Intelligent detection method for unbalance loading state in operation process of railway wagon

Abstract

The invention relates to the technical field of wagon operation monitoring, in particular to an intelligent detection method for an unbalance loading state in the operation process of a railway wagon, which comprises the following steps of: acquiring an instantaneous load response track of each side axle box relative to a track in the operation process to form a left-right path differential response sequence; the method is used for reflecting wheel-rail contact abnormal distribution characteristics caused by unbalance loading. Inputting the left and right path differential response sequence into an abnormal disturbance code construction module, marking key load disturbance points by using a multi-window trajectory disturbance detection mechanism, and generating a corresponding adaptive disturbance code sequence; and outputting the unbalance loading state type of the current vehicle and the corresponding offset side direction. According to the method, the consistency judgment of the disturbance form and the physical configuration is realized, the unbalance loading grade and direction two-dimensional diagnosis result is finally output, and compared with a traditional threshold value method, the method has the advantages of low false alarm rate, high adaptability and good interpretability, and is suitable for various railway wagon marshalling structures and operation conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of freight train operation monitoring, and in particular to an intelligent detection method for a load imbalance state of a railway freight train during operation. BACKGROUND

[0002] The load state of a railway freight train during operation has a direct impact on vehicle stability, track safety and operation life. Load imbalance (i.e., uneven force on the left and right wheel sets or front and rear axle groups of the vehicle) is a key hidden factor causing track side wear, wheel lock, derailment and structural fatigue. In particular, in the scenarios of long formation, multi-cargo compartment heterogeneous loading or irregular cargo transportation, the load imbalance phenomenon is more common and hidden. Timely and accurate identification of the load imbalance state during operation is a core requirement for intelligent operation and maintenance and state supervision.

[0003] Existing railway freight train load imbalance detection technologies mainly include static weighing and distribution detection methods. Such methods rely on ground weighing systems or formation station weight ratio devices to measure the weight of each axle group of the vehicle body when the train is stationary. Some systems attempt to infer the load imbalance direction by measuring the lateral acceleration of the vehicle body or the lateral displacement of the track. However, such methods are often sensitive to track structure state, curve radius and vehicle speed changes, and can only roughly reflect the imbalance trend of the force, making it difficult to achieve axle-level and lateral quantitative identification, and there is a high false alarm rate and poor adaptability.

[0004] Existing load imbalance detection methods generally lack a high-frequency dynamic perception mechanism during operation, cannot track the perturbation response induced by load imbalance in real time, and fail to build a perturbation model with time structure and direction characteristics, making the identification results one-sided and lacking temporal logic. It is difficult to integrate train running speed and actual load configuration information for reasoning and judgment, and to eliminate disturbance artifacts and structural bias. SUMMARY

[0005] The present application provides an intelligent detection method for a load imbalance state of a railway freight train during operation, a railway freight train load imbalance state detection method with high-frequency perception, disturbance modeling and multi-source intelligent reasoning capabilities, which can accurately identify the load imbalance level and direction during train operation, and improve the safety of vehicle operation and the intelligent level of maintenance.

[0006] An intelligent detection method for a load imbalance state of a railway freight train during operation, comprising the following steps: S1, extracting an axle box dynamic load response path: setting strain sensing nodes on the left and right axle boxes of the railway freight train, and obtaining the instantaneous load response trajectory of each side axle box relative to the track during operation in real time, forming a left-right path differential response sequence, which is used to reflect the abnormal distribution characteristics of wheel-rail contact caused by load imbalance; S2, constructing asymmetric disturbance coding: inputting the left-right path difference response sequence into an abnormal disturbance coding construction module, which labels key load disturbance points by using a multi-window trajectory disturbance detection mechanism and generates a corresponding adaptive disturbance coding sequence to represent the asymmetric evolution mode in the trajectory response; S3, bias load state reasoning and judgment: performing similarity matching between the disturbance coding sequence and a pre-constructed standard bias load disturbance coding dictionary, combining the current running speed of the vehicle and the load distribution configuration data, and outputting the bias load state type and the corresponding offset side direction of the current vehicle.

[0007] Optionally, the strain sensing nodes are symmetrically arranged in the middle force area of each axle box body of the railway wagon, and each group includes three Wheatstone full-bridge strain gauges distributed in an isosceles triangle, which are respectively used to measure the strain components of the axle box in the vertical, longitudinal and transverse directions; multi-dimensional strain data of each axle box are synchronously collected through the strain sensing nodes, and the multi-dimensional strain data are input into a load solving model to obtain three-dimensional instantaneous load responses of each side axle box relative to the track.

[0008] Optionally, based on the three-dimensional instantaneous load responses, vertical load response trajectories of the left and right side axle boxes are respectively constructed in time sequence, and the vertical load response trajectories of the left and right sides are subjected to timestamp alignment processing; the difference values of the left and right side vertical load responses at the same time are calculated to form the left-right path difference response sequence, wherein a positive value in the difference response sequence indicates that the right side load is biased, and a negative value indicates that the left side load is biased.

[0009] Optionally, the abnormal disturbance coding construction module performs parallel analysis on the left-right path difference response sequence by using a multi-scale sliding window group, wherein the multi-scale sliding window group includes at least three windows with different time lengths, and the window length is adjusted according to the current running speed of the vehicle, a short window is used to capture fast disturbance when running at high speed, and a long window is used to identify slow trends when running at low speed.

[0010] Optionally, in each sliding window, candidate disturbance points in the difference response sequence are identified by using a local extreme value detection algorithm, and key load disturbance points are screened based on an adaptive threshold value, wherein the adaptive threshold value is calculated according to the statistical variance and peak factor of the sequence in the window.

[0011] Optionally, for each key load disturbance point, a disturbance feature vector is extracted, including disturbance amplitude level, disturbance duration, disturbance rising slope and disturbance frequency, and the disturbance feature vector is mapped into a fixed-length disturbance coding unit by using a binary coding rule, and the disturbance coding units are combined into an adaptive disturbance coding sequence in time sequence, wherein each coding unit includes a position identifier to reflect the time sequence relationship in the original sequence.

[0012] Optionally, a multi-level matching strategy is adopted in S3, a standard bias load disturbance coding dictionary is traversed through coding structure, candidate standard modes with the same disturbance point distribution as the current disturbance coding sequence are quickly screened, and a dynamic time warping algorithm is used to calculate the matching degree of the current sequence and each candidate standard mode, and the top three standard modes with the highest matching degree are selected as the preliminary matching result.

[0013] Optionally, the preliminary matching result is corrected in reliability based on the current running speed of the vehicle, wherein the disturbance mode changing rapidly is given priority in the high-speed running state, and the long-term trend disturbance mode is mainly analyzed in the low-speed running state, and a speed-adaptive matching degree weighting calculation model is established.

[0014] Optionally, the matching degree after weighting of the matching degree weighting calculation model is verified and calibrated in combination with the vehicle load distribution configuration data, and the mis-matching caused by the normal load distribution characteristics is excluded, wherein the load distribution configuration data includes the cargo loading position, the mass distribution ratio and the vehicle self-weight distribution parameter.

[0015] Optionally, S3 further comprises outputting a state judgment result including a bias load level and a bias direction according to the matching degree result after final calibration, in combination with the offset side direction features recorded in the disturbance coding sequence, wherein the bias load level includes normal, mild bias load, moderate bias load and severe bias load, and the bias direction includes left bias load and right bias load.

[0016] The beneficial effects of the present application are as follows: The present application realizes high-precision perception of vertical, longitudinal and transverse multi-axis strain by arranging three-directional dynamic strain sensing nodes on the axle boxes on the left and right sides of the railway wagon, constructs a least square load solution model suitable for the axle box stress scene based on the strain-stress-load inversion path, and converts the original strain signal into a physically interpretable three-dimensional instantaneous load response in real time, thereby improving the detection sensitivity of the abnormal load distribution.

[0017] The present application firstly proposes a multi-scale sliding window disturbance detection, a feature vector discrete coding and an adaptive disturbance sequence construction mechanism, standardizes the complex bias load disturbance features in the left and right path difference response sequence, adjusts the window length and the train running speed in linkage, selects the adaptive threshold of the disturbance key point, and encodes the four-dimensional disturbance features (amplitude, duration, slope and frequency) in a fixed format, so that the asymmetric evolution law and the directional variation trend of the bias load disturbance under multiple working conditions are effectively captured. The disturbance coding sequence not only has time reversibility, but also has feature comparability, improves the describability and traceability of the bias load behavior, and provides a structured input for subsequent intelligent identification.

[0018] This invention constructs a multi-source perception fusion-based off-center load state inference model by pre-building a standard off-center load disturbance coding dictionary, introducing structural indexing, dynamic time warping matching, and a multi-level confidence correction mechanism, and combining real-time vehicle operating speed and load configuration data. The speed adaptive weighting factor dynamically strengthens the matching weight of high-frequency / low-frequency disturbance modes, while the load distribution calibration mechanism eliminates mismatched results that are clearly inconsistent with the loading center of gravity, achieving consistent judgment of disturbance morphology and physical configuration. Finally, it outputs two-dimensional diagnostic results for off-center load level and direction. Compared with the traditional threshold method, it has the advantages of low false alarm rate, strong adaptability, and good interpretability, and is applicable to various railway freight car formation structures and operating conditions. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention 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 only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Fig. 1 This is a schematic diagram of the detection method flow according to an embodiment of the present invention; Fig. 2 This is a schematic diagram illustrating the construction of the perturbation coding sequence according to an embodiment of the present invention; Fig. 3 This is a schematic diagram illustrating the off-center load state reasoning and judgment in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0022] like Figs. 1-3 As shown, an intelligent detection method for off-center loading during railway freight car operation includes the following steps: S1, Extract the dynamic load response path of the axle box: Strain sensing nodes are set on the left and right axle boxes of railway freight cars to obtain the instantaneous load response trajectory of each axle box relative to the track in real time during operation, forming a differential response sequence of the left and right paths, which is used to reflect the abnormal distribution characteristics of wheel-rail contact caused by off-center loading.

[0023] S1 specifically includes the following: S11, Deploy dynamic strain sensing nodes: In the stress-bearing area at the center of each axle box of the railway freight car, two sets of dynamic strain sensing nodes are symmetrically arranged. Each set of nodes includes three Wheatstone full-bridge strain gauges distributed in an isosceles triangle pattern, used to synchronously collect strain components in the following three directions: Vertical strain ; Longitudinal strain ; Transverse strain .

[0024] The axle box is a key component of the railway freight car bogie. It is located between the axle and the bogie frame. The axle box is directly mounted on the bearings at both ends of the axle and is connected to the axle through the bearings. At the same time, it is connected to the bogie frame through the suspension.

[0025] Railway freight cars in operation: 1. Conventional vertical strain (caused by vehicle static load + track disturbance): The vehicle's own weight and load will cause the axle box to always bear a static load vertically downward. During operation, track irregularities, differences in sleeper stiffness, joint impacts, etc. will superimpose to generate dynamic vertical vibration. When the load is uneven, the left and right wheelsets will be subjected to uneven forces, which will be directly reflected in the difference in vertical strain of the left and right axle boxes.

[0026] 2. Lateral strain (caused by passing through curves or off-center loading): When a truck passes through a curve or the center of gravity shifts due to off-center loading, the force between the wheel and rail will introduce a lateral component, causing the wheelset to shift and tilt to a certain extent. The axle box will be eccentrically loaded, which will induce observable lateral strain. Under off-center loading, the lateral strain often changes in conjunction with the vertical strain.

[0027] 3. Longitudinal strain (caused by traction / braking dynamic response): When a truck starts, brakes, passes through a slope, or is coupled, the longitudinal impact force transmitted by the coupler and braking device will act on the axle box. These loads manifest as periodic or pulsed longitudinal strain peaks. If there is an off-center load, the uneven distribution of inertia will cause the longitudinal response to exhibit an asymmetric mode.

[0028] The strain calculation is as follows: Wheatstone full bridge amplifies voltage changes caused by minute resistance variations, improving the accuracy of strain measurements; When the axle box is subjected to external disturbances such as train load, track irregularities, or curves during operation, stress field changes will occur in the sensor installation area; These stresses will cause micro-elastic deformation in the metal region where the strain gauge is located; The metal grid wires of the strain gauge will stretch / compress as the substrate deforms, and its resistance will change. After the resistance change is converted into a voltage signal, it is output through a full-bridge circuit, and then amplified / sampled to restore the strain value.

[0029] S12, Multidimensional Strain Data Acquisition and Load Calculation: Simultaneous acquisition of data from all dynamic strain sensing nodes is performed using a sampling frequency of at least 1kHz to obtain a multidimensional strain data sequence. ;in, This represents the multidimensional strain data of the i-th axle box at time t. , , These are the vertical, longitudinal, and transverse strain values, respectively.

[0030] The multidimensional strain data is input into the axle box load calculation model. The three-dimensional instantaneous load response of each axle box is obtained by solving: ;in, This represents the three-dimensional load response of the i-th axle box at time t. For vertical loads, For longitudinal loads, It is a lateral load.

[0031] The strain signal acquired by the sensor is the deformation per unit length, which cannot directly reflect the load magnitude. To analyze the corresponding load, the strain needs to be converted into the actual load using the axle box load calculation model. This process includes the conversion between strain and stress, projection of force components, and load calculation, ultimately obtaining the load components of the axle box in three directions at each moment. The axle box load calculation model is a calculation module based on the inversion principles of structural mechanics and materials mechanics. Its goal is to deduce the magnitude of the external load from the strain measurement value. The specific process can be divided into the following steps: 1. Strain-stress transformation Hooke's Law, which states that the elastic modulus, strain value and stress of a material are related, is used to convert strain into stress, that is, the force per unit area.

[0032] 2. Although stress is information at a single point, the axle box, as a force-bearing unit with geometric structural characteristics, reflects the overall load condition of its load-bearing capacity through combinations of stress responses in different directions and locations. Therefore, it is necessary to consider: The orientation and location of the sensors, such as which one measures the main vertical direction and which one is on the slope; The influence of local structural geometry of the axle box, such as plate thickness and shape, on the direction of deformation; The projection of stress in different directions; This establishes a set of linear response relationships: the impact of the projection of the external load in three dimensions on each strain gauge; this relationship forms the sensitivity matrix A. The sensitivity matrix A can be modeled using finite element simulation, matching the strain response with the simulated loading of the axle box in different directions, or calibrated experimentally by sequentially loading known magnitudes onto a standard loading platform. , , Record the corresponding strain changes and fit the response sensitivity of each strain gauge to triaxial load.

[0033] 3. Using the least squares method for inversion and solving, after completing the structural mapping between stress and load, the entire problem is abstracted into a standard system of linear equations: ;in, Here, A represents the load-strain sensitivity matrix constructed from geometric modeling and experimental calibration, and F represents the three-dimensional load vector (vertical / longitudinal / lateral) to be solved. If the error is measurement or modeling, perform a least-squares fit on the system of equations to find the objective function that minimizes the following: Its least-squares analytical solution is: .

[0034] S13, Constructing the vertical load response trajectories on the left and right sides: Extracting loads in three directions to accurately determine the most critical vertical load. During the left and right differential analysis, extract the vertical load components for each axle box on the left and right sides. Construct the left vertical load response trajectory according to the time series. Response trajectory of vertical load on the right side By using timestamp alignment, we ensure that the response values ​​compared are from the same moment.

[0035] The purpose of this section is to record the changes in vertical load over time for the left and right axle boxes of railway freight cars during operation, forming a continuous and comparable time series. From the load calculation model above, the vertical load component of each axle box at each moment is extracted. Combine the vertical loads of all axle boxes on the left side into a single trajectory. The right side is merged into The two represent the force changes of the left and right wheelsets over time, respectively. Since the sampling data comes from multiple sensors, it is necessary to align the timestamps of each time point to ensure that the load data on the left and right sides are corresponding values ​​at the same time, so that they can be compared meaningfully.

[0036] S14, Calculate the differential response sequence: For each sampling time t, calculate the difference in the vertical load responses on the left and right sides, and define the differential response sequence. as follows: ;in, A value greater than 0 indicates that the load on the right side is too large, and there is a tendency for the load to be eccentrically loaded to the right. A value less than 0 indicates that the load on the left side is too large, showing a tendency to be unbalanced to the left. A value approximately equal to 0 indicates that the forces on the left and right sides are basically balanced, and the vehicle is in a balanced state. This is the left and right path differential response sequence, which is used for subsequent off-center load disturbance analysis and pattern recognition.

[0037] S2, Constructing asymmetric perturbation coding: Input the left and right path differential response sequences into the asymmetric perturbation coding construction module. The asymmetric perturbation coding construction module uses a multi-window trajectory perturbation detection mechanism to mark key load perturbation points and generate corresponding adaptive perturbation coding sequences to represent the asymmetric evolution mode in the trajectory response.

[0038] S2 is as follows: S21, Parallel analysis of multi-scale sliding window groups: Constructing a multi-scale sliding window set: ;in, For the k-th sliding window, there are different time lengths. The unit is the number of sampling points. This indicates that it contains at least three scales, including the sliding window length. With train speed Dynamic adjustment to meet: ; This is a window adjustment factor used to ensure that data segments of equivalent spatiotemporal length are extracted at different speeds. Let t be the vehicle speed at time t. When the train speed increases, the window length should decrease to ensure that the analyzed trajectory covers the same physical space distance. When the train speed decreases, the window length should increase to obtain sufficient information. This method ensures that regardless of the speed, each window analyzes the trajectory features within the same distance, thus maintaining consistency.

[0039] The above describes how multiple sliding windows are used to analyze differential load sequences in parallel at different time scales, thereby capturing the disturbance characteristics caused by off-center loading more comprehensively. A sliding window extracts a continuous data segment from time series data to analyze local features. Multiple scales mean using multiple windows of different lengths; for example, a short window observes instantaneous fluctuations, while a long window captures trend changes. Here, at least three sliding window scales are set: Short window Suitable for detecting rapid disturbances, such as instantaneous shocks; Mid-time window : Balancing the magnitude and persistence of change; Long window : Captures long-lasting biased loading trends or slowly changing disturbances.

[0040] Railway freight cars exhibit different off-center loading characteristics under different operating conditions. When running at high speed, the track impact frequency is high and the disturbance changes rapidly, requiring a shorter time window to capture these high-frequency disturbances in time. When running at low speed, the disturbance develops more slowly and the fluctuation range is larger, requiring a longer time window to identify these slow changes. Therefore, in order not to miss any possible abnormal signals, a multi-scale parallel analysis mechanism is adopted to observe each time point from multiple time spans.

[0041] S22, Screening candidate disturbance point detection and key point selection: Identify truly meaningful abnormal disturbance points from the trajectory differential response sequence, remove invalid fluctuations such as noise and jitter, and retain only key disturbance features related to off-center load.

[0042] The whole process is divided into two small steps: 1) First, identify potential disturbance points, i.e., candidate disturbance points; 2) Then, the truly important perturbation points, i.e., the key perturbation points, are selected through an adaptive method.

[0043] Candidate perturbation point identification: Local extremum detection is employed. Within each sliding window, the difference response sequence D(t) is traversed to find so-called local extrema. Extrema are those relatively prominent positions in the sequence, such as peaks or valleys. These points appear as "sharp" or "recessed" locations on the image, often representing sudden changes in stress or abnormal fluctuations. All such local extrema points are temporarily identified as candidate perturbation points; that is, they may be perturbation signals generated by off-center loading, or they may simply be normal fluctuations.

[0044] Screening of key disturbance points: An adaptive threshold is used. Instead of using a fixed threshold to determine the importance of these points, an adaptive threshold is dynamically set based on the volatility and kurtosis of the data within the current window. The adaptive threshold is determined by two indicators: Standard deviation: measures the overall strength of fluctuations within a given window; Kurometry: measures the presence of unusually sharp abrupt changes; the higher the kurtosis, the more likely an anomalous perturbation is to occur. These two values ​​are combined using weighting parameters to arrive at a dynamically adjusted detection threshold. Once the threshold is set, all candidate extreme points exceeding this threshold are filtered out and identified as critical load disturbance points. If only a fixed threshold is used, misjudgments may occur due to the different fluctuation amplitudes under different operating speeds and track conditions. However, by adopting an adaptive calculation method within a window, it is possible to automatically determine what constitutes a true anomaly based on actual trajectory fluctuations, thereby effectively avoiding false alarms or missed detections caused by changes in the environment or differences in operating conditions.

[0045] The specific calculations are as follows: For each window Differential response sequence within Perform the following operations: S221, Local Extremum Detection: Identifying a set of candidate perturbation points that satisfy the conditions for local maxima or minima. for Local extrema within the interior; S222, Adaptive threshold filtering of key perturbation points: Defining the window Internal disturbance detection threshold for: ;in, This represents the standard deviation of the difference response sequence within the window. Indicates the peak factor within the window. The weighting coefficient is set based on the vehicle structure and noise level.

[0046] The filtering rules are as follows: That is, extreme points exceeding the threshold are identified as critical load disturbance points.

[0047] S23, Perturbation Feature Vector Extraction and Encoding: Each key perturbation point detected in the previous step is transformed into a set of structured features that can be recognized, compared, and recorded by the digital system, and encoded into a standard format; complex perturbations are simplified into structured numerical representations, facilitating subsequent analysis, unifying perturbation description methods, and facilitating comparison of the similarity between multiple perturbations; specifically including the extraction and encoding of perturbation feature vectors for each key perturbation point. Instead of focusing solely on when and how large the event occurred, it extracts multiple key dimensions, known as the perturbation feature vector, which is represented as: ;in: The disturbance amplitude level is defined as follows: The level in the global amplitude distribution indicates which level the intensity of this disturbance point belongs to in the global difference sequence. It can be classified by a pre-defined amplitude range (small, medium, large, and extremely large) to reflect whether the uneven force caused by the off-center loading is significant.

[0048] The duration of the disturbance refers to the width within the oscillation envelope of the preceding and following disturbances. It indicates the time span from the appearance to the end of the disturbance and can be judged by the left and right boundaries of the oscillation envelope. A long duration may indicate stable off-center loading or structural abnormalities in the vehicle body.

[0049] The slope of the disturbance describes how quickly the disturbance rises from its inception to its maximum magnitude. A rapid rise usually indicates a sudden event or a sharp change in force. It is calculated as the slope from the first 10% of the disturbance to the peak point.

[0050] The frequency of similar disturbances occurring per unit time indicates the number of times a disturbance with similar characteristics to the current disturbance occurs per unit time. Frequent occurrences may indicate some repetitive mechanical problem or track structure influence.

[0051] To make features easier for the program to recognize and match, they are mapped to a uniform binary code format, using a predefined binary encoding template: Each feature is represented using a fixed-length binary number, for example, each feature uses 3 bits; The total encoding length is fixed, for example, 12 bits; The codes of the four features are concatenated to form a fixed-length perturbation coding unit. The structure of the coding unit is as follows: In order: , , The encoding.

[0052] S24, combine the coding units of all key perturbation points in chronological order to form a complete adaptive perturbation coding sequence: Each encoded unit carries a timestamp. Or location information, to preserve its temporal relationship in the original response sequence.

[0053] In this sequence, each coding unit retains its corresponding time point or position in the original sequence, so that it is possible to reconstruct when and with what characteristics each perturbation occurred.

[0054] S3, Off-center load state reasoning and judgment: The disturbance coding sequence is matched with the pre-built standard off-center load disturbance coding dictionary for similarity. Combined with the vehicle running speed and load distribution configuration data, the off-center load state type and corresponding offset side direction of the current vehicle are output.

[0055] S3 specifically includes the following: S31, Rapid screening of coding structure and dynamic time warping matching: In actual railway operation, although the specific values ​​of disturbance sequences under different off-center load conditions are different, they usually have certain structural characteristics in terms of the number of disturbances, the timing of occurrence, and the directional distribution. Therefore, a set of standard off-center load disturbance coding templates are pre-constructed to represent typical light, moderate, and heavy off-center load conditions. In the actual detection process, the disturbance coding sequence of the current vehicle needs to be matched with this set of standard templates to determine which known pattern is closest, thereby determining the level and direction of the off-center load.

[0056] Specifically, it includes: S311, Initial Screening Stage: This stage employs structure matching, i.e., rapid screening of candidate templates. Instead of performing complex calculations on all templates, it first uses simple structural features to quickly select a subset of standard templates similar to the current perturbation structure. These structural features include the number of perturbation points in the current perturbation sequence, the approximate time period in which these perturbation points appear, and whether the skewness is predominantly left-biased or right-biased. Specifically, the perturbation point structure is quickly matched by traversing the standard bias perturbation coding dictionary. Using the number of perturbation codes, the temporal distribution of perturbation, and the directional distribution of perturbation as indices, a set of candidate templates with similar perturbation structures is selected: This formula represents selecting those sequences from the entire standard template dictionary that are consistent with the current perturbation coding sequence. A set of templates that are structurally similar can be used as a candidate template set for the next step of fine matching. Structural similarity means that the number of perturbations is about the same, the time distribution is similar, and the offset direction is consistent.

[0057] S312, Fine-tuning Stage: Dynamic Time Warping (DTW) is used for one-by-one comparison. For candidate templates that have passed the initial screening, each one is more precisely aligned and compared with the current perturbation coding sequence. The tool used is the Dynamic Time Warping algorithm, which can compare the overall shape of two time series without requiring them to be completely synchronized. For example, two perturbation patterns may occur one after the other, but their feature order is similar. DTW can intelligently align them and measure their differences. Specifically, Dynamic Time Warping matches candidate templates... For each perturbation coding sequence in the sequence, calculate its relationship with the current perturbation coding sequence. Matching distance between: This formula represents the expression for each candidate template obtained from the initial screening. Calculate the candidate template and the current perturbation coding sequence DTW matching distance between The smaller this value, the more similar the two sequences are.

[0058] The three candidate templates with the smallest matching distance are selected as the initial matching results: This formula represents selecting the top three templates with the smallest matching degree (i.e., the DTW distance) from all candidate templates as the standard off-center loading mode most likely to match the current disturbance.

[0059] S32. In actual operation, vehicle speed directly affects the manifestation of off-center load disturbances. At high speeds, the abnormal force on the track-wheel pair caused by off-center loads will manifest as a shorter and more severe disturbance; at low speeds, the same off-center load will manifest as a slower but longer-lasting trend change. Therefore, even if two disturbance sequences are similar in shape, their meanings may be completely different if they occur at different speeds. To avoid misjudgments caused by speed differences, a speed-sensitive correction mechanism needs to be introduced to weight and adjust the reliability of the matching results, as follows: Based on the matching distance between each candidate standard pattern and the current perturbation sequence Define the original matching degree as: In the formula, the smaller the matching distance, the more similar the matches, and the higher the matching degree. The closer to 1, the greater the matching distance, and the faster the matching degree decays. Introduce the vehicle's current operating speed Construct a speed-weighted factor: ; The rule is as follows: If the current template is a fast-perturbation type, and the vehicle speed is indeed very fast, that is... ,but: Weighted amplification of matching degree; If the current template is a slow disturbance type, and the vehicle speed is also low, that is... ,but: It is also given weighted enhancement; In other cases, such as when the velocity and disturbance type do not match, no weighting is applied, and the original values ​​are retained.

[0060] in, This represents the matching distance attenuation coefficient. This represents the weighting coefficients, used to strengthen the template that matches the current velocity state. Using the operating speed threshold to distinguish between high-speed and low-speed states, the final weighted matching degree is: .

[0061] S33, Load Configuration Calibration and Mismatch Removal: Based on the vehicle's actual load configuration data, which is fundamental information describing the current vehicle loading status, the data mainly includes three aspects: first, the loading position of each cargo; second, the mass of each cargo, used to estimate the overall load offset trend; and third, the center of gravity distribution or mass weight of the vehicle's own structure, such as which part has a larger self-weight, as shown below: ;in, This indicates the loading position of the k-th storage location. This represents the mass of the k-th item. This represents the distribution weight of the vehicle's weight in the k-th segment, using empirical rules or a load calibration model. The weighted matching degree is corrected, and off-center loading patterns that are significantly inconsistent with the current load distribution are removed, outputting the final reliable matching result. .

[0062] The load calibration model is essentially a logical system that cross-validates disturbance assessments with actual load conditions. The assessment process is as follows: 1. The perturbation matching results provide a preliminary judgment: for example, if the current perturbation sequence is highly similar to a certain standard template, a conclusion of moderate right skewness is drawn, and the weighted matching degree... Relatively high.

[0063] 2. Obtain the current vehicle load configuration data L: cargo loading location, cargo weight, and whether the vehicle structure itself is unbalanced.

[0064] 3. Estimate the theoretical off-center load trend caused by the current load: Based on the total mass distribution on the left and right sides, calculate the direction of mass center of gravity offset, use an index Δload to represent the center of gravity offset, and evaluate its intensity level (slight, moderate, severe) as the theoretical state label Ltheory.

[0065] 4. Compare the disturbance matching judgment with the load theory trend: If the direction of the disturbance judgment result is consistent with the load offset trend, the matching is reliable; if the direction is opposite or the difference is large, it may be a misjudgment.

[0066] More specifically, the calibration model's correction mechanism for matching accuracy is as follows: Increase credibility when the direction is consistent: If the direction of the disturbance judgment is consistent with the direction of the load offset, the matching degree is enhanced. For example, if the initial judgment is right offset and the center of gravity of the load is indeed right offset, the credibility is increased by multiplying by the amplification factor. If the direction is inconsistent, the credibility is weakened or the object is eliminated: if the direction of the disturbance judgment is opposite to the direction of the load offset, the matching degree is weakened or the object is eliminated directly; Off-center load rating calibration: If the matching rating is severe off-center load, but the load offset is very small, the rating weight is reduced or it is labeled as a minor disturbance that is structurally reasonable.

[0067] S34, Output off-center load level and offset direction: based on the template corresponding to the maximum final matching degree. Combining the offset direction sign in the current perturbation coding sequence (such as the left or right offset marker recorded in the coding), the following state judgment result is output: 1. Off-center load level: Normal / Slight off-center load / Moderate off-center load / Severe off-center load; the judgment is based on the maximum matching degree. The labeling level of the area or corresponding template; 2. Offset direction: Left-side offload / Right-side offload; determined by a comprehensive analysis of the statistical results of the perturbation direction in the perturbation coding sequence.

[0068] The standard biased load perturbation coding dictionary is essentially a pre-built knowledge base used to store digital descriptions of perturbation behavior under typical biased load conditions. Its structure can be understood as a template library, where each template records the representation of a certain type of biased load condition in the perturbation coding space.

[0069] Each template item contains the following parts: Part 1, Coding Sequence: This is the perturbation coding sequence generated from historical data under typical off-center loading conditions. It consists of multiple perturbation coding units and reflects the number, temporal distribution, intensity level, and directionality of perturbation points.

[0070] Part Two, Corresponding Off-center Load Level Label: Indicates the severity of the off-center load corresponding to this template, namely normal, slight off-center load, moderate off-center load, and severe off-center load.

[0071] Part 3, Offset Direction Label: This indicates that the disturbance sequence is mainly characterized by left-side or right-side off-center loading.

[0072] Disturbance type: This is an optional field used to indicate whether the template is suitable for high-speed impact or low-speed trend disturbance scenarios.

[0073] Part Four, Statistical Attribute Summary, facilitates rapid matching, including the total number of disturbance points, time span, average amplitude level, and main disturbance time periods.

[0074] The construction process of the standard off-center perturbation coding dictionary mainly involves three steps: 1. Data collection and labeling: Collect a large amount of data containing off-center loading behavior from the existing railway freight car operation history. These data are required to be labeled with specific off-center loading levels and offset directions by manual or empirical models, including real operating data under different load structures, speed conditions, and track environments.

[0075] 2. Apply the S1–S2 process in the application scheme to generate a perturbation coding sequence. For each piece of original running data, extract the left and right difference responses, perform multi-window perturbation detection, feature vector extraction and encoding according to the aforementioned S1 and S2 steps; generate a complete perturbation coding sequence and record information such as timestamps and feature point positions.

[0076] 3. Through cluster analysis, manual screening, or comparison based on experience templates, remove perturbation patterns that are too repetitive or too similar, select representative perturbation coding sequences as typical patterns, and manually or automatically assign labels to each template: off-center load level, offset direction, applicable speed range, etc., ultimately forming a structured dictionary that can be stored in multiple formats.

[0077] During the detection phase, the disturbance coding sequence of the current vehicle needs to be compared with all templates in the dictionary. The process is as follows: 1. Fast structure matching (dictionary indexing) 2. Coding Sequence Similarity Calculation (DTW) 3. Select the template with the highest matching degree.

[0078] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0079] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligently detecting the unbalanced load state of a railway wagon during operation, characterized in that, The method comprises the following steps: S1, a strain sensing node is arranged on the axle box body of the left and right sides of the railway wagon, and the instantaneous load response trajectory of each side axle box relative to the track during operation is obtained in real time, a left-right path differential response sequence is formed, and the abnormal distribution characteristics of wheel-rail contact caused by the unbalanced load are reflected; S2, the left-right path differential response sequence is input into an abnormal disturbance coding construction module, the abnormal disturbance coding construction module labels the key load disturbance points by using a multi-window trajectory disturbance detection mechanism, and generates a corresponding adaptive disturbance coding sequence to represent the asymmetric evolution mode in the trajectory response; S3, the disturbance coding sequence is matched with a pre-constructed standard unbalanced load disturbance coding dictionary, the speed and load distribution configuration data of the vehicle are combined, and the unbalanced load state type and the corresponding offset side direction of the current vehicle are output.

2. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 1, characterized in that, The strain sensing node is symmetrically arranged in the middle stress area of each axle box body of the railway wagon, each group includes three Wheatstone full-bridge strain gauges arranged in an isosceles triangle, which are used to measure the strain components of the axle box in the vertical, longitudinal and transverse directions; the multi-dimensional strain data of each axle box is synchronously collected through the strain sensing node, and the multi-dimensional strain data is input into a load solving model to obtain the three-dimensional instantaneous load response of each side axle box relative to the track.

3. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 2, characterized in that, Based on the three-dimensional instantaneous load response, the vertical load response trajectories of the left and right side axle boxes are constructed in time sequence respectively, and the vertical load response trajectories of the left and right sides are timestamped and aligned; The difference values of the left and right vertical load responses at the same time are calculated to form the left-right path differential response sequence, wherein the positive value in the differential response sequence indicates that the right side load is larger, and the negative value indicates that the left side load is larger.

4. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 1, characterized in that, The abnormal disturbance coding construction module adopts a multi-scale sliding window group to perform parallel analysis on the left-right path differential response sequence, wherein the multi-scale sliding window group includes at least three windows with different time lengths, and the window length is adjusted according to the current running speed of the vehicle, a short window is used to capture fast disturbance when running at high speed, and a long window is used to identify slow trends when running at low speed.

5. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 4, characterized in that, In each sliding window, candidate disturbance points in the differential response sequence are identified by a local extreme value detection algorithm, and key load disturbance points are screened based on an adaptive threshold, and the adaptive threshold is calculated according to the statistical variance and peak factor of the sequence in the window.

6. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 5, characterized in that, For each key load disturbance point, its disturbance feature vector is extracted, including disturbance amplitude level, disturbance duration, disturbance rising slope and disturbance frequency, and the disturbance feature vector is mapped to a fixed-length disturbance coding unit by using a binary coding rule, and the disturbance coding units are combined into an adaptive disturbance coding sequence in time sequence, wherein each coding unit includes a position identifier to reflect its time sequence relationship in the original sequence.

7. The intelligent method for detecting the unbalance load state of a railway wagon during operation according to claim 1, characterized in that, The S3 adopts a multi-level matching strategy, traverses the standard bias disturbance coding dictionary through the coding structure, quickly screens the candidate standard modes with the same disturbance point distribution as the current disturbance coding sequence, and then calculates the matching degree of the current sequence and each candidate standard mode by using the dynamic time warping algorithm, and selects the top three standard modes with the highest matching degree as the preliminary matching result.

8. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 7, characterized in that, The preliminary matching result is corrected in reliability based on the current running speed of the vehicle, wherein the fast-changing disturbance mode is given priority in the high-speed running state, and the long-term trend disturbance mode is analyzed in the low-speed running state, and a speed-adaptive matching degree weighting calculation model is established.

9. The intelligent detection method for the unbalance load state of a railway wagon during operation according to claim 8, characterized in that, The matching degree after weighting by the matching degree weighting calculation model is verified and calibrated in combination with the vehicle load distribution configuration data to exclude the mis-matching caused by the normal load distribution characteristics, wherein the load distribution configuration data includes the cargo loading position, mass distribution ratio and vehicle self-weight distribution parameters.

10. The intelligent method for detecting the unbalance load state of a railway wagon during operation according to claim 9, characterized in that, The S3 also includes outputting the state judgment result including the bias level and the offset direction according to the final calibrated matching degree result in combination with the offset side direction features recorded in the disturbance coding sequence, wherein the bias level includes normal, mild bias, moderate bias and severe bias, and the offset direction includes left bias and right bias.

Images