Railway electric switch machine state monitoring system

Abstract

The invention discloses a state monitoring system for a railway electric switch machine. The state monitoring system comprises a multi-mode sensor module, a data preprocessing module, an intelligent analysis and decision module and a visualization and interaction module. The multi-mode sensor module is distributed at a key part of the switch machine and is used for acquiring multi-dimensional data such as the rotating speed of a power transmission shaft, displacement and vibration of an actuating rod, an expression notch image, temperature and humidity, air pressure and the like; the data preprocessing module performs noise reduction through a Kalman adaptive filtering algorithm and fuses asynchronous data by using a space-time alignment technology; the intelligent analysis and decision-making module constructs a hierarchical index system, determines a weight in combination with an analytic hierarchy process and an entropy weight method, dynamically evaluates a health level through a fuzzy grey clustering algorithm, predicts and represents a gap size trend by adopting an ARIMA model, performs monthly early warning on jamming faults by applying a random forest model, and optimizes and maintains resource configuration according to historical data; the visualization and interaction module develops a Web mobile terminal interface and supports multi-level health state display, real-time alarm and maintenance work order generation.

Description

Technical Field

[0001] The present invention relates to the field of track equipment detection, and particularly to a railway electric switch machine condition monitoring system. Background Art

[0002] In the railway transportation system, as a core signal device, the reliability of the operation state of the switch machine is directly related to the safety and efficiency of train operation. However, there are many drawbacks in the traditional railway switch machine condition monitoring methods, mainly relying on manual inspections or single sensors, facing a series of severe challenges in actual operation.

[0003] Firstly, the problem of signal distortion caused by electromagnetic interference seriously affects the monitoring accuracy. There are a large number of electrical devices in the railway environment, such as the high-voltage power supply system during train operation, communication base stations, etc. These devices will generate strong electromagnetic interference during operation. Traditional monitoring sensors, especially some sensors relying on electrical signal transmission, are extremely vulnerable to the influence of these electromagnetic interferences. For example, in areas where trains pass frequently, the signals collected by sensors monitoring the operating current and voltage of the switch machine often show fluctuations, distortions, etc. This makes the monitoring data unable to truly reflect the actual operating state of the switch machine, leading to wrong judgments by maintenance personnel based on these distorted data, further affecting the timely discovery and handling of switch machine failures, and increasing the safety hazards of railway operation. Secondly, the complex mechanical structure of the switch machine results in high maintenance costs. The switch machine internally contains many complex components such as motors, reducers, friction couplings, locking mechanisms, etc. These components cooperate with each other to jointly complete the functions of turnout conversion and locking. Due to the complexity of its structure, in daily maintenance, maintenance personnel need to have professional technical knowledge and rich experience to conduct detailed inspections, debugging, and maintenance of each component. Moreover, the structures of different models of switch machines are different, further increasing the difficulty and complexity of maintenance. Furthermore, the lack of fault prediction ability makes the maintenance work often in a lagging state. The traditional monitoring method mainly conducts maintenance after the switch machine fails, lacking the ability to predict potential faults in advance. Due to the inability to timely obtain the operating state data of the switch machine, it is difficult to accurately analyze and predict its future operating trend.

[0004] The traditional railway switch machine condition monitoring methods are difficult to solve these problems, seriously restricting the safety and efficiency of railway transportation, and there is an urgent need for a more advanced and reliable monitoring system to solve these problems. Summary of the Invention

[0005] To solve the above problems, the object of the present invention is to provide a state monitoring system for railway electric switch machines, which mainly consists of a multi-modal sensor module, a data preprocessing module, an intelligent analysis and decision-making module, and a visualization and interaction module. The multi-modal sensor module synchronously collects the rotational speed of the power transmission shaft of the switch machine, the displacement of the operating rod, the vibration during operation, the indication gap, the internal environment temperature and humidity, and the external environment air pressure data. The data preprocessing module is used to eliminate the noise of the multi-modal sensor module, fuse multi-source data by using spatio-temporal alignment technology, and adopt the Kalman adaptive filtering algorithm to eliminate noise. The intelligent analysis and decision-making module realizes the dynamic evaluation of the health level of the switch machine by constructing a hierarchical index system including electrical characteristics, mechanical characteristics, and environmental parameters, combining the variable weight combination weighting method (analytic hierarchy process + entropy weight method) and the fuzzy grey clustering algorithm; at the same time, an autoregressive integrated moving average model is used to predict the change trend of the indication gap size, and a random forest model is used to give a monthly warning of jamming faults, and the maintenance resource allocation is optimized based on historical data. The visualization and interaction module develops a Web mobile interface, supports the display of health status at the network level, line level, and equipment level, and also has functions such as real-time alarm, trend analysis, and maintenance work order generation, providing strong assistance for operation and maintenance decision-making.

[0006] To achieve the above object, the state monitoring system for railway electric switch machines provided by the present invention is implemented as follows:

[0007] A railway electric switch machine status monitoring system includes a multi-modal sensor module, a data preprocessing module, an optical fiber transmission module, an intelligent analysis and decision-making module, and a visualization and interaction module. The multi-modal sensor module is distributed at each key part of the switch machine to obtain comprehensive and accurate data. The data preprocessing module uses a TMS320F28335 minimum system board, which is used to receive the data collected by the multi-modal sensor module and eliminate the noise of the data transmitted by the multi-modal sensor module. In the TMS320F28335 minimum system board, spatio-temporal alignment technology is used to fuse multi-source data, and the Kalman adaptive filtering algorithm is used to eliminate noise. The intelligent analysis and decision-making module uses an STM32F103ZET6 minimum system board, receives the data processed by the data preprocessing module, and constructs a hierarchical index system including electrical characteristics, mechanical characteristics, and environmental parameters in the STM32F103ZET6 minimum system board. The variable weight combination weighting method (analytic hierarchy process + entropy weight method) and the fuzzy grey clustering algorithm are added to realize the dynamic assessment of the health level of the switch machine. At the same time, an autoregressive integrated moving average model is added in the STM32F103ZET6 minimum system board to predict the change trend of the gap size, and a random forest model is used to give a monthly warning of jamming faults, optimize the maintenance resource allocation according to historical data, and the STM32F103ZET6 minimum system board controls the optical fiber transmission module to transmit the data to the visualization and interaction module. A Web mobile interface is developed in the visualization and interaction module, which supports the display of the health status at the network level, line level, and equipment level, and also has functions such as real-time alarm, trend analysis, and maintenance work order generation, providing strong assistance for operation and maintenance decision-making.

[0008] The multi-modal sensor module of the present invention includes a speed sensor, a displacement sensor, a vibration sensor, a camera, a temperature and humidity sensor, and a barometric pressure sensor. The speed sensor is installed on the power transmission shaft of the switch machine to monitor the rotation speed of the power transmission shaft of the switch machine in real time. The displacement sensor is installed on the operating rod of the switch machine to measure the linear displacement of the operating rod. The vibration sensor is installed on the outer shell of the switch machine to sense the vibration of the switch machine during operation. By monitoring the vibration of the switch machine, it is judged whether there is looseness, wear or other abnormalities in the internal mechanical structure. The camera is installed on the outer side of the switch machine to take pictures of the gap image of the indicating rod of the switch machine. Therefore, the temperature and humidity sensor is installed inside the switch machine to monitor the internal environmental temperature and humidity of the switch machine, and the barometric pressure sensor is installed on the surface of the switch machine to monitor the external environmental barometric pressure. The data information collected by the speed sensor, displacement sensor, vibration sensor, camera, temperature and humidity sensor, and barometric pressure sensor is transmitted to the TMS320F28335 minimum system board for preprocessing.

[0009] The minimum system board of TMS320F28335, the minimum system board of STM32F103ZET6, and the optical fiber transmission module are placed in a metal aluminum box, and the metal aluminum box is installed beside the switch machine.

[0010] The present invention uses the Canny operator to perform edge detection on the image of the switch rod notch collected by the camera, and uses the pixel integral projection method to calculate the size of the switch rod notch. The specific scheme is as follows:

[0011] S1. Preprocess the collected color image. First, convert it from the RGB color space to the grayscale space, which is expressed as:

[0012] G(x, y) = 0.299R(x,y) + 0.587G(x,y) + 0.114B(x,y) (1)

[0013] This operation can reduce the amount of data and facilitate subsequent analysis;

[0014] S2. Use Gaussian filtering to smooth the grayscale image. The convolution kernel of Gaussian filtering is determined by the Gaussian function, which is expressed as:

[0015]

[0016] where σ is the standard deviation, (m, n) is the coordinate within the convolution kernel, and the filtered image S(x, y) can be obtained through convolution operation, which is expressed as:

[0017]

[0018] This can effectively remove the noise in the image;

[0019] S3. Use the Canny operator to perform edge detection on the smoothed image S(x, y). The Canny operator first calculates the gradient magnitude M(x, y) and gradient direction θ(x, y) of the image, which are expressed as:

[0020]

[0021] where G x (x, y) and G y (x, y) are the gradients of the image in the x and y directions respectively

[0022] S4. Perform non-maximum suppression, only retain the pixels at the local gradient maximum, and then determine the true edges through the double-threshold method to obtain the edge image E(x, y);

[0023] S5. Use the pixel integral projection method to calculate the size of the switch rod notch. Perform a horizontal projection on the edge image E(x, y), and the horizontal projection function is:

[0024]

[0025] Where Y is the height of the image. By analyzing the horizontal projection curve P(x), a suitable threshold T is set. When P(x) < T, the starting and ending positions x of the notch are determined. start and x end , then the pixel size of the notch is:

[0026] L = x end - x start (6)

[0027] The solution of the data preprocessing module of the present invention for eliminating the noise of the multi-modal sensor module is as follows:

[0028] S1. Noise characteristic analysis and identification: During a representative period when the switch machine is in normal operation, for various sensors in the multi-modal sensor module, the output data is synchronously collected. For the data sequence collected by each type of sensor, statistical quantities such as the mean, variance, and standard deviation are calculated respectively. Combining the noise characteristics in the actual operating environment of the switch machine, the measurement accuracy of the sensor, and the sensitivity requirement of the monitoring system for noise judgment, a data deviation threshold coefficient k for determining noise is determined. When the newly collected data point x new satisfies when x new - μ > kσ, this data point is marked as a data point suspected of being contaminated by noise, where μ represents the mean of the data sequence collected during the normal operation of the sensor, and μ represents the standard deviation of the data sequence collected during the normal operation of the sensor;

[0029] S2. Noise data repair and filtering processing: For the data points marked as suspected noise, the Kalman adaptive filtering algorithm is used for repair. This algorithm performs optimal estimation on the dynamic system containing noise by constructing a state space model. Its adaptive mechanism can dynamically adjust the process noise covariance matrix Q and the observation noise covariance matrix R according to real-time data. The specific process is as follows:

[0030] First, define the state vector of the sensor signal where x k is the current value of the signal, is the signal change rate, and the state transition equation is: X k = FX k-1 + W k-1 where F is the state transition matrix and W k-1 is the process noise; then calculate the observation equation, the observation vector Z k = x k + V k , where V k is the observation noise and follows a normal distribution. By calculating the residual e k = Zk -HX k , where H is the observation matrix and the covariance Dynamically update the process noise covariance matrix Q and the observation noise covariance matrix R to make the filtering effect adapt to the change of noise characteristics;

[0031] S3. Spatiotemporal alignment of multi-source data: For asynchronous data from different sensors, use timestamp synchronization technology for spatiotemporal alignment:

[0032] Timestamp unification: Attach timestamps accurate to milliseconds to all sensor data, use the switch machine action event as the synchronization trigger benchmark, and then perform interpolation and completion. For low-frequency data, use cubic spline interpolation to complete the missing values at high-frequency data time points, forming a multi-dimensional dataset D = {D1, D2, D3} with time alignment, where D1 is electrical data, including data such as rotational speed, current, and voltage, D2 is mechanical data, including data such as displacement, vibration, and notch size, and D3 is environmental data, including data such as temperature, humidity, and air pressure.

[0033] The dynamic evaluation scheme of the switch machine health level by the intelligent analysis and decision-making module of the present invention is as follows:

[0034] S1. Construct a hierarchical index system: Define the first-level indicators: Determine the three core dimensions of electrical characteristics, mechanical characteristics, and environmental parameters; Refine the second-level indicators: including electrical characteristics: set the stability of the operating current, voltage fluctuation amplitude, motor speed deviation rate, etc.; Mechanical characteristics: cover the displacement accuracy of the operating rod, vibration amplitude of the power transmission shaft, change rate of the indication notch size, etc.; Environmental parameters: include the extreme values of internal temperature and humidity, external air pressure change frequency, etc., to form quantifiable underlying indicators;

[0035] S2. Determine the subjective weight based on the analytic hierarchy process: First, invite experts to score the importance of the same-level indicators, thereby forming a judgment matrix A = (a ij ) n×n ; After obtaining the judgment matrix, by solving the maximum eigenvalue λ max and its corresponding eigenvector ω, and normalizing the eigenvector, the subjective weight ω AHP can be obtained; To ensure the rationality of the expert scoring and the scientificity of the weight determination, a consistency test is also required, that is, by the formula calculate the consistency index, and then use to further judge, so as to ensure that the weight determination process meets the logical and actual requirements, where R I is the random consistency index.

[0036] S3. Determine the objective weight based on the entropy weight method: First, normalize the data of the third-level indicators to eliminate the influence of dimensions, specifically through the formula is realized, where xij represents the j-th index value of the i-th sample; on this basis, further calculate the information entropy of index i, that is, use the formula to quantify the information disorder degree of index data, where m is the number of samples, and p ij represents the proportion of the j-th index value x′ in the i-th sample ij in the sum of all sample values of this index; finally, according to the information entropy result, through the formula calculate the objective weight, so that the weight determination can reflect the discrete characteristics and information contribution degree of the index data itself, and complete the determination of the objective weight based on the entropy weight method, where E j is the information entropy of the j-th index, n represents the total number of indexes, and S j represents the objective weight of the j-th index;

[0037] S4. Variable weight combination weighting: Introduce the working condition adjustment coefficient α ∈ [0, 1], and dynamically fuse the subjective weight C determined by the analytic hierarchy process through this coefficient j and the objective weight S determined by the entropy weight method j , to obtain the final weight W j , which is expressed as:

[0038] W i = αC i +(1 - α)S i (7)

[0039] S5. Fuzzy grey clustering algorithm evaluation: To ensure effective analysis of data under the same dimension, it is necessary to normalize the data to the [0, 1] interval again. Taking the health level standard mode as the reference sequence X0, calculate the correlation coefficient between the sequence X to be evaluated i and X0, which is expressed as:

[0040]

[0041] In the formula, ρ is the resolution coefficient, taking 0.5, min i min k |x0(k)-x i (k)| represents the minimum absolute difference between all sequences X i and the reference sequence X0 at all times k, and max i max k |x0(k)-x i (k)| represents the maximum absolute difference; through determine the membership degree of the sequence to each health level Y l , where, ω kis the weight at each moment k, n is the total number of moments of the data, and L is the total number of health levels. According to the membership degree, the health levels of the switch machine are divided into categories such as normal, warning, and failure, and the evaluation results and confidence levels are output.

[0042] The intelligent analysis and decision-making module of the present invention for the prediction scheme of the change trend of the indicating gap size is as follows:

[0043] S1. Data collection and preprocessing: Obtain long-time series image data of the indicating rod gap of the switch machine from the camera in the multi-modal sensor module. At the same time, collect auxiliary information such as the running time and the number of operations of the switch machine corresponding thereto. For the collected image data, use the Canny operator for edge detection and the pixel integral projection method to calculate the size of the indicating rod gap, obtain the indicating gap size value at each time point, and organize it into a time series data set. Conduct a stationarity test on the time series data. If the data is not stationary, use the differencing method to convert it into a stationary time series, and record the number of differencing times at the same time.

[0044] S2. Model construction and training: Use the autoregressive integrated moving average model for predicting the change trend of the indicating gap size. Divide the preprocessed time series data into a training set and a test set. The training set is used for model parameter estimation, and the test set is used for evaluating the prediction performance of the model. Divide it in the ratio of 80% of the data as the training set and 20% as the test set. Then use the training set data to train the autoregressive integrated moving average model, and estimate the parameters of the model, including the autoregressive coefficient and the moving average coefficient, by the maximum likelihood estimation method.

[0045] S3. Model evaluation and optimization: Use the test set data to evaluate the prediction performance of the trained autoregressive integrated moving average model, and use indicators such as the root mean square error and the mean absolute error to measure the error between the model prediction value and the true value; according to the evaluation index results, if the model prediction performance does not meet the expectations, optimize the model.

[0046] S4. Prediction and result output: When the model evaluation meets the requirements, use the trained autoregressive integrated moving average model to predict the indicating gap size within a future period of time. The prediction duration is set to one week in the future. Output the prediction results in a visual form to the visualization and interaction module, and display the change trend of the indicating gap size over time in the form of a curve. At the same time, mark the prediction value and the historical true value on the curve to facilitate the maintenance personnel to intuitively understand the change situation of the indicating gap size and discover potential abnormal trends in advance.

[0047] The intelligent analysis and decision-making module of the present invention for the monthly warning scheme of the jamming failure is as follows:

[0048] S1. Data collection and feature engineering: Collect the historical operation data of the switch machine, including data such as rotational speed, displacement, vibration, temperature and humidity, air pressure, etc. collected by the multi-modal sensor module, as well as the recorded data of the jamming failure of the switch machine. Extract features from the collected data. For the features related to the jamming failure, extract features such as the number of sudden changes in the rotational speed of the power transmission shaft, the duration of abnormal displacement of the operating rod, the frequency of vibration amplitude exceeding the threshold, and the rate of change of temperature and humidity. Normalize the extracted features to eliminate the influence of different feature dimensions and enable the analysis of each feature on the same scale;

[0049] S2. Model training and selection: Use the random forest model for monthly early warning of jamming failures. The random forest model consists of multiple decision trees. Multiple decision trees are constructed by sampling the training data with replacement, and then the prediction results of multiple decision trees are integrated for the final decision; Divide the historical data into a training set and a test set, with 80% as the training set and 20% as the test set. Use the training set data to train the random forest model and adjust the parameters of the model, such as the number of decision trees, the maximum depth, the minimum sample split number, etc., to improve the prediction performance of the model. Through the method of cross-validation, evaluate the performance of the model under different parameter combinations and select the parameter combination that makes the model performance optimal;

[0050] S3. Model evaluation and verification: Use the test set data to evaluate the trained random forest model. Use indicators such as accuracy, recall rate, and F1 value to measure the prediction ability of the model for jamming failures, and then intuitively display the prediction results of the model through the confusion matrix, analyze the prediction accuracy of the model for jamming failures and normal states, and find out the sample features that the model is prone to misjudge, providing a basis for further optimizing the model;

[0051] S4. Implementation of monthly early warning: At the beginning of each month, collect the operation data of the switch machine in the previous month, process the data according to the methods of feature extraction and normalization, input the processed data into the trained random forest model for prediction. According to the model prediction results, if it is predicted that a certain switch machine has a risk of jamming failure, send a warning message to the operation and maintenance personnel through the visualization and interaction module. The warning message includes the location, number, predicted failure type, and the possibility of failure occurrence of the switch machine.

[0052] The intelligent analysis and decision-making module of the present invention optimizes the maintenance resource allocation plan based on historical data as follows:

[0053] S1. Data collection and collation: Collect the historical failure data of the switch machine, including the time of failure occurrence, failure type, maintenance records, including: maintenance personnel, maintenance duration, parts used for maintenance, etc., the model of the switch machine, service life, etc. Clean and collate the collected data, and remove duplicate data, incorrect data, and incomplete data;

[0054] S2. Fault Analysis and Classification: Analyze the historical fault data, classify and statistically analyze it according to the fault types, calculate indicators such as the occurrence frequency, average maintenance duration, and maintenance cost of various faults. Based on indicators such as fault frequency and maintenance cost, determine the key fault types. Key fault types are usually those with a relatively high occurrence frequency and a relatively large maintenance cost, and these fault types have a greater impact on the consumption of maintenance resources.

[0055] S3. Resource Demand Prediction: Based on the historical fault data and the operating status data of the switch machine, use the autoregressive integrated moving average model to predict the number of occurrences of various faults in the future for a certain period of time. According to the average maintenance duration and maintenance cost of various faults, combined with the predicted number of fault occurrences, calculate the human, material, and financial resources required to maintain the switch machine in the future for a certain period of time.

[0056] S4. Optimal Resource Allocation: According to the results of resource demand prediction, combined with the reserve situation and usage status of the existing maintenance resources, formulate an optimized maintenance resource allocation plan. For key fault types, prioritize the guarantee of the required maintenance resources, such as increasing the inventory of key components, arranging experienced maintenance personnel to be responsible for relevant maintenance work, etc. Establish a dynamic adjustment mechanism for maintenance resources, and adjust the allocation of maintenance resources in real time according to the actual fault occurrence situation and the progress of maintenance work.

[0057] Since the present invention constructs an innovative structure in which the multi-modal sensor, data preprocessing, intelligent analysis and decision-making, and visualization interaction modules cooperate, a series of remarkable beneficial effects can be produced:

[0058] The multi-modal sensor accurately collects data such as the rotational speed of the power transmission shaft of the switch machine, the displacement of the operating rod, vibration, indication gap, internal environment temperature and humidity, and external environment air pressure from multiple dimensions. At the same time, the data preprocessing module synchronously reduces noise, greatly improving the monitoring accuracy. The intelligent analysis and decision-making module, relying on advanced algorithms, can detect potential hazards in advance and optimize resource allocation, effectively reducing the maintenance cost; predicting faults with a specific model effectively enhances the fault prediction ability. The visualization interaction module creates a Web mobile interface, which helps the operation and maintenance personnel intuitively master the status of the switch machine and achieve efficient decision-making. Brief Description of the Drawings

[0059] [[ID=·17]] Figure 1 It is a schematic installation structure diagram of a railway electric switch machine status monitoring system of the present invention;

[0060] Figure 2 It is a working principle diagram of a railway electric switch machine status monitoring system of the present invention;

[0061] Figure 3 It is a flow chart of a solution for image recognition of the indication rod gap of a railway electric switch machine by a railway electric switch machine status monitoring system of the present invention;

[0062] Figure 4 This is the flowchart of the solution for the data preprocessing module of a railway electric switch machine status monitoring system of the present invention to eliminate the noise of the multi-modal sensor module;

[0063] Figure 5 This is the flowchart of the dynamic assessment solution for the health level of the switch machine by the intelligent analysis and decision-making module of a railway electric switch machine status monitoring system of the present invention;

[0064] Figure 6 This is the flowchart of the prediction solution for the change trend of the gap size represented by the intelligent analysis and decision-making module of a railway electric switch machine status monitoring system of the present invention;

[0065] Figure 7 This is the flowchart of the monthly early warning solution for the jamming fault by the intelligent analysis and decision-making module of a railway electric switch machine status monitoring system of the present invention;

[0066] Figure 8 This is the flowchart of the solution for the intelligent analysis and decision-making module of a railway electric switch machine status monitoring system of the present invention to optimize the maintenance resource allocation based on historical data.

[0067] Description of main component symbols.

[0068] Detailed implementation manners

[0069] The present invention will be further described in detail below in conjunction with the embodiments and with reference to the accompanying drawings.

[0070] Please refer to Figures 1 to 8 As shown, a railway electric switch machine status monitoring system in the present invention includes a multi-modal sensor module, a data preprocessing module, an optical fiber transmission module 9, an intelligent analysis and decision-making module, and a visualization and interaction module 10.

[0071] As Figure 1As shown in the figure, the multimodal sensor modules are distributed in each key part of the switch machine to obtain comprehensive and accurate data. The data preprocessing module uses the TMS320F28335 minimum system board 1 to receive the data collected by the multimodal sensor modules and eliminate the noise of the data transmitted by the multimodal sensor modules. The spatio-temporal alignment technology is used on the TMS320F28335 minimum system board 1 to fuse multi-source data, and the Kalman adaptive filtering algorithm is used to eliminate noise. The intelligent analysis and decision-making module uses the STM32F103ZET6 minimum system board 2 to receive the data processed by the data preprocessing module, and constructs a hierarchical index system including electrical characteristics, mechanical characteristics, and environmental parameters in the STM32F103ZET6 minimum system board 2, and adds the variable weight combination weighting method (analytic hierarchy process + entropy weight method) and the fuzzy grey clustering algorithm to realize the dynamic evaluation of the health level of the switch machine. At the same time, the autoregressive integrated moving average model prediction is added to the STM32F103ZET6 minimum system board 2 to represent the change trend of the notch size, and the random forest model is used to give monthly early warnings for jamming faults, optimize the maintenance resource allocation according to historical data, and the STM32F103ZET6 minimum system board 2 controls the optical fiber transmission module 9 to transmit the data to the visualization and interaction module 10. A Web mobile interface is developed in the visualization and interaction module 10, which supports the display of the health status at the network level, line level, and equipment level, and also has functions such as real-time alarm, trend analysis, and maintenance work order generation, providing strong assistance for operation and maintenance decision-making.

[0072] As Figure 1As shown in the figure, the multi-modal sensor module includes a speed sensor 3, a displacement sensor 4, a vibration sensor 5, a camera 6, a temperature and humidity sensor 7, and a barometric pressure sensor 8. The speed sensor 3 is installed on the power transmission shaft of the switch machine to monitor the rotation speed of the power transmission shaft of the switch machine in real time. Abnormal rotation speed may indicate problems such as motor failure and wear of transmission components. The displacement deviation of the action rod may cause the switch to fail to switch in place, affecting train operation safety. Therefore, the displacement sensor 4 is installed on the action rod of the switch machine to measure the linear displacement of the action rod. The vibration sensor 5 is installed on the outer shell of the switch machine to sense the vibration condition during the operation of the switch machine. By monitoring the vibration condition of the switch machine, it is judged whether there are looseness, wear or other abnormalities in the internal mechanical structure. Abnormal vibration may be an early signal of component damage. The camera 6 is installed on the outer side of the switch machine to capture the image of the notch of the indicating rod of the switch machine. The camera 6 can be installed at a position with a horizontal distance of 15 - 20 cm from the indicating notch and a vertical height 10 cm higher than the center of the indicating notch, with a downward shooting angle of 20 degrees. This can clearly present the size and position change of the indicating notch and avoid interference caused by light reflection. The change of temperature and humidity will affect the performance of internal electrical components and the lubrication condition of mechanical components in the switch machine. Therefore, the temperature and humidity sensor 7 is installed inside the switch machine to monitor the ambient temperature and humidity inside the switch machine. Since the change of barometric pressure will affect the sealing performance and pneumatic components inside the switch machine, the barometric pressure sensor 8 is installed on the surface of the switch machine to monitor the external ambient barometric pressure. The data information collected by the speed sensor 3, the displacement sensor 4, the vibration sensor 5, the camera 6, the temperature and humidity sensor 7, and the barometric pressure sensor 8 is transmitted to the TMS320F28335 minimum system board 1 for preprocessing.

[0073] As Figure 3 shown in the figure, the present invention uses the Canny operator to perform edge detection on the image of the notch of the indicating rod of the switch machine collected by the camera 6, and uses the pixel integral projection method to calculate the size of the notch of the indicating rod. The specific scheme is as follows:

[0074] S1. Preprocess the collected color image. First, convert it from the RGB color space to the grayscale space, expressed as:

[0075] G(x, y) = 0.299R(x,y) + 0.587G(x,y) + 0.114B(x,y) (1)

[0076] This operation can reduce the data volume and facilitate subsequent analysis;

[0077] S2. Use Gaussian filtering to smooth the grayscale image. The convolution kernel of Gaussian filtering is determined by the Gaussian function, expressed as:

[0078]

[0079] where σ is the standard deviation, (m, n) are the coordinates within the convolution kernel, and the filtered image S(x, y) can be obtained through convolution operation, expressed as:

[0080]

[0081] This can effectively remove the noise in the image;

[0082] S3. Apply the Canny operator to perform edge detection on the smoothed image S(x, y). The Canny operator first calculates the gradient magnitude M(x, y) and gradient direction θ(x, y) of the image, expressed as:

[0083]

[0084] where G x (x, y) and G y (x, y) are the gradients of the image in the x and y directions respectively

[0085] S4. Perform non-maximum suppression, only retaining the pixels at the local gradient maximum, and then determine the true edges through the double-threshold method to obtain the edge image E(x, y);

[0086] S5. Use the pixel integral projection method to calculate the size of the rod notch. Perform a horizontal projection on the edge image E(x, y). The horizontal projection function is:

[0087]

[0088] where Y is the height of the image. By analyzing the horizontal projection curve P(x), set an appropriate threshold T. When P(x) < T, determine the starting and ending positions x start and x end , then the pixel size of the notch is:

[0089] L = x end -x start (6)

[0090] As Figure 4 shown, the scheme for the data preprocessing module to eliminate the noise of the multi-modal sensor module is:

[0091] S1. Noise characteristic analysis and identification: During a representative period when the switch machine is in normal operation, output data of various sensors in the multi-modal sensor module are synchronously collected. The acquisition frequency is determined according to the accuracy requirements of each sensor, the signal change rate, and the actual monitoring needs, ensuring that the acquired data can fully reflect the characteristics of the sensor signals during the normal operation of the switch machine. For the data sequences collected by each type of sensor, statistical quantities such as mean, variance, and standard deviation are calculated respectively. These statistical features are used to describe the central tendency and dispersion degree of the data of this type of sensor in the normal state. Combining the noise characteristics in the actual operating environment of the switch machine, the measurement accuracy of the sensor, and the sensitivity requirements of the monitoring system for noise judgment, a data deviation threshold coefficient k for determining noise is determined. When the newly acquired data point x new satisfies when x new - μ > kσ, this data point is marked as a data point suspected of being contaminated by noise, where μ represents the mean of the data sequence collected during the normal operation of the sensor, and μ represents the standard deviation of the data sequence collected during the normal operation of the sensor;

[0092] S2. Noise data repair and filtering processing: For the data points marked as suspected noise, the Kalman adaptive filtering algorithm is used for repair. This algorithm implements optimal estimation for the dynamic system with noise by constructing a state space model. Its adaptive mechanism can dynamically adjust the process noise covariance matrix Q and the observation noise covariance matrix R according to real-time data. The specific process is as follows:

[0093] First, define the state vector of the sensor signal where x k is the current value of the signal, is the signal change rate, and the state transition equation is: X k = FX k-1 + W k-1 where F is the state transition matrix and W k-1 is the process noise; then calculate the observation equation, the observation vector Z k = x k + V k , where V k is the observation noise and follows a normal distribution. By calculating the residual e k = Z k - HX k in real time, where H is the observation matrix, and the covariance dynamically updates the process noise covariance matrix Q and the observation noise covariance matrix R to make the filtering effect adapt to the change of noise characteristics;

[0094] S3. Spatiotemporal alignment of multi-source data: For asynchronous data of different sensors, timestamp synchronization technology is used for spatiotemporal alignment:

[0095] Timestamp Unification: Attach timestamps accurate to milliseconds to all sensor data. Using the switch machine operation event as the synchronization trigger benchmark, then perform interpolation and completion. For low-frequency data, use cubic spline interpolation to complete the missing values at high-frequency data time points, forming a multi-dimensional data set D = {D1, D2, D3} with time alignment, where D1 is electrical data, including data such as rotational speed, current, and voltage; D2 is mechanical data, including data such as displacement, vibration, and notch size; D3 is environmental data, including data such as temperature, humidity, and air pressure.

[0096] As Figure 5 shown, the dynamic evaluation scheme of the intelligent analysis and decision-making module for the health level of the switch machine is as follows:

[0097] S1. Construct a hierarchical index system: Define the first-level index: Determine the three core dimensions of electrical characteristics, mechanical characteristics, and environmental parameters; Refine the second-level index: including electrical characteristics: set stability of operating current, voltage fluctuation amplitude, motor speed deviation rate, etc.; Mechanical characteristics: cover displacement accuracy of the operating rod, vibration amplitude of the power transmission shaft, change rate of the indication notch size, etc.; Environmental parameters: include internal temperature and humidity extremes, external air pressure change frequency, etc., to form quantifiable underlying indicators;

[0098] S2. Determine the subjective weight based on the analytic hierarchy process: First, invite experts to score the importance of the same-level indicators, thus forming a judgment matrix A = (a ij ) n×n ; After obtaining the judgment matrix, by solving the maximum eigenvalue λ max and its corresponding eigenvector ω, and normalizing the eigenvector, the subjective weight ω AHP can be obtained; To ensure the rationality of expert scoring and the scientificity of weight determination, a consistency test is also required, that is, calculate the consistency index through the formula , and then use to further judge, so as to ensure that the weight determination process meets logical and actual requirements, where R I is the random consistency index.

[0099] S3. Determine the objective weight based on the entropy weight method: First, normalize the data of the third-level indicators to eliminate the influence of dimensions, specifically through the formula to achieve, where x ij represents the jth index value of the ith sample; On this basis, further calculate the information entropy of index i, that is, use the formula to quantify the information disorder degree of the index data, where m is the number of samples, and p ij represents the proportion of the jth index value x′ ij of the ith sample in the total sum of all sample values of this index; Finally, according to the information entropy result, through the formula Calculate the objective weight so that the determination of the weight can reflect the discrete characteristics and information contribution degree of the index data itself, and complete the determination of the objective weight based on the entropy weight method, where E j is the information entropy of the j-th index, n represents the total number of indexes, and S j represents the objective weight of the j-th index;

[0100] S4. Variable weight combination weighting: Introduce the working condition adjustment coefficient α ∈ [0, 1], and dynamically fuse the subjective weight C j determined by the analytic hierarchy process and the objective weight S j determined by the entropy weight method through this coefficient to obtain the final weight W j , which is expressed as:

[0101] W i = αC i + (1 - α)S i (7)

[0102] S5. Fuzzy grey clustering algorithm evaluation: To ensure effective analysis of data under the same dimension, it is necessary to normalize the data to the [0, 1] interval again. Taking the health level standard mode as the reference sequence X0, calculate the correlation coefficient between the sequence X i to be evaluated and X0, which is expressed as:

[0103]

[0104] In the formula, ρ is the resolution coefficient, taking 0.5, min i min k |x0(k) - x i (k)| represents the minimum absolute difference between all sequences X i and the reference sequence X0 at all times k, and max i max k |x0(k) - x i (k)| represents the maximum absolute difference; through determine the membership degree of the sequence to each health level Y l , where ω k is the weight at each time k, n is the total number of data times, L is the total number of health levels, and the health level of the switch machine is divided into categories such as normal, warning, and failure according to the membership degree, and the evaluation result and confidence level are output.

[0105] As Figure 6 shown, the intelligent analysis and decision-making module's prediction scheme for the change trend of the gap size is:

[0106] S1. Data collection and preprocessing: Obtain long-time series image data of the switch rod notch from the camera 6 in the multi-modal sensor module. At the same time, collect auxiliary information such as the running time and the number of actions of the switch machine corresponding to it. For the collected image data, use the Canny operator for edge detection and the pixel integral projection method to calculate the size of the switch rod notch, obtain the size value of the representation notch at each time point, and organize it into a time series dataset. Conduct a stationarity test on the time series data. If the data is non-stationary, use the differencing method to transform it into a stationary time series, and record the number of differencing times. For example, judge the stationarity of the data by calculating the autocorrelation function and partial autocorrelation function of the data. If the autocorrelation function and partial autocorrelation function do not decay rapidly to 0 after a certain number of lags, it indicates that the data is non-stationary, and perform first-order or multi-order differencing on it until the data becomes stationary.

[0107] S2. Model construction and training: Use the autoregressive integrated moving average model to predict the change trend of the representation notch size. Divide the preprocessed time series data into a training set and a test set. The training set is used for model parameter estimation, and the test set is used to evaluate the prediction performance of the model. Divide the data in the ratio of 80% as the training set and 20% as the test set. For example, if there are 1000 time points of representation notch size data, select the first 800 data as the training set and the last 200 data as the test set. Then use the training set data to train the autoregressive integrated moving average model, and estimate the parameters of the model, including the autoregressive coefficient and the moving average coefficient, by the maximum likelihood estimation method;

[0108] S3. Model evaluation and optimization: Use the test set data to evaluate the prediction performance of the trained autoregressive integrated moving average model, and use indicators such as the root mean square error and the mean absolute error to measure the error between the model prediction value and the true value; According to the evaluation index results, if the prediction performance of the model does not meet the expectations, optimize the model; For example, if the root mean square error and the mean absolute error values are large, it indicates that the model prediction error is large. Adjust the parameters and then retrain the model and evaluate the performance until the model reaches a satisfactory prediction performance.

[0109] S4. Prediction and Result Output: After the model evaluation meets the requirements, use the trained autoregressive integrated moving average model to predict the representation gap size within a certain period in the future. The prediction duration is set to one week in the future. Output the prediction results in a visual form to the visualization and interaction module 10, and display the change trend of the representation gap size over time in the form of a curve. At the same time, mark the predicted values and historical true values on the curve to facilitate the maintenance personnel to intuitively understand the change of the representation gap size and discover potential abnormal trends in advance. For example, on the Web mobile interface, use time as the horizontal axis and the representation gap size as the vertical axis to plot the historical data curve and the predicted data curve. The predicted data curve is distinguished from the historical data curve by different colors or line styles, and at the same time, mark the values at key time points on the curve.

[0110] As Figure 7 shown, the monthly early warning plan for the jamming fault by the intelligent analysis and decision-making module is as follows: S1. Data Collection and Feature Engineering: Collect the historical operation data of the switch machine, including data such as rotational speed, displacement, vibration, temperature, humidity, air pressure, etc. collected by the multi-modal sensor module, and the recorded data of the jamming fault of the switch machine. Extract features from the collected data. For features related to the jamming fault, extract features such as the number of sudden changes in the rotational speed of the power transmission shaft, the abnormal duration of the displacement of the action rod, the frequency of the vibration amplitude exceeding the threshold, and the change rate of temperature and humidity. For example, calculate the change amount of the rotational speed of the power transmission shaft within a short period (such as within 1 second). If the change amount exceeds the set threshold, it is recorded as one rotational speed mutation; calculate the deviation duration of the displacement of the action rod from the normal displacement range. When the duration exceeds a certain length (such as 5 seconds), it is considered that the displacement is abnormally continuous. Normalize the extracted features to eliminate the influence of different feature dimensions and analyze each feature on the same scale.

[0111] S2. Model Training and Selection: Use the random forest model for monthly early warning of jamming faults. The random forest model consists of multiple decision trees. Multiple decision trees are constructed by sampling the training data with replacement, and then the prediction results of multiple decision trees are integrated for the final decision. Divide the historical data into a training set and a test set, with 80% as the training set and 20% as the test set. Use the training set data to train the random forest model and adjust the parameters of the model, such as the number of decision trees, the maximum depth, the minimum sample split number, etc., to improve the prediction performance of the model. Evaluate the performance of the model under different parameter combinations through the method of cross-validation, and select the parameter combination that makes the model performance optimal. For example, adjust the number of decision trees (such as from 50 to 200), the maximum depth (such as from 3 to 10), and the minimum sample split number (such as from 2 to 10) within a certain range. Use 5-fold cross-validation to calculate indicators such as the accuracy rate, recall rate, and F1 value of the model on the training set for each parameter combination, and select the parameter combination that makes these indicators comprehensively optimal.

[0112] S3. Model Evaluation and Verification: Use the test set data to evaluate the trained random forest model. Adopt metrics such as accuracy, recall rate, and F1 value to measure the model's prediction ability for jamming faults. Then, visually display the model's prediction results through a confusion matrix, analyze the prediction accuracy of the model for jamming faults and normal states, and find the sample characteristics that the model is prone to misjudge, providing a basis for further optimizing the model. For example, generate a confusion matrix based on the test set data, observe the distribution of TP, FP, FN, and TN (true negative cases, that is, the number of samples correctly predicted as normal states) in the matrix. If it is found that there are more misjudgments in a certain category, analyze the characteristics of the samples in this category. For example, if the range of some feature values is different from other samples, further process these features or adjust the model parameters accordingly.

[0113] S4. Monthly Warning Implementation: At the beginning of each month, collect the operation data of the switch machine in the previous month, process the data according to the methods of feature extraction and normalization, input the processed data into the trained random forest model for prediction. According to the model prediction results, if it is predicted that a certain switch machine has a risk of jamming fault, send a warning message to the operation and maintenance personnel through the visualization and interaction module 10. The warning message includes the location, number, predicted fault type, and the possibility of the fault occurrence of the switch machine. For example, pop up a warning window on the Web mobile interface, showing that "the switch machine at XX Station on Line XX has a risk of jamming fault, and the probability of the fault occurrence is XX%", and at the same time mark the location of the switch machine on the map to facilitate the operation and maintenance personnel to conduct inspections and handling in a timely manner.

[0114] As Figure 8 shown, the intelligent analysis and decision-making module optimizes the maintenance resource allocation plan based on historical data as follows:

[0115] S1. Data Collection and Sorting: Collect the historical fault data of the switch machine, including the fault occurrence time, fault type, maintenance records, including: maintenance personnel, maintenance duration, parts used for maintenance, etc., the model of the switch machine, service life, etc. Clean and sort the collected data, remove duplicate data, error data, and incomplete data. For example, check whether the fields such as maintenance personnel and maintenance duration in the maintenance records are filled in completely. If there are missing values, supplement or delete the corresponding records according to the actual situation; check whether the format of the fault occurrence time is unified, and perform format conversion if it is not unified;

[0116] S2. Fault Analysis and Classification: Analyze historical fault data, classify and count them according to fault types, and calculate indicators such as the occurrence frequency, average repair duration, and repair cost of each type of fault. For example, count the number of occurrences of various types of faults such as electrical faults, mechanical faults, and faults caused by environmental factors, and calculate the occurrence frequency of each type of fault; sum up the repair durations of each type of fault and divide by the number of occurrences of that type of fault to obtain the average repair duration; count the component costs, labor costs, etc. consumed in repairing each type of fault, and calculate the repair cost = Determine the key fault types based on indicators such as fault frequency and repair cost. Key fault types are usually those with a relatively high occurrence frequency and a relatively large repair cost, and these fault types have a greater impact on the consumption of maintenance resources; for example, by comparing the frequencies and repair costs of various types of faults, it is found that the occurrence frequency of mechanical faults accounts for 40% of the total faults, and the repair cost accounts for 50% of the total repair cost, then mechanical faults are determined as the key fault type.

[0117] S3. Resource Requirement Prediction: Based on historical fault data and the operating status data of the switch machine, use the autoregressive integrated moving average model to predict the number of occurrences of various types of faults in a future period. For example, use the exponential smoothing method to predict the number of occurrences of electrical faults. According to the historical data of the number of occurrences of electrical faults, set the smoothing coefficient to 0.3, and calculate the predicted values for future periods through the exponential smoothing formula F t =αA t-1 +(1 - α)F t-1 , where F t is the predicted value for the t-th period, A t-1 is the actual value for the (t - 1)-th period, F t-1 is the predicted value for the (t - 1)-th period, and α is the smoothing coefficient. Based on the average repair duration and repair cost of each type of fault, combined with the predicted number of occurrences of faults, calculate the human, material, and financial resources required to maintain the switch machine in a future period. For example, if it is predicted that the number of occurrences of mechanical faults in the next month is 10 times, the average repair duration for each mechanical fault is 3 hours, the labor cost per hour for repair personnel is 100 yuan, and the average component cost for each repair is 500 yuan, then the labor cost required for mechanical fault repair in the next month is 10×3×100 = 3000 yuan, the component cost required is 10×500 = 5000 yuan, and at the same time, the number of repair personnel required can be calculated based on the repair duration and the working time arrangement of the repair personnel;

[0118] S4. Optimized Allocation of Resources: Based on the results of resource demand forecasting, combined with the reserve situation and usage status of existing maintenance resources, formulate an optimized maintenance resource allocation plan. For key failure types, prioritize the guarantee of the required maintenance resources, such as increasing the inventory of key components and arranging experienced maintenance personnel to be responsible for relevant maintenance work. For example, if it is predicted that the number of mechanical failures will be relatively high in the future and the inventory of mechanical failure repair parts is insufficient, a certain quantity of common parts can be purchased in advance to ensure that the maintenance work can be carried out in a timely manner; for experienced maintenance personnel, reasonably arrange their work tasks so that they are mainly responsible for the repair of key failure types such as mechanical failures to improve the maintenance efficiency. Establish a dynamic adjustment mechanism for maintenance resources, and adjust the allocation of maintenance resources in real time according to the actual occurrence of failures and the progress of maintenance work. For example, if it is found during actual maintenance that the occurrence frequency of a certain type of failure is higher than the predicted value, promptly allocate corresponding maintenance resources from other projects to ensure the smooth progress of maintenance work; if the repair time of a certain type of failure is shortened, the input of maintenance resources for this type of failure can be reduced accordingly, and the saved resources can be allocated to other places where they are needed.

[0119] The visualization and interaction module 10 of the present invention is placed on the computer side, and a Web mobile interface is developed in the visualization and interaction module 10, which supports the display of the health status at the network level, line level, and device level, and also has functions such as real-time alarm, trend analysis, and maintenance work order generation, providing strong assistance for operation and maintenance decision-making.

[0120] The working principle and process of the present invention are as follows:

[0121] Such as Figure 2As shown in the figure, the multi-modal sensor module synchronously collects multi-dimensional data such as the rotational speed of the power transmission shaft, the displacement of the operating rod, vibration signals, the image of the indication notch, internal temperature and humidity, and external air pressure through sensors such as speed, displacement, vibration, camera 6, temperature and humidity, and air pressure installed at key parts of the switch machine, and transmits the data to the data preprocessing module; this module first marks suspected noise data through noise characteristic analysis, then uses the Kalman adaptive filtering algorithm to repair the noise, and at the same time uses timestamp synchronization and cubic spline interpolation method to perform spatio-temporal alignment on asynchronous data to form a unified multi-dimensional data set; the intelligent analysis and decision-making module constructs a hierarchical index system including electrical, mechanical, and environmental parameters based on the preprocessed data, determines the index weights by combining the analytic hierarchy process and the entropy weight method, dynamically evaluates the health level of the switch machine through the fuzzy grey clustering algorithm, and at the same time uses the autoregressive integrated moving average model to predict the change trend of the indication notch size and the random forest model to conduct monthly early warning of jamming faults, and optimizes the maintenance resource allocation according to historical fault data; finally, the visualization and interaction module 10 presents the health status, prediction results, warning information, etc. in a Web mobile interface, supports the state display from the line network level to the device level, real-time alarm, trend analysis and maintenance work order generation, and provides full-process assistance for operation and maintenance decision-making.

Claims

1. A railway electric switch machine status monitoring system, characterized in that: It includes a multi-modal sensor module, a data preprocessing module, an optical fiber transmission module, an intelligent analysis and decision-making module, and a visualization and interaction module. The multi-modal sensor module is distributed at various key parts of the switch machine to obtain comprehensive and accurate data. The data preprocessing module uses the TMS320F28335 minimum system board to receive the data collected by the multi-modal sensor module and eliminate the noise of the data transmitted by the multi-modal sensor module. The TMS320F28335 minimum system board uses spatio-temporal alignment technology to fuse multi-source data and adopts the Kalman adaptive filtering algorithm to eliminate noise. The intelligent analysis and decision-making module uses the STM32F103ZET6 minimum system board to receive the data processed by the data preprocessing module, and constructs a hierarchical index system including electrical characteristics, mechanical characteristics, and environmental parameters in the STM32F103ZET6 minimum system board. The variable weight combination weighting method (analytic hierarchy process + entropy weight method) and the fuzzy grey clustering algorithm are added to realize the dynamic assessment of the health level of the switch machine. At the same time, the autoregressive integrated moving average model is added to the STM32F103ZET6 minimum system board to predict the change trend of the notch size, and the random forest model is used to give monthly warnings for jamming faults, optimize the maintenance resource allocation according to historical data, and the STM32F103ZET6 minimum system board controls the optical fiber transmission module to transmit the data to the visualization and interaction module. A Web mobile interface is developed in the visualization and interaction module, which supports the display of the health status at the line network level, line level, and equipment level, and also has functions such as real-time alarm, trend analysis, and maintenance work order generation, providing strong assistance for operation and maintenance decision-making.

2. The railway electric switch machine status monitoring system according to claim 1, wherein: The multi-modal sensor module includes a speed sensor, a displacement sensor, a vibration sensor, a camera, a temperature and humidity sensor, and a barometric pressure sensor. The speed sensor is installed on the power transmission shaft of the switch machine to monitor the rotation speed of the power transmission shaft of the switch machine in real time. The displacement sensor is installed on the operating rod of the switch machine to measure the linear displacement of the operating rod. The vibration sensor is installed on the outer shell of the switch machine to sense the vibration condition during the operation of the switch machine. By monitoring the vibration condition of the switch machine, it is judged whether there is looseness, wear or other abnormalities in the internal mechanical structure. The camera is installed on the outer side of the switch machine to take pictures of the notch image of the indicating rod of the switch machine. Therefore, the temperature and humidity sensor is installed inside the switch machine to monitor the internal environmental temperature and humidity of the switch machine, and the barometric pressure sensor is installed on the surface of the switch machine to monitor the external environmental barometric pressure. The data information collected by the speed sensor, displacement sensor, vibration sensor, camera, temperature and humidity sensor, and barometric pressure sensor is transmitted to the TMS320F28335 minimum system board for preprocessing.

3. The railway electric switch machine status monitoring system according to claim 2, characterized in that: The Canny operator is used to perform edge detection on the notch image of the indicating rod of the switch machine collected by the camera, and the pixel integral projection method is used to calculate the size of the notch of the indicating rod. The specific scheme is as follows: S1. Preprocess the collected color image. First, convert it from the RGB color space to the grayscale space, which is expressed as: G(x, y) = 0.299R(x, y) + 0.587G(x, y) + 0.114B(x, y) (1) This operation can reduce the data volume and facilitate subsequent analysis; S2. Use Gaussian filtering to smooth the grayscale image. The convolution kernel of Gaussian filtering is determined by the Gaussian function and is expressed as: where σ is the standard deviation, and (m, n) are the coordinates within the convolution kernel. The filtered image S(x, y) can be obtained through convolution operation and is expressed as: This can effectively remove the noise in the image; S3. Use the Canny operator to perform edge detection on the smoothed image S(x, y). The Canny operator first calculates the gradient magnitude M(x, y) and gradient direction θ(x, y) of the image, which are expressed as: where G x (x, y) and G y (x, y) are the gradients of the image in the x and y directions respectively S4. Perform non-maximum suppression, only retain the pixels at the local gradient maximum, and then determine the real edges through the double-threshold method to obtain the edge image E(x, y); S5. Use the pixel integral projection method to calculate the size of the rod notch. Perform horizontal projection on the edge image E(x, y), and the horizontal projection function is: Where Y is the height of the image. By analyzing the horizontal projection curve P(x), a suitable threshold T is set. When P(x) < T, the starting and ending positions x of the notch are determined start and x end , then the pixel size of the notch is: L = x end -x start (6) 4. The railway electric switch machine status monitoring system according to claim 1, characterized in that: The solution for the data preprocessing module to eliminate the noise of the multi-modal sensor module is as follows: S1. Noise characteristic analysis and identification: During a representative period when the switch machine is in normal operating conditions, synchronously collect the output data of various sensors in the multi-modal sensor module. For the data sequences collected by each type of sensor, calculate statistical quantities such as their mean, variance, and standard deviation respectively. Combining the noise characteristics in the actual operating environment of the switch machine, the measurement accuracy of the sensors, and the sensitivity requirements of the monitoring system for noise judgment, determine a data deviation threshold coefficient k for noise determination. When the newly collected data point x new satisfies that when x new - μ > kσ, mark this data point as a data point suspected of being contaminated by noise, where μ represents the mean of the data sequence collected during the normal operation of the sensor, and μ represents the standard deviation of the data sequence collected during the normal operation of the sensor; S2. Noise data repair and filtering processing: For the data points marked as suspected noise, use the Kalman adaptive filtering algorithm for repair. This algorithm implements optimal estimation for the dynamic system with noise by constructing a state space model. Its adaptive mechanism can dynamically adjust the process noise covariance matrix Q and the observation noise covariance matrix R according to real-time data. The specific process is as follows: First, define the state vector of the sensor signal where x k is the current value of the signal, is the signal change rate, and the state transition equation is: X k = FX k-1 + W k-1 where F is the state transition matrix and W k-1 is the process noise; then calculate the observation equation. The observation vector Z k = x k + V k , where V k is the observation noise and follows a normal distribution. By calculating the residual e k = Z k - HX k , where H is the observation matrix, and the covariance dynamically updates the process noise covariance matrix Q and the observation noise covariance matrix R to adapt the filtering effect to the changes in the noise characteristics; S3. Spatiotemporal alignment of multi-source data: For the asynchronous data of different sensors, use the timestamp synchronization technology for spatiotemporal alignment: Timestamp unification: Attach timestamps accurate to milliseconds to all sensor data. Based on the switch machine action event as the synchronization trigger benchmark, then perform interpolation and completion. For low-frequency data, use the cubic spline interpolation method to complete the missing values at the high-frequency data time points, forming a spatiotemporally aligned multi-dimensional dataset D = {D1, D2, D3}, where D1 is electrical data, including data such as rotational speed, current, and voltage; D2 is mechanical data, including data such as displacement, vibration, and notch size; D3 is environmental data, including data such as temperature, humidity, and air pressure.

5. The railway electric switch machine status monitoring system according to claim 1, characterized in that: The solution for the intelligent analysis and decision-making module to dynamically evaluate the health level of the switch machine is as follows: S1. Build a hierarchical index system: Define the first-level indicators: Determine the three core dimensions of electrical characteristics, mechanical characteristics, and environmental parameters; Refine the second-level indicators: including electrical characteristics: set the stability of the operating current, voltage fluctuation amplitude, motor speed deviation rate, etc.; Mechanical characteristics: cover the displacement accuracy of the operating rod, vibration amplitude of the power transmission shaft, change rate of the indication notch size, etc.; Environmental parameters: include the extreme values of internal temperature and humidity, external air pressure change frequency, etc., to form quantifiable underlying indicators; S2. Determine the subjective weight based on the analytic hierarchy process: First, invite experts to score the importance of the same-level indicators, thus forming a judgment matrix A = (a ij ) n×n ; After obtaining the judgment matrix, by solving the maximum eigenvalue λ max and its corresponding eigenvector ω, and normalizing the eigenvector, the subjective weight ω AHP can be obtained; To ensure the rationality of the experts' scoring and the scientificity of the weight determination, a consistency test is also required, that is, by using the formula to calculate the consistency index, and then using for further judgment, so as to ensure that the weight determination process meets the logical and actual requirements, where R is the random consistency index; S3. Determine the objective weight based on the entropy weight method: First, normalize the data of the third-level indicators to eliminate the influence of dimensions, specifically through the formula to achieve, where x ij represents the value of the j-th indicator of the i-th sample; on this basis, further calculate the information entropy of indicator i, that is, use the formula to quantify the degree of information disorder of the indicator data, where m is the number of samples, and p ij represents the proportion of the value x′ ij of the j-th indicator in the i-th sample in the total sum of all sample values of this indicator; finally, according to the information entropy result, use the formula to calculate the objective weight, so that the weight determination can reflect the discrete characteristics and information contribution degree of the indicator data itself, and complete the determination of the objective weight based on the entropy weight method, where E j is the information entropy of the j-th indicator, n represents the total number of indicators, and S j represents the objective weight of the j-th indicator; S4. Variable weight combination weighting: Introduce the working condition adjustment coefficient α ∈ [0, 1], and dynamically fuse the subjective weight C determined by the analytic hierarchy process through this coefficient j and the objective weight S determined by the entropy weight method j to obtain the final weight W j which is expressed as: W i = αC i + (1 - α)S i (7) S5. Fuzzy Grey Clustering Algorithm Evaluation: To ensure effective analysis of data under the same dimension, it is necessary to normalize the data to the [0, 1] interval again. Taking the healthy level standard mode as the reference sequence X0, calculate the correlation coefficient between the sequence X to be evaluated and X0, which is expressed as: i And the correlation coefficient of X0, which is expressed as: where ρ is the resolution coefficient, taking 0.5, min i min k |x0(k)-x i (k)| represents the minimum absolute difference between all sequences X i and the reference sequence X0 at all times k, max i max k |x0(k)-x i (k)| represents the maximum absolute difference; by determining the membership degrees of the sequences to each health level Y l where ω k is the weight at each time k, n is the total number of data times, L is the total number of health levels, and the health levels of the switch machine are classified into categories such as normal, warning, and failure according to the membership degrees, and the evaluation results and confidence levels are output.

6. The railway electric switch machine status monitoring system according to claim 1 and claim 2, characterized in that: The solution for the intelligent analysis and decision-making module to predict the change trend of the indication notch size is as follows: S1. Data collection and preprocessing: Obtain long-time series image data of the switch machine's indicating rod notch from the camera in the multimodal sensor module. Meanwhile, collect auxiliary information such as the running time and number of operations of the switch machine corresponding to the image data. For the collected image data, use the Canny operator for edge detection and the pixel integral projection method to calculate the size of the indicating rod notch, obtain the size value of the indicating notch at each time point, and organize it into a time series dataset. Conduct a stationarity test on the time series data. If the data is non-stationary, use the differencing method to transform it into a stationary time series, and record the number of differencing times. S2. Model construction and training: Use the autoregressive integrated moving average model to predict the change trend of the indicating notch size. Divide the preprocessed time series data into a training set and a test set. The training set is used for model parameter estimation, and the test set is used to evaluate the prediction performance of the model. Divide the data in the ratio of 80% as the training set and 20% as the test set, and then use the training set data to train the autoregressive integrated moving average model. Estimate the parameters of the model, including the autoregressive coefficient and the moving average coefficient, through the maximum likelihood estimation method. S3. Model evaluation and optimization: Use the test set data to evaluate the prediction performance of the trained autoregressive integrated moving average model. Use indicators such as the root mean square error and the mean absolute error to measure the error between the predicted value and the true value of the model. According to the results of the evaluation indicators, if the prediction performance of the model does not meet the expectations, optimize the model. S4. Prediction and result output: When the model evaluation meets the requirements, use the trained autoregressive integrated moving average model to predict the indicating notch size within a certain period in the future. The prediction duration is set to one week in the future. Output the prediction results in a visual form to the visualization and interaction module, and display the change trend of the indicating notch size over time in the form of a curve. At the same time, mark the predicted value and the historical true value on the curve to facilitate the maintenance personnel to intuitively understand the change situation of the indicating notch size and discover potential abnormal trends in advance.

7. The railway electric switch machine status monitoring system according to claim 1, wherein: The monthly warning plan for the jamming fault by the intelligent analysis and decision-making module is as follows: S1. Data collection and feature engineering: Collect the historical operation data of the switch machine, including data such as the rotation speed, displacement, vibration, temperature and humidity, and air pressure collected by the multimodal sensor module, as well as the recorded data of the jamming fault of the switch machine. Extract features from the collected data. For the features related to the jamming fault, extract features such as the number of sudden changes in the rotation speed of the power transmission shaft, the duration of abnormal displacement of the operating rod, the frequency of vibration amplitude exceeding the threshold, and the change rate of temperature and humidity. Normalize the extracted features to eliminate the influence of different feature dimensions and analyze each feature on the same scale. S2. Model Training and Selection: The random forest model is used for monthly early warning of jamming faults. The random forest model consists of multiple decision trees. Multiple decision trees are constructed by sampling the training data with replacement, and then the prediction results of multiple decision trees are integrated for the final decision. The historical data is divided into a training set and a test set, with 80% as the training set and 20% as the test set. The training set data is used to train the random forest model, and the parameters of the model, such as the number of decision trees, the maximum depth, and the minimum sample split number, are adjusted to improve the prediction performance of the model. Through the method of cross-validation, the performance of the model under different parameter combinations is evaluated, and the parameter combination that makes the model performance optimal is selected. S3. Model Evaluation and Verification: The trained random forest model is evaluated using the test set data. Metrics such as accuracy, recall rate, and F1 value are used to measure the prediction ability of the model for jamming faults. Then, the prediction results of the model are visually displayed through a confusion matrix, and the prediction accuracy of the model for jamming faults and normal states is analyzed to find the sample characteristics that the model is prone to misjudge, providing a basis for further optimizing the model. S4. Monthly Early Warning Implementation: At the beginning of each month, the operation data of the switch machine in the previous month is collected, and the data is processed according to the method of feature extraction and normalization. The processed data is input into the trained random forest model for prediction. According to the model prediction results, if it is predicted that a certain switch machine has a risk of jamming fault, an early warning message is sent to the operation and maintenance personnel through the visualization and interaction module. The early warning message includes the location, number, predicted fault type, and the possibility of fault occurrence of the switch machine.

8. The railway electric switch machine status monitoring system according to claim 1, wherein: The intelligent analysis and decision-making module optimizes the maintenance resource allocation plan based on historical data as follows: S1. Data Collection and Sorting: Collect the historical fault data of the switch machine, including the fault occurrence time, fault type, maintenance records, including: maintenance personnel, maintenance duration, spare parts used in maintenance, etc., the model of the switch machine, service life, etc. The collected data is cleaned and sorted to remove duplicate data, error data, and incomplete data. S2. Fault Analysis and Classification: Analyze the historical fault data, classify and count it according to the fault type, and calculate indicators such as the frequency of occurrence of various faults, average maintenance duration, and maintenance cost. According to indicators such as fault frequency and maintenance cost, determine the key fault types. The key fault types are usually the fault types with higher occurrence frequency and larger maintenance cost, and these fault types have a greater impact on the consumption of maintenance resources. S3. Resource Demand Prediction: Based on the historical fault data and the operation status data of the switch machine, the autoregressive integrated moving average model is used to predict the number of occurrences of various faults in the future for a period of time. According to the average maintenance duration and maintenance cost of various faults, combined with the predicted number of fault occurrences, calculate the human, material, and financial resources required for maintaining the switch machine in the future for a period of time. S4. Optimized resource allocation: Based on the results of resource demand prediction, combined with the reserve situation and usage status of existing maintenance resources, formulate an optimized maintenance resource allocation plan. For key failure types, prioritize the guarantee of the required maintenance resources, such as increasing the inventory of key components, arranging experienced maintenance personnel to be responsible for relevant maintenance work, etc. Establish a dynamic adjustment mechanism for maintenance resources, and adjust the allocation of maintenance resources in real time according to the actual occurrence of failures and the progress of maintenance work.

Images