decision trees, using the feature importance evaluation function of the random forest model, evaluating the contribution score of each feature to the model prediction performance includes the following steps:

[0062] S3531. For each decision tree, traverse all nodes, and for each node that is split based on the feature, calculate the reduction in impurity caused by the split;

[0063] S3532. For each feature, the impurity reduction amount caused by all split points of each feature on all trees is accumulated to obtain the total impurity reduction amount of the feature in the entire random forest model;

[0064] S3533. Divide the total impurity reduction by the total number of trees in the random forest to calculate the average impurity reduction of the feature, and use the average impurity reduction as the importance score of the feature;

[0065] S3534. Add the importance scores of all features to obtain a total importance score;

[0066] S3535. Divide the importance score of each feature by the total score to obtain a normalized importance score;

[0067] S3536. Sort all features by their importance scores according to the normalized importance scores;

[0068] S3537. Select the highest ranked feature as the feature that contributes most to the model's prediction performance.

[0069] Optionally, extracting features from multi-source sensor data collected in real time and inputting the extracted features into a fault diagnosis model for fault diagnosis includes the following steps:

[0070] S41, obtaining multi-source sensor data of the explosion-proof junction box monitored by the multi-source sensor in real time;

[0071] S42, extracting features from multi-source sensor data, integrating all extracted features, and forming a feature vector;

[0072] S43, inputting the feature vector into the fault diagnosis model, and outputting the diagnosis result of the fault diagnosis according to the input feature vector;

[0073] The step of inputting the feature vector into the fault diagnosis model and outputting the fault diagnosis result according to the input feature vector comprises the following steps:

[0074] S431, loading a pre-trained fault diagnosis model;

[0075] S432, providing the characteristic vector as input data to a fault diagnosis model;

[0076] S433, the fault diagnosis model receives the input feature vector, performs calculations using the patterns learned during training, identifies known fault patterns based on the calculation results, and outputs a specific fault category.

[0077] Optionally, constructing a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generating alarm information and maintenance suggestions based on the diagnostic results includes the following steps:

[0078] S51, real-time monitoring and collection of multi-source sensor data of explosion-proof junction boxes;

[0079] S52, using a pre-trained fault diagnosis model to identify faults based on the processed data, and automatically generating alarm information for a specific fault category according to the diagnosis result;

[0080] S53. Build a visual display interface to display the operation status and diagnosis results of the junction box, and integrate a user feedback mechanism.

[0081] According to another aspect of the present invention, there is also provided a sensor-based explosion-proof junction box fault diagnosis system, the system comprising: a data acquisition module, a sample extraction module, a model building module, a fault diagnosis module and a visualization and alarm module;

[0082] The data acquisition module is connected to the model building module through the sample extraction module, and the model building module is connected to the visualization and alarm module through the fault diagnosis module;

[0083] A data acquisition module, used to obtain multi-source sensor data reflecting the status of the explosion-proof junction box;

[0084] A sample extraction module is used to extract time series samples of preprocessed historical multi-source sensor data using a sliding time window method;

[0085] Model building module, used to extract multi-domain features reflecting fault types from time series samples, remove redundant and irrelevant features using Pearson correlation coefficient and principal component analysis, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance;

[0086] The fault diagnosis module is used to extract features from multi-source sensor data collected in real time and input the extracted features into the fault diagnosis model for fault diagnosis;

[0087] The visualization and alarm module is used to build a visualization display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnostic results.

[0088] Compared with the prior art, this application has the following beneficial effects:

[0089] 1. The present invention significantly improves the quality of data by detecting and removing outliers, invalid values ​​and missing values, ensuring that the basic data for subsequent analysis and model training is more accurate and reliable. The execution of time synchronization eliminates the problem of inconsistent sampling frequencies and timestamps between different sensors, provides a unified time reference for all sensor data, and ensures consistency and comparability between data. Normalization processing unifies data of different ranges and units to the same scale, eliminates the impact of different dimensions, makes model training more efficient, and improves the adaptability and accuracy of the model to features of different scales. The sliding time window method is used to effectively capture the temporal characteristics of the data, providing richer information for fault diagnosis model learning, thereby improving the diagnostic accuracy of the model. The oversampling method balances the number of samples of different fault categories, effectively alleviates the problem of data imbalance, improves the model's diagnostic ability for minority categories, and enhances the generalization performance of the model.

[0090] 2. The present invention uses the comprehensive extraction of time domain, frequency domain, and time-frequency domain features. The model can capture subtle changes in fault signals from multiple angles, providing rich information for accurate diagnosis. The Pearson correlation coefficient is used to screen features that are significantly related to the fault type, and the principal component analysis is used for dimensionality reduction, which effectively removes redundant and irrelevant features and improves the operational efficiency and accuracy of the model. The feature set after dimensionality reduction reduces the computational complexity of the model while retaining the most diagnostically valuable information, making the model more concise and effective. The feature set after dimensionality reduction reduces the computational complexity of the model while retaining the most diagnostically valuable information, making the model more concise and effective. The random forest model is used to evaluate the diagnostic performance of the feature combination, and the optimal feature combination can be determined based on actual performance, thereby constructing a high-accuracy fault diagnosis model. The selection of model parameters and feature combinations is based on a data-driven approach, which makes the diagnostic model highly adaptable and flexible, and can perform self-optimization according to different fault types and equipment characteristics.

[0091] 3. The present invention can dynamically monitor the operating status of the explosion-proof junction box by collecting multi-source sensor data in real time, timely reflect the real-time status of the equipment, and provide support for fault prevention and timely response. It uses feature extraction technology, such as time domain, frequency domain and time-frequency domain analysis, to extract meaningful information from multi-source data, ensuring the accuracy and effectiveness of fault diagnosis. The random forest model, with its advantages of integrated learning, improves the accuracy and stability of fault diagnosis by integrating the judgments of multiple decision trees, and is suitable for processing complex data features and various types of fault diagnosis.

[0092] 4. The visual display interface constructed by the present invention not only displays the equipment operation status and fault diagnosis results in real time, but also automatically generates alarm information and maintenance suggestions based on the diagnosis results, which greatly improves the efficiency of equipment management and maintenance. Through timely fault detection and diagnosis, as well as an effective alarm mechanism, it helps to prevent the occurrence of potential faults and ensure the safety and stability of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0094] Figure 1 is a flow chart of a sensor-based explosion-proof junction box fault diagnosis method according to an embodiment of the present invention;

[0095] Figure 2 The invention is a principle block diagram of a sensor-based explosion-proof junction box fault diagnosis system according to an embodiment of the invention.

[0096] In the figure:

[0097] 1. Data acquisition module; 2. Sample extraction module; 3. Model building module; 4. Fault diagnosis module; 5. Visualization and alarm module. DETAILED DESCRIPTION

[0098] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0099] According to an embodiment of the present invention, a sensor-based explosion-proof junction box fault diagnosis method and system are provided.

[0100] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a sensor-based explosion-proof junction box fault diagnosis method is provided, and the explosion-proof junction box fault diagnosis method includes the following steps:

[0101] S1. Acquire historical multi-source sensor data reflecting the status of the explosion-proof junction box and perform preprocessing.

[0102] It needs to be explained that, first of all, according to the monitoring requirements of the explosion-proof junction box, the appropriate type of sensor should be selected. The sensors include temperature sensors, pressure sensors, current and voltage sensors, humidity sensors, vibration sensors, etc. Each sensor is used to monitor a specific parameter in the operation of the junction box. When selecting a sensor, its accuracy, response time, working environment range (such as temperature, pressure, etc.) and whether it is suitable for an explosion-proof environment should be considered; when arranging sensors, it is necessary to ensure that they can accurately monitor the key parts and key parameters in the junction box; for example, temperature sensors should be placed near components that are prone to overheating, current and voltage sensors should be connected to appropriate locations in the circuit, and data should be transmitted through preset communication protocols (such as Modbus, CAN bus, Ethernet, etc.). The implementation of data acquisition usually involves the use of microcontrollers and PLCs (programmable logic controllers) to continuously read sensor data and transmit the data to a cloud platform through an industrial communication network.

[0103] S2. Use the sliding time window method to extract time series samples of preprocessed historical multi-source sensor data.

[0104] Preferably, extracting the time series samples of the preprocessed historical multi-source sensor data using the sliding time window method comprises the following steps:

[0105] S21, detecting and removing abnormal values ​​and invalid values ​​in the historical multi-source sensor data, and deleting missing values ​​in the historical multi-source sensor data;

[0106] S22. Time-synchronize historical multi-source sensor data to eliminate the impact of inconsistent sampling frequencies and timestamps of different sensors (assign an accurate timestamp to each sensor data point, then align data from different sensors based on the timestamp, and synchronize the clocks of all sensors and data records, effectively eliminating the impact of inconsistent sampling frequencies and timestamps of different sensors);

[0107] S23. Normalize the historical multi-source sensor data, set a time window of fixed length, slide on the time axis with a fixed step size, extract the historical multi-source sensor data within the window, and form a time series sample (convert the data from different sensors with different ranges and units to the same scale, usually [0, 1] or [-1, 1] interval, to eliminate the influence of different dimensions. Commonly used normalization techniques include minimum-maximum normalization and Z-score normalization. Minimum-maximum normalization adjusts the data by subtracting the minimum value and dividing it by the difference between the maximum and minimum values, while Z-score normalization adjusts the data by subtracting the minimum value and dividing it by the difference between the maximum and minimum values. The data is converted by adding the mean and dividing it by the standard deviation. For each sensor data set, the maximum value, minimum value, mean value and standard deviation of the whole or window are calculated, and then the above normalization formula is applied. By setting a window of fixed length on the time axis and sliding the window with a fixed step size, a series of time series samples are extracted from the continuous data stream to capture the timing characteristics of the data and provide rich information for subsequent analysis. The first window is set from the beginning of the data sequence, and the data in the window is extracted as a sample. Then the window is moved according to the set step size, and the extraction process is repeated until the end of the data sequence is reached);

[0108] S24. Use oversampling method to balance the number of samples of different fault categories.

[0109] Preferably, using the oversampling method to balance the number of samples of different fault categories includes the following steps:

[0110] S241, for a fault category whose sample number is less than a preset reference category, randomly select a sample as a start;

[0111] S242, generating a new time series sample for the currently selected sample, and randomly selecting a time point in the time series of the time series sample;

[0112] S243, randomly selecting a time period before and after the selected time point, and cutting the data fragment of the time period (randomly selecting a time point from the time series sample, the time point is used as the reference point for subsequent operations, and a time period is selected around it for cutting and inserting operations. Before and after the selected time point, the length of a time period is randomly determined. This time period can be a fixed length or a length randomly selected from a preset range to increase the diversity of results. According to the determined time period, the corresponding data fragment is cut from the original time series sample to ensure that the cutting operation does not destroy the continuity of the data or introduce unreasonable deviations. An appropriate position is selected to insert the cut data fragment into the same time series sample or another sample. The insertion position can be randomly selected, but the continuity and logical consistency of the data should be taken into consideration);

[0113] S244, inserting the cut data segment into a random position of the time series sample to form a new time series sample;

[0114] S245. Repeat steps S222-S244 until the number of samples in the category is the same as the number of reference categories.

[0115] It should be explained that by detecting and removing outliers and invalid values ​​and processing missing values, data quality can be significantly improved. Time synchronization is the key because it solves the problem of inconsistent sampling frequencies and timestamps of different sensors and ensures the temporal consistency of data. Normalization and the method of extracting samples using a sliding time window method allow structured time series samples to be obtained from continuous sensor data streams. Time series samples are crucial for subsequent pattern recognition and fault diagnosis. Finally, the sample set is balanced by oversampling methods, especially the number of samples for different fault categories, which can alleviate the model bias problem caused by sample imbalance. New samples are generated by randomly selecting time points in time series samples and cutting and inserting data fragments. This innovative data enhancement method not only improves the diversity of data, but also helps to improve the model's recognition ability for minority categories.

[0116] S3. Extract multi-domain features reflecting the fault type from time series samples, use Pearson correlation coefficient and principal component analysis to remove redundant and irrelevant features, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance.

[0117] Preferably, extracting multi-domain features reflecting the fault type from the time series samples, using the Pearson correlation coefficient and principal component analysis method to remove redundant and irrelevant features, obtaining a feature combination with the best diagnostic performance, and constructing a fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps:

[0118] S31. Extract multi-domain features including time domain features, frequency domain features and time-frequency domain features from time series samples (extract multi-domain features from time series samples, including time domain features such as mean, variance, kurtosis, etc., frequency domain features such as frequency components, energy spectrum density, etc. and time-frequency domain features such as wavelet transform coefficients, use mathematical models and algorithms in different domains to fully capture the characteristics of the data and provide rich information for diagnosis);

[0119] S32, using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable, screening out a feature subset associated with the preset target variable;

[0120] S33, normalizing the feature subset (normalizing the selected feature subset to ensure that all features are on the same scale to avoid excessive impact of some features on the model due to their large value range);

[0121] S34, using principal component analysis to perform dimensionality reduction processing on the normalized feature subset;

[0122] S35. Use the random forest model to evaluate the diagnostic performance of the feature combination on the feature subset after dimensionality reduction, determine the feature combination with the best diagnostic performance based on the evaluation results, and build the final fault diagnosis model based on the feature combination with the best diagnostic performance.

[0123] In addition, it should be noted that the complete process of extracting multi-domain features from time series samples, using Pearson correlation coefficient and principal component analysis (PCA) for feature screening and dimensionality reduction, and finally using random forest to evaluate the diagnostic performance of feature combination and construct the final fault diagnosis model, firstly, by extracting time domain, frequency domain and time-frequency domain features to comprehensively capture the characteristics of the fault signal, then using Pearson correlation coefficient to select feature subsets that are significantly related to the fault type, and through normalization processing to ensure that the features are in the same dimension, PCA is applied for effective feature dimensionality reduction to reduce computational complexity while retaining key information. Finally, the feature combination with the best diagnostic performance is determined through the evaluation of the random forest model, and a fault diagnosis model is constructed based on this combination.

[0124] The Pearson Correlation Coefficient is a statistical indicator that measures the degree of linear correlation between two quantitative variables. Its value range is between -1 and +1, where +1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 means that there is no linear relationship between the two variables. The Pearson Correlation Coefficient reflects the trend of one variable changing when another variable changes.

[0125] Principal component analysis (PCA) is a powerful data dimensionality reduction tool that transforms related variables into a set of linearly independent variables, called principal components, through orthogonal transformation. The purpose of PCA is to reduce the dimensionality of a dataset while retaining the most important variability in the data.

[0126] Preferably, using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable, screening out the feature subset associated with the preset target variable comprises the following steps:

[0127] S321. For each feature, calculate the Pearson correlation coefficient between each feature and the target variable (the selection of the preset target variable is based on the needs of the actual application scenario and the identification requirements for a specific fault type. The main purpose of building a fault diagnosis model is to accurately identify or predict the value of the preset target variable, that is, to determine the health status of the equipment or the specific fault type. By extracting and analyzing multi-domain features in the time series samples, such as time domain, frequency domain and time-frequency domain features, the model aims to capture patterns or features related to a specific fault type);

[0128] S322. Based on the calculated Pearson correlation coefficient, a threshold is set, and features associated with the target variable are screened according to the set threshold;

[0129] Among them, the formula for calculating the Pearson correlation coefficient between each feature and the target variable is:

[0130]

[0131] In the formula, r represents the Pearson correlation coefficient between each feature and the target variable;

[0132] m a represents the observed value of the eigenvalue;

[0133] n a represents the observed value of the target variable;

[0134] represents the average value of the eigenvalues;

[0135] represents the mean value of the target variable.

[0136] Preferably, the dimensionality reduction process of the normalized feature subset using principal component analysis includes the following steps:

[0137] S341, calculating the covariance matrix of the normalized feature subset, and calculating all eigenvalues ​​and corresponding eigenvectors based on the covariance matrix;

[0138] S343, according to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components;

[0139] S344, constructing a projection matrix using the eigenvector;

[0140] S345, using a projection matrix to transform the normalized data into a new feature space composed of principal components, to obtain a feature subset after dimensionality reduction;

[0141] Among them, the formula for calculating the covariance matrix of the normalized feature subset is:

[0142]

[0143] In the formula, C ij represents the covariance between feature i and feature j;

[0144] n represents the number of samples in the feature subset;

[0145] k represents the sum index;

[0146] X ki Represents the value of the i-th feature of the k-th sample;

[0147] X kj Represents the value of the jth feature of the kth sample.

[0148] Preferably, using the random forest model, evaluating the diagnostic performance of the feature combination on the feature subset after dimensionality reduction, determining the feature combination with the best diagnostic performance according to the evaluation results, and constructing the final fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps:

[0149] S351, initializing the random forest model and setting the parameters of the random forest model;

[0150] S352, using the training set to train the random forest model, during the training process, constructing multiple decision trees by randomly selecting features and data samples;

[0151] S353. In multiple decision trees, use the feature importance evaluation function of the random forest model to evaluate the contribution score of each feature to the model prediction performance;

[0152] S354. Filter the features that contribute most to model performance according to the contribution score;

[0153] S355, Use the test set to evaluate the diagnostic performance of the random forest model;

[0154] S356, based on the contribution score of the performance evaluation results of the random forest model, screen the best feature combination;

[0155] S357. Use the determined optimal feature combination and the adjusted model parameters to retrain the random forest model and construct the final fault diagnosis model.

[0156] Preferably, in multiple decision trees, using the feature importance evaluation function of the random forest model, evaluating the contribution score of each feature to the model prediction performance includes the following steps:

[0157] S3531. For each decision tree, traverse all nodes, and for each node that is split based on the feature, calculate the reduction in impurity caused by the split (in the random forest model, calculating the reduction in impurity caused by the split is a key step in evaluating the importance of the feature. This process involves selecting the best split feature and threshold for the data set of each node when building the decision tree, based on the goal of reducing the impurity of the data set. The reduction in impurity is calculated as the difference in impurity before and after the split. By accumulating the reduction in impurity caused by all splits based on the same feature, the total contribution of the feature can be evaluated);

[0158] S3532. For each feature, the impurity reduction amount caused by all split points of each feature on all trees is accumulated to obtain the total impurity reduction amount of the feature in the entire random forest model;

[0159] S3533. Divide the total impurity reduction by the total number of trees in the random forest to calculate the average impurity reduction of the feature, and use the average impurity reduction as the importance score of the feature;

[0160] S3534. Add the importance scores of all features to obtain a total importance score;

[0161] S3535. Divide the importance score of each feature by the total score to obtain a normalized importance score;

[0162] S3536. Sort all features by their importance scores according to the normalized importance scores;

[0163] S3537. Select the highest ranked feature as the feature that contributes most to the model's prediction performance.

[0164] It should be explained that the diagnostic performance of the feature combination is evaluated by using the random forest model, and the best feature combination is determined for building the fault diagnosis model. This process starts with initializing the random forest model and setting its parameters. Multiple decision trees are constructed by randomly selecting features and data samples during the training process. The contribution of each feature is then determined using the feature importance evaluation function of the random forest. The features that contribute most to the model performance are screened out based on the contribution. Finally, the model is retrained based on the features to build the final fault diagnosis model. The impurity reduction of all nodes in the decision tree is evaluated, the total impurity reduction of the features is accumulated, and then the average impurity reduction of the features is calculated as the feature importance score. After the score is normalized, it can be determined which features contribute the most to the model prediction performance, and then the best feature combination is selected. This method based on feature importance evaluation and selection not only optimizes the performance of the model, but also improves the interpretability of the model, because it can be clearly known which features have the greatest impact on the prediction results.

[0165] S4. Extract features from multi-source sensor data collected in real time, and input the extracted features into a fault diagnosis model for fault diagnosis.

[0166] Preferably, extracting features from multi-source sensor data collected in real time and inputting the extracted features into a fault diagnosis model for fault diagnosis comprises the following steps:

[0167] S41. Acquire multi-source sensor data of the explosion-proof junction box that is monitored in real time by multi-source sensors (multi-source sensors deployed around the explosion-proof junction box, such as temperature, humidity, pressure and other sensors, monitor relevant parameters in real time. The data can be transmitted to the data processing center in real time through a wired or wireless network. The collected multi-source sensor data is pre-processed, such as denoising, normalization, etc., and then relevant time domain, frequency domain and time-frequency domain features are extracted according to actual application requirements. The features include the mean, standard deviation, energy, frequency component, etc. of the signal);

[0168] S42, extracting features from multi-source sensor data collected in real time, integrating all extracted features, and forming a feature vector;

[0169] S43, inputting the feature vector into the fault diagnosis model, and outputting the diagnosis result of the fault diagnosis according to the input feature vector;

[0170] The step of inputting the feature vector into the fault diagnosis model and outputting the fault diagnosis result according to the input feature vector comprises the following steps:

[0171] S431, loading a pre-trained fault diagnosis model;

[0172] S432, providing the characteristic vector as input data to a fault diagnosis model;

[0173] S433, the fault diagnosis model receives the input feature vector, performs calculations using the patterns learned during training, identifies known fault patterns based on the calculation results, and outputs a specific fault category.

[0174] The fault diagnosis model receives the input feature vector, performs calculations using the patterns learned during training, identifies known fault modes based on the calculation results, and outputs a specific fault category, including the following steps:

[0175] Receive feature vectors from preprocessing steps (such as feature extraction) and verify the integrity and format of the feature vectors to ensure that they meet the input specifications required by the model;

[0176] The fault diagnosis model uses the internal random forest model to analyze and process the input feature vector according to the weights and patterns learned during training;

[0177] The fault diagnosis model determines the fault mode that the feature vector best matches through a classification algorithm based on its calculation results;

[0178] Once a specific failure mode is identified, the model will output the corresponding failure category; the output format can be a label of the failure type, and may also include the probability or confidence score of the failure occurring.

[0179] The following steps are included in determining the most suitable fault mode of the feature vector by using the classification algorithm:

[0180] Select appropriate classification algorithms based on the characteristics of the data and the expected output, such as support vector machines (SVMs), decision trees, deep neural networks, etc.;

[0181] Ensure that the input feature vector has been properly preprocessed, including normalization, standardization, or other transformations, to meet the input requirements of the model;

[0182] The processed feature vector is input into the trained classification model, which uses the learned patterns and weights to evaluate the input feature vector and determine its most likely fault category;

[0183] The model outputs specific fault categories, including labels of fault types and associated confidence or probability scores;

[0184] Verify the results of the model output to confirm its accuracy and reliability. It should be explained that the fault diagnosis model is essentially a random forest model, which consists of multiple decision trees. Each tree randomly selects different data samples and features for learning during the training process. When constructing each decision tree, the random forest algorithm not only increases diversity by randomly selecting data subsets (row sampling), but also further increases randomness by randomly selecting feature subsets (column sampling) at each decision node. When the random forest model is used for prediction, the new input feature vector will be sent to each decision tree, and each tree will give a prediction result. The final prediction result of random forest is obtained by summarizing the results of all decision trees. For classification tasks, the "majority voting" method is usually used; for regression tasks, the average of all tree prediction results is calculated. First, key features are extracted from multi-source sensor data collected by monitoring equipment (such as explosion-proof junction boxes). Features can be time domain, frequency domain or time-frequency domain features, reflecting the operating status of the equipment. The extracted features and known equipment fault records are used to train the random forest model. By adjusting the model parameters (such as the number of trees, the depth of the tree, the number of features considered when splitting nodes, etc.) during the training process, the performance of the model can be optimized. Random forest can provide feature importance scores to help identify which features are most critical for fault diagnosis, which is very useful for understanding the root cause of the fault and improving feature engineering. For newly collected sensor data, the same feature set is extracted and input into the trained random forest model for fault diagnosis. The diagnostic performance of the model can be determined by evaluating indicators such as the accuracy and recall rate of the model.

[0185] S5. Build a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnostic results.

[0186] Preferably, constructing a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generating alarm information and maintenance suggestions according to the diagnostic results includes the following steps:

[0187] S51, real-time monitoring and collection of multi-source sensor data of explosion-proof junction boxes;

[0188] S52, using a pre-trained fault diagnosis model to identify faults based on the processed data, and automatically generating alarm information for a specific fault category according to the diagnosis result;

[0189] S53. Build a visual display interface to display the running status and diagnostic results of the junction box, and integrate the user feedback mechanism. It should be explained that the intuitive user interface is designed through front-end technology (such as React or Vue with ECharts or D3.js), and the smooth transmission and processing of data is ensured through back-end real-time data processing. The user interface must include key information such as equipment status dashboard, fault diagnosis results, alarm information and maintenance suggestions. The back-end is responsible for receiving sensor data, feature extraction and fault diagnosis, and triggering corresponding alarm rules according to the diagnosis results, automatically generating alarm information and maintenance suggestions. Adding user authority management and historical data analysis functions can further enhance security and practicality. Through intelligent analysis of historical maintenance data, the system can also provide suggestions for optimizing maintenance and equipment performance, and improve the overall monitoring maintenance efficiency and equipment reliability.

[0190] In addition, it should be noted that various types of sensors, such as temperature sensors, humidity sensors, pressure sensors, etc., are deployed inside and outside the explosion-proof junction box to monitor and record key operating parameters; a data acquisition unit (such as a PLC or dedicated data acquisition hardware) is used to convert the analog signals of the sensors into digital signals for easy processing and transmission; the collected data is transmitted to a central monitoring system or cloud platform in real time through a network (such as Ethernet, wireless network, etc.); when processing data from different sensors, the data is fused to improve the accuracy and reliability of fault detection; the collected and pre-processed data is input into a trained fault diagnosis model. These models may be based on machine learning techniques, such as random forests, neural networks, etc.; the model analyzes the input data, identifies abnormal patterns in the data, and matches these patterns with known fault types; once the model identifies a specific fault category, it automatically triggers the alarm mechanism, sends an alarm and necessary fault information to the operator, and designs an intuitive user interface to display all key monitoring data and fault alarms. The interface design should take into account the user's operational convenience and information readability, and use elements such as charts, dashboards, and real-time images to dynamically display data. According to another embodiment of the present invention, Figure 2 As shown, a sensor-based explosion-proof junction box fault diagnosis system includes: a data acquisition module 1, a sample extraction module 2, a model building module 3, a fault diagnosis module 4 and a visualization and alarm module 5;

[0191] The data acquisition module 1 is connected to the model building module 3 via the sample extraction module 2, and the model building module 3 is connected to the visualization and alarm module 5 via the fault diagnosis module 4;

[0192] The data acquisition module 1 is used to obtain multi-source sensor data reflecting the status of the explosion-proof junction box;

[0193] Sample extraction module 2, used to extract time series samples of pre-processed historical multi-source sensor data using a sliding time window method;

[0194] Model building module 3 is used to extract multi-domain features reflecting fault types from time series samples, remove redundant and irrelevant features using Pearson correlation coefficient and principal component analysis, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance;

[0195] The fault diagnosis module 4 is used to extract features from multi-source sensor data collected in real time, and input the extracted features into the fault diagnosis model for fault diagnosis;

[0196] The visualization and alarm module 5 is used to construct a visualization display interface, display the operating status and diagnosis results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnosis results.

[0197] In summary, with the help of the above technical solution of the present invention, the present invention significantly improves the quality of data by detecting and removing outliers, invalid values ​​and missing values, ensuring that the basic data for subsequent analysis and model training is more accurate and reliable, and performing time synchronization eliminates the problem of inconsistent sampling frequencies and timestamps between different sensors, providing a unified time reference for all sensor data, ensuring consistency and comparability between data, and normalization processing unifies data of different ranges and units to the same scale, eliminating the impact of different dimensions, making model training more efficient, and improving the adaptability and accuracy of the model to features of different scales. The sliding time window method is used to effectively capture the temporal characteristics of the data, providing a richer The information is used for fault diagnosis model learning, thereby improving the diagnostic accuracy of the model. The oversampling method balances the number of samples of different fault categories, effectively alleviates the problem of data imbalance, improves the model's diagnostic ability for a minority of categories, and enhances the generalization performance of the model. The present invention uses the comprehensive extraction of time domain, frequency domain, and time-frequency domain features. The model can capture subtle changes in fault signals from multiple angles, providing rich information for accurate diagnosis. The Pearson correlation coefficient is used to screen features significantly related to the fault type, and the principal component analysis is used for dimensionality reduction processing, which effectively removes redundant and irrelevant features and improves the model's operational efficiency and accuracy. The feature set after dimensionality reduction processing reduces the model's computational complexity while retaining the most diagnostic The feature set after dimensionality reduction reduces the computational complexity of the model while retaining the most valuable information for diagnosis, making the model more concise and effective. The random forest model is used to evaluate the diagnostic performance of the feature combination, and the best feature combination can be determined based on the actual performance, thereby building a high-accuracy fault diagnosis model. The selection of model parameters and feature combinations is based on a data-driven method, making the diagnosis model highly adaptable and flexible, and able to self-optimize according to different fault types and equipment characteristics. The present invention can dynamically monitor the operating status of the explosion-proof junction box by collecting multi-source sensor data in real time, timely reflect the real-time status of the equipment, and provide support for fault prevention and timely response. The invention adopts feature extraction technology, such as time domain, frequency domain and time-frequency domain analysis, to extract meaningful information from multi-source data, thereby ensuring the accuracy and effectiveness of fault diagnosis. The random forest model, with its advantages of integrated learning, improves the accuracy and stability of fault diagnosis by integrating the judgments of multiple decision trees, and is suitable for processing complex data features and various types of fault diagnosis. The visual display interface constructed by the invention not only displays the equipment operation status and fault diagnosis results in real time, but also automatically generates alarm information and maintenance suggestions according to the diagnosis results, thereby greatly improving the efficiency of equipment management and maintenance. Through timely fault detection and diagnosis, as well as an effective alarm mechanism, it helps to prevent the occurrence of potential faults and ensure the safety and stability of equipment operation.

[0198] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A sensor-based explosion-proof junction box fault diagnosis method, characterized in that: The explosion-proof junction box fault diagnosis method comprises the following steps: S1, obtaining historical multi-source sensor data reflecting the status of the explosion-proof junction box and performing preprocessing; S2, extracting time series samples of preprocessed historical multi-source sensor data using the sliding time window method; S3. Extract multi-domain features reflecting the fault type from the time series samples, use the Pearson correlation coefficient and principal component analysis method to remove redundant and irrelevant features, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance; S4, extracting features from multi-source sensor data collected in real time, and inputting the extracted features into a fault diagnosis model for fault diagnosis; S5. Build a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnostic results.

2. A sensor-based explosion-proof junction box fault diagnosis method according to claim 1, characterized in that: The method of extracting the time series samples of the preprocessed historical multi-source sensor data by using the sliding time window method comprises the following steps: S21, detecting and removing abnormal values ​​and invalid values ​​in multi-source sensor data, and deleting missing values ​​in historical multi-source sensor data; S22, time synchronization of historical multi-source sensor data to eliminate the impact of inconsistent sampling frequencies and timestamps of different sensors; S23, normalizing the historical multi-source sensor data, setting a time window of fixed length, sliding on the time axis with a fixed step length, extracting the historical multi-source sensor data within the window, and forming a time series sample; S24. Use oversampling method to balance the number of samples of different fault categories.

3. A sensor-based explosion-proof junction box fault diagnosis method according to claim 2, characterized in that: The method of balancing the number of samples of different fault categories by using the oversampling method comprises the following steps: S241, for a fault category whose sample number is less than a preset reference category, randomly select a sample as a start; S242, generating a new time series sample for the currently selected sample, and randomly selecting a time point in the time series of the time series sample; S243, randomly selecting a time period before and after the selected time point, and cutting a data segment of the time period; S244, inserting the cut data segment into a random position of the time series sample to form a new time series sample; S245. Repeat steps S222-S244 until the number of samples in the category is the same as the number of reference categories.

4. The sensor-based explosion-proof junction box fault diagnosis method according to claim 1, characterized in that: The method of extracting multi-domain features reflecting the fault type from the time series samples, removing redundant and irrelevant features by using the Pearson correlation coefficient and the principal component analysis method, obtaining the feature combination with the best diagnostic performance, and constructing the fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps: S31, extracting multi-domain features including time domain features, frequency domain features and time-frequency domain features from the time series samples; S32, using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable, screening out a feature subset associated with the preset target variable; S33, normalizing the feature subset; S34, using principal component analysis to perform dimensionality reduction processing on the normalized feature subset; S35. Use the random forest model to evaluate the diagnostic performance of the feature combination on the feature subset after dimensionality reduction, determine the feature combination with the best diagnostic performance based on the evaluation results, and build the final fault diagnosis model based on the feature combination with the best diagnostic performance.

5. A sensor-based explosion-proof junction box fault diagnosis method according to claim 4, characterized in that: The method of using the correlation between the multi-domain features extracted by the Pearson correlation coefficient and the preset target variable to screen out the feature subset associated with the preset target variable comprises the following steps: S321. For each feature, calculate the Pearson correlation coefficient between each feature and the target variable; S322. Based on the calculated Pearson correlation coefficient, a threshold is set, and features associated with the target variable are screened according to the set threshold; The formula for calculating the Pearson correlation coefficient between each feature and the target variable is: In the formula, r represents the Pearson correlation coefficient between each feature and the target variable; m a represents the observed value of the eigenvalue; n a represents the observed value of the target variable; represents the average value of the eigenvalues; represents the mean value of the target variable.

6. A sensor-based explosion-proof junction box fault diagnosis method according to claim 5, characterized in that: The method of reducing the dimension of the normalized feature subset by principal component analysis comprises the following steps: S341, calculating the covariance matrix of the normalized feature subset, and calculating all eigenvalues ​​and corresponding eigenvectors based on the covariance matrix; S343, according to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components; S344, constructing a projection matrix using the eigenvector; S345, using a projection matrix to transform the normalized data into a new feature space composed of principal components, to obtain a feature subset after dimensionality reduction; The formula for calculating the covariance matrix of the normalized feature subset is: In the formula, C ij represents the covariance between feature i and feature j; n represents the number of samples in the feature subset; k represents the sum index; X ki Represents the value of the i-th feature of the k-th sample; X kj Represents the value of the jth feature of the kth sample.

7. A sensor-based explosion-proof junction box fault diagnosis method according to claim 6, characterized in that: The method of using the random forest model to evaluate the diagnostic performance of the feature combination on the feature subset after dimensionality reduction, determining the feature combination with the best diagnostic performance according to the evaluation results, and constructing the final fault diagnosis model based on the feature combination with the best diagnostic performance includes the following steps: S351, initializing the random forest model and setting the parameters of the random forest model; S352, using the training set to train the random forest model, during the training process, constructing multiple decision trees by randomly selecting features and data samples; S353. In multiple decision trees, use the feature importance evaluation function of the random forest model to evaluate the contribution score of each feature to the model prediction performance; S354. Filter the features that contribute most to model performance according to the contribution score; S355, Use the test set to evaluate the diagnostic performance of the random forest model; S356, based on the contribution score of the performance evaluation results of the random forest model, screen the best feature combination; S357, using the determined optimal feature combination and the adjusted model parameters, retraining the random forest model to construct a final fault diagnosis model; In the above, in multiple decision trees, using the feature importance evaluation function of the random forest model, evaluating the contribution score of each feature to the model prediction performance includes the following steps: S3531. For each decision tree, traverse all nodes, and for each node that is split based on the feature, calculate the reduction in impurity caused by the split; S3532. For each feature, the impurity reduction amount caused by all split points of each feature on all trees is accumulated to obtain the total impurity reduction amount of the feature in the entire random forest model; S3533. Divide the total impurity reduction by the total number of trees in the random forest to calculate the average impurity reduction of the feature, and use the average impurity reduction as the importance score of the feature; S3534. Add the importance scores of all features to obtain a total importance score; S3535. Divide the importance score of each feature by the total score to obtain a normalized importance score; S3536. Sort all features by their importance scores according to the normalized importance scores; S3537. Select the highest ranked feature as the feature that contributes most to the model's prediction performance.

8. The sensor-based explosion-proof junction box fault diagnosis method according to claim 1, characterized in that: The method of extracting features from multi-source sensor data collected in real time and inputting the extracted features into a fault diagnosis model for fault diagnosis comprises the following steps: S41, obtaining multi-source sensor data of the explosion-proof junction box monitored by the multi-source sensor in real time; S42, extracting features from multi-source sensor data collected in real time, integrating all extracted features, and forming a feature vector; S43, inputting the feature vector into a fault diagnosis model, and the fault diagnosis model gives a fault diagnosis result according to the input feature vector; The step of inputting the feature vector into the fault diagnosis model, and the fault diagnosis model providing a fault diagnosis result according to the input feature vector, comprises the following steps: S431, loading a pre-trained fault diagnosis model; S432, providing the characteristic vector as input data to a fault diagnosis model; S433, the fault diagnosis model receives the input feature vector, performs calculations using the patterns learned during training, identifies known fault patterns based on the calculation results, and outputs a specific fault category.

9. The sensor-based explosion-proof junction box fault diagnosis method according to claim 1, characterized in that: The construction of a visual display interface to display the operating status and diagnostic results of the explosion-proof junction box in real time and automatically generate alarm information and maintenance suggestions according to the diagnostic results includes the following steps: S51, real-time monitoring and collection of multi-source sensor data of explosion-proof junction boxes; S52, using a pre-trained fault diagnosis model to identify faults based on the processed data, and automatically generating alarm information for a specific fault category according to the diagnosis result; S53. Build a visual display interface to display the operation status and diagnosis results of the junction box, and integrate a user feedback mechanism.

10. A sensor-based explosion-proof junction box fault diagnosis system, used to implement the sensor-based explosion-proof junction box fault diagnosis method according to any one of claims 1 to 9, characterized in that: The system includes: data acquisition module, sample extraction module, model building module, fault diagnosis module and visualization and alarm module; The data acquisition module is connected to the model building module through the sample extraction module, and the model building module is connected to the visualization and alarm module through the fault diagnosis module; The data acquisition module is used to obtain historical multi-source sensor data reflecting the status of the explosion-proof junction box; The sample extraction module is used to extract time series samples of pre-processed historical multi-source sensor data using a sliding time window method; The model building module is used to extract multi-domain features reflecting the fault type from the time series samples, remove redundant and irrelevant features by using the Pearson correlation coefficient and principal component analysis method, obtain the feature combination with the best diagnostic performance, and build a fault diagnosis model based on the feature combination with the best diagnostic performance; The fault diagnosis module is used to extract features from multi-source sensor data collected in real time, and input the extracted features into the fault diagnosis model for fault diagnosis; The visualization and alarm module is used to construct a visualization display interface, display the operating status and diagnosis results of the explosion-proof junction box in real time, and automatically generate alarm information and maintenance suggestions based on the diagnosis results.

Images