Railway wagon loading abnormality early warning method and system based on artificial intelligence

Abstract

This application provides an artificial intelligence-based method and system for early warning of loading anomalies in railway freight cars. The method includes: collecting dynamic pressure distribution data at the car connections and multiple wheel support areas during freight car operation; fusing the phase offset of the dynamic pressure distribution data at the car connections and the amplitude differences of the dynamic pressure distribution data in the multiple wheel support areas to generate spatial pressure gradient features for each car of the freight car; inputting the spatial pressure gradient features and the ground state of the pressure distribution of the freight car under historical normal operating conditions into a pre-trained residual network to output abnormal fluctuation components between the car connections and multiple wheel support areas; based on the abnormal fluctuation components, combined with the current operating speed of the freight car and track curvature parameters, determining the type of loading anomaly, and issuing an early warning according to the warning level corresponding to the type of loading anomaly. This application improves the accuracy and timeliness of early warning for the safe operation of freight trains.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based method and system for early warning of abnormal loading of railway freight cars. Background Technology

[0002] During railway freight transport, uneven loading or overloading can easily lead to abnormal conditions such as uneven loading and overloading, seriously affecting train operation safety. Traditional detection methods are insufficient for real-time monitoring of the operational status.

[0003] Currently, there are monitoring methods based on track scales and on-board weighing sensors. These methods collect static pressure data as freight cars pass by by installing weighing devices at specific locations on the track or deploying strain gauge sensors at the bottom of the car, and then using preset thresholds to determine whether an overload has occurred. Some improved solutions incorporate machine learning algorithms to train on historical weighing data, enabling preliminary classification of abnormal conditions.

[0004] This scheme relies on fixed-position track scales or single-point weighing sensors, which can only acquire static load data at discrete time points. At the same time, it does not consider the dynamic coupling force at the car connection and the collaborative force characteristics of multiple wheel support areas, resulting in insufficient accuracy in identifying off-center load conditions. Furthermore, it lacks fusion analysis with dynamic parameters such as running speed and track curvature, leading to a high false alarm rate. Summary of the Invention

[0005] This application provides an artificial intelligence-based method and system for early warning of abnormal loading of railway freight cars, in order to solve the problems of low accuracy and poor timeliness of early warning of freight train operation safety in the prior art.

[0006] Firstly, this application provides an artificial intelligence-based method for early warning of abnormal loading of railway freight cars, including:

[0007] Collect dynamic pressure distribution data at the connection points of railway freight cars and the support areas of multiple sets of wheels during operation;

[0008] By integrating the phase offset of the dynamic pressure distribution data at the car connection and the amplitude difference of the dynamic pressure distribution data of the multiple wheel support areas, spatial pressure gradient characteristics of each car of a railway freight car are generated.

[0009] The spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions are input into a pre-trained residual network to output the abnormal fluctuation components between the car connection and multiple wheel support areas.

[0010] Based on the abnormal fluctuation components, combined with the current operating speed of the railway freight cars and the track curvature parameters, the loading anomaly type is determined, and an early warning is issued according to the early warning method corresponding to the early warning level of the loading anomaly type.

[0011] Optionally, the step of inputting the spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions into a pre-trained residual network, and outputting the abnormal fluctuation components between the car connection and multiple sets of wheel support areas, includes:

[0012] The spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions are input into a pre-trained residual network. Through the decomposition module in the pre-trained residual network, the spatial pressure gradient features are decomposed into a first equilibrium parameter component parallel to the longitudinal axis of the car and a second equilibrium parameter component perpendicular to the longitudinal axis of the car.

[0013] By using the matching module in the pre-trained residual network, the corresponding components of the pressure distribution ground state are hierarchically matched to generate dynamic coupling difference quantities.

[0014] The dynamic coupling difference is weighted and fused using the fusion module in the pre-trained residual network to output the abnormal fluctuation component.

[0015] Optionally, the step of hierarchically matching the corresponding components of the pressure distribution ground state using the matching module in the pre-trained residual network to generate a dynamically coupled difference quantity includes:

[0016] In the residual network, the pressure distribution ground state of railway freight cars under historical normal operating conditions is hierarchically decomposed by the matching module into longitudinal ground state components corresponding to the first equilibrium parameter components and transverse ground state components corresponding to the second equilibrium parameter components.

[0017] Hierarchical matching is performed on the first equilibrium parameter component and the longitudinal ground state component to obtain the cumulative difference. Hierarchical matching is performed on the second equilibrium parameter component and the transverse ground state component to generate the transverse force imbalance compensation coefficient.

[0018] Based on the cumulative difference and the lateral force imbalance compensation coefficient, a dynamic coupling difference is generated.

[0019] Optionally, the step of performing hierarchical matching of the first equilibrium parameter component and the longitudinal ground state component to obtain the cumulative difference, and performing hierarchical matching of the second equilibrium parameter component and the transverse ground state component to generate the transverse force imbalance compensation coefficient, includes:

[0020] The first balance parameter component is divided into multiple equidistant segments according to the length of the carriage, and each equidistant segment corresponds to a longitudinal ground state component reference value in the longitudinal ground state component;

[0021] For each equidistant segment, calculate the instantaneous difference between the first equilibrium parameter component corresponding to the equidistant segment and the reference value of the corresponding longitudinal ground state component;

[0022] The instantaneous differences of all equidistant sections are recursively accumulated according to the direction of carriage travel to generate a cumulative difference.

[0023] The second balance parameter component is divided into a left rail contact component and a right rail contact component according to the wheel set position;

[0024] Calculate the left rail offset between the left rail contact component and the left rail reference value of the lateral ground state component, and the right rail offset between the right rail contact component and the right rail reference value of the lateral ground state component, respectively.

[0025] Based on the reverse variation characteristics of the left rail offset and the right rail offset of the same wheel set, the lateral force imbalance compensation coefficient is calculated.

[0026] Optionally, generating the dynamic coupling difference based on the accumulated difference and the lateral force imbalance compensation coefficient includes:

[0027] The cumulative difference is mapped to the transverse plane of the carriage according to the corresponding wheel set position;

[0028] Based on the aforementioned lateral force imbalance compensation coefficient, adjust the distribution weight of the mapped cumulative difference in the left and right rail directions;

[0029] Integrate the adjusted cumulative differences for all wheel set positions to generate a dynamically coupled difference.

[0030] Optionally, determining the loading anomaly type based on the abnormal fluctuation components, combined with the current operating speed of the railway freight car and track curvature parameters, includes:

[0031] The abnormal fluctuation components are decomposed into vertical feature vectors and horizontal feature vectors;

[0032] Based on the current operating speed of the railway freight car and the track curvature parameters, the longitudinal feature vector and the lateral feature vector are corrected respectively;

[0033] Based on the frequency distribution of the corrected vertical feature vector, determine the overload level code corresponding to the overload level interval;

[0034] Based on the magnitude difference of the corrected lateral feature vector, the off-center load direction code corresponding to the off-center load direction interval is determined;

[0035] The loading anomaly type is matched from the preset loading anomaly classification rule base to the loading anomaly type that corresponds to both the overload level code and the off-center load direction code.

[0036] Optionally, the step of correcting the longitudinal feature vector and the lateral feature vector based on the current operating speed of the railway freight car and the track curvature parameters includes:

[0037] The current operating speed of the railway freight car is converted into a speed correction factor that is consistent with the longitudinal axis of the car.

[0038] The track curvature parameters of railway freight cars are decomposed into the reference curvature of straight sections and the incremental curvature of curved sections, and the incremental curvature of curved sections is converted into track curvature correction coefficients corresponding to the direction of the lateral feature vector.

[0039] The longitudinal feature vector is corrected according to the running speed correction coefficient to generate a corrected longitudinal feature vector;

[0040] The lateral feature vector is corrected based on the operating speed correction coefficient and the track curvature correction coefficient to generate a corrected lateral feature vector.

[0041] Secondly, this application provides an artificial intelligence-based early warning system for abnormal loading of railway freight cars, comprising:

[0042] The data acquisition module is used to collect dynamic pressure distribution data at the connection points of railway freight cars and the support areas of multiple sets of wheels during operation.

[0043] The generation module is used to fuse the phase offset of the dynamic pressure distribution data at the car connection and the amplitude difference of the dynamic pressure distribution data of the multiple sets of wheel support areas to generate the spatial pressure gradient characteristics of each car of the railway freight car.

[0044] The input module is used to input the spatial pressure gradient features and the pressure distribution ground state of railway freight cars under historical normal operating conditions into a pre-trained residual network, and output the abnormal fluctuation components between the car connection and multiple wheel support areas.

[0045] The determination module is used to determine the loading anomaly type based on the abnormal fluctuation component, combined with the current operating speed of the railway freight car and the track curvature parameters, and to issue an early warning according to the early warning method corresponding to the early warning level of the loading anomaly type.

[0046] Thirdly, this application provides a computing device including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute an artificial intelligence-based early warning method for abnormal loading of railway freight cars as described in any of the first aspects.

[0047] Fourthly, this application provides a computer storage medium storing computer program instructions thereon, which, when executed by a processor, implement the artificial intelligence-based early warning method for abnormal loading of railway freight cars as described in any one of the first aspects.

[0048] This application provides an artificial intelligence-based method for early warning of abnormal loading in railway freight cars. The method includes: collecting dynamic pressure distribution data at the car-to-car connection points and multiple wheel support areas during railway freight car operation; fusing the phase offset of the dynamic pressure distribution data at the car-to-car connection points and the amplitude difference of the dynamic pressure distribution data in the multiple wheel support areas to generate spatial pressure gradient features for each car of the railway freight car; inputting the spatial pressure gradient features and the ground state of the railway freight car's pressure distribution under historical normal operating conditions into a pre-trained residual network to output abnormal fluctuation components between the car-to-car connection points and multiple wheel support areas; based on the abnormal fluctuation components, combined with the current operating speed and track curvature parameters of the railway freight car, determining the type of loading anomaly, and issuing an early warning according to the warning level corresponding to the type of loading anomaly.

[0049] The technical solution provided in this application has the following beneficial effects:

[0050] This application achieves full-dimensional perception of the dynamic coupling force at the carriage connection and the contact pressure in the wheel support area through multi-region synchronous monitoring, providing a complete data foundation for anomaly detection; it integrates phase offset and amplitude difference to establish a collaborative feature representation reflecting the imbalance between longitudinal force transmission and lateral force distribution in the car body, improving the timeliness of freight train operation safety warnings; through hierarchical matching of historical ground state and real-time features, it accurately separates the dynamic fluctuation components caused by loading anomalies; and by dynamically adjusting the judgment threshold based on operating speed and track curvature, it achieves graded and accurate early warning under complex operating conditions.

[0051] Furthermore, the residual network described in this application decouples the spatial pressure gradient features into longitudinal / lateral components through a decomposition module, generates a dynamic coupling difference quantity by hierarchically comparing it with the historical ground state through a matching module, and finally outputs the abnormal fluctuation component by a weighted fusion module.

[0052] Furthermore, this design improves the separation accuracy of off-center load and overload anomalies through physical dimension decoupling and hierarchical feature matching, and in particular improves the detection sensitivity of lateral force imbalance under curve conditions.

[0053] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart illustrating an artificial intelligence-based early warning method for abnormal loading of railway freight cars, provided as an embodiment of this application;

[0056] Figure 2 A schematic diagram of the structure of an artificial intelligence-based early warning system for abnormal loading of railway freight cars is provided in this application embodiment;

[0057] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0059] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0060] Researchers have found that existing methods for monitoring the loading of railway freight cars struggle to capture multi-dimensional stress characteristics under dynamic operating conditions in real time, and lack analysis of the synergistic effects between the car connections and the wheel-rail contact area, resulting in insufficient accuracy in detecting off-center loading and overload. Based on this, this application provides an artificial intelligence-based early warning method for abnormal loading of railway freight cars. This method synchronously collects dynamic pressure data from the car connections and multiple wheel support areas, fuses phase shift and amplitude differences to construct spatial pressure gradient features, and uses a residual network to separate abnormal fluctuation components. Finally, it combines operating environment parameters to achieve accurate classification and graded early warning of loading anomalies. The technical solution of this application is applicable to real-time safety monitoring scenarios for heavy-haul railway freight cars under complex track conditions.

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] Figure 1 A flowchart of an artificial intelligence-based early warning method for abnormal loading of railway freight cars is provided for embodiments of this application, as shown below. Figure 1 As shown, the method includes:

[0063] Step 101: Collect dynamic pressure distribution data at the connection points of railway freight cars and the support areas of multiple sets of wheels during operation.

[0064] In this step, the car connection specifically refers to the mechanical connection between adjacent cars, and the dynamic pressure distribution data at the car connection is used to characterize the dynamic coupling force between the car bodies. Multiple wheel support areas refer to the distribution area of ​​support points where each pair of wheels of a railway freight car contacts the rail, and the dynamic pressure distribution data of these multiple wheel support areas is used to characterize the contact pressure distribution between the wheels and the rail. The dynamic pressure distribution data comes from a sensor network symmetrically arranged around the longitudinal axis of the car, covering the support areas of at least two pairs of adjacent wheels, and includes time-series data with timestamps, pressure values, and spatial location codes, with a sampling frequency ≥100Hz.

[0065] In this embodiment, a multi-axis force sensor array is first deployed at the car body connection point, using dynamic pressure sensing technology to collect real-time interaction force data between the coupler and the buffer device. Simultaneously, wheel-rail force sensor groups are installed on the left and right rail contact surfaces of each wheel group, acquiring the wheel-rail contact pressure distribution through high-frequency sampling. All sensor data undergoes anti-interference filtering and is then time-stamped by a data synchronization module to ensure spatiotemporal consistency of data from each measuring point. The collected raw pressure data is labeled with spatial location attributes according to preset coding rules, forming a dynamic pressure dataset with three-dimensional coordinate information, providing standardized input for subsequent feature fusion.

[0066] For example, when a four-axle freight train is running on a freight line, a force sensor installed at the car body connection point detects that the peak dynamic pressure at connection point 1 is 1.2 times a certain standard value (calculated by directly measuring the original voltage signal through the pressure sensor and then applying the calibration formula "pressure value = voltage reading × conversion coefficient"). Simultaneously, at wheel set 2, the pressure difference between the left and right rails is detected to reach 80% of a certain safety threshold (this threshold is determined based on the wheel and axle material strength standard). All sensor data are collected synchronously at fixed sampling intervals and, after verification, form a complete dataset containing timestamps and spatial coordinates.

[0067] Step 102: Combine the phase offset of the dynamic pressure distribution data at the car connection point with the amplitude difference of the dynamic pressure distribution data of the multiple wheel support areas to generate the spatial pressure gradient characteristics of each car in the railway freight car.

[0068] In this step, the phase offset represents the time delay difference (in milliseconds) of the pressure waveform at the car body connection, reflecting the delay in dynamic force transmission between the car bodies. The amplitude difference represents the difference in peak pressure between the left and right rails of the same group of wheels (in MPa), characterizing the degree of imbalance in the lateral load distribution. The spatial pressure gradient feature includes a first equilibrium parameter component parallel to the longitudinal axis of the car body and a second equilibrium parameter component perpendicular to the longitudinal axis of the car body.

[0069] In this embodiment, cross-correlation analysis is performed on the time-series pressure data collected at the car body connection to calculate the phase delay of the pressure waveform at different measuring points, revealing the dynamic force transmission characteristics between the car bodies. Simultaneously, extreme values ​​are extracted from the left and right rail pressure data of each group of wheel support areas to calculate the real-time amplitude difference reflecting the lateral load distribution. The phase offset and amplitude difference are input into a spatial mapping algorithm, and the longitudinal force transmission distribution map and lateral pressure imbalance map of the car body are reconstructed through three-dimensional interpolation. Feature fusion technology is used to convert the two types of distribution maps into a two-dimensional vector sequence containing longitudinal gradient intensity and lateral imbalance degree, ultimately outputting spatial pressure gradient features with clear physical meaning.

[0070] For example, based on the collected data, the system calculates that there is a phase difference of a certain time unit in the pressure waveforms of the measuring points before and after the No. 1 connection (the time offset corresponding to the maximum correlation coefficient is obtained through a cross-correlation algorithm). At the same time, the extreme difference in pressure between the left and right rails of the No. 2 wheel set is converted into a specific value in a certain unit of measurement. After these parameters are input into the spatial feature construction module, a longitudinal distribution map showing the maximum pressure gradient located in the area of ​​the No. 3 wheelset is generated (the gradient intensity is calculated through a two-dimensional interpolation algorithm), as well as a transverse pressure cloud map with obvious right deviation (the degree of imbalance is calculated according to the formula "imbalance coefficient = (left rail pressure - right rail pressure) / average pressure" based on the proportion of left and right pressure differences).

[0071] Step 103: Input the spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions into the pre-trained residual network, and output the abnormal fluctuation components between the car connection and multiple wheel support areas.

[0072] In this step, the pressure distribution ground state represents a standard pressure pattern library statistically obtained under historical normal operating conditions (stored hierarchically by velocity / curvature). The residual network contains a dedicated neural network with a three-level structure of decomposition-matching-fusion (256 nodes in the input layer). Anomalous fluctuation components are used to reflect anomalous force characteristics deviating from the ground state (including longitudinal and lateral feature vectors).

[0073] In this embodiment, real-time spatial pressure gradient features are input into a pre-trained residual network. The network first separates the input data into a longitudinal component parallel to the carriage axis and a lateral component perpendicular to the axis through a convolutional feature decomposition module. Then, each component is compared with the standard ground state data under the same working conditions in the historical database layer by layer to calculate the feature deviation at each network level. Finally, the deviation data of each level are integrated through a weighted fusion module to generate a composite anomaly index that simultaneously includes the longitudinal abnormal fluctuation intensity and the lateral abnormal region distribution, and the output is a standardized abnormal fluctuation code.

[0074] For example, the obtained feature data is compared with the normal pressure patterns of the same type of truck in a certain speed range in the historical database. The residual network first identifies that the longitudinal pressure fluctuation in the area of ​​wheel set 2 exceeds a certain percentage of the ground state (calculated by the sum of squared differences point by point), while the lateral offset at connection 1 reaches a certain multiple of the ground state value (a relative ratio after standardization). The network finally outputs anomaly codes, marking the main abnormal areas as wheel set 2 (the longitudinal anomaly level is determined by a preset fluctuation amplitude classification table) and the right side of connection 1 (the lateral anomaly degree is classified according to the ratio of offset to a safety threshold).

[0075] Step 104: Based on the abnormal fluctuation components, combined with the current operating speed of the railway freight car and the track curvature parameters, determine the loading anomaly type, and issue an early warning according to the early warning method corresponding to the early warning level of the loading anomaly type.

[0076] In this step, the operating speed represents the amplification factor of the anomaly's effect (formula: (v / v0)^2, v0 = 100 km / h). The track curvature parameter represents the adjustment coefficient of the curve to the lateral force (formula: 1 + 0.5 * |1 / R|, where R is the radius). The warning levels include four levels (Level I: Minor; Level IV: Emergency).

[0077] In this embodiment, the abnormal fluctuation code output by the residual network is analyzed for spectral characteristics. The longitudinal abnormal main frequency component is extracted and the speed influence coefficient is calculated in combination with the real-time running speed. The overload level is determined by a preset frequency-load relationship mapping table. Simultaneously, the spatial distribution pattern of the lateral abnormal area is analyzed, and the curvature compensation coefficient is calculated by superimposing the current track curvature parameters. The off-center load direction is determined based on the difference ratio between the left and right track anomalies. Finally, the overload level and off-center load direction parameters are input into the graded early warning rule base, matching the corresponding early warning level and triggering audible and visual alarms and data reporting, completing the closed-loop processing from anomaly detection to early warning output.

[0078] For example, based on the anomaly code, the system combines the current vehicle speed (obtained directly from the onboard speed sensor) and track curvature (from the digital map of the route) for a comprehensive judgment. The speed influence coefficient is calculated as a certain value (calculated using the formula "coefficient = (actual speed / reference speed) squared"), and the track curvature compensation coefficient is calculated as a certain value (determined based on the ratio of the reciprocal of the curve radius to the reference curvature). Finally, if an overload of a certain level is determined combined with an off-center load in a certain direction (obtained by consulting a pre-set anomaly level comparison table), a corresponding level of warning signal is triggered. The warning level is determined according to the anomaly index weighting formula "comprehensive index = longitudinal anomaly × speed coefficient + lateral anomaly × curvature coefficient".

[0079] This technical solution achieves real-time and accurate monitoring of the loading status of railway freight cars through a complete process of multi-source sensor collaborative monitoring, dynamic feature fusion analysis, and intelligent early warning decision-making. The method innovatively combines the collaborative force characteristics of the car connection point and the wheel-rail contact area, and fully considers the influence of operating environment parameters, improving the accuracy of identifying off-center loading and overload anomalies and the timeliness of early warning under complex operating conditions. The entire system operates stably and reliably, effectively ensuring the safe and efficient operation of railway freight, and is particularly suitable for the real-time safety monitoring needs of long and heavy-haul trains under various track conditions.

[0080] To address the issue of inaccurate dynamic feature extraction in railway freight car loading anomaly detection, in some embodiments, step 103: inputting the spatial pressure gradient features and the ground state of the railway freight car's pressure distribution under historical normal operating conditions into a pre-trained residual network, and outputting the abnormal fluctuation components between the car connection and multiple wheel support areas, includes:

[0081] Step 201: Input the spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions into a pre-trained residual network. Through the decomposition module in the pre-trained residual network, decompose the spatial pressure gradient features into a first equilibrium parameter component parallel to the longitudinal axis of the carriage and a second equilibrium parameter component perpendicular to the longitudinal axis of the carriage.

[0082] In step 201, the first balance parameter component refers to the pressure variation characteristics along the length of the carriage, reflecting the longitudinal force transmission state. The second balance parameter component refers to the pressure distribution characteristics perpendicular to the carriage axis, characterizing the lateral load balance.

[0083] In this embodiment, the residual network decomposition module employs spatial convolution technology. First, it performs directional filtering on the input spatial pressure gradient features. Then, it extracts continuous pressure change patterns parallel to the carriage axis using a vertical convolution kernel to generate the first component. Simultaneously, it uses a horizontal convolution kernel to capture pressure difference patterns perpendicular to the axis to generate the second component. The two components retain mechanical information from different dimensions in the original feature map.

[0084] Step 202: Using the matching module in the pre-trained residual network, perform hierarchical matching of the corresponding components of the pressure distribution ground state to generate dynamic coupling difference quantities.

[0085] In step 202, the dynamic coupling difference refers to the matching difference between real-time features and historical ground states at each network level, including the longitudinal difference spectrum and the lateral difference map.

[0086] In this embodiment, the matching module employs a hierarchical correlation analysis method to perform multi-scale similarity comparison between the first component and the historical longitudinal ground state, calculating the difference coefficient for each frequency band; simultaneously, it performs region-by-region matching between the second component and the historical lateral ground state to generate a spatial difference distribution. Through cross-level feature fusion technology, the two types of differences are uniformly quantified into a physically meaningful coupled difference index.

[0087] Step 203: The dynamic coupling difference is weighted and fused using the fusion module in the pre-trained residual network to output the abnormal fluctuation component.

[0088] In step 203, weighted fusion refers to applying dynamic weights based on the spatial distribution characteristics of the differential quantities to highlight key abnormal areas. The specific process of weighted fusion of dynamically coupled differential quantities is as follows: First, a longitudinal weight coefficient is generated based on the force transmission attenuation rate of the first balance parameter component in each detection section along the longitudinal axis of the carriage. Simultaneously, a lateral weight coefficient is generated based on the amplitude difference ratio of the second balance parameter component in the left and right rail directions. Then, the dynamic coupled differential quantities are multiplied by the corresponding weight coefficients in both the longitudinal and lateral dimensions of the carriage and the track, respectively. The dot product result in the longitudinal dimension reflects the degree of abnormal force transmission in the car body structure, while the dot product result in the lateral dimension reflects the degree of imbalance in wheel-rail contact force. Finally, the weighted results of the two dimensions are dynamically mixed according to the actual operating speed of the carriage to generate an abnormal fluctuation component that simultaneously contains longitudinal force anomalies and lateral force imbalance characteristics. The mixing ratio of longitudinal and lateral features in the abnormal fluctuation component increases with the increase of the operating speed, with the weight ratio of longitudinal features increasing.

[0089] In this embodiment, the fusion module first performs frequency domain energy analysis on the longitudinal difference spectrum to determine the weight coefficients for each frequency band; simultaneously, it performs regional detection on the lateral difference map to generate spatial attention weights. An adaptive weighting algorithm is used to integrate the two types of difference quantities, and the final output includes fluctuation components that simultaneously contain both frequency domain anomaly intensity and spatial anomaly distribution.

[0090] Here is a specific example:

[0091] During the operation of a four-axle freight car on a certain freight line, the system inputs the spatial pressure gradient features constructed from the dynamic pressure data of the No. 1 connection (peak value is 1.2 times the standard value) and the pressure difference data between the left and right rails of the No. 2 wheel set (reaching 80% of the safety threshold) into the residual network. The decomposition module first extracts the pressure fluctuation features along the axis of the car at the No. 1 connection (the first balance parameter component, whose fluctuation amplitude is calculated by "current pressure value / ground state average value" to obtain a change coefficient of 1.15) through longitudinal convolution operation. At the same time, it extracts the pressure distribution features of the No. 2 wheel set perpendicular to the axis through lateral convolution operation (the second balance parameter component, whose imbalance degree is calculated by "(left rail pressure - right rail pressure) / ground state average pressure" to obtain an imbalance ratio of 0.25). The matching module performs a hierarchical comparison of the first component with the longitudinal ground state under historical straight-line conditions, identifying mid-frequency differences (the difference coefficient is calculated to be 0.18 using the mean square error of the layer-by-layer feature maps). Simultaneously, it performs regional matching of the second component with the lateral ground state under historical curved-road conditions, identifying an abnormal right-rail pressure region (the difference is calculated to be 0.32 using the spatial correlation coefficient). The fusion module assigns a weight of 0.6 to the longitudinal difference (determined based on the frequency band energy ratio) and a weight of 0.4 to the lateral difference (determined based on the region area ratio), ultimately outputting the abnormal fluctuation component to mark a complex anomaly requiring close attention in the area of ​​wheel set 2.

[0092] In this embodiment of the application, the solution achieves accurate separation of abnormal features under complex working conditions through multi-dimensional feature decoupling and intelligent matching, effectively avoiding misjudgment caused by environmental interference in traditional methods, and enabling the system to accurately identify the real dangerous state and provide timely warnings.

[0093] To further improve the accuracy of railway freight car anomaly detection, in some embodiments, step 202: the hierarchical matching of the corresponding components of the pressure distribution ground state through the matching module in the pre-trained residual network to generate a dynamically coupled difference quantity includes:

[0094] Step 301: In the residual network, the pressure distribution ground state of railway freight cars under historical normal operating conditions is hierarchically decomposed by the matching module into longitudinal ground state components corresponding to the first equilibrium parameter components and transverse ground state components corresponding to the second equilibrium parameter components.

[0095] In step 301, the longitudinal ground state component refers to the reference pressure fluctuation pattern along the carriage axis under historical normal operating conditions. The lateral ground state component refers to the reference pressure distribution pattern perpendicular to the carriage axis under historical normal operating conditions.

[0096] In this embodiment, the matching module uses multi-scale decomposition technology to perform wavelet transform processing on the historical pressure distribution ground state, extracting the longitudinal pressure fluctuation features of different frequency bands as longitudinal ground state components. At the same time, it extracts the lateral pressure distribution features of each region as lateral ground state components through spatial grid division, and establishes a complete benchmark feature library.

[0097] Step 302: Perform hierarchical matching on the first equilibrium parameter component and the longitudinal ground state component to obtain the cumulative difference; perform hierarchical matching on the second equilibrium parameter component and the transverse ground state component to generate the transverse force imbalance compensation coefficient.

[0098] In step 302, the cumulative difference refers to the cumulative difference between the real-time longitudinal characteristics and the historical longitudinal ground state across all frequency bands. The lateral force imbalance compensation coefficient refers to the correction parameter required when the real-time lateral characteristics deviate from the historical lateral ground state.

[0099] In this embodiment, wavelet packet transform is performed on the first balance parameter component to decompose it into the same frequency band as the longitudinal ground state component, and the cumulative value of the energy difference of each frequency band is calculated as the cumulative difference amount; at the same time, regional grid matching is performed on the second balance parameter component, and the lateral force imbalance compensation coefficient of each region is calculated by a preset compensation formula based on the degree of deviation between the left and right rail pressure difference in each grid cell and the historical benchmark.

[0100] Step 303: Generate dynamic coupling difference based on the accumulated difference and the lateral force imbalance compensation coefficient.

[0101] In step 303, the dynamic coupling difference refers to the composite anomaly index after integrating longitudinal differences and lateral corrections.

[0102] In this embodiment, the accumulated difference is mapped to the longitudinal axis of the carriage according to its spatial location, and the lateral force imbalance compensation coefficient is applied to the corresponding area. The two types of parameters are fused by a spatial superposition algorithm to generate a dynamic coupled difference matrix that includes both frequency domain anomaly features and spatial distribution features.

[0103] Here is a specific example:

[0104] In the operation monitoring of a four-axle truck on a freight line, the system inputs the extracted first balance parameter component (longitudinal fluctuation characteristics at connection 1, variation coefficient 1.15) into the matching module. First, it performs a three-level wavelet decomposition on the historical longitudinal ground state to obtain low-frequency, mid-frequency, and high-frequency reference components. By calculating the energy ratio between the current component and the reference component layer by layer (formula: "current layer energy / reference layer energy"), the difference coefficients for each layer are 0.9, 1.3, and 1.1, respectively. After summing these, the cumulative longitudinal difference is obtained as 1.8 (0.9 + 1.3 + 1.1 - 1.5, where 1.5 is the normal fluctuation threshold). Simultaneously, for the second balance parameter component (lateral imbalance ratio of wheel set 2, 0.25), the matching module compares it with the lateral ground state (reference imbalance ratio 0.15) at the same location under historical curve conditions. Based on the formula "compensation coefficient = (current imbalance ratio - reference imbalance ratio) / reference imbalance ratio", the lateral force imbalance compensation coefficient is calculated to be 0.67. Finally, the longitudinal cumulative difference is mapped to the No. 2 wheel group area according to the weight distribution of the car body position (the weight coefficient of 0.7 is determined by the sensor density in this area), and after superimposing the lateral compensation coefficient, a dynamic coupling difference of 1.26 (1.8×0.7+0.67×0.3) is generated.

[0105] In this embodiment of the application, the scheme achieves accurate quantification and localization of abnormal features through hierarchical ground state matching and multi-parameter fusion, effectively distinguishing between normal operating condition fluctuations and real abnormal signals, and providing a reliable quantitative basis for subsequent early warning decisions.

[0106] To further improve the accuracy and reliability of railway freight car anomaly detection, in some embodiments, step 302: performing hierarchical matching of the first balance parameter component and the longitudinal ground state component to obtain the cumulative difference, and performing hierarchical matching of the second balance parameter component and the lateral ground state component to generate the lateral force imbalance compensation coefficient, includes:

[0107] Step 401: Divide the first balance parameter component into multiple equidistant segments according to the length of the carriage, and each equidistant segment corresponds to a longitudinal ground state component reference value in the longitudinal ground state component.

[0108] In step 401, the equidistant section refers to a detection unit that is evenly divided along the length of the carriage. The longitudinal ground state component reference value refers to the baseline pressure fluctuation amplitude of the corresponding section under historical normal operating conditions.

[0109] In this embodiment, the system divides the longitudinal axis into fixed-length detection sections according to the structural characteristics of the carriage. The length of each section is determined based on the sensor distribution density, and the system retrieves the reference pressure fluctuation data of the corresponding section from the historical database as a reference value.

[0110] Step 402: For each equidistant segment, calculate the instantaneous difference between the first equilibrium parameter component corresponding to the equidistant segment and the corresponding longitudinal ground state component reference value.

[0111] In step 402, the instantaneous difference refers to the instantaneous deviation between the real-time detected pressure fluctuation and the historical benchmark at the current moment.

[0112] In this embodiment of the application, a point-to-point subtraction operation is performed on the real-time pressure data of each segment and the corresponding historical benchmark data to calculate the instantaneous difference sequence reflecting the current degree of anomaly.

[0113] Step 403: The instantaneous differences of all equidistant sections are recursively accumulated according to the direction of carriage travel to generate a cumulative difference.

[0114] In step 403, the length direction of the cargo box refers to the longitudinal axis direction of the cargo box structure (the fixed physical extension direction from the front to the rear of the vehicle), which is static. The direction of travel of the cargo box refers to the actual movement direction of the freight truck during operation (which may deviate from the length direction of the cargo box, such as when driving on a curve), which is dynamic.

[0115] In this embodiment, a sliding window integral algorithm is used to weight and accumulate the instantaneous differences of each segment according to the direction of carriage travel. The window size is adaptively adjusted according to the train speed, and finally a cumulative difference quantity reflecting the overall abnormal trend is generated.

[0116] Step 404: Divide the second balance parameter components into left rail contact components and right rail contact components according to the wheel set position.

[0117] In step 404, the wheel set position is the logical calibration of the aforementioned support areas. Each support area is converted into a calculable parameterized position through coordinate encoding. For example: the first wheel set position = 10-12 meters from the front of the train, and the second wheel set position = 20-22 meters from the front of the train. Multiple wheel support areas refer to the collection of physical areas in the railway freight car chassis structure where all wheels contact the rails, forming the spatial basis for sensor network deployment. For example, a four-axle freight car contains four independent support areas, each corresponding to the rail contact surface of one wheelset (left and right wheels). The left rail contact component and right rail contact component refer to the pressure distribution characteristics of a single wheel set on the left and right rails, respectively.

[0118] In this embodiment of the application, the lateral pressure data is separated into independent components of the left and right rails based on the geometric position information of the wheel set, and a corresponding relationship matrix is ​​established.

[0119] Step 405: Calculate the left rail offset between the left rail contact component and the left rail reference value of the lateral ground state component, and the right rail offset between the right rail contact component and the right rail reference value of the lateral ground state component.

[0120] In step 405, the left rail offset and the right rail offset reflect the degree of deviation between the left and right rail pressures and the historical benchmark, respectively.

[0121] In this embodiment of the application, a quantified offset index is obtained by calculating the relative rate of change between the real-time left and right rail pressure values ​​and the historical benchmark values.

[0122] Step 406: Calculate the lateral force imbalance compensation coefficient based on the reverse variation characteristics of the left rail offset and the right rail offset of the same wheel set.

[0123] In step 406, the specific formation process of the reverse change characteristic is as follows: When the left rail offset of the same wheel set increases positively, its right rail offset will inevitably show a symmetrical change of decreasing negatively. This reverse change characteristic originates from the rigid connection characteristics of the railway freight car wheelset structure—the left and right wheels are fixedly connected by the axle to form a motion coupling body. Under the action of the lateral force of the track, the increase in the contact pressure of the left rail will inevitably lead to an equal decrease in the contact pressure of the right rail. By real-time monitoring of whether the algebraic sum of the left and right rail offsets of the same wheel set exceeds the elastic deformation threshold of the wheel axle structure, when it is detected that the left rail offset increases by ΔP while the right rail offset decreases by ΔP simultaneously, it is determined to be a typical reverse change characteristic. At this time, a lateral force imbalance compensation coefficient is generated based on the ratio of ΔP to the elastic coefficient of the wheel axle material. This coefficient is proportional to the absolute value of the algebraic difference between the left and right rail offsets. The lateral force imbalance compensation coefficient refers to the correction parameter calculated based on the left and right rail offset characteristics.

[0124] In this embodiment of the application, when the left and right rail offsets of the same wheel set are detected to show opposite trends, the dynamic compensation coefficient is calculated based on the ratio of the absolute value of the difference between the two to the historical maximum allowable deviation.

[0125] Here is a specific example:

[0126] In the real-time monitoring of a four-axle freight truck on a certain freight line, the system divides the truck bed longitudinally into 6 equidistant sections. In the third section (corresponding to the position of wheel set 2), the first balance parameter component was detected to be 1.3 times the reference value (calculated by the ratio of the real-time reading of the pressure sensor in this section to the historical reference value). The instantaneous difference value of this section was calculated to be 0.3 (1.3-1.0). The instantaneous differences of the first 3 sections were accumulated along the direction of travel (0.2 for the first section, 0.25 for the second section, and 0.3 for the third section) to obtain a cumulative difference of 0.75. At the same time, the system separated the second balance parameter component of wheel set 2 according to the left and right rails. The left rail contact component was measured to be 1.2 times the reference value, and the right rail was 0.9 times. The left rail offset was calculated as +20% ((1.2-1.0) / 1.0×100%), and the right rail offset was calculated as -10% ((0.9-1.0) / 1.0×100%). Based on the reverse variation characteristics of the left and right rail offsets (total difference 30%), the lateral force imbalance compensation coefficient of 1.2 (30% / 25%) is calculated according to the formula "compensation coefficient = actual total difference / safety threshold" (safety threshold is set to 25%).

[0127] In this embodiment of the application, the solution achieves precise location and quantitative evaluation of abnormal features through refined area division and dynamic parameter calculation, which improves the accuracy and reliability of abnormal detection under complex working conditions and provides a strong guarantee for the safe operation of trains.

[0128] To further improve the accuracy and reliability of railway freight car anomaly detection, in some embodiments, step 303: generating a dynamically coupled difference quantity based on the accumulated difference quantity and the lateral force imbalance compensation coefficient, includes:

[0129] Step 501: Map the accumulated difference amount to the transverse plane of the carriage according to the corresponding wheel set position.

[0130] In step 501, the transverse plane of the carriage refers to a two-dimensional projection plane that unfolds the carriage along its width, used to visually display the force distribution at the position of each wheel set.

[0131] In this embodiment of the application, the system establishes a coordinate system based on the actual layout position of the wheel set at the bottom of the carriage, maps the accumulated difference amount to the transverse plane according to the spatial position, and forms a heat map reflecting the distribution of longitudinal anomalies in the width direction of the carriage.

[0132] Step 502: Adjust the distribution weight of the mapped cumulative difference in the left and right rail directions according to the lateral force imbalance compensation coefficient.

[0133] In step 502, the left and right rail directions refer to the direction of the line connecting the left and right rails (parallel to the lateral extension direction of the track but physically connected), based on the actual rail contact surface. The physical distribution direction used to specifically calibrate the wheel set pressure data is the detection direction corresponding to the physical rail. Distribution weight refers to the process of dynamically correcting longitudinal anomalies based on the degree of lateral force imbalance.

[0134] In this embodiment of the application, a dynamic weighted algorithm based on the compensation coefficient is adopted. When the compensation coefficient for the lateral force imbalance of a certain track is greater than 1, the weight of the cumulative difference on that side is increased accordingly (weight adjustment coefficient = compensation coefficient × basic weight) to highlight the abnormal concentrated area.

[0135] Step 503: Integrate the adjusted cumulative differences corresponding to all wheel set positions to generate dynamic coupling differences.

[0136] In this embodiment of the application, the difference in the position of each wheel group after adjustment is smoothed by a spatial interpolation algorithm to generate a dynamic coupling difference, and the peak region is extracted as the key focus.

[0137] Here is a specific example:

[0138] In the operation monitoring of a four-axle freight car on a certain freight line, the system maps the calculated cumulative differences of the four wheel sets (0.9 for wheel set 1, 1.8 for wheel set 2, 1.2 for wheel set 3, and 1.0 for wheel set 4) to the transverse plane of the freight car according to their actual positions. For wheel set 2, the calculated transverse force imbalance compensation coefficient of 0.67 (calculated based on the deviation of the pressure difference between the left and right rails of this set from the historical benchmark) is corrected according to the formula "adjusted difference = original difference × (1 + compensation coefficient × orientation factor)" (the orientation factor is 0.6 for the left rail and 0.4 for the right rail). The adjusted difference on the left rail side of wheel set 2 is 1.8 × (1 + 0.67 × 0.6) = 2.52, and on the right rail side it is 1.8 × (1 + 0.67 × 0.4) = 2.28. After integrating all wheel set data, the system generates a final dynamic coupling difference of 1.85 using a weighted average algorithm (weights are allocated according to the number of sensors in each area).

[0139] In this embodiment of the application, the scheme achieves precise coupled analysis of longitudinal anomalies and lateral imbalances by combining spatial mapping and dynamic weighting, which effectively improves the accuracy of anomaly location under complex working conditions and provides reliable technical support for the safe operation of railway freight cars.

[0140] To further improve the accuracy of railway freight car loading anomaly classification, in some embodiments, step 104: determining the loading anomaly type based on the abnormal fluctuation component, combined with the current operating speed of the railway freight car and track curvature parameters, includes:

[0141] Step 601: Decompose the abnormal fluctuation component into a longitudinal feature vector and a transverse feature vector.

[0142] In step 601, the longitudinal feature vector refers to the set of parameters reflecting the abnormal fluctuation characteristics in the direction of the car's axis. The lateral feature vector refers to the set of parameters reflecting the abnormal distribution characteristics perpendicular to the direction of the car's axis. In the example of the abnormal fluctuation component before decomposition, this component is a multi-dimensional vector group containing time series data, in the form of [timestamp, pressure value, spatial location code]. The pressure value includes both the dynamic coupling pressure at the car connection point (e.g., the peak force of the coupler is 120kN±15%) and the contact pressure of the four wheel support areas (e.g., 85MPa for the left rail of wheel 1, 78MPa for the right rail of wheel 1...92MPa for the right rail of wheel 4). The longitudinal feature vector obtained after decomposition is in the form of [timestamp, longitudinal force value, car section code]. The code contains a dynamic force sequence distributed along the length of the carriage (e.g., section 1 +108kN, section 2 +95kN... section 6 +82kN). The lateral feature vector is in the form of [timestamp, left and right rail pressure difference, wheel group number], containing the real-time unbalanced force at each wheel group position (e.g., left-right rail pressure difference of wheel group 1 +7MPa, wheel group 2 -5MPa... wheel group 4 +10MPa). The longitudinal feature vector reflects the abnormal force transmission of the car body structure, and the lateral feature vector characterizes the imbalance of wheel-rail contact force distribution.

[0143] In this embodiment, principal component analysis is used to decouple the abnormal fluctuation components by direction, extract the spectral features along the length of the carriage as the longitudinal feature vector, and extract the pressure difference distribution features between the left and right rails as the lateral feature vector.

[0144] Step 602: Based on the current operating speed of the railway freight car and the track curvature parameters, the longitudinal feature vector and the lateral feature vector are corrected respectively.

[0145] In this embodiment, the longitudinal feature vector is corrected using the velocity square weighting method (correction coefficient = actual velocity square / reference velocity square), and the transverse feature vector is corrected using the inverse radius of curvature weighting method (correction coefficient = 1 + standard radius of curvature / actual radius of curvature).

[0146] Step 603: Determine the overload level code corresponding to the overload level interval based on the frequency distribution of the corrected longitudinal feature vector.

[0147] In step 603, the overload level code refers to the quantitative level identifier based on the degree of longitudinal anomaly.

[0148] In this embodiment, the main frequency component of the longitudinal feature vector is extracted by Fourier transform, and the specific overload level code is obtained by matching according to the preset "frequency-load" correspondence table.

[0149] Step 604: Determine the off-center load direction code corresponding to the off-center load direction interval based on the magnitude difference of the corrected lateral feature vector.

[0150] In step 604, the off-center load direction coding refers to the directional identifier determined based on the lateral anomaly distribution characteristics.

[0151] In this embodiment, the average left and right rail pressure difference of each wheel set in the lateral feature vector is calculated, and the off-center load direction code is generated by combining the track curvature direction and the direction discrimination algorithm.

[0152] Step 605: Match the loading anomaly type that corresponds to both the overload level code and the off-center load direction code from the preset loading anomaly classification rule library.

[0153] In step 605, loading the anomaly classification rule base refers to a predefined set of anomaly type discrimination criteria.

[0154] In this embodiment, a two-dimensional lookup table structure is used to store the combination relationship between overload level code and off-center load direction code, and the final anomaly type identifier is determined by parallel matching.

[0155] Here is a specific example:

[0156] During the operation of a four-axle freight car on a certain freight line, the system decomposes and processes the detected abnormal fluctuation components (including data showing that the pressure peak at connection point 1 exceeds the standard value by 20% and the pressure difference between the left and right rails of wheel set 2 reaches the safety threshold of 80%). First, the longitudinal feature vector is extracted through spectrum analysis, and the dominant frequency is measured to be 1.3 times the baseline value (calculated using Fast Fourier Transform). Simultaneously, the pressure difference between the left and right rails of wheel set 2 is obtained from the lateral feature vector as 1.25 times the standard value (calculated using the formula "measured difference / baseline difference"). Combined with the current vehicle speed (measured by a speed sensor as 90% of the baseline speed), a speed correction coefficient of 0.81 is calculated using the formula "correction coefficient = (actual speed / baseline speed) squared". After longitudinal feature correction, the coefficient is 1.3 × 0.81 ≈ 1.05. Then, based on the track curvature radius (obtained from the line database as 80% of the baseline radius), a curvature correction coefficient of 1.25 is calculated using the formula "curvature correction coefficient = baseline radius / actual radius". After lateral feature correction, the coefficient is 1.25 × 1.25 = 1.56. According to preset standards, a vertical overload of 1.05 times corresponds to a level 2 overload code, and a horizontal overload of 1.56 times corresponds to a right-side off-load code. Finally, the system matches the "level 2 overload combined with right-side off-load" anomaly type from the rule base, and then initiates the corresponding level of early warning procedure.

[0157] In this embodiment of the application, the scheme achieves accurate identification of loading anomaly types through multi-dimensional feature decoupling and environmental parameter compensation, effectively overcomes the interference of speed and curvature factors on the detection results, improves the accuracy of anomaly classification, and provides a reliable guarantee for the safe transportation of heavy-haul railways.

[0158] To further improve the accuracy of railway freight car anomaly feature analysis, in some embodiments, step 602: the correction of the longitudinal feature vector and the lateral feature vector based on the current operating speed of the railway freight car and the track curvature parameters, respectively, includes:

[0159] Step 701: Convert the current operating speed of the railway freight car into a speed correction factor that is consistent with the longitudinal axis direction of the car.

[0160] In step 701, the specific process of converting the current operating speed of the railway freight car into an operating speed correction coefficient is as follows: First, a speed-correction coefficient mapping function is established. This function uses the train's maximum design speed (e.g., 120 km / h) as the reference value. When the real-time speed reaches the reference value, the correction coefficient is 1.0; when it is lower than the reference value, it decreases according to a quadratic function relationship (e.g., 0.25 for 60 km / h); and when it exceeds the reference value, it increases linearly (e.g., 1.25 for 150 km / h). Simultaneously, the real-time monitoring value of the longitudinal vibration frequency of the carriage is introduced for dynamic adjustment. When the vibration frequency exceeds the safety threshold, a decay factor of 0.8 is applied to the correction coefficient. Example: When the freight car travels at 100 km / h (reference speed 120 km / h) and the vibration frequency is normal, the speed correction coefficient calculated through the quadratic function is 0.69 ((100 / 120)²≈0.69). "Consistent with the longitudinal axis of the carriage" means that the direction of action of the physical quantity is completely collinear with the longitudinal axis of the carriage (e.g., traction / braking force). The "operating speed correction factor" in this application must be defined as "in the same direction" (because speed is a parameter of the overall motion of the vehicle body). Parallel to the longitudinal axis of the carriage means that the direction of the physical quantity is parallel to the longitudinal axis of the carriage, but spatial offset is allowed (such as the longitudinal pressure distribution of multiple sets of wheels). The "first balance parameter component" in this application must be defined as "parallel to the axis" (because the pressure distribution may come from wheel sets in different spatial locations). The operating speed correction factor is an adjustment parameter that reflects the degree of influence of vehicle speed on longitudinal force characteristics.

[0161] In this embodiment of the application, the real-time vehicle speed is obtained by a speed sensor, and the current speed is mapped to the corresponding correction coefficient value according to a preset speed-correction coefficient lookup table (established based on the vehicle dynamics model). The coefficient has a non-linear positive correlation with the vehicle speed.

[0162] Step 702: Decompose the track curvature parameters of the railway freight car into the reference curvature of the straight section and the incremental curvature of the curved section, and convert the incremental curvature of the curved section into a track curvature correction coefficient corresponding to the direction of the lateral feature vector.

[0163] In step 702, the track curvature correction coefficient refers to a compensation parameter reflecting the degree of influence of the curve on the lateral force characteristics. The specific generation process is as follows: the track curvature parameters of the railway freight car are decomposed into the reference curvature of the straight segment and the incremental curvature of the curved segment; based on the reference curvature of the straight segment, a basic correction amount is generated; according to the relationship between the incremental curvature of the curved segment and the track bending radius, a dynamic adjustment coefficient is generated; the basic correction amount is multiplied by the dynamic adjustment coefficient to generate the track curvature correction coefficient associated with the direction of the second fluctuation characteristic vector. The track bending radius refers to the radius of curvature of the track centerline of the curved segment (unit: meters), which is a fundamental parameter in railway line design. It has an inverse conversion relationship with the track curvature parameter (unit: 1 / meter): track curvature parameter = 1 / track bending radius.

[0164] In this embodiment of the application, the curvature data of the current section is extracted from the line database, decomposed into straight line reference value and curve increment value, and the correction coefficient corresponding to the incremental curvature is calculated by the curvature radius conversion formula. This coefficient is proportional to the curvature increment.

[0165] Step 703: Correct the longitudinal feature vector according to the running speed correction coefficient to generate the corrected longitudinal feature vector.

[0166] In step 703, the corrected longitudinal feature vector refers to the standardized abnormal features after eliminating the influence of velocity.

[0167] In this embodiment, each component value of the original longitudinal feature vector is divided by the running speed correction coefficient to obtain standardized feature data that is independent of speed.

[0168] Step 704: Correct the lateral feature vector according to the running speed correction coefficient and the track curvature correction coefficient to generate the corrected lateral feature vector.

[0169] In step 704, the corrected lateral feature vector refers to the standardized abnormal feature after eliminating the combined effects of speed and curve.

[0170] In this embodiment, the original lateral feature vector is first corrected by speed (divided by speed correction coefficient), and then corrected by curvature (multiplied by curvature correction coefficient) to finally obtain feature data reflecting the actual load distribution. Specifically, the pressure difference between the left and right rails in the lateral feature vector is multiplied by the running speed correction coefficient to amplify the unbalanced force characteristics under high-speed conditions (e.g., original pressure difference of wheel group 1 + 7MPa × 1.25 coefficient → +8.75MPa); the direction of incremental curvature of the track curve section is identified. If it is a right turn, the pressure difference of the left rail is multiplied by (1 + track curvature correction coefficient), and the pressure difference of the right rail is multiplied by (1 - track curvature correction coefficient). If it is a left turn, the operation is reversed; the elastic deformation limit constraint of the wheel axle material (e.g., ±15MPa) is applied to the corrected pressure difference, and the excess part is compressed to the safe range proportionally; the processed pressure difference of each wheel group is regenerated into a lateral feature vector in the format of [timestamp, corrected pressure difference, wheel group number].

[0171] Here is a specific example:

[0172] During the operation of a four-axle freight truck on a certain freight line, the system detected that the current operating speed was 90% of the reference speed (obtained directly through the onboard speed measurement device). According to the speed correction formula "correction coefficient = (actual speed / reference speed) squared", a speed correction coefficient of 0.81 was calculated. Simultaneously, the current track curvature parameters were retrieved from the route database. The reference curvature for straight sections was zero, and the radius corresponding to the incremental curvature of curved sections was 80% of the reference radius. Based on the curvature correction formula "correction coefficient = reference radius / actual radius", a curvature correction coefficient of 1.25 was calculated. The extracted longitudinal feature vector value of 1.15 (calculated by the ratio of the pressure peak at connection 1 to the reference value) is corrected by speed to obtain a corrected longitudinal feature vector of 1.42 (1.15 / 0.81). The lateral feature vector value of 0.25 (calculated by the ratio of the pressure difference between the left and right rails of wheel set 2 to the reference value) is corrected by a combination of speed correction factor of 0.81 to obtain 0.31, and then multiplied by curvature correction factor of 1.25 to obtain a corrected lateral feature vector of 0.39.

[0173] In this embodiment of the application, the scheme effectively eliminates the interference of speed and curve factors on the detection results through dual environmental parameter compensation, making the abnormal feature analysis results more objective and accurate, and providing a reliable basis for subsequent abnormal classification.

[0174] Figure 2 A schematic diagram of an artificial intelligence-based early warning system for abnormal loading of railway freight cars is provided as an embodiment of this application, as shown below. Figure 2 As shown, the system includes:

[0175] The data acquisition module 21 is used to collect dynamic pressure distribution data at the connection points of the carriages and the support areas of multiple sets of wheels during the operation of railway freight cars.

[0176] The generation module 22 is used to fuse the phase offset of the dynamic pressure distribution data at the connection of the carriages and the amplitude difference of the dynamic pressure distribution data of the multiple sets of wheel support areas to generate the spatial pressure gradient characteristics of each carriage of the railway freight car.

[0177] Input module 23 is used to input the spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions into a pre-trained residual network, and output the abnormal fluctuation components between the car connection and multiple wheel support areas.

[0178] The determination module 24 is used to determine the loading anomaly type based on the abnormal fluctuation component, combined with the current operating speed of the railway freight car and the track curvature parameters, and to issue an early warning according to the early warning method of the early warning level to which the loading anomaly type belongs.

[0179] Figure 2 The aforementioned AI-based railway freight car loading anomaly early warning system can perform... Figure 1 The implementation principle and technical effects of the AI-based railway freight car loading anomaly early warning method described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the AI-based railway freight car loading anomaly early warning system described in the above embodiments have been detailed in the relevant method embodiments and will not be elaborated upon here.

[0180] In one possible design, Figure 2 The illustrated embodiment of an artificial intelligence-based early warning system for abnormal loading of railway freight cars can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0181] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0182] The processing component 32 is described above Figure 1 The embodiment describes an artificial intelligence-based method for early warning of abnormal loading of railway freight cars.

[0183] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the above-described method.

[0184] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0185] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0186] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0187] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0188] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0189] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is an artificial intelligence-based method for early warning of abnormal loading of railway freight cars.

[0190] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0191] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0192] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for early warning of abnormal loading of railway freight cars based on artificial intelligence, characterized in that, include: Collect dynamic pressure distribution data at the connection points of railway freight cars and the support areas of multiple sets of wheels during operation; By integrating the phase offset of the dynamic pressure distribution data at the car connection and the amplitude difference of the dynamic pressure distribution data of the multiple wheel support areas, a spatial pressure gradient feature of each car of a railway freight car is generated. The phase offset refers to the time delay difference of the pressure waveform at the car connection, which is used to reflect the delay of dynamic force transmission between the car bodies. The amplitude difference refers to the difference in the peak pressure of the left and right rails of the same group of wheels, which is used to characterize the degree of imbalance in the lateral distribution of load. The spatial pressure gradient feature includes a first balance parameter component parallel to the longitudinal axis of the car and a second balance parameter component perpendicular to the longitudinal axis of the car. The spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions are input into a pre-trained residual network to output the abnormal fluctuation components between the car connection and multiple wheel support areas. Based on the abnormal fluctuation components, combined with the current operating speed of the railway freight cars and the track curvature parameters, the loading anomaly type is determined, and an early warning is issued according to the early warning method corresponding to the early warning level of the loading anomaly type.

2. The method according to claim 1, characterized in that, The step of inputting the spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions into a pre-trained residual network, and outputting the abnormal fluctuation components between the car connection and multiple wheel support areas, includes: The spatial pressure gradient features and the ground state of the pressure distribution of railway freight cars under historical normal operating conditions are input into a pre-trained residual network. Through the decomposition module in the pre-trained residual network, the spatial pressure gradient features are decomposed into a first equilibrium parameter component parallel to the longitudinal axis of the car and a second equilibrium parameter component perpendicular to the longitudinal axis of the car. By using the matching module in the pre-trained residual network, the corresponding components of the pressure distribution ground state are hierarchically matched to generate dynamic coupling difference quantities. The dynamic coupling difference is weighted and fused using the fusion module in the pre-trained residual network to output the abnormal fluctuation component.

3. The method according to claim 2, characterized in that, The step involves using a matching module in a pre-trained residual network to perform hierarchical matching of the corresponding components of the pressure distribution ground state, generating a dynamically coupled difference quantity, including: In the residual network, the pressure distribution ground state of railway freight cars under historical normal operating conditions is hierarchically decomposed by the matching module into longitudinal ground state components corresponding to the first equilibrium parameter components and transverse ground state components corresponding to the second equilibrium parameter components. Hierarchical matching is performed on the first equilibrium parameter component and the longitudinal ground state component to obtain the cumulative difference. Hierarchical matching is performed on the second equilibrium parameter component and the transverse ground state component to generate the transverse force imbalance compensation coefficient. Based on the cumulative difference and the lateral force imbalance compensation coefficient, a dynamic coupling difference is generated.

4. The method according to claim 3, characterized in that, The step of performing hierarchical matching of the first equilibrium parameter component and the longitudinal ground state component to obtain the cumulative difference, and performing hierarchical matching of the second equilibrium parameter component and the lateral ground state component to generate the lateral force imbalance compensation coefficient includes: The first balance parameter component is divided into multiple equidistant segments according to the length of the carriage, and each equidistant segment corresponds to a longitudinal ground state component reference value in the longitudinal ground state component; For each equidistant segment, calculate the instantaneous difference between the first equilibrium parameter component corresponding to the equidistant segment and the reference value of the corresponding longitudinal ground state component; The instantaneous differences of all equidistant sections are recursively accumulated according to the direction of carriage travel to generate a cumulative difference. The second balance parameter component is divided into a left rail contact component and a right rail contact component according to the wheel set position; Calculate the left rail offset between the left rail contact component and the left rail reference value of the lateral ground state component, and the right rail offset between the right rail contact component and the right rail reference value of the lateral ground state component, respectively. Based on the reverse variation characteristics of the left rail offset and the right rail offset of the same wheel set, the lateral force imbalance compensation coefficient is calculated.

5. The method according to claim 3, characterized in that, The generation of the dynamic coupling difference based on the accumulated difference and the lateral force imbalance compensation coefficient includes: The cumulative difference is mapped to the transverse plane of the carriage according to the corresponding wheel set position; Based on the aforementioned lateral force imbalance compensation coefficient, adjust the distribution weight of the mapped cumulative difference in the left and right rail directions; Integrate the adjusted cumulative differences for all wheel set positions to generate a dynamically coupled difference.

6. The method according to claim 1, characterized in that, The method for determining the loading anomaly type based on the abnormal fluctuation components, combined with the current operating speed of the railway freight car and track curvature parameters, includes: The abnormal fluctuation components are decomposed into vertical feature vectors and horizontal feature vectors; Based on the current operating speed of the railway freight car and the track curvature parameters, the longitudinal feature vector and the lateral feature vector are corrected respectively; Based on the frequency distribution of the corrected vertical feature vector, determine the overload level code corresponding to the overload level interval; Based on the magnitude difference of the corrected lateral feature vector, the off-center load direction code corresponding to the off-center load direction interval is determined; The loading anomaly type is matched from the preset loading anomaly classification rule base to the loading anomaly type that corresponds to both the overload level code and the off-center load direction code.

7. The method according to claim 6, characterized in that, The correction of the longitudinal feature vector and the lateral feature vector based on the current operating speed of the railway freight car and the track curvature parameters includes: The current operating speed of the railway freight car is converted into a speed correction factor that is consistent with the longitudinal axis of the car. The track curvature parameters of railway freight cars are decomposed into the reference curvature of straight sections and the incremental curvature of curved sections, and the incremental curvature of curved sections is converted into track curvature correction coefficients corresponding to the direction of the lateral feature vector. The longitudinal feature vector is corrected according to the running speed correction coefficient to generate a corrected longitudinal feature vector; The lateral feature vector is corrected based on the operating speed correction coefficient and the track curvature correction coefficient to generate a corrected lateral feature vector.

8. An artificial intelligence-based early warning system for abnormal loading of railway freight cars, characterized in that, include: The data acquisition module is used to collect dynamic pressure distribution data at the connection points of railway freight cars and the support areas of multiple sets of wheels during operation. The generation module is used to fuse the phase offset of the dynamic pressure distribution data at the car connection and the amplitude difference of the dynamic pressure distribution data of the multiple wheel support areas to generate the spatial pressure gradient characteristics of each car of the railway freight car. The phase offset refers to the time delay difference of the pressure waveform at the car connection, which is used to reflect the delay of dynamic force transmission between the car bodies. The amplitude difference refers to the difference in the peak pressure of the left and right rails of the same group of wheels, which is used to characterize the degree of imbalance in the lateral distribution of load. The spatial pressure gradient characteristics include a first balance parameter component parallel to the longitudinal axis of the car and a second balance parameter component perpendicular to the longitudinal axis of the car. The input module is used to input the spatial pressure gradient features and the pressure distribution ground state of railway freight cars under historical normal operating conditions into a pre-trained residual network, and output the abnormal fluctuation components between the car connection and multiple wheel support areas. The determination module is used to determine the loading anomaly type based on the abnormal fluctuation component, combined with the current operating speed of the railway freight car and the track curvature parameters, and to issue an early warning according to the early warning method corresponding to the early warning level of the loading anomaly type.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement an artificial intelligence-based early warning method for abnormal loading of railway freight cars as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements an artificial intelligence-based early warning method for abnormal loading of railway freight cars as described in any one of claims 1 to 7.

Images