Interface point switch state monitoring system and method based on image recognition

Abstract

The invention discloses an interface switch machine state monitoring system and method based on image recognition, and relates to the technical field of computer vision, the system is composed of a plurality of functional modules, and the system comprises a data acquisition module for acquiring the interface switch machine state and weather data; the interface switch machine state comprises stretching time, a switch machine image and a laser vibration frequency spectrum, and the weather data comprises temperature, humidity and particulate matter concentration; the feature extraction module is used for receiving the state of the interface switch machine and the weather data, carrying out data fusion and generating a first feature by utilizing feature engineering; the first feature comprises an interface switch machine telescopic feature, an interface switch machine vibration feature and a corrosion image feature; the data analysis and anomaly detection module is used for extracting the telescopic characteristics and the vibration characteristics of the interface switch machine, calculating the deviation and the vibration amplitude of the locking gap, and comparing the deviation and the vibration amplitude with the gap threshold value and the vibration threshold value respectively; and when the gap locking deviation is smaller than the gap threshold value and the vibration amplitude is smaller than the vibration threshold value, continuous monitoring is conducted.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and specifically to an interface switch machine status monitoring system and method based on image recognition. Background Art

[0002] As a core execution device in the railway signal system, the switch machine is responsible for controlling the track switching to ensure that trains can safely and accurately complete the track transition. The accuracy of track switching is one of the important indicators to measure the railway operation efficiency. The switch machine is the key to ensuring the accuracy of track switching. It moves the turnout from one position to another through mechanical operation, thereby guiding the train into the correct track. The accuracy of track switching also affects the smoothness and safety of high-speed trains. For high-speed trains, even a slight track deviation may cause a huge impact force, which will not only damage the vehicle itself but also threaten the lives of passengers.

[0003] The status maintenance of traditional interface switch machines relies on manual periodic inspections and threshold alarms, which have problems such as detection lag and strong subjectivity. When inexperienced technicians conduct regular inspections on the switch machine, they fail to accurately identify slight signs of poor electrical contact. As the usage time increases, this poor contact gradually deteriorates, eventually leading to signal transmission interruption and affecting the normal operation of the train dispatching system. At the same time, the interface switch machine is exposed to a complex environment for a long time, and the transmission components are easily eroded by moisture, resulting in metal corrosion, mechanical jamming, poor electrical contact and other faults. When not identified in time, it will pose a safety hazard to the use of the switch machine. Summary of the Invention

[0004] (1) Technical Problems to be Solved Aiming at the deficiencies of the prior art, the present invention provides an interface switch machine status monitoring system and method based on image recognition to at least solve the problems in the prior art that the status maintenance of the interface switch machine relies on manual periodic inspections and threshold alarms, resulting in detection lag, and the transmission components of the interface switch machine are easily eroded by moisture, leading to metal corrosion, mechanical jamming, poor electrical contact and other faults, which pose a safety hazard to the use of the switch machine when not identified in time.

[0005] (2) Technical Solutions To achieve the above objectives, the present invention is realized through the following technical solutions: An interface switch machine status monitoring system based on image recognition, comprising: A data acquisition module for acquiring the status of the interface switch machine and weather data; the status of the interface switch machine includes the telescopic time, the switch machine image, and the laser vibration spectrum, and the weather data includes temperature, humidity, and particulate matter concentration; The feature extraction module receives the status of the interface switch machine and weather data, performs data fusion, and generates the first features using feature engineering; the first features include the telescopic features of the interface switch machine, the vibration features of the interface switch machine, and the rust image features; The data analysis and anomaly detection module extracts the telescopic features and vibration features of the interface switch machine, calculates the locking notch deviation and vibration amplitude, and compares them with the notch threshold and vibration threshold respectively; When the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuous monitoring is carried out, and the difference score and rust evaluation index are obtained to calculate the status score of the switch machine; The prediction and optimization module issues multi-level warnings and interventions based on the status score of the switch machine.

[0006] Furthermore, the process of obtaining the telescopic features of the interface switch machine is as follows: Obtain the telescopic time and the image of the switch machine, and use CNN to extract the telescopic features of the interface switch machine. The telescopic features of the interface switch machine include: the damaged image of the switch machine and the displacement image of the switch machine; CNN extracts the telescopic features of the interface switch machine: Through the preprocessing of the switch machine image, use the segmentation and regression algorithm for multi-task network design, extract the damaged image of the switch machine and the displacement image of the switch machine, train the configuration, and verify the deployment.

[0007] Furthermore, the vibration features of the interface switch machine are: Using data fusion and feature engineering, synchronize the vibration spectrum data and weather data at the time stamp, extract the spectrum features from the laser vibration spectrum, extract the environmental features from the weather data, and fuse the spectrum features and environmental features, and use machine learning algorithms to generate the vibration features of the interface switch machine.

[0008] Furthermore, the rust image features are: Use color space analysis to extract the rust area; obtain the texture color, perform hierarchical fusion, and generate the rust image features.

[0009] Furthermore, the process of calculating the locking notch deviation and vibration amplitude is as follows: By calibrating the reference position of the telescopic process of the interface switch machine and obtaining the telescopic features of the interface switch machine, perform deterioration correction. The formula is: In the formula, is the deterioration correction value; is the maximum number of action times; is the current action time point; k is the current number of actions, and k = 1, 2, 3...M; N is the number of sampling points within the window; The displacement sensor reading for the k-th action; Generate the locking notch deviation based on the degradation correction value, with the formula: In the formula, is the locking notch deviation; is the time point of the last action; is the steady-state value; Calculate the vibration amplitude: In the formula, is the vibration amplitude; is the side frequency amplitude; is the integration window time; is the mechanical vibration weight; is the dust damping coefficient; is the temperature; is the symbolic representation of the derivative; is the vibration time.

[0010] Further, compare the locking notch deviation and the vibration amplitude with the notch threshold and the vibration threshold respectively: Based on historical data, obtain the notch threshold and the vibration threshold respectively; The comparison process is: When the locking notch deviation is greater than or equal to the notch threshold, or the vibration amplitude is greater than or equal to the vibration threshold, an alarm is issued; When the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuous monitoring is carried out and a difference score is obtained.

[0011] Further, the process of obtaining the difference score is: Respectively obtain the difference between the locking notch deviation and the notch threshold, and the difference between the vibration amplitude and the vibration threshold, and use a non-linear algorithm to calculate the difference score, with the formula: In the formula, is the difference score; is the locking notch deviation threshold; is the vibration amplitude threshold.

[0012] Further, obtain the rust evaluation index and the turnout machine status score: By obtaining the rust image features, calculate the proportion of the rusty red area in the component surface, with the formula: In the formula, cx is the rust evaluation index; is the rust area; is the non-rusty area; is the edge fragmentation weight; Q is the edge fractal dimension damage index; The turnout machine status score is obtained as follows: By obtaining the difference score and the rust evaluation index, calculate the turnout machine status score. The formula is: In the formula, R is the turnout machine status score; is the difference score weight; is the rust evaluation index weight; is the difference score feature mapping function value; is the rust evaluation index feature mapping function value.

[0013] Furthermore, the process of multi-level warning and intervention based on the turnout machine status score is as follows: Obtain historical data and the score range. By substituting the turnout machine status score into the historical data score range, set the Eq alarm threshold, and execute the preset disposal strategy.

[0014] An interface turnout machine status monitoring method based on image recognition includes the following steps: Step 1: Obtain the interface turnout machine status and weather data; the interface turnout machine status includes the telescopic time, the turnout machine image, and the laser vibration spectrum, and the weather data includes temperature, humidity, and particulate matter concentration; Step 2: Receive the interface turnout machine status and weather data, and perform data fusion, and generate the first feature by using feature engineering; the first feature includes the interface turnout machine telescopic feature, the interface turnout machine vibration feature, and the rust image feature; Step 3: Extract the interface turnout machine telescopic feature and the interface turnout machine vibration feature, calculate the locking notch deviation and the vibration amplitude, and compare them with the notch threshold and the vibration threshold respectively; when the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuously monitor, and obtain the difference score and the rust evaluation index, and calculate the turnout machine status score; Step 4: Perform multi-level warning and intervention based on the turnout machine status score.

[0015] (III) Beneficial effects The present invention provides an interface turnout machine status monitoring system and method based on image recognition, which has the following beneficial effects: (1) By adopting advanced computer vision and deep learning algorithms, this system can automatically identify various states of the switch machine, including normal working state, slight wear, severe damage, etc.; compared with the traditional manual inspection method, this automated detection method not only improves the detection speed, but also greatly enhances the accuracy of fault detection, and immediately issues an alarm when an abnormal situation is detected; enabling maintenance personnel to take actions before the problem becomes more serious, thus effectively avoiding train operation accidents caused by equipment failures. (2) Through the analysis of historical data and pattern recognition, the system can predict which components are about to fail, which can not only reduce unnecessary preventive maintenance work, but also ensure that critical components are replaced or repaired in a timely manner, extending the service life of the equipment; in addition to image recognition technology, the system also integrates various data sources such as vibration sensors, temperature and humidity sensors to form a comprehensive monitoring platform, which not only provides more comprehensive status information, but also can verify the existence of faults from multiple angles, enhancing the reliability of the system. Brief Description of the Drawings

[0016] Figure 1 It is a schematic diagram of the system flow of the present invention; Figure 2 It is a schematic diagram of the overall method of the present invention. Detailed Embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1: Please refer to Figure 1 , this embodiment provides an interface switch machine status monitoring system based on image recognition, and this system includes: A data acquisition module, which acquires the status of the interface switch machine and weather data. The status of the interface switch machine includes the retraction and extension time, the image of the switch machine, and the laser vibration spectrum; the weather data includes temperature, humidity, and particulate matter concentration; and outputs the status of the interface switch machine and the weather data. Acquisition of interface switch machine status data: Retraction and extension time: Using an optoelectronic encoder installed on the actuator rod or motor drive shaft of the switch machine, by detecting the rotation angle or displacement change of the gear, and at the same time using the PLC timing module for timing, accurately record the start and end times of the action. PLC timing module: triggers the timer through switch signals (such as "turnout switching action started" and "turnout switching action completed") and calculates the action time difference; Switch machine image: An industrial multispectral camera supporting visible light and infrared imaging (such as the FLIR series) is mounted on the side bracket of the switch machine. This provides a bird's-eye view of dynamic components (requiring an anti-obstruction design) and key switch locations (locking notches, indicator rods, and gearboxes). This extends the dynamic range in strong light or backlight conditions to avoid overexposure or underexposure. The camera automatically captures key frames when the switch starts or completes, in conjunction with the switch machine's motion signals. Laser vibration spectrum: A laser Doppler vibrometer is used to non-contactly measure the surface vibration of the switch machine housing or gearbox. A PCB Piezotronics triaxial accelerometer is also installed in contact with key bearings to measure the vibration of key bearings. Fast Fourier transform (FFT) is used to convert time-domain vibration signals into frequency-domain features, and resonance peak detection is used to identify natural frequency deviations of the switch machine (such as high-frequency harmonic anomalies caused by gear wear). Weather data: Temperature and humidity: Utilize an industrial-grade temperature and humidity sensor with built-in anti-condensation coating, close to the switch machine housing surface to avoid direct sunlight, and collect temperature and humidity in real time; Particle concentration: A laser scattering particle sensor is used to detect solid particles in the range of 0.3 to 10 μm. A sintered metal filter is installed at the air inlet of the laser scattering particle sensor to prevent clogging by large particles during dusty weather. The particle concentration is also measured. Calibration is also performed regularly on-site using a standard dust concentration detector. A feature extraction module acquires the interface switch machine status and weather data, performs data fusion, and uses feature engineering to generate the first feature, which includes the interface switch machine expansion and contraction feature, the interface switch machine vibration feature, and the rust image feature; Interface switch telescopic features: Obtain the expansion and contraction time and switch machine image, and use CNN to extract the expansion and contraction features of the interface switch machine. The expansion and contraction features of the interface switch machine include: switch machine damage image and switch machine displacement image; Images of switch machine damage include cracks in the switch machine casing, peeling of the surface coating, rust on metal parts, and deformation of joints. Cracks in the switch machine casing are mainly concentrated in areas of mechanical stress concentration, indicating possible structural fatigue problems caused by long-term use. Peeling of the surface coating and rust on metal parts reflect the equipment's insufficient weather resistance in harsh environments. Deformation of joints may be caused by improper operation or external impact. The displacement image of the switch machine intuitively reflects the real-time position changes of the key components of the switch machine (such as push rods, connecting rods, etc.) during the turnout conversion process, records the key frames from the initial static state to the fully converted state, and the precise displacement trajectories and relative position relationships of each component during the turnout conversion; CNN extraction interface for the telescopic features of the switch machine: Through preprocessing of the switch machine images, a multi-task network design is carried out using segmentation and regression algorithms to extract the damaged images and displacement images of the switch machine, train the configuration, and verify the deployment; Preprocessing of the switch machine images: Four types of damaged areas are labeled (cracks in the switch machine housing, peeling of the surface coating, rust on metal components, and deformation of the connection parts) and stored as Mask semantic segmentation labels. Classification labels are assigned according to the types of damaged areas (e.g., crack_01 represents the crack number of the switch machine housing); the displacement coordinates of the key components are regressed (e.g., push rods, connecting rods), and Bounding Box annotations are made. Combining the telescopic time, the real-time position changes during the turnout conversion are named according to the time stamp; Mask semantic segmentation label: Assign a class label to each pixel in the switch machine image to distinguish and mark different damage types and components, thereby supporting precise image analysis and feature extraction; Bounding Box annotation: Use a rectangular box in the switch machine image to mark the position and range of a specific object or component to indicate its precise position and size in the switch machine image, facilitating subsequent detection and analysis; Extracting the damaged images and displacement images of the switch machine: In the multi-task network design, two tasks are simultaneously completed through a shared CNN backbone network: segmentation of the damaged areas of the switch machine and regression and segmentation of the displacement coordinates of the key components. This can not only share the underlying features to reduce computational redundancy but also combine the correlation information of the two types of tasks to improve the robustness of the model; The pre-trained ResNet-50 is used as the shared feature extractor to convert the input switch machine image into a high-dimensional feature map; the advantage of sharing is that after extraction, there is a correlation between the damaged images and displacement images of the switch machine in terms of static and dynamic features; for example, cracks often appear in the mechanical stress concentration areas with frequent displacement. The shared backbone network can promote the feature learning of the two types of tasks; Training configuration: Prepare the labeled dataset, including the damaged images of switch machines with Mask semantic segmentation labels and the displacement images of switch machines with BoundingBox annotations; select a convolutional neural network (CNN) architecture such as ResNet, U-Net, or EfficientNet and adjust it according to the task requirements; set the training hyperparameters such as the learning rate, batch size, number of training epochs, and the optimizer; to prevent overfitting, adopt data augmentation techniques (such as random cropping, rotation, flipping, etc.) and regularization methods (such as Dropout or L2 regularization); Learning rate: Determines the step size for updating the model parameters, affecting the convergence speed and stability; Batch size: The number of samples used in each iteration, affecting the memory usage and the accuracy of gradient estimation; Number of training epochs: The number of times the entire dataset is trained, affecting the depth of model learning; Optimizer type: Such as Adam or SGD, the method used to minimize the loss function, affecting the training efficiency and effect Regularization method: By adding additional constraints or penalty terms to the loss function, it prevents the model from overfitting, thereby improving the generalization ability of the model; Verification and deployment: Integrate the model into the target system, evaluate it, monitor its performance, and make iterative improvements as needed to ensure that the model can continuously and stably provide accurate predictions or analysis results. Test and verify the trained model in the actual application environment to ensure that its performance and stability meet the expectations, and make necessary adjustments and optimizations based on the feedback to achieve reliable operation and maintenance. The code example for model lightweighting is as follows: ```python from torchvision.models import mobilenet_v3_small # Replace the backbone with MobileNetV3 self.backbone = mobilenet_v3_small(pretrained=True).features ``` - Edge inference acceleration: ```bash # Use TensorRT to convert the model trtexec --onnx=model.onnx --saveEngine=model.engine --fp16 ``` Example: Detect the long-term displacement anomaly of a switch machine in a railway hub. Using the displacement image sequence continuously captured for 20 hours as input, the CNN segmentation network detected a 0.8 mm crack at the push rod connection (IoU = 0.72). Displacement regression showed that the maximum stroke of the push rod decreased by 5% (the threshold alarm range was set at ±3%); the system was linked with vibration spectrum analysis to confirm that the abnormal vibration frequency was related to crack propagation; an early warning signal was output, indicating that the push rod assembly needed to be replaced, and a heat map of the crack location was attached; Vibration characteristics of the interface switch machine: Using data fusion and feature engineering, synchronize the vibration spectrum data and weather data at the time stamp, extract spectrum features from the laser vibration spectrum, extract environmental features from the weather data, and fuse the spectrum features and environmental features. Use machine learning algorithms (such as principal component analysis PCA, recursive feature elimination RFE, etc.) to generate the vibration characteristics of the interface switch machine; Synchronize the vibration spectrum data and weather data at the time stamp: Determine a unified time reference, use UTC (Coordinated Universal Time) as the standard time, ensure that the time recorded by all devices is based on the same time standard, align according to the time stamp. If there is no record within a specific time window, use the data from the previous time period to fill in; Extract spectrum features and environmental features: Spectrum feature extraction: Filter and denoise the original vibration signal of the vibration spectrum data to remove unnecessary interference signals, obtain time domain features, and obtain the mean, variance, peak value, skewness, and kurtosis through the processed vibration spectrum data; construct a temperature-vibration sensitivity matrix, which was measured in the experiment: Environmental feature extraction: Obtain the temperature, humidity, and particulate matter concentration in the weather data; extract the mean, standard deviation, maximum value, and minimum value of the temperature, humidity, and particulate matter concentration from the weather data; select a time window in seconds and calculate the environmental features within each time window; Data fusion: Merge the vibration spectrum data and weather data according to the time stamp to form a comprehensive data set; If the vibration spectrum data and weather data are recorded once per second, directly merge the vibration spectrum data and weather data into one or more lines of records according to the time stamp; if the recording frequencies are different, appropriate resampling or interpolation processing is required to make the vibration spectrum data and weather data match at the same time point according to the time stamp; Machine learning algorithms to generate the vibration characteristics of the interface switch machine: Use a machine learning model to evaluate the importance of each feature, and screen out the key features that have a significant impact on the model's prediction performance; reduce the number of redundant features through principal component analysis (PCA) or other dimensionality reduction techniques, simplify the model structure, and improve computational efficiency; construct a supervised learning model (such as support vector machine, neural network, etc.), and use the labeled historical data for training, with the goal of accurately predicting the state of the switch machine (normal, slightly worn, severely damaged, etc.); evaluate the generalization ability of the model through cross-validation methods to prevent overfitting; use methods such as grid search or Bayesian optimization to find the best combination of model hyperparameters; Rust image features: Utilize color space analysis to extract the rust area in the HSV / Lab space, and use the local binary pattern (LBP) or gray-level co-occurrence matrix (GLCM) to quantify the surface roughness; obtain the texture color and perform hierarchical fusion to extract rust features; For example, threshold segmentation of red / orange hues, as well as the proportion of the rust spot area, the maximum rust spot size, and the perimeter / area ratio of the rust area, thereby reflecting the diffusion degree, and obtaining the centroid position and spatial density of the rust area; For example, concatenate the low-level feature of texture color and the high-level feature of deep learning embedding, and use a spatial attention network to dynamically weight the importance of different features; The data analysis and anomaly detection module obtains the telescopic features and vibration features of the interface switch machine, and calculates the locking notch deviation and vibration amplitude; compare the locking notch deviation and vibration amplitude with the notch threshold and vibration threshold respectively. When the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuously monitor, obtain the difference score, and at the same time obtain the rust evaluation index to calculate the state score of the switch machine; Calculate the locking notch deviation: By calibrating the reference position of the telescopic process of the interface switch machine and obtaining the telescopic features of the interface switch machine, calculate the locking notch deviation in real time. Under the condition of no equipment failure, record the standard locking position and the average displacement value when the action rod reaches the position. After each action is completed, take the steady-state displacement value for deterioration correction. The formula is: In the formula, is the deterioration correction value; is the maximum number of actions; is the current action time point; k is the current number of actions, and k takes values of k = 1, 2, 3...M; N is the number of sampling points within the window. When the sampling rate ≥ 100Hz, N = 300; is the displacement sensor reading for the kth action; Obtain the locking notch deviation by obtaining the deterioration correction value. The formula is: In the formula, is the deviation of the locking notch; is the time point of the last operation; is the steady-state value, the steady-state observation window; Calculate the vibration amplitude: In the formula, is the vibration amplitude; is the side frequency amplitude; is the integration window time; is the mechanical vibration weight; is the dust damping coefficient; is the temperature; is the symbolic representation of the derivative, used to express the rate of change of the function with time; is the vibration time, taking 3 seconds as a time period; The historical data is based on past experiments, papers, and public data; Notch threshold: Based on the 3σ principle of normal historical data, that is, the mean ± 3 times the standard deviation. Every Y operations, recalculate the mean and standard deviation under normal working conditions, and introduce a temperature correction factor (the influence of thermal expansion and contraction of metals on the displacement reference); For example: Collect a large amount of locking notch deviation data under normal working conditions, and perform preprocessing and statistical analysis: the mean is 0.5mm; the standard deviation is 0.1mm; choose to use 3 times the standard deviation as the threshold, and the formula is: Notch threshold = mean + 3 * standard deviation = 0.5mm + 3 * 0.1mm = 0.8mm While ensuring the detection sensitivity, reduce the possibility of false alarms; Vibration threshold: Choose a fixed multiple of the standard deviation as the threshold; for example, if the mean of the vibration amplitude is 0.5g and the standard deviation is 0.1g, set the threshold to 0.5g + 3 * 0.1g = 0.8g; Collect a large amount of vibration amplitude data under normal working conditions, and perform preprocessing and statistical analysis to obtain the following results: the mean is 0.5g; the standard deviation is 0.1g; choose to use 3 times the standard deviation as the threshold, and the formula is: Vibration threshold = mean + 3 * standard deviation = 0.5g + 3 * 0.1g = 0.8g The setting of the vibration threshold reduces the possibility of false alarms while ensuring the detection sensitivity; the process of comparing the locking notch deviation and the vibration amplitude with the notch threshold and the vibration threshold respectively is: When the deviation of the locking notch is greater than or equal to the notch threshold, or the vibration amplitude is greater than or equal to the vibration threshold, an alarm is issued; When the deviation of the locking notch is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuous monitoring is carried out and a difference score is obtained; The obtaining of the difference score is as follows: The difference between the locking notch deviation and the notch threshold, and the difference between the vibration amplitude and the vibration threshold are respectively obtained, and a non-linear algorithm is used to calculate the difference score. The formula is: In the formula, is the difference score; is the locking notch deviation threshold; is the vibration amplitude threshold; Obtaining the rust evaluation index: By obtaining the characteristics of the rust image, the proportion of the rust-red area in the surface of the component is calculated. The formula is: In the formula, cx is the rust evaluation index; is the rust area; is the non-rust area; is the edge fragmentation weight; Q is the edge fractal dimension damage index; The obtaining of the switch machine status score is as follows: By obtaining the difference score and the rust evaluation index, the switch machine status score is further calculated. The formula is: In the formula, R is the switch machine status score; is the difference score weight; is the rust evaluation index weight; is the difference score feature mapping function value; is the rust evaluation index feature mapping function value; The prediction and optimization module issues multi-level warnings and intervenes by obtaining the switch machine status score; The switch machine status score is obtained, historical data and score intervals are obtained. By substituting the score intervals of historical data, the Eq alarm threshold is set, which is defined according to the normal operation range, minor abnormality, moderate abnormality and severe abnormality of the equipment. The historical data table is as follows: In this way, the status of the switch machine can be effectively monitored, and corresponding measures can be taken in case of different levels of abnormalities to ensure the safe operation of the equipment.

[0019] Embodiment 2: Please refer to Figure 2, based on Embodiment 1, this embodiment also provides a method for monitoring the status of an interface switch machine based on image recognition, including the following specific steps: Step 1: Obtain the status of the interface switch machine and weather data; the status of the interface switch machine includes the telescopic time, the image of the switch machine, and the laser vibration spectrum, and the weather data includes temperature, humidity, and particulate matter concentration; Step 2: Receive the status of the interface switch machine and weather data, and perform data fusion to generate the first feature using feature engineering; the first feature includes the telescopic feature of the interface switch machine, the vibration feature of the interface switch machine, and the rust image feature; Step 3: Extract the telescopic feature and vibration feature of the interface switch machine, calculate the locking notch deviation and the vibration amplitude, and compare them with the notch threshold and the vibration threshold respectively; when the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuously monitor, and obtain the difference score and the rust evaluation index to calculate the status score of the switch machine; Step 4: Perform multi-level warnings and interventions based on the status score of the switch machine.

[0020] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0021] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0022] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application.

Claims

1. An interface switch machine status monitoring system based on image recognition, characterized in that The system includes: A data acquisition module that acquires the status of the interface switch machine and weather data; the status of the interface switch machine includes the telescopic time, the switch machine image, and the laser vibration spectrum, and the weather data includes temperature, humidity, and particulate matter concentration; A feature extraction module that receives the status of the interface switch machine and weather data, performs data fusion, and generates first features using feature engineering; the first features include the telescopic feature of the interface switch machine, the vibration feature of the interface switch machine, and the rust image feature; A data analysis and anomaly detection module that extracts the telescopic feature of the interface switch machine and the vibration feature of the interface switch machine, calculates the locking notch deviation and the vibration amplitude, and compares them with the notch threshold and the vibration threshold respectively; When the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuous monitoring is carried out, and the difference score and the rust evaluation index are obtained to calculate the switch machine status score; A prediction and optimization module that issues multi-level warnings and interventions based on the switch machine status score.

2. The state monitoring system of an interface switch machine based on image recognition according to claim 1, wherein, The process of obtaining the telescopic feature of the interface switch machine is as follows: The telescopic time and the switch machine image are obtained, and the CNN is used to extract the telescopic feature of the interface switch machine. The telescopic feature of the interface switch machine includes: the damaged image of the switch machine and the displacement image of the switch machine; The CNN extracts the telescopic feature of the interface switch machine: Through the preprocessing of the switch machine image, a multi-task network design is carried out using the segmentation and regression algorithm to extract the damaged image of the switch machine and the displacement image of the switch machine, and the training configuration is verified and deployed.

3. The interface switch machine status monitoring system based on image recognition according to claim 2, wherein, The vibration feature of the interface switch machine is: Using data fusion and feature engineering, the vibration spectrum data and weather data are synchronized at the time stamp, the spectrum feature is extracted from the laser vibration spectrum, the environmental feature is extracted from the weather data, and the spectrum feature and the environmental feature are fused, and a machine learning algorithm is used to generate the vibration feature of the interface switch machine.

4. The interface switch machine status monitoring system based on image recognition according to claim 3, wherein: The rust image feature is: Using color space analysis, the rust area is extracted; the texture color is obtained, and hierarchical fusion is carried out to generate the rust image feature.

5. The state monitoring system of an interface switch machine based on image recognition according to claim 4, characterized in that: The process of calculating the locking notch deviation and the vibration amplitude is as follows: By calibrating the reference position of the telescopic process of the interface switch machine and obtaining the telescopic feature of the interface switch machine, deterioration correction is carried out, and the formula is: In the formula, is the deterioration correction value; is the maximum number of operation times; is the current operation time point; k is the current number of operations, and k = 1, 2, 3... M; N is the number of sampling points within the window; is the displacement sensor reading for the k-th operation; The locking notch deviation is generated based on the deterioration correction value, and the formula is: In the formula, is the locking notch deviation; is the last action time point; is the steady state value; Calculate the vibration amplitude: In the formula, is the vibration amplitude; is the side frequency amplitude; is the integration window time; is the mechanical vibration weight; is the dust damping coefficient; is the temperature; is the symbolic representation of the derivative; is the vibration time.

6. The interface switch machine status monitoring system based on image recognition according to claim 5, wherein: The locking notch deviation and the vibration amplitude are compared with the notch threshold and the vibration threshold respectively: Based on historical data, the notch threshold and the vibration threshold are obtained respectively; The comparison process is: When the locking notch deviation is greater than or equal to the notch threshold, or the vibration amplitude is greater than or equal to the vibration threshold, an alarm is issued; When the locking notch deviation is less than the notch threshold and the vibration amplitude is less than the vibration threshold, continuous monitoring is carried out, and the difference score is obtained.

7. The status monitoring system of the switch machine based on image recognition according to claim 6, wherein: The process of obtaining the difference score is: The difference between the locking notch deviation and the notch threshold, and the difference between the vibration amplitude and the vibration threshold are obtained respectively, and a non-linear algorithm is used to calculate the difference score, and the formula is: Wherein, is the difference score; is the deviation threshold of the locking notch; is the vibration amplitude threshold.

8. The state monitoring system for an interface switch machine based on image recognition according to claim 7, wherein: Obtaining the rust evaluation index and the switch machine status score: By obtaining the rust image feature, the proportion of the rust-red area in the component surface is calculated, and the formula is: Wherein, \(c_x\) is the rust evaluation index; is the rust area; is the non-rusted area; is the edge fragmentation weight; \(Q\) is the edge fractal dimension damage index; The switch machine status score is obtained as: By obtaining the difference score and the rust evaluation index, the switch machine status score is calculated, and the formula is: Wherein, R is the turnout machine state score; is the difference score weight; is the rust evaluation index weight; is the difference score feature mapping function value; is the rust evaluation index feature mapping function value.

9. The state monitoring system of the interface switch machine based on image recognition according to claim 8, characterized in that: The process of multi-level warning and intervention based on the switch machine status score is as follows: Obtain historical data and score intervals. By bringing the switch machine status score into the historical data score interval, set the Eq alarm threshold, and execute the preset disposal strategy.

10. A method for monitoring the status of an interface switch machine based on image recognition, using the system according to any one of claims 1 to 9, characterized in that: It includes the following steps: Step 1: Obtain the status of the interface switch machine and weather data; the status of the interface switch machine includes the telescopic time, the image of the switch machine, and the laser vibration spectrum, and the weather data includes temperature, humidity, and particulate matter concentration; Step 2: Receive the status of the interface switch machine and weather data, and perform data fusion to generate the first feature using feature engineering; the first feature includes the telescopic feature of the interface switch machine, the vibration feature of the interface switch machine, and the rust image feature; Step 3: Extract the telescopic feature of the interface switch machine and the vibration feature of the interface switch machine, calculate the locking gap deviation and the vibration amplitude, and compare them with the gap threshold and the vibration threshold respectively; when the locking gap deviation is less than the gap threshold and the vibration amplitude is less than the vibration threshold, continuously monitor, and obtain the difference score and the rust evaluation index to calculate the switch machine status score; Step 4: Perform multi-level warning and intervention based on the switch machine status score.

Images