A sound barrier risk identification prediction method, system and computer

Abstract

The present application relates to the technical field of sound barrier, provide a kind of sound barrier risk identification prediction method, system and computer, comprising: sound barrier information, environmental condition information and train operation condition information are combined into condition data coding, and the component category of several components is extracted;Collect appearance image, carry out component positioning, obtain area data coding and several appearance clues score;Obtain several sensitive area weight values, form sensitive area weight graph;Several acceleration sensors are set, and several vibration acceleration signals are collected, and several vibration acceleration signals are event sliced to form event signal;obtain damage index based on event signal, to obtain evaluation result according to damage index.By using the above method, the problem that monitoring point is fixed in the prior art, lacks the comprehensive analysis of the dynamic correlation between working condition, sensitive area and risk is solved.

Description

Technical Field

[0001] This invention relates to the field of sound barrier technology, and in particular to a sound barrier risk identification and prediction method, system and computer. Background Technology

[0002] Noise barriers are widely used for noise reduction along high-speed railways, conventional railways, and urban rail transit lines. A noise barrier is composed of various components, typically including columns, foundations, screen units, and connectors. Connectors include fasteners, clamps, covers, and fall arrestors. Noise barriers have long linear extensions, numerous connection nodes, and are exposed to complex environments.

[0003] During long-term service, sound barriers are subjected to repeated effects of unsteady aerodynamic pressure pulsations induced by high-speed train passage, structural vibrations, and alternating fatigue loads. These effects are compounded by wind, rain, temperature cycles, corrosion, material aging, and construction / assembly deviations, making them prone to defects such as weakened preload on connectors, loose bolts, missing connectors, screen swaying, and abnormal noises. These defects are often insidious and progressive, lacking obvious visual characteristics in their early stages. If not detected in time, they may further develop into safety risks such as component detachment and encroachment.

[0004] Current maintenance methods mainly rely on manual inspections, sampling and re-tightening, and post-incident repairs. These methods suffer from low coverage efficiency, reliance on experience, difficulty in quantifying damage levels, and inability to account for the impact of different sound barrier types and operating conditions. While some monitoring methods can obtain structural response or appearance information, they often have fixed monitoring points and lack the ability to locate sensitive areas requiring focused monitoring. The location of sensitive areas will change under the influence of operating conditions such as vehicle speed, frequency, oncoming traffic, wind, rain, and temperature. Current technologies lack comprehensive analysis of the dynamic correlation between operating conditions, sensitive areas, and risks, resulting in insufficient early warning and prediction capabilities. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method, system, and computer for identifying and predicting risks associated with sound barriers. This invention seeks to solve the technical problem of insufficient early warning and prediction capabilities in existing technologies due to fixed monitoring points and a lack of comprehensive analysis of the dynamic correlation between operating conditions, sensitive areas, and risks.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: A method for identifying and predicting the risks of sound barriers includes the following steps: Acquire the sound barrier information, environmental condition information, and train operation condition information of the sound barrier to be identified; combine the sound barrier information, environmental condition information, and train operation condition information into an operation condition data code; and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. The appearance image of the sound barrier to be identified is acquired, and the components in the appearance image are located based on several component categories to obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. Based on several component categories, several working condition data codes, and several appearance clue scores, several sensitive area weight values ​​are obtained, and the several component categories, several sensitive area weight values, the working condition data codes, and the area data codes are combined to form a sensitive area weight map. Based on the sensitive area weight map, several acceleration sensors are set on the sound barrier to be identified, and several vibration acceleration signals are collected. Event slices are performed on the several vibration acceleration signals to form event signals. A damage index is obtained based on the event signal, and an evaluation result is obtained based on the damage index.

[0007] Furthermore, the sound barrier information includes the sound barrier type, column spacing, sound barrier foundation form, screen unit type, connector data, and mileage location; the environmental condition information includes temperature, wind speed, and rainfall status; and the train operation condition information includes train speed range, meeting status, passage time period, and train frequency.

[0008] Furthermore, the step of locating components in the appearance image based on several component categories to obtain the region data encoding and several appearance cue scores corresponding to the sound barrier to be identified includes: The appearance image is preprocessed to obtain the image to be identified; The image to be identified is identified based on several component categories to obtain several identification results corresponding to several components, and the several identification results are converted into region data encoding, wherein the identification results include bounding boxes or masks; Based on several identification results, the appearance of several components is judged to obtain several appearance clues, which are normal appearance clues, missing connector clues, abnormal gap clues, or structural deformation clues. Several area proportions are obtained based on several of the aforementioned recognition results; Based on the area ratio and all the appearance cues, an appearance cue score corresponding to the component is generated.

[0009] Furthermore, the step of obtaining several sensitive area weight values ​​based on several component categories, several working condition data codes, and several appearance cue scores includes: Several basic weights are set according to several component categories; Based on several appearance cue scores, several basic weights are corrected to obtain several cue correction weights; Obtain historical maintenance data, and adjust the correction weights of several clues based on the historical maintenance data to obtain several historical correction weights; Based on the operating condition data encoding, several historical correction weights are adaptively scaled according to operating conditions to obtain several sensitive area weight values.

[0010] Furthermore, the step of performing event slicing on several of the vibration acceleration signals to form an event signal includes: Obtain the slicing method identifier, and determine the slicing method based on the slicing method identifier; Several event moments are obtained using the slicing method described above; If the slicing method is vehicle passage information slicing, then vehicle passage information is obtained, and several event times are determined based on the vehicle passage information; If the slicing method is vibration threshold slicing, then several event moments are determined based on the short-time energy, peak value, or energy mutation rate of the vibration acceleration signal; The vibration acceleration signal is sliced ​​into several event segments based on several event moments. Extract the leading segment of the event fragment to obtain several baseline segments, and obtain the background noise statistical signal based on the several baseline segments; An event signal is established based on all the event fragments described.

[0011] Furthermore, the step of establishing an event signal based on all the event fragments includes: Quality control is performed on all the event fragments to obtain several event fragments to be aligned; Several anchor points are selected from the several event times, and all event segments are time-aligned based on the several anchor points to obtain the event signal.

[0012] Furthermore, the step of obtaining a damage index based on the event signal, and then obtaining an evaluation result based on the damage index, includes: Acquire several baseline signals and several baseline damage index threshold sets under several working conditions; Extract the overall constraint change features and impact nonlinear features of the event signal. The overall constraint change features include one or more of the following: main frequency drift, resonance peak amplitude change, frequency band energy ratio, energy centroid change, spectral entropy change, and inter-measurement point transfer rate amplitude change. The impact nonlinear features include one or more of the following: peak value, kurtosis, impact factor, short-time energy, pulse count, and energy mutation rate. Based on the background noise statistical signal or the baseline signal, the overall constraint change characteristics and the impact nonlinear characteristics are standardized, and the damage index is estimated by feature weighted fusion or learning model. Based on the operating condition data encoding, a set of baseline damage index thresholds to be compared is selected from several sets of baseline damage index thresholds. Based on the set of baseline damage index thresholds to be compared and the damage index, the damage level is determined. Based on the overall constraint variation characteristics and the impact nonlinear characteristics, the damage type is determined; Based on the output of all event fragments and the total number of event fragments, a consistency index is obtained, and a damage confidence level is obtained based on the consistency index. The damage level, the damage category, and the damage confidence level constitute the evaluation result.

[0013] Furthermore, after the step of obtaining a damage index based on the event signal to obtain an evaluation result based on the damage index, the method further includes: Based on the assessment results, a maintenance strategy is output, which includes re-tightening, anti-loosening replacement, local reinforcement, component replacement, or emergency treatment. Based on the damage index, the baseline damage index threshold set, the consistency index, and the damage confidence, it is determined whether to perform encrypted verification. If encrypted verification is performed, several risk areas are selected from the sensitive area weight map according to the sorting of the sensitive area weight values, and several accelerometers are deployed in the several risk areas to obtain updated evaluation results.

[0014] A sound barrier risk identification and prediction system, employing the sound barrier risk identification and prediction method described in the above technical solution, the system comprising: The operating condition module is used to acquire the sound barrier information, environmental condition information and train operation condition information of the sound barrier to be identified, combine the sound barrier information, environmental condition information and train operation condition information into operating condition data code, and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. The positioning module is used to acquire the appearance image of the sound barrier to be identified, locate the components in the appearance image based on several component categories, and obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. The weighting module is used to obtain several sensitive area weight values ​​based on several component categories, several working condition data codes, and several appearance clue scores, and to form a sensitive area weight map by combining several component categories, several sensitive area weight values, the working condition data codes, and the area data codes. The signal module is used to set up a number of acceleration sensors on the sound barrier to be identified according to the sensitive area weight map, and to collect a number of vibration acceleration signals, and to perform event slicing on the number of vibration acceleration signals to form event signals; An evaluation module is used to obtain a damage index based on the event signal, so as to obtain an evaluation result based on the damage index.

[0015] A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the sound barrier risk identification and prediction method as described in the above technical solution.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: By collecting information on the sound barrier, environmental conditions, and train operation, a file recording specific operating conditions is established and combined into operating condition data codes. Through image acquisition and recognition, each sound barrier component on the sound barrier to be identified is distinguished and located. Based on the identification of the sound barrier components, the regional data codes of the sound barrier to be identified are established, and appearance clues are judged and scored for local component areas. Sensitive area weight values ​​for each component area on the sound barrier to be identified are obtained in a targeted manner. By comprehensively considering actual operating conditions and the sound barrier's own information, operating condition data codes and regional data codes can be established specifically and dynamically for different sections and types of sound barriers on railways or highways to obtain the most suitable sensitive area weight values, enhancing attention to key parts of the sound barrier. Establishing the sensitive area weight map is beneficial for targeted... By deploying different monitoring points for sound barriers of varying road sections, structures, and characteristics, optimal monitoring results can be achieved. This transforms the monitoring and verification of sound barriers from passive coverage at fixed points to proactive focusing on sensitive and risk-prioritized locations, significantly improving inspection coverage efficiency and resource utilization. Combining vehicle traffic information and collected signal characteristics under actual working conditions, event triggering and slice alignment are performed on the vibration acceleration signals. The overall monitoring results of the sound barriers to be identified are integrated into event signals, and different types of features are extracted. Damage indices are then judged based on specific working conditions using a baseline damage index threshold, reducing evaluation errors caused by changes in working conditions. Furthermore, features are standardized based on background signals from the sound barriers during operation without triggering special events, allowing for further verification of the evaluation results. All of these methods facilitate dynamic adjustment of monitoring points on the sound barriers, leading to accurate and reliable prediction results. Attached Figure Description

[0017] Figure 1 This is a flowchart of the sound barrier risk identification and prediction method in the first embodiment of the present invention; Figure 2 This is a structural block diagram of the sound barrier risk identification and prediction system in the second embodiment of the present invention; The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0018] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0019] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] Please see Figure 1 The sound barrier risk identification and prediction method in the first embodiment of the present invention includes the following steps: Step S10: Obtain the sound barrier information, environmental condition information and train operation condition information of the sound barrier to be identified; combine the sound barrier information, environmental condition information and train operation condition information into operation condition data code; and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. Preferably, the component categories include at least one or more of column bases, connection nodes, ends, columns, and screen units. The vibration response of the sound barrier is non-stationary under conditions such as trains passing at different speeds, passing oncoming trains, and changes in wind, rain, and temperature. The amplitude, frequency band energy, and impact characteristics of the same structure may vary significantly under different conditions. The operating condition data encoding is used to characterize the combined structural-environmental-operating conditions, combining multi-dimensional information into a file represented by field codes. By establishing the operating condition file and combining the operating condition data encoding, it is beneficial to perform hierarchical and graded processing in subsequent steps of processing thresholds, processing features, and processing prediction models, reducing the risk of misjudgment caused by changes in operating conditions, and enabling targeted identification and prediction of sound barrier structural loosening and sound barrier risks under different structures and different climatic environments.

[0022] In step S10, the sound barrier information includes the sound barrier type, column spacing, sound barrier foundation form, screen unit type, connector data and mileage location; the environmental condition information includes temperature, wind speed and rainfall status; and the train operation condition information includes train speed range, meeting status, passage time period and train frequency.

[0023] Preferably, the connector data includes connector type and quantity. Fields are extracted from the sound barrier information, environmental condition information, and train operation condition information. Continuous variables in the environmental condition information are intervalized or binned, and each field is then encoded. Finally, the codes are combined in a preset order to form the operation condition data code. The speed range in the train operation condition information is the result of intervalizing the speed. Specifically, for example, the sound barrier type can be encoded as SB1, the column spacing as PS2, and the sound barrier foundation form as F1. The screen unit type is coded as P3, the connector data is coded as C8, the mileage position is coded as K125600, the vehicle speed range is coded as V3, the oncoming traffic status is coded as M2, the temperature is coded as T1 after being divided into ranges, the wind speed is coded as W2 after being divided into ranges, and the rainfall status is coded as R0. The final combined operating condition data is coded as “SB1-PS2-F1-P3-C8-K125600-V3-M2-T1-W2-R0”, and the field of the operating condition data encoding is named “Condition_ID”.

[0024] Step S20: Acquire the appearance image of the sound barrier to be identified, locate the components in the appearance image based on several component categories, and obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. Preferably, the data collection method can be manual inspection and photography, vehicle-mounted inspection and photography, or drone inspection and photography. Loose and missing sound barrier connectors are common at column bases, connection nodes, and ends. By collecting and identifying the appearance images, intuitive clues such as missing parts, abnormal gaps, and misalignment can be provided. Targeted component positioning and appearance clue assessment are beneficial for providing a priori basis for establishing the weight of sensitive areas and can also be used for subsequent review of the priority of sensitive areas.

[0025] Step S20 includes: S210: Preprocess the appearance image to obtain the image to be identified; Preferably, the preprocessing includes distortion correction, deblurring, brightness normalization, and sharpness screening.

[0026] S220: The image to be identified is identified based on several component categories to obtain several identification results corresponding to several components, and the several identification results are converted into region data encoding, wherein the identification results include bounding boxes or masks; Preferably, several components are initially located using a target detection method to obtain several bounding boxes. Then, semantic segmentation and recognition are performed on the bounding box regions to obtain several masks. Geometric correction is performed by combining the column spacing, the geometric relationship between adjacent components, or the design template to generate the region data code. The region data code records the location, category, and other information of multiple component regions on the sound barrier to be identified.

[0027] S230: Based on several identification results, determine the appearance of several components to obtain several appearance clues, the appearance clues being normal appearance clues, missing connector clues, abnormal gap clues, or structural deformation clues. Preferably, judging the appearance of several components based on several recognition results can be achieved by at least one of target detection, region segmentation, edge detection, template matching, contour fitting, and geometric deviation calculation. The missing connector clues include missing bolts, missing nuts, missing pressure plates, and missing connecting plates. The abnormal gap clues include increased connection gaps, discontinuous contact boundaries, and abnormal shadow gaps. The structural deformation clues include misalignment, tilting, local bulging, and end offset.

[0028] S240: Based on several of the aforementioned recognition results, several area proportions are obtained; Preferably, the area ratio is the ratio of the area of ​​the abnormal region to the area of ​​the corresponding component region, which can be calculated using the mask area and the bounding box area of ​​the abnormal region.

[0029] S250: Based on the area ratio and all the appearance cues, generate an appearance cue score corresponding to the component.

[0030] Preferably, based on all the appearance clues, the occurrence frequency of each type of clue is counted according to the missing, gap, and deformation categories. The appearance clues of the component can be evaluated based on the occurrence frequency and area ratio of the appearance clues. In addition, the appearance clue score can be generated by combining morphological features and deviation values ​​from the structural reference. Morphological features may include length, width, aspect ratio, tilt angle, gap width, misalignment height, offset distance, contour irregularity, and symmetry deviation. The structural reference can be derived from design drawings, standard templates, healthy sample images, images of adjacent similar components, or historical archived images. The deviation value is calculated by the geometric difference, contour difference, or position difference between the current detection result and the structural reference.

[0031] Step S30: Based on several component categories, several working condition data codes, and several appearance clue scores, several sensitive area weight values ​​are obtained, and the several component categories, several sensitive area weight values, the working condition data codes, and the area data codes are combined to form a sensitive area weight map; Preferably, the sensitive area weight map is a data set composed of multiple area entries. Each entry includes at least the area data code, the component category, the sensitive area weight value, and the mileage location. After the sensitive area weight map is formed, a data set of sensitive area entries that require priority can be output according to the sorting of the sensitive area weight values. It is understandable that, since the sound barrier structure extends along the route and has a large number of components that are widely distributed, it is difficult to balance coverage and cost by using fixed-point long-term monitoring. The sensitive area weight map helps to prioritize the allocation of monitoring and verification resources to high-risk areas that require key attention, thereby improving monitoring coverage efficiency.

[0032] Step S30 includes: S310: Set several basic weights according to several of the aforementioned component categories; Preferably, a component category library is preset, and the basic weight is set for each component. During actual calculation, the weight of the corresponding component category is retrieved from the library.

[0033] S320: Based on the scores of the appearance cues, the basic weights are corrected to obtain the cues-corrected weights. Preferably, the appearance clue score can be incorporated into a correction coefficient, which is used to correct the base weight.

[0034] S330: Obtain historical maintenance data, and correct several clue correction weights based on the historical maintenance data to obtain several historical correction weights; Preferably, the historical maintenance data includes data such as historical anomaly frequency and post-repair recurrence rate. The historical anomaly frequency and post-repair recurrence rate are normalized, and a correction coefficient is constructed based on the two normalized values ​​to correct the clue correction weight.

[0035] S340: Based on the operating condition data encoding, several historical correction weights are subjected to operating condition adaptive scaling to obtain several sensitive area weight values.

[0036] Preferably, the scaling factor corresponding to the current working condition and the current component category is obtained based on the working condition data encoding, and the historical correction weight value is scaled by the scaling factor to finally obtain the sensitive area weight value.

[0037] Step S40: Based on the sensitive area weight map, set up a number of acceleration sensors on the sound barrier to be identified, and collect a number of vibration acceleration signals. Perform event slicing on the number of vibration acceleration signals to form event signals. Preferably, when the sound barrier has loose or missing connectors, it is prone to vibration characteristics such as increased impact, abrupt constraint changes, and altered transmission paths when a train passes. By deploying accelerometers in sensitive areas to collect the vibration acceleration signals, a data foundation is provided for subsequent processing of the vibration signals. Furthermore, the deployment of the accelerometers adopts a combination of long-term deployment at sentinel points and subsequent encrypted deployment and verification after triggering. The accelerometers are bound to the data encoding of the area, and the installation position, installation direction, and fixing method are recorded. The sampling frequency is greater than or equal to 1000Hz, which can be set according to the sound barrier structure, target frequency band, and event triggering strategy. The event slice is a long-term continuous sampling... The vibration acceleration signal is segmented according to specific events to obtain several signal segments of single events, such as vehicle passing events and threshold triggering events. Understandably, in long-term monitoring, vibration signals usually contain background noise, environmental disturbances, and the superposition of multiple events. Segmenting and combining multiple slices from continuous vibration signals is beneficial for feature extraction, consistency analysis, and risk and damage assessment. By establishing the sensitive area weight map, it is beneficial to deploy different monitoring points for sound barriers of different road sections, different structures, and different characteristics to obtain the best monitoring effect. This transforms the monitoring and verification of sound barriers from passive coverage of fixed points to active focusing on sensitive and risk-prioritized areas.

[0038] Step S40 includes: S410: Obtain the slicing method identifier, and determine the slicing method based on the slicing method identifier; S420: Obtain several event moments using the slicing method; S430: If the slicing method is vehicle passage information slicing, then obtain the vehicle passage information and determine several event times based on the vehicle passage information; Preferably, the start and end times of the train passing the sensor are determined based on the train passing information, that is, a number of the event times are determined.

[0039] S440: If the slicing method is vibration threshold slicing, then a number of event moments are determined based on the short-time energy, peak value, or energy mutation rate of the vibration acceleration signal; Preferably, a threshold is set, and an event window is determined based on the short-term energy, peak value, or energy mutation rate exceeding the threshold, thereby determining several event moments.

[0040] S450: The vibration acceleration signal is sliced ​​according to several event moments to obtain several event segments; By determining the time related to the start and end of the event from several event moments, several event windows are determined, and the event windows are sliced ​​to obtain the event segments.

[0041] S460: Extract the leading segment of the event fragment to obtain several baseline segments, and obtain a background noise statistical signal based on the several baseline segments; The signal preceding the event window is used as the leading segment corresponding to the event segment.

[0042] S470: Establish an event signal based on all the event fragments.

[0043] Preferably, the event signal is subjected to filtering and denoising, frequency selection enhancement, and envelope processing. The filtering and denoising includes at least one of bandpass filtering, wavelet denoising, and outlier removal. The frequency selection enhancement includes at least one of target frequency band energy enhancement and spectral kurtosis enhancement. The envelope processing includes at least one of Hilbert envelope demodulation and short-time energy envelope, which is beneficial to improving the quality of the event signal.

[0044] The S470 also includes: S4710: Perform quality control on all the event fragments to obtain several event fragments to be aligned; Preferably, the quality control includes saturation detection, packet loss detection, abnormal drift detection, and signal-to-noise ratio screening, where event segments that do not meet the quality conditions are removed or marked as low-confidence inputs.

[0045] S4720: Select several anchor points from the several event times, and time-align all the event segments based on the several anchor points to obtain the event signal.

[0046] Preferably, the trigger time or energy peak time of the event can be used as the anchor time. Time alignment can align multiple event segments from different vibration acceleration signals obtained at different measuring points according to the same event, or it can be based on multiple events at the same measuring point, with timing correction performed under a unified reference time. Specifically, it can be divided into coarse alignment and fine alignment. First, coarse alignment is performed based on the vehicle passage trigger time or threshold trigger time, and then fine alignment is performed based on the energy peak time, the main impact peak time, or the position of maximum cross-correlation. Understandably, aligning signal segments is beneficial to reducing the impact of trigger delay, propagation time difference, and sampling start deviation on subsequent feature extraction, and improving the comparability between different events and different measuring points.

[0047] Step S50: Obtain the damage index based on the event signal, and obtain the evaluation result based on the damage index.

[0048] Step S50 includes: S510: Acquire several baseline signals and several baseline damage index threshold sets under several working conditions; Preferably, the baseline represents a health event, and the health event is specifically selected from events collected after the system is put into operation or after maintenance and repair. Threshold sets are established based on health events under different operating conditions.

[0049] S520: Extract the overall constraint change features and impact nonlinear features of the event signal. The overall constraint change features include one or more of the following: main frequency drift, resonance peak amplitude change, frequency band energy ratio, energy centroid change, spectral entropy change, and inter-measurement point transfer rate amplitude change. The impact nonlinear features include one or more of the following: peak value, kurtosis, impact factor, short-time energy, pulse count, and energy mutation rate. S530: Based on the background noise statistical signal or the baseline signal, the overall constraint change characteristics and the impact nonlinear characteristics are standardized, and the damage index is estimated by feature weighted fusion or learning model. Preferably, the standardization process first transforms features with different dimensions and amplitude ranges to a comparable unified scale. Then, using noise statistics based on the baseline segment or the background noise statistics of the leading segment, the features are converted into the degree of deviation from the relative noise. In addition to standardization, the features can also be normalized. Specifically, the operating condition is stratified according to the intervals divided by the continuous changes in the operating condition information, such as different speed ranges, oncoming traffic conditions, temperature compartments, or wind speed compartments. Different operating condition layers have different reference standards. Then, the corresponding operating condition layer is found according to the operating condition data encoding for normalization. For two features, the standardized and normalized values ​​are weighted and calculated to obtain the desired result. The damage index is described below. For example, the standardized and normalized overall constraint change characteristic value is denoted as F_cons, and the standardized and normalized impact nonlinear characteristic value is denoted as F_imp. The weighting coefficient assigned to F_cons is 0.4, and the weighting coefficient assigned to F_imp is 0.6. The two characteristic values ​​are also calculated by weighting the standardized and normalized values ​​of their respective characteristics. Specifically, F_cons is calculated by weighting the standardized and normalized main frequency drift, resonance peak amplitude change, and energy centroid change. If F_cons is 0.62 and F_imp is 0.78 in a certain event, the final calculated damage index is 0.716.

[0050] S540: Select a set of baseline damage index thresholds to be compared from a set of several baseline damage index thresholds based on the working condition data encoding; determine the damage level based on the set of baseline damage index thresholds to be compared and the damage index. Preferably, a baseline standard is determined based on the specific working conditions to identify the corresponding set of thresholds. Multiple thresholds can be used to classify the damage risk level within each working condition layer into mild, moderate, and severe. When the baseline standard deviates significantly over a long period or experiences significant seasonal drift, new healthy samples can be added to update the threshold classification. Understandably, using the baseline damage index threshold set to discriminate the damage index based on specific working conditions helps reduce the error in assessment results caused by changes in working conditions.

[0051] S550: Based on the overall constraint variation characteristics and the impact nonlinear characteristics, determine the damage type; Preferably, the overall constraint change characteristics and the impact nonlinear characteristics are preferentially compared with the corresponding characteristic results generated by health baseline events under the same region and the same working condition. Alternatively, they can be compared with the characteristics of health baseline events of the same component category and adjacent regions, or historical health samples can be selected for comparison. When the impact nonlinear characteristics are significantly strong, the damage type is determined to be loosening. When the overall constraint change characteristics are significantly strong and there are clues of missing appearance, it is determined to be connector missing. Specifically, a threshold comparison method can be used for comparison. For example, when F_imp is higher than the impact threshold and F_cons does not show obvious abrupt changes, it is determined to be an increased risk of loosening. When F_cons is higher than the constraint threshold and there are clues of missing connectors in the appearance, it is determined to be an increased risk of missing connectors.

[0052] S560: Based on the output of all the event fragments and the total number of the event fragments, obtain a consistency index, and obtain a damage confidence level based on the consistency index. The damage level, the damage category, and the damage confidence level constitute the evaluation result.

[0053] Preferably, the consistency index includes a multi-event consistency index and a multi-measurement point consistency index. The multi-event consistency index represents the number of times the output results are consistent in several events under the same regional data encoding. For example, the damage category and the damage level that occur most frequently are selected as the combination of output results, and the proportion of their occurrence to the total number of events is used as the multi-event consistency index. The multi-measurement point consistency index represents the output consistency of different measurement points.

[0054] Following step S50, the method further includes: S570: Based on the evaluation results, output a maintenance strategy, which may be re-tightening, anti-loosening replacement, local reinforcement, component replacement, or emergency treatment; S580: Based on the damage index, the baseline damage index threshold set, the consistency index, and the damage confidence, determine whether to perform encrypted verification; Preferably, any one of the following can trigger encrypted verification: the damage index exceeds the threshold of the corresponding working condition; the damage index deviates significantly from the baseline; the damage confidence level is lower than a preset threshold; or the consistency index does not meet the consistency judgment.

[0055] S590: If encrypted verification is performed, several risk areas are selected from the sensitive area weight map according to the sorting of the sensitive area weight values, and several accelerometers are encrypted and deployed in several of the risk areas to obtain updated evaluation results.

[0056] Preferably, the densification of monitoring points can be implemented by temporarily increasing the number of monitoring points, reducing the spacing between monitoring points, increasing the sampling frequency, or extending the observation period. Understandably, standardizing the features based on the background signal from the sound barrier during its operation, without triggering any special events, and further verifying the evaluation results are all beneficial for dynamically adjusting the monitoring points on the sound barrier and obtaining accurate and reliable prediction results.

[0057] Preferably, a database is established, which records the working condition data code, the regional data code, the sensitive area weight value, the appearance clue score, the overall constraint change characteristics, the impact nonlinear characteristics, the damage index, the damage level, and the maintenance strategy. It also records maintenance feedback, event identifiers, timestamps, model versions, and feature versions. Understandably, the database aggregates accumulable data resources from the working condition-sensitive area-risk link.

[0058] Further, training samples are established based on the database. The input features of the samples include at least one of operating condition features, structural features, regional features, weighted features, and historical statistical features. The historical statistical features include at least one of the abnormality percentage within a preset time window, damage index distribution statistics, and maintenance frequency. Each sample corresponds to the risk status of a certain regional data code under a certain operating condition data code. Maintenance feedback is used as a monitoring signal to label the training samples. Samples confirmed to be loose, missing, or requiring handling are marked as high-risk, while samples confirmed to have no abnormalities are marked as low-risk. Based on the training samples, a GBDT model is trained, and the model version and feature version are fixed. The GBDT model is a gradient boosting decision tree GBDT model, which can calibrate the model's output probability. During online prediction, a feature vector of the sensitive area is constructed according to the given operating condition data code and input into the GBDT model. The GBDT model outputs the risk probability and risk value corresponding to the data codes of each region. It sorts the risk values ​​and outputs a sensitive region set TopK_Pred and its risk level. The methods for using the sensitive region weight values ​​in prediction include: using them as input features of the GBDT model; adjusting the weights based on the risk values ​​output by the GBDT model; and using weight sorting as a fallback to output TopK_Pred when there are insufficient samples or insufficient model confidence. Maintenance feedback records are obtained, and the parameters of the GBDT model are updated based on these records. The sensitive region weight values ​​are then updated by backfeeding based on the maintenance feedback records. Specifically, when maintenance confirms that a region is loose or has missing connectors, the weight of that region is increased by a gain factor, and the weights are diffused to its adjacent connecting nodes or related regions of the same component category. When maintenance confirms a false alarm, the weight of that region is decreased by a decay factor.

[0059] Please see Figure 2 The system provided in the second embodiment of the present invention applies the sound barrier risk identification and prediction method described in the first embodiment above, and the system includes: The operating condition module 10 is used to acquire the sound barrier information, environmental condition information and train operation condition information of the sound barrier to be identified, combine the sound barrier information, environmental condition information and train operation condition information into operating condition data encoding, and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. In the operating condition module 10, the sound barrier information includes the sound barrier type, column spacing, sound barrier foundation form, screen unit type, connector data and mileage location; the environmental condition information includes temperature, wind speed and rainfall status; and the train operation condition information includes train speed range, meeting status, passage time period and train frequency.

[0060] The positioning module 20 is used to acquire the appearance image of the sound barrier to be identified, locate the components in the appearance image based on several component categories, and obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. The positioning module 20 includes: The first unit is used to preprocess the appearance image to obtain the image to be identified; The second unit is used to identify the image to be identified based on several component categories, obtain several identification results corresponding to several components, and convert the several identification results into region data encoding, wherein the identification results include bounding boxes or masks; The third unit is used to judge the appearance of several components based on several recognition results to obtain several appearance clues, which are normal appearance clues, missing connector clues, abnormal gap clues or structural deformation clues. The fourth unit is used to obtain several area proportions based on several of the recognition results; The fifth unit is used to generate an appearance clue score corresponding to the component based on the area ratio and all the appearance clues.

[0061] The weighting module 30 is used to obtain several sensitive area weight values ​​based on several component categories, several working condition data codes and several appearance clue scores, and to form a sensitive area weight map by combining several component categories, several sensitive area weight values, working condition data codes and area data codes. The weighting module 30 includes: The sixth unit is used to set several basic weights according to several component categories; The seventh unit is used to modify several basic weights based on several appearance cue scores to obtain several cue modification weights. The eighth unit is used to acquire historical maintenance data, and to modify the correction weights of several clues based on the historical maintenance data to obtain several historical correction weights. The ninth unit is used to perform adaptive scaling of several historical correction weights based on the operating condition data encoding to obtain several sensitive area weight values.

[0062] The signal module 40 is used to set up a number of acceleration sensors on the sound barrier to be identified according to the sensitive area weight map, and to collect a number of vibration acceleration signals, and to perform event slicing on the number of vibration acceleration signals to form event signals; The signal module 40 includes: The tenth unit is used to obtain the slicing method identifier and determine the slicing method based on the slicing method identifier; The eleventh unit is used to obtain several event moments through the slicing method; The twelfth unit is used to obtain vehicle passage information if the slicing method is vehicle passage information slicing, and to determine several event times based on the vehicle passage information; The thirteenth unit is used to determine several event moments based on the short-time energy, peak value, or energy mutation rate of the vibration acceleration signal if the slicing method is vibration threshold slicing. The fourteenth unit is used to slice the vibration acceleration signal into several event segments based on several event moments. The fifteenth unit is used to extract the leading segment of the event segment to obtain several baseline segments, and to obtain a background noise statistical signal based on the several baseline segments; The sixteenth unit is used to establish an event signal based on all the event fragments.

[0063] The sixteenth unit is specifically used for: Quality control is performed on all the event fragments to obtain several event fragments to be aligned; Several anchor points are selected from the several event times, and all event segments are time-aligned based on the several anchor points to obtain the event signal.

[0064] Evaluation module 50 is used to obtain a damage index based on the event signal, so as to obtain an evaluation result based on the damage index.

[0065] The evaluation module 50 includes: Unit 17 is used to acquire several baseline signals and several baseline damage index threshold sets under several working conditions; The eighteenth unit is used to extract the overall constraint change characteristics and impact nonlinear characteristics of the event signal. The overall constraint change characteristics include one or more of the following: main frequency drift, resonance peak amplitude change, frequency band energy ratio, energy centroid change, spectral entropy change, and inter-measurement point transfer rate amplitude change. The impact nonlinear characteristics include one or more of the following: peak value, kurtosis, impact factor, short-time energy, pulse count, and energy mutation rate. The nineteenth unit is used to standardize the overall constraint change characteristics and the impact nonlinear characteristics based on the background noise statistical signal or the baseline signal, and estimate the damage index through feature weighted fusion or learning model. The twentieth unit is used to select a set of baseline damage index thresholds to be compared from a set of several baseline damage index thresholds based on the working condition data encoding, and to determine the damage level based on the set of baseline damage index thresholds to be compared and the damage index. The twenty-first unit is used to determine the damage type based on the overall constraint change characteristics and the impact nonlinear characteristics; The twenty-second unit is used to obtain a consistency index based on the output of all the event fragments and the total number of the event fragments, and to obtain a damage confidence level based on the consistency index. The damage level, the damage category, and the damage confidence level constitute the evaluation result.

[0066] The twenty-third unit is used to output maintenance strategies based on the evaluation results. The maintenance strategies include re-tightening, anti-loosening replacement, local reinforcement, component replacement, or emergency treatment. The twenty-fourth unit is used to determine whether to perform encrypted verification based on the damage index, the baseline damage index threshold set, the consistency index and the damage confidence. The twenty-fifth unit is used to select several risk areas from the sensitive area weight map according to the sorting of the sensitive area weight values ​​if encryption verification is to be performed, and to encrypt and deploy several acceleration sensors in several of the risk areas to obtain updated evaluation results.

[0067] The third embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the sound barrier risk identification and prediction method as described in the first embodiment above.

[0068] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0069] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for identifying and predicting the risk of a sound barrier, characterized in that, Includes the following steps: Acquire the sound barrier information, environmental condition information, and train operation condition information of the sound barrier to be identified; combine the sound barrier information, environmental condition information, and train operation condition information into an operation condition data code; and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. The appearance image of the sound barrier to be identified is acquired, and the components in the appearance image are located based on several component categories to obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. Based on several component categories, several working condition data codes, and several appearance clue scores, several sensitive area weight values ​​are obtained, and the several component categories, several sensitive area weight values, the working condition data codes, and the area data codes are combined to form a sensitive area weight map. Based on the sensitive area weight map, several acceleration sensors are set on the sound barrier to be identified, and several vibration acceleration signals are collected. Event slices are performed on the several vibration acceleration signals to form event signals. A damage index is obtained based on the event signal, and an evaluation result is obtained based on the damage index.

2. The sound barrier risk identification and prediction method according to claim 1, characterized in that, The sound barrier information includes the sound barrier type, column spacing, sound barrier foundation form, screen unit type, connector data, and mileage location. The environmental condition information includes temperature, wind speed, and rainfall status. The train operation condition information includes train speed range, meeting status, passage time period, and train frequency.

3. The sound barrier risk identification and prediction method according to claim 1, characterized in that, The step of locating components in the appearance image based on several component categories to obtain the region data encoding and several appearance cue scores corresponding to the sound barrier to be identified includes: The appearance image is preprocessed to obtain the image to be identified; The image to be identified is identified based on several component categories to obtain several identification results corresponding to several components, and the several identification results are converted into region data encoding, wherein the identification results include bounding boxes or masks; Based on several identification results, the appearance of several components is judged to obtain several appearance clues, which are normal appearance clues, missing connector clues, abnormal gap clues, or structural deformation clues. Several area proportions are obtained based on several of the aforementioned recognition results; Based on the area ratio and all the appearance cues, an appearance cue score corresponding to the component is generated.

4. The sound barrier risk identification and prediction method according to claim 1, characterized in that, The step of obtaining several sensitive area weight values ​​based on several component categories, several working condition data codes, and several appearance cue scores includes: Several basic weights are set according to several component categories; Based on several appearance cue scores, several basic weights are corrected to obtain several cue correction weights; Obtain historical maintenance data, and adjust the correction weights of several clues based on the historical maintenance data to obtain several historical correction weights; Based on the operating condition data encoding, several historical correction weights are adaptively scaled according to operating conditions to obtain several sensitive area weight values.

5. The sound barrier risk identification and prediction method according to claim 1, characterized in that, The step of performing event slicing on several vibration acceleration signals to form an event signal includes: Obtain the slicing method identifier, and determine the slicing method based on the slicing method identifier; Several event moments are obtained using the slicing method described above; If the slicing method is vehicle passage information slicing, then vehicle passage information is obtained, and several event times are determined based on the vehicle passage information; If the slicing method is vibration threshold slicing, then several event moments are determined based on the short-time energy, peak value, or energy mutation rate of the vibration acceleration signal; The vibration acceleration signal is sliced ​​into several event segments based on several event moments. Extract the leading segment of the event fragment to obtain several baseline segments, and obtain the background noise statistical signal based on the several baseline segments; An event signal is established based on all the event fragments described.

6. The sound barrier risk identification and prediction method according to claim 5, characterized in that, The step of establishing an event signal based on all the event fragments includes: Quality control is performed on all the event fragments to obtain several event fragments to be aligned; Several anchor points are selected from the several event times, and all event segments are time-aligned based on the several anchor points to obtain the event signal.

7. The sound barrier risk identification and prediction method according to claim 5, characterized in that, The step of obtaining a damage index based on the event signal, and then obtaining an evaluation result based on the damage index, includes: Acquire several baseline signals and several baseline damage index threshold sets under several working conditions; Extract the overall constraint change features and impact nonlinear features of the event signal. The overall constraint change features include one or more of the following: main frequency drift, resonance peak amplitude change, frequency band energy ratio, energy centroid change, spectral entropy change, and inter-measurement point transfer rate amplitude change. The impact nonlinear features include one or more of the following: peak value, kurtosis, impact factor, short-time energy, pulse count, and energy mutation rate. Based on the background noise statistical signal or the baseline signal, the overall constraint change characteristics and the impact nonlinear characteristics are standardized, and the damage index is estimated by feature weighted fusion or learning model. Based on the operating condition data encoding, a set of baseline damage index thresholds to be compared is selected from several sets of baseline damage index thresholds. Based on the set of baseline damage index thresholds to be compared and the damage index, the damage level is determined. Based on the overall constraint variation characteristics and the impact nonlinear characteristics, the damage type is determined; Based on the output of all event fragments and the total number of event fragments, a consistency index is obtained, and a damage confidence level is obtained based on the consistency index. The damage level, the damage category, and the damage confidence level constitute the evaluation result.

8. The sound barrier risk identification and prediction method according to claim 7, characterized in that, After the step of obtaining a damage index based on the event signal to obtain an evaluation result based on the damage index, the method further includes: Based on the assessment results, a maintenance strategy is output, which includes re-tightening, anti-loosening replacement, local reinforcement, component replacement, or emergency treatment. Based on the damage index, the baseline damage index threshold set, the consistency index, and the damage confidence, it is determined whether to perform encrypted verification. If encrypted verification is performed, several risk areas are selected from the sensitive area weight map according to the sorting of the sensitive area weight values, and several accelerometers are deployed in the several risk areas to obtain updated evaluation results.

9. A sound barrier risk identification and prediction system, employing the method described in any one of claims 1 to 8, characterized in that, The system includes: The operating condition module is used to acquire the sound barrier information, environmental condition information and train operation condition information of the sound barrier to be identified, combine the sound barrier information, environmental condition information and train operation condition information into operating condition data code, and extract the component categories of several components of the sound barrier to be identified from the sound barrier information. The positioning module is used to acquire the appearance image of the sound barrier to be identified, locate the components in the appearance image based on several component categories, and obtain the area data code and several appearance clue scores corresponding to the sound barrier to be identified. The weighting module is used to obtain several sensitive area weight values ​​based on several component categories, several working condition data codes, and several appearance clue scores, and to form a sensitive area weight map by combining several component categories, several sensitive area weight values, the working condition data codes, and the area data codes. The signal module is used to set up a number of acceleration sensors on the sound barrier to be identified according to the sensitive area weight map, and to collect a number of vibration acceleration signals, and to perform event slicing on the number of vibration acceleration signals to form event signals; An evaluation module is used to obtain a damage index based on the event signal, so as to obtain an evaluation result based on the damage index.

10. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the sound barrier risk identification and prediction method as described in any one of claims 1 to 8.

Images