Railway track state inspection and evaluation method based on Internet of Things perception

Abstract

The invention provides a railway track state inspection and evaluation method based on Internet of Things sensing, which comprises the following steps: periodically triggering an Internet of Things sensor to collect vibration, image, displacement and other multi-source data, extracting steel rail frequency deviation, rail surface damage level and geometric deformation trend characteristics through frequency spectrum and image recognition and time sequence analysis, and fusing the characteristics into state characteristic vectors; and after comparison and grading with a standard range, a report containing positioning, severity and maintenance suggestions is generated and pushed to a public service terminal. According to the method, the railway track state can be efficiently and accurately evaluated, potential defects are found in time, targeted maintenance suggestions are provided, the railway track inspection efficiency and safety are improved, and the manual inspection cost and the maintenance risk are reduced.

Description

Technical Field

[0001] This invention relates to the field of railway track maintenance technology, and more specifically, to a railway track condition inspection and evaluation method based on Internet of Things (IoT) sensing. Background Technology

[0002] In the railway transportation sector, track safety and stability are key factors ensuring the safe operation of trains. Traditional railway track inspections mainly rely on regular manual checks, which is not only inefficient but also susceptible to human error, making it difficult to detect minor track damage and potential hazards in a timely manner. In recent years, with the development of Internet of Things (IoT) technology, methods for monitoring track conditions using sensors have emerged. These methods deploy sensors beside or on the track to collect data such as track vibration, images, and displacement in real time, enabling preliminary monitoring of track conditions. However, existing technologies have at least the following problems or shortcomings: Firstly, the multi-source and heterogeneous nature of data acquisition makes data fusion and feature extraction difficult, making it hard to comprehensively and accurately reflect track conditions; secondly, existing assessment methods often rely on single data sources or simple threshold judgments, failing to effectively identify track damage under complex operating conditions and lacking dynamic assessment and precise positioning capabilities for track conditions, thus failing to meet the high efficiency and accuracy requirements of railway transportation for track maintenance. Summary of the Invention

[0003] This invention provides a railway track condition inspection and evaluation method based on Internet of Things (IoT) sensing, comprising:

[0004] Multi-source sensing datasets are collected through an IoT sensor network deployed along the track and on the track.

[0005] Spectral analysis is performed on the vibration data in the multi-source sensing dataset to extract the natural frequency offset features of the rail.

[0006] Surface defect identification is performed on the image data in the multi-source sensing dataset, and track surface damage level features are extracted;

[0007] Time-series analysis is performed on the displacement data in the multi-source sensing dataset to extract the trend characteristics of track geometric deformation;

[0008] The rail's natural frequency offset feature, the rail surface damage level feature, and the track's geometric deformation trend feature are fused together to form the track state feature vector;

[0009] The orbital state feature vector is compared with a preset standard orbital feature vector range, and the orbital state evaluation level is determined based on the comparison result.

[0010] Based on the track condition assessment level and the track condition feature vector, a comprehensive assessment report is generated that includes defect location, severity, and maintenance recommendations.

[0011] Further, the step of performing spectral analysis on the vibration data in the multi-source sensing dataset to extract the natural frequency offset features of the rail includes:

[0012] The vibration spectrum is obtained by applying a fast Fourier transform to the vibration data from the trackside accelerometer.

[0013] Within the preset natural frequency range of the healthy steel rail, identify all peak points in the vibration spectrum whose amplitude exceeds the background noise threshold;

[0014] From all the peak points, select at least one peak point with the largest amplitude and determine its frequency value as the target natural frequency peak.

[0015] The reference natural frequency is determined based on historical measurement data of healthy steel rails;

[0016] Calculate the percentage offset of the target natural frequency peak relative to the reference natural frequency;

[0017] The percentage offset is quantified as the rail's inherent frequency offset characteristic.

[0018] Furthermore, the specific formula for calculating the percentage offset of the target natural frequency peak relative to the reference natural frequency is as follows:

[0019]

[0020] in, For the offset percentage, The target's inherent frequency peak value. The reference natural frequency is denoted as .

[0021] Further, the step of identifying surface defects and extracting track surface damage level features from the image data in the multi-source sensing dataset includes:

[0022] The image data is acquired using a line scan camera deployed on the inspection vehicle;

[0023] The image data is input into a pre-trained track surface damage recognition model, which outputs track surface damage category information, damage contour pixel size information, and damage location information in the image.

[0024] Based on the injury category information, a preset injury category basic level mapping table is queried to obtain the basic severity level;

[0025] Based on the pixel size information of the damage outline, a preset damage size correction table is queried to correct the basic severity level, resulting in a corrected severity level.

[0026] The corrected severity level is encoded as a numerical feature within a predetermined range, which serves as the track surface damage level feature.

[0027] Furthermore, the step of performing time-series analysis on the displacement data in the multi-source sensing dataset to extract track geometric deformation trend features includes:

[0028] The displacement data is acquired from the track geometry measurement sensor, and the displacement data includes time-series measurements of track gauge, level, elevation, and alignment parameters.

[0029] The timing measurement values ​​of each parameter within the preset time window are subjected to noise reduction and smoothing processing.

[0030] Based on the smoothed time-series measurements, the average rate of change and measurement standard deviation of each parameter within the time window are calculated.

[0031] For each parameter, a trend change coefficient is calculated based on its average rate of change and the measurement standard deviation;

[0032] The orbital geometric deformation trend characteristics are generated based on the rate of change, the standard deviation, and the trend coefficient.

[0033] Furthermore, the trend change coefficient The calculation formula is:

[0034]

[0035] in, The average rate of change, The standard deviation of the measurement is... and These are the pre-set weighting coefficients for the rate of change and the standard deviation, respectively.

[0036] Furthermore, the process of fusing the rail's natural frequency offset characteristics, the rail surface damage level characteristics, and the track's geometric deformation trend characteristics to form the track state feature vector includes:

[0037] Based on the predefined feature importance, a corresponding fusion weight is assigned to each sub-feature among the rail natural frequency offset feature, the rail surface damage level feature, and the track geometric deformation trend feature;

[0038] Based on the assigned fusion weights, each feature is weighted.

[0039] Normalize all the weighted eigenvalues ​​to make them fall within the same numerical range;

[0040] All the normalized feature values ​​are concatenated in a predetermined order to form a multidimensional orbital state feature vector.

[0041] Further, the step of comparing the orbital state feature vector with a preset standard orbital feature vector range and determining the orbital state evaluation level based on the comparison result includes:

[0042] The value of each feature dimension in the orbital state feature vector is compared step by step with multiple preset level thresholds corresponding to that feature dimension. The multiple level thresholds include attention threshold, abnormality threshold and severity threshold.

[0043] Record the highest severity level achieved by each feature dimension after comparison;

[0044] Based on the highest severity level achieved across all feature dimensions, a preset global judgment rule is applied to determine the final orbital status assessment level, which includes normal, alert, abnormal, and severe.

[0045] Furthermore, the application uses a preset global judgment rule to determine the final orbital state evaluation level. The specific rule is as follows: if any feature dimension exceeds the severity threshold, the evaluation level is severe; if no feature dimension exceeds the severity threshold, but at least one feature dimension exceeds the abnormal threshold, the evaluation level is abnormal; if no feature dimension exceeds the abnormal threshold, but at least one feature dimension exceeds the attention threshold, the evaluation level is attention; if all feature dimensions are within the normal threshold range, the evaluation level is normal.

[0046] Furthermore, the step of associating the acquisition time corresponding to the track state feature vector with the sensor's geographical location information to generate defect location information includes:

[0047] Obtain the sensor geographic location code and data acquisition timestamp carried by each sensing data unit in the multi-source sensing dataset;

[0048] Based on the geographic location code of the sensor, the corresponding line number, mileage marker and geographic coordinates are queried from the railway engineering geographic information system;

[0049] The data collection timestamp, the route number, the mileage marker, and the geographic coordinates are bound together to generate the defect location information.

[0050] The embodiments of the present invention have at least the following beneficial effects:

[0051] 1. By deploying an IoT sensor network to collect multi-source sensing data and combining it with techniques such as spectrum analysis, image recognition, and time series analysis for in-depth data processing, accurate extraction and fusion of multi-dimensional features such as rail inherent frequency offset, rail surface damage level, and track geometric deformation trend have been achieved. This multi-dimensional and refined feature extraction and fusion method can comprehensively and accurately reflect the actual condition of the track, effectively solving the problem of incomplete and inaccurate track condition assessment caused by single data in traditional methods, thus providing a more reliable basis for railway track maintenance decisions.

[0052] 2. Based on a step-by-step comparison of track status feature vectors with preset standard ranges, combined with global judgment rules, dynamic assessment and graded early warning of track status are achieved. This method can quickly determine the assessment level of track status, including normal, warning, abnormal, and severe, according to different feature dimensions and threshold settings. This enables railway maintenance personnel to understand the changing trends of track status in a timely and accurate manner and take corresponding maintenance measures accordingly. This not only improves the efficiency of track maintenance but also reduces the safety risks caused by untimely or inaccurate track status assessment, effectively solving the problems of lag and strong subjectivity in existing track status assessment technologies.

[0053] 3. By associating the acquisition time and sensor geographic location information with the track state feature vector, precise location of track defects was achieved. This method binds the specific location information of the defect with the acquisition time, generating detailed defect location information, including the track number, mileage marker, and geographic coordinates. This enables maintenance personnel to quickly and accurately locate the specific position of track defects, greatly improving the efficiency and focus of maintenance work, avoiding problems such as wasted maintenance resources and extended maintenance time caused by inaccurate defect location, and enhancing the overall efficiency of railway track maintenance. Attached Figure Description

[0054] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and non-limiting manner, wherein:

[0055] Figure 1 This is a flowchart illustrating a railway track condition inspection and evaluation method based on Internet of Things (IoT) sensing, provided as an embodiment of the present invention. Detailed Implementation

[0056] The principles and spirit of the present invention will now be described with reference to several exemplary embodiments. It should be understood that the above embodiments are provided merely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way.

[0057] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating a railway track condition inspection and evaluation method based on Internet of Things (IoT) sensing, provided as an embodiment of the present invention. Figure 1 As shown, a railway track condition inspection and evaluation method based on IoT sensing includes:

[0058] S1. Collect multi-source sensing datasets through an IoT sensor network deployed along the track and on the track.

[0059] S2. Perform spectral analysis on the vibration data in the multi-source sensing dataset to extract the natural frequency offset characteristics of the rail.

[0060] S3. Perform surface defect identification on the image data in the multi-source sensing dataset and extract track surface damage level features;

[0061] S4. Perform time-series analysis on the displacement data in the multi-source sensing dataset to extract the track geometric deformation trend characteristics;

[0062] S5. The rail's natural frequency offset feature, the rail surface damage level feature, and the track geometric deformation trend feature are fused together to form the track state feature vector;

[0063] S6. Compare the orbital state feature vector with a preset standard orbital feature vector range, and determine the orbital state evaluation level based on the comparison result;

[0064] S7. Based on the track condition assessment level and the track condition feature vector, generate a comprehensive assessment report that includes defect location, severity, and maintenance recommendations.

[0065] Multi-source sensing datasets are collected through an IoT sensor network deployed along and on the railway track. This IoT sensor network refers to a collection of various sensor devices distributed along and on the railway track. These sensors can collect various data related to track conditions in real time. For example, trackside accelerometers are used to collect track vibration data, line scan cameras are used to collect track surface image data, and track geometry measurement sensors are used to collect track displacement data.

[0066] The aforementioned sensors are connected via Internet of Things (IoT) technology to form a collaborative network that synchronously collects data according to a preset inspection cycle, ensuring the integrity and timeliness of the data.

[0067] Specifically, trackside accelerometers are primarily used to monitor track vibration. By capturing the vibration signals generated by the rails when trains pass, they provide data support for subsequent analysis of the rails' natural frequency shifts. Linear array cameras, mounted on the inspection vehicle, can capture high-resolution images of the track surface to identify damage such as cracks and wear. Track geometry measurement sensors are responsible for measuring the track's geometric parameters, including gauge, level, elevation, and alignment. Changes in these parameters reflect the track's geometric deformation.

[0068] During data acquisition, each sensor is accompanied by its own geolocation code and data acquisition timestamp. The geolocation code is used to determine the sensor's specific location on the railway line, while the timestamp is used to record the precise time of data acquisition, providing a basis for subsequent defect location and data synchronization. After preliminary screening and processing, the data collected by the sensors forms a multi-source sensing dataset, providing a data foundation for subsequent feature extraction and track condition assessment.

[0069] In some embodiments, performing spectral analysis on the vibration data in the multi-source sensing dataset to extract the rail's natural frequency offset features includes:

[0070] The vibration spectrum is obtained by applying a fast Fourier transform to the vibration data from the trackside accelerometer.

[0071] Within the preset natural frequency range of the healthy steel rail, identify all peak points in the vibration spectrum whose amplitude exceeds the background noise threshold;

[0072] From all the peak points, select at least one peak point with the largest amplitude and determine its frequency value as the target natural frequency peak.

[0073] The reference natural frequency is determined based on historical measurement data of healthy steel rails;

[0074] Calculate the percentage offset of the target natural frequency peak relative to the reference natural frequency;

[0075] The percentage offset is quantified as the rail's inherent frequency offset characteristic.

[0076] The aforementioned spectral analysis is a technique that converts time-domain signals into frequency-domain signals. Its purpose is to extract frequency characteristics related to the health status of the rail from complex vibration signals. Vibration data primarily comes from trackside accelerometers, which can capture the rail's vibration response in real time under the influence of train passage or other external factors. By analyzing the spectral characteristics of this vibration data, the natural frequency variations of the rail can be identified, and the rail's health status can be assessed.

[0077] Specifically, in the frequency domain, by identifying peak points in the vibration spectrum where the amplitude exceeds the background noise threshold, signals related to the natural frequency of the rail can be found. The background noise threshold is a benchmark value used in spectrum analysis to distinguish effective signals from environmental noise, and is usually set according to the noise level in the actual measurement environment.

[0078] The preset natural frequency range for healthy rails is derived from statistical analysis of historical measurement data from a large number of healthy rails, and is used to define the possible range of natural frequencies. Within this range, the peak point with the largest amplitude is selected as the target natural frequency peak, and the frequency value of this peak point will be used for subsequent frequency offset calculations. The reference natural frequency is obtained through statistical analysis of historical measurement data of healthy rails, and it represents the natural frequency of the rail under normal conditions. By calculating the percentage offset between the target natural frequency peak and the reference natural frequency, the degree of change in the rail's natural frequency can be quantified, thus serving as an important characteristic parameter for assessing the health status of the rail.

[0079] The background noise threshold can be dynamically adjusted based on the measurement results of ambient noise to ensure accurate identification of valid vibration signals. For example, in noisy environments, the background noise threshold can be appropriately increased to avoid misinterpreting noise as a valid signal. Setting the natural frequency range of healthy rails requires a large amount of rail measurement data under different environmental conditions. A reasonable frequency range is determined through statistical analysis, typically centered on the average natural frequency of healthy rails, with a certain percentage fluctuation above and below, such as ±5%. When calculating the percentage frequency offset, the parameter in the formula represents the target natural frequency peak value. It is the frequency value corresponding to the maximum peak point selected from the vibration spectrum, the reference natural frequency. This is the natural frequency of a healthy steel rail, obtained through statistical analysis of historical data. (Offset percentage) The relative change in the rail's natural frequency is obtained by calculating the ratio of the difference between the current frequency and the reference frequency to the reference frequency. This calculation process can intuitively reflect changes in the rail's health status and provide crucial information for subsequent track condition assessments.

[0080] In some embodiments, the specific formula for calculating the percentage offset of the target natural frequency peak relative to the reference natural frequency is as follows:

[0081]

[0082] in, For the offset percentage, The target's inherent frequency peak value. The reference natural frequency is denoted as .

[0083] The percentage offset reflects the relative change between the currently measured natural frequency of the rail and the reference frequency of a healthy rail. Specifically, the target natural frequency peak value... This is the maximum frequency peak value in the current rail vibration signal, obtained through spectral analysis. It represents the dominant vibration frequency of the rail under its current condition. Reference natural frequency. This is based on statistical analysis of historical measurement data from a large number of healthy steel rails, reflecting the natural frequency of the rails under normal, undamaged conditions. Offset percentage. It is obtained by calculating the ratio of the difference between the current frequency and the reference frequency to the reference frequency, and is used to measure the degree of change in the natural frequency of the rail. In practical applications, the reference natural frequency can be determined by taking the average of multiple measurements of the same type of rail at different locations and times to ensure its representativeness and accuracy.

[0084] To improve the accuracy of frequency offset percentage calculations, the reference natural frequency can be updated periodically. For example, healthy rails can be remeasured every six months or year to account for the long-term effects of environmental factors such as temperature and humidity on the rail's natural frequency. During the calculation process, vibration signals can be sampled and averaged multiple times to reduce interference from random noise. Furthermore, to further improve the reliability of the assessment, other auxiliary parameters, such as rail temperature and train load conditions, can be incorporated to correct the offset percentage. For instance, in high-temperature environments, the rail's natural frequency may change slightly due to thermal expansion and contraction; in this case, a temperature correction factor can be introduced to adjust the calculated offset percentage. Through these optimization measures, the actual health status of the rails can be more accurately reflected, providing a more reliable basis for railway maintenance.

[0085] In some embodiments, the step of identifying surface defects and extracting track surface damage level features from image data in the multi-source sensing dataset includes:

[0086] The image data is acquired using a line scan camera deployed on the inspection vehicle;

[0087] The image data is input into a pre-trained track surface damage recognition model, which outputs track surface damage category information, damage contour pixel size information, and damage location information in the image.

[0088] Based on the injury category information, a preset injury category basic level mapping table is queried to obtain the basic severity level;

[0089] Based on the pixel size information of the damage outline, a preset damage size correction table is queried to correct the basic severity level, resulting in a corrected severity level.

[0090] The corrected severity level is encoded as a numerical feature within a predetermined range, which serves as the track surface damage level feature.

[0091] Image data refers to images of the track surface acquired through a line scan camera. These images contain detailed information about the track surface, such as cracks, wear, and spalling. The track surface damage recognition model is a model built based on machine learning or deep learning algorithms to automatically identify the type and extent of damage in the images. In this way, complex image information can be transformed into specific damage features.

[0092] A line-scan camera is a high-resolution imaging device, typically mounted on an inspection vehicle, capable of continuously capturing images of the track surface at high speed. This image data is input into a pre-trained track surface damage recognition model. This model outputs damage category information, pixel size information of the damage outline, and location information of the damage in the image. Damage category information refers to the type of damage identified by the model, such as cracks, wear, and spalling; pixel size information of the damage outline refers to the size of the damaged area in the image, usually in pixels; damage location information refers to the specific location of the damage in the image, used for subsequent localization and analysis. Based on the above information, by querying a pre-set damage category basic level mapping table and damage size correction table, the basic severity level and the corrected severity level can be obtained. Finally, the corrected severity level is encoded as a numerical feature, serving as the track surface damage level feature. The tables are formulated based on railway industry standards and practical experience, used to correspond damage types and sizes with specific severity levels.

[0093] A track surface damage recognition model can be constructed using a Convolutional Neural Network (CNN) architecture from deep learning. The model training process involves collecting a large amount of track surface image data, labeling the damage types and locations, and then inputting this data into the CNN model for training. During training, the model automatically learns the mapping relationship between image features and damage types. In practical applications, the image data input to the model needs to be preprocessed, such as grayscale conversion and normalization, to improve the model's recognition accuracy. For the damage category basic level mapping table and damage size correction table, specific level thresholds can be set according to different damage types and size ranges. For example, for crack damage, those less than 5 mm in length are defined as basic level 1, and those greater than 10 mm are defined as basic level 3. In this way, complex image information can be transformed into specific numerical features, providing a more accurate basis for track condition assessment.

[0094] In some embodiments, the step of performing time-series analysis on the displacement data in the multi-source sensing dataset to extract track geometric deformation trend features includes:

[0095] The displacement data is acquired from the track geometry measurement sensor, and the displacement data includes time-series measurements of track gauge, level, elevation, and alignment parameters.

[0096] The timing measurement values ​​of each parameter within the preset time window are subjected to noise reduction and smoothing processing.

[0097] Based on the smoothed time-series measurements, the average rate of change and measurement standard deviation of each parameter within the time window are calculated.

[0098] For each parameter, a trend change coefficient is calculated based on its average rate of change and the measurement standard deviation;

[0099] The orbital geometric deformation trend characteristics are generated based on the rate of change, the standard deviation, and the trend coefficient.

[0100] Displacement data refers to parameters about the track's geometric state collected by track geometry measurement sensors, including gauge, level, elevation, and alignment. The time-series changes of these parameters reflect the track's geometric deformation in time and space. Time-series analysis is a data analysis method used to extract useful information from time-series data and identify trends and patterns in the data. In this way, complex displacement data can be transformed into specific geometric deformation trend characteristics, providing important evidence for track condition assessment.

[0101] Track geometry measurement sensors monitor changes in track geometric parameters in real time, outputting time-series measurements of parameters such as gauge, level, elevation, and alignment. During time-series analysis, the time-series measurements of each parameter are first denoised and smoothed to eliminate random noise and outliers. The smoothed data more clearly reflects the true changing trends of the track geometric parameters. Next, based on the smoothed time-series measurements, the average rate of change and standard deviation of each parameter within a preset time window are calculated. The average rate of change reflects the overall trend of the parameter within the time window, while the standard deviation reflects the degree of data fluctuation. By calculating the trend change coefficient, the above information can be integrated to generate track geometric deformation trend characteristics. These characteristics can intuitively reflect changes in the track's geometric state, providing important basis for subsequent track condition assessment.

[0102] When performing time series analysis, moving averages or exponential smoothing can be used to denoise and smooth displacement data. Moving averages smooth the data by calculating the average value within a time window, while exponential smoothing assigns higher weight to recent data, more flexibly reflecting data trends. When calculating the trend change coefficient, the rate of change weighting coefficient can be set according to actual needs. and standard deviation weighting coefficient For example, if the geometry of the orbit is mainly affected by long-term variations, the [geometric deformation] can be appropriately increased. The value; if you need to pay more attention to short-term fluctuations in the data, you can increase it. The value of .

[0103] In some embodiments, the trend change coefficient The calculation formula is:

[0104]

[0105] in, The average rate of change, The standard deviation of the measurement is... and These are the pre-set weighting coefficients for the rate of change and the standard deviation, respectively.

[0106] Trend change coefficient The calculation is a quantitative method based on the average rate of change and measurement standard deviation of track geometric parameters, using weighting coefficients to comprehensively evaluate the trend of track geometric deformation. The average rate of change refers to the degree of change of track geometric parameters, such as gauge, level, elevation, and orientation, within a preset time window, reflecting the dynamic trend of track geometry over time. The measurement standard deviation indicates the degree of fluctuation of this parameter within the time window, reflecting the stability and consistency of the data. By introducing a weighting coefficient for the rate of change... and standard deviation weighting coefficient The contribution ratio of both to the trend change coefficient can be adjusted according to actual needs, allowing for a more flexible assessment of the severity and urgency of orbital geometric deformation.

[0107] Average rate of change It is calculated from the time-series measurements of smoothed track geometry parameters, representing the average rate of change of the parameters within a preset time window. For example, for the gauge parameter, its average rate of change can be obtained by calculating the ratio of the difference in gauge measurements within the time window to the time interval. Measurement standard deviation This is a statistic that measures the degree of fluctuation of the parameter within a time window, reflecting the dispersion of the measured value. (Rate of Change Weighting Coefficient) and standard deviation weighting coefficient These parameters are set based on track maintenance experience and actual needs, and are used to adjust the importance of the average rate of change and the standard deviation of measurement in the trend change coefficient. For example, if the track maintenance strategy focuses more on the long-term trend of track geometry parameters, then... A higher value; if more attention is paid to the impact of short-term fluctuations on orbital safety, it can be increased. The weighting. In this way, the trend change coefficient... It can comprehensively reflect the dynamic characteristics of track geometry deformation, providing a quantitative basis for track condition assessment.

[0108] In some embodiments, fusing the rail's natural frequency offset characteristics, the rail surface damage level characteristics, and the track's geometric deformation trend characteristics to form the track state feature vector includes:

[0109] Based on the predefined feature importance, a corresponding fusion weight is assigned to each sub-feature among the rail natural frequency offset feature, the rail surface damage level feature, and the track geometric deformation trend feature;

[0110] Based on the assigned fusion weights, each feature is weighted.

[0111] Normalize all the weighted eigenvalues ​​to make them fall within the same numerical range;

[0112] All the normalized feature values ​​are concatenated in a predetermined order to form a multidimensional orbital state feature vector.

[0113] The track condition feature vector is obtained by fusing the rail natural frequency offset feature, rail surface damage level feature, and track geometric deformation trend feature. The rail natural frequency offset feature reflects the change in rail structural integrity, the rail surface damage level feature quantifies the severity of rail surface damage, and the track geometric deformation trend feature describes the dynamic changes in track geometry. These features have different importance in track condition assessment, therefore, it is necessary to assign appropriate fusion weights to each feature according to the actual situation. For example, for high-speed railway tracks, the rail surface damage level feature has a greater impact on train operation safety, so it can be assigned a higher weight. The fusion weights can be set based on historical data, expert experience, or optimized through machine learning algorithms. In the feature fusion process, each feature is first weighted, and then the weighted feature values ​​are normalized to eliminate differences in the dimensions and numerical ranges of different features. The normalized feature values ​​are then concatenated in a predetermined order to form a multi-dimensional track condition feature vector.

[0114] Preferably, during feature fusion, a standardization method can be used to normalize each feature value. For example, the min-max normalization method can be used to scale each feature value to the [0,1] interval to ensure that different features are comparable after fusion. For the allocation of fusion weights, a weight allocation model can be introduced, which can dynamically adjust the weights based on track type, operating environment, and historical maintenance data. For example, by analyzing the contribution ratio of different features to track faults, a machine learning algorithm, such as a support vector machine or neural network, can be used to train the weight allocation model. Input parameters include feature type, track operating speed, train load, etc.

[0115] In some embodiments, comparing the orbital state feature vector with a preset standard orbital feature vector range and determining the orbital state evaluation level based on the comparison result includes:

[0116] The value of each feature dimension in the orbital state feature vector is compared step by step with multiple preset level thresholds corresponding to that feature dimension. The multiple level thresholds include attention threshold, abnormality threshold and severity threshold.

[0117] Record the highest severity level achieved by each feature dimension after comparison;

[0118] Based on the highest severity level achieved across all feature dimensions, a preset global judgment rule is applied to determine the final orbital status assessment level, which includes normal, alert, abnormal, and severe.

[0119] The standard orbital feature vector range refers to a reasonable range pre-defined based on the feature data of a healthy orbit, used as a benchmark for assessing orbital status. The attention threshold, anomaly threshold, and severity threshold involved in the comparison process are grading standards used to distinguish different degrees of severity in orbital status.

[0120] The track condition feature vector is a fusion of multiple dimensions, including rail natural frequency offset, rail surface damage level, and track geometric deformation trend. The value of each feature dimension reflects the track's condition information in different aspects. The standard track feature vector range is derived from historical data of a large number of healthy tracks, setting a reasonable numerical range for each feature dimension. During the comparison process, the value of each feature dimension is sequentially compared with the corresponding attention threshold, anomaly threshold, and severity threshold. For example, for the rail natural frequency offset feature, if its value exceeds the severity threshold, it indicates that the rail's natural frequency change has reached a level that may affect track safety; if it is within the normal threshold range, it indicates that the track condition is good. In this way, the health status of each feature dimension can be independently assessed, and the highest severity level reached for each dimension can be recorded, providing a basis for the final track condition assessment.

[0121] In some embodiments, for high-speed railway tracks, due to their higher requirements for track condition, the anomaly and severity thresholds can be appropriately lowered to more sensitively detect potential problems. Simultaneously, the global judgment rules can be optimized based on the actual operational needs of the track. For example, if a certain feature dimension reaches the severity threshold, even if other dimensions are normal, the track condition can be directly determined as severe to ensure track safety. Furthermore, to improve the accuracy of the assessment, machine learning algorithms can be introduced to optimize the thresholds. By collecting a large amount of track condition data and its corresponding maintenance records, a classification model is trained. The input parameters include the values ​​of each feature dimension and the corresponding thresholds, and the output is the track condition level. The model can automatically adjust the thresholds based on historical data to better adapt to the maintenance needs of different tracks.

[0122] In some embodiments, the application uses a preset global judgment rule to determine the final orbital state evaluation level. The specific rule is as follows: if any feature dimension exceeds the severity threshold, the evaluation level is severe; if no feature dimension exceeds the severity threshold, but at least one feature dimension exceeds the abnormal threshold, the evaluation level is abnormal; if no feature dimension exceeds the abnormal threshold, but at least one feature dimension exceeds the attention threshold, the evaluation level is attention; if all feature dimensions are within the normal threshold range, the evaluation level is normal.

[0123] The global judgment rule is based on the logical basis of classifying the severity levels of each feature dimension in the orbital state feature vector. During the evaluation process, the value of each feature dimension is first compared with the corresponding attention threshold, anomaly threshold, and severity threshold, recording the highest severity level achieved for each feature dimension. For example, if the value of a feature dimension exceeds the severity threshold, the highest severity level for that dimension is severe; if it does not exceed the severity threshold but exceeds the anomaly threshold, it is abnormal; and so on. Finally, based on the highest severity levels of all feature dimensions, the overall orbital state evaluation level is determined according to preset rules. For example, if any feature dimension reaches the severity level, the overall evaluation level is severe; if there is no severity level but an anomaly level, the overall evaluation level is abnormal; if there is no anomaly level but an attention level, the overall evaluation level is attention; if all feature dimensions are within the normal range, the overall evaluation level is normal.

[0124] In practical applications, for high-speed railway tracks, due to their higher requirements for safety and stability, the rules can be appropriately adjusted so that when a key feature dimension, such as the rail surface damage level, reaches an abnormal level, the overall assessment level is directly determined to be severe, ensuring timely maintenance measures are taken. Furthermore, to further improve the accuracy and adaptability of the assessment, a data-driven model can be introduced to optimize the global judgment rules. For example, by collecting a large amount of track status data and its corresponding maintenance records, a decision tree or random forest model can be trained. The input parameters include the highest severity level of each feature dimension, and the output is the overall track status assessment level. The model can automatically learn the optimal judgment rules under different combinations of feature dimensions based on historical data, thus better adapting to the actual maintenance needs of different tracks.

[0125] In some embodiments, the step of associating the acquisition time corresponding to the track state feature vector with the sensor geographical location information to generate defect location information includes:

[0126] Obtain the sensor geographic location code and data acquisition timestamp carried by each sensing data unit in the multi-source sensing dataset;

[0127] Based on the geographic location code of the sensor, the corresponding line number, mileage marker and geographic coordinates are queried from the railway engineering geographic information system;

[0128] The data collection timestamp, the route number, the mileage marker, and the geographic coordinates are bound together to generate the defect location information.

[0129] Sensor geolocation coding is a unique location identifier pre-assigned to each sensor by the railway engineering geographic information system (GIS), containing information such as line number, mileage marker, and geographic coordinates. The line number distinguishes different railway lines, the mileage marker identifies the sensor's specific location on that line, and the geographic coordinates provide the sensor's precise latitude and longitude. The data acquisition timestamp is the time information recorded when the sensor collects data, used to identify the specific moment of data acquisition. When generating defect location information, the sensor geolocation code and data acquisition timestamp carried by each sensing data unit are first obtained from the multi-source sensing dataset. Then, based on this information, the corresponding detailed location information, including line number, mileage marker, and geographic coordinates, is queried from the railway engineering GIS system. By binding this information with the acquisition timestamp, defect location information containing both the defect location and time can be generated, providing clear guidance for track maintenance.

[0130] In practical applications, the accuracy of sensor geolocation coding can be improved using high-precision positioning equipment, such as differential GPS, to ensure that the error in defect location is within an acceptable range. Simultaneously, the accuracy of the data acquisition timestamp should also be as high as possible to ensure accurate reflection of the time of defect occurrence. When generating defect location information, a data verification mechanism can be introduced to check the consistency between sensor geolocation coding and data acquisition timestamps, avoiding location deviations caused by data errors. Furthermore, to facilitate rapid response by maintenance personnel, defect location information can be integrated with the real-time alarm function of the railway maintenance management system. Once a defect is detected, detailed information including the defect location, time, and severity is immediately pushed to maintenance personnel, enabling rapid location and efficient maintenance of track defects.

[0131] The above embodiments of the present invention have the following beneficial effects:

[0132] 1. By deploying an IoT sensor network to collect multi-source sensing data and combining it with techniques such as spectrum analysis, image recognition, and time series analysis for in-depth data processing, accurate extraction and fusion of multi-dimensional features such as rail inherent frequency offset, rail surface damage level, and track geometric deformation trend have been achieved. This multi-dimensional and refined feature extraction and fusion method can comprehensively and accurately reflect the actual condition of the track, effectively solving the problem of incomplete and inaccurate track condition assessment caused by single data in traditional methods, thus providing a more reliable basis for railway track maintenance decisions.

[0133] 2. Based on a step-by-step comparison of track status feature vectors with preset standard ranges, combined with global judgment rules, dynamic assessment and graded early warning of track status are achieved. This method can quickly determine the assessment level of track status, including normal, warning, abnormal, and severe, according to different feature dimensions and threshold settings. This enables railway maintenance personnel to understand the changing trends of track status in a timely and accurate manner and take corresponding maintenance measures accordingly. This not only improves the efficiency of track maintenance but also reduces the safety risks caused by untimely or inaccurate track status assessment, effectively solving the problems of lag and strong subjectivity in existing track status assessment technologies.

[0134] 3. By associating the acquisition time and sensor geographic location information with the track state feature vector, precise location of track defects was achieved. This method binds the specific location information of the defect with the acquisition time, generating detailed defect location information, including the track number, mileage marker, and geographic coordinates. This enables maintenance personnel to quickly and accurately locate the specific position of track defects, greatly improving the efficiency and focus of maintenance work, avoiding problems such as wasted maintenance resources and extended maintenance time caused by inaccurate defect location, and enhancing the overall efficiency of railway track maintenance.

[0135] Furthermore, the storage medium in the embodiments of this application stores program instructions capable of implementing all the above methods. These program instructions can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0136] The above description is merely an explanation of some preferred embodiments of the present invention and the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments of the present invention is not limited to the specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A railway track state inspection evaluation method based on Internet of Things perception, characterized in that, The method comprises the following steps: Collecting a multi-source perception data set through an Internet of Things sensor network deployed on the track side and the track; Performing spectral analysis on vibration data in the multi-source perception data set to extract a rail inherent frequency offset feature; Performing surface defect recognition on image data in the multi-source perception data set to extract a rail surface damage level feature; Performing time series analysis on displacement data in the multi-source perception data set to extract a track geometric deformation trend feature; Fusing the rail inherent frequency offset feature, the rail surface damage level feature, and the track geometric deformation trend feature to form a track state feature vector; Comparing the track state feature vector with a preset standard track feature vector range to determine a track state evaluation level according to a comparison result; Generating a comprehensive evaluation report including defect positioning, severity, and maintenance recommendations based on the track state evaluation level and the track state feature vector.

2. The method of claim 1, wherein, The method of performing spectral analysis on vibration data in the multi-source perception data set to extract a rail inherent frequency offset feature comprises the following steps: Applying fast Fourier transform to the vibration data from the track side acceleration sensor to obtain a vibration spectrum; Identifying all peak points in the vibration spectrum whose amplitudes exceed a background noise threshold in a preset healthy rail inherent frequency interval; From all the peak points, filtering out at least one peak point with the largest amplitude, and determining the frequency value of the peak point as a target inherent frequency peak value; Determining a reference inherent frequency based on historical measurement data of a healthy rail; Calculating a percentage offset of the target inherent frequency peak value relative to the reference inherent frequency; Quantifying the percentage offset as the rail inherent frequency offset feature.

3. The method of claim 2, wherein, The specific calculation formula of the percentage offset of the target inherent frequency peak value relative to the reference inherent frequency is: wherein, is the offset percentage, is the target natural frequency peak, is the reference natural frequency.

4. The method of claim 1, wherein, The method of performing surface defect recognition on image data in the multi-source perception data set to extract a rail surface damage level feature comprises the following steps: Collecting the image data by using a linear array camera deployed on a track inspection vehicle; Inputting the image data into a pre-trained rail surface damage recognition model, which outputs category information of rail surface damage, pixel size information of damage contour, and position information of damage in the image; According to the category information of the damage, querying a preset basic damage category level mapping table to obtain a basic severity level; According to the pixel size information of the damage contour, querying a preset damage size correction table to correct the basic severity level to obtain a corrected severity level; Encoding the corrected severity level into a numerical feature in a predetermined numerical range as the rail surface damage level feature.

5. The method of claim 1, wherein, The method of performing time series analysis on displacement data in the multi-source perception data set to extract a track geometric deformation trend feature comprises the following steps: Obtaining the displacement data from a track geometric measurement sensor, which includes time series measurement values of track gauge, level, height, and track direction parameters; Performing denoising and smoothing processing on the time series measurement values of each parameter in a preset time window; Based on the smoothed time series measurement values, calculating an average change rate and a measurement standard deviation of each parameter in the time window; For each parameter, a trend change coefficient is calculated based on the average change rate and the measurement standard deviation thereof; Based on the change rate, the standard deviation and the trend coefficient, the track geometry deformation trend feature is generated.

6. The method of claim 5, wherein, The trend change coefficient The calculation formula is: wherein, is the average rate of change, is the measurement standard deviation, and are preset rate of change weight coefficient and standard deviation weight coefficient, respectively.

7. The method according to claim 1 or 5, characterized in that, The track state feature vector is composed of the rail inherent frequency offset feature, the rail surface damage level feature and the track geometry deformation trend feature by fusing them, which comprises: According to the pre-defined feature importance, a corresponding fusion weight is assigned to each sub-feature in the rail inherent frequency offset feature, the rail surface damage level feature and the track geometry deformation trend feature; Based on the assigned fusion weight, each feature is weighted processed; All feature values after weighted processing are normalized to be in the same numerical interval; All normalized feature values are spliced in a predetermined order to form a multi-dimensional track state feature vector.

8. The method of claim 1, wherein, The track state feature vector is compared with the pre-set standard track feature vector range, and the track state evaluation level is determined according to the comparison result, which comprises: The numerical value of each feature dimension in the track state feature vector is compared with a plurality of level thresholds corresponding to the feature dimension, which includes attention threshold, abnormal threshold and serious threshold; The highest severity level reached by each feature dimension after comparison is recorded; Based on the highest severity level reached by all feature dimensions, a pre-set global decision rule is applied to determine the final track state evaluation level, which includes normal, attention, abnormal and serious.

9. The method of claim 8, wherein, The specific rule of applying the pre-set global decision rule to determine the final track state evaluation level is: if any feature dimension exceeds the serious threshold, the evaluation level is serious; if no feature dimension exceeds the serious threshold, but at least one feature dimension exceeds the abnormal threshold, the evaluation level is abnormal; if no feature dimension exceeds the abnormal threshold, but at least one feature dimension exceeds the attention threshold, the evaluation level is attention; if all feature dimensions are within the normal threshold range, the evaluation level is normal.

10. The method of claim 1, wherein, The acquisition time corresponding to the track state feature vector and the sensor geographic location information are associated to generate defect positioning information, which comprises: The sensor geographic location code and data acquisition timestamp carried by each sensing data unit in the multi-source perception data set are obtained; Based on the sensor geographic location code, the corresponding line number, mileage pile number and geographic coordinates are queried from the railway maintenance geographic information system; The data acquisition timestamp, the line number, the mileage pile number and the geographic coordinates are bound to generate the defect positioning information.

Images