A ballast shoulder track bed mechanical state intelligent detection method and system using a tamping device

Abstract

The application discloses a kind of ball shoulder track bed mechanical state intelligent detection method and system using ramming device, comprising the following steps: obtaining the vibration acceleration signal of tamper bottom plate, tamper bottom plate and ball shoulder track bed interaction force signal, vertical relative displacement of operation car running part and sleeper and the lateral displacement of sleeper;The vibration acceleration signal, force signal and displacement signal collected are sliced and filtered, respectively to obtain the sliced vibration acceleration signal, sliced force signal and sliced displacement signal after filtering;The sliced vibration acceleration signal, sliced force signal and sliced displacement signal after filtering are simultaneously input into the pre-trained ball shoulder track bed mechanical state intelligent detection model, to obtain the real ball shoulder track bed supporting stiffness, ball shoulder track bed compactness and rail row lateral resistance, the method can intelligently detect the mechanical state of ball shoulder track bed after ramming operation, with the advantages of high efficiency, intelligence, low cost and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mechanical state detection of railway ballast track bed, and particularly relates to a ballast shoulder track bed mechanical state intelligent detection method and system using a tamping device. BACKGROUND

[0002] By the end of 2024, the railway operating mileage in China has reached 162,000 kilometers, among which the ballast track bed track has become the main part of the railway track with an absolute advantage, playing an irreplaceable backbone role. As an important part of the ballast track, the granular track bed will appear broken, wear and tear and track bed degradation under the coupling effect of long-term train dynamic load and complex geology, natural environment, etc. The uneven settlement caused by the continuous accumulation of plastic deformation of the track bed changes the line geometry, reduces the smoothness of the line, and seriously affects the safety of train operation.

[0003] Tamping operation is one of the necessary procedures for ballast track maintenance, and is also an effective means to improve the compactness of the ballast shoulder of the track bed after tamping operation and the lateral stability of the track bed. However, in the research of ballast track bed maintenance, researchers pay more attention to the effect of tamping and stabilization operation, and ignore the important function of tamping operation which can quickly restore the lateral resistance of the track bed after tamping operation. In fact, the ballast shoulder area of the ballast track bed has an important influence on maintaining the lateral resistance of the track bed and maintaining good track geometry. However, the current tamping operation mostly relies on engineering experience and uses uniform operation parameters throughout the line, which cannot obtain the mechanical state of the ballast shoulder track bed and the whole track bed before and after tamping operation in time, resulting in the blindness of the existing maintenance operation. Therefore, it is urgent to propose an intelligent detection method and system for the mechanical state of the ballast shoulder track bed according to the characteristics of the tamping operation of the railway large machinery, to provide key basis and technical support for the intelligent maintenance of the railway. SUMMARY

[0004] The technical problem to be solved by the present application is that the existing railway large-scale mechanical tamping operation is blind, and the mechanical state of the ballast shoulder track bed after operation cannot be obtained in time, resulting in that the evaluation of operation effect depends on engineering experience or expert judgment, and the objectivity is insufficient. The purpose of the present application is to establish a multi-index detection system based on the fusion of acceleration measurement, force measurement and displacement measurement, to obtain various response signals generated during tamping operation, and then to establish an intelligent detection model of the mechanical state of the ballast shoulder track bed based on a hybrid expert agent and pre-train it on a simulation data set. Secondly, the model is fine-tuned using field measurement data to obtain a trained intelligent detection model of the mechanical state of the ballast shoulder track bed. Finally, the acceleration signal, force signal and displacement signal measured during the actual tamping operation are input to obtain the real mechanical state of the ballast shoulder track bed after operation. The present application can realize data fusion of various response signals, intelligent detection of track bed supporting stiffness, track bed compactness and rail row lateral resistance, thereby providing a basis for evaluating tamping operation effect and improving the intelligent level of railway operation and maintenance.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] In the first aspect, the present application provides an intelligent detection method for the mechanical state of a ballast shoulder track bed using a tamping device, comprising:

[0007] S1, obtaining the vibration acceleration signal of the tamping device bottom plate, the vibration acceleration signal including vertical vibration acceleration signal and lateral acceleration signal, which are collected by an acceleration sensor and a data acquisition system;

[0008] S2, obtaining the interaction force signal between the tamping device bottom plate and the ballast shoulder track bed, the force signal including vertical interaction force and lateral interaction force signal, which are collected by a thin film pressure sensor and a data acquisition system;

[0009] S3, obtaining the vertical relative displacement between the tamping vehicle running part and the sleeper and the lateral displacement of the sleeper, the displacement signal being collected by a laser displacement sensor, a CCD camera and a data acquisition system;

[0010] S4, slicing and filtering the collected vibration acceleration signal, force signal and displacement signal to obtain the filtered sliced vibration acceleration signal, sliced force signal and sliced displacement signal respectively;

[0011] S5, inputting the filtered sliced vibration acceleration signal, sliced force signal and sliced displacement signal into the pre-trained intelligent detection model of the mechanical state of the ballast shoulder track bed to obtain the real mechanical state of the ballast shoulder track bed, the mechanical state of the ballast shoulder track bed including track bed supporting stiffness, track bed compactness and rail row lateral resistance.

[0012] Optionally, in step S1, the vibration acceleration signal of the tamper plate is acquired, specifically including: installing double-axis acceleration sensors above the tamper plates on the left and right sides of the working vehicle; installing a thin film pressure sensor inside the tamper plate; setting a threshold value of the interaction force between the tamper plate and the ballast shoulder track bed delta ; when the tamper plate falls, if the interaction force between the tamper plate and the ballast shoulder track bed F is greater than the preset threshold value delta , the vibration acceleration signal is synchronously collected at this time; when the tamper plate is lifted, if the interaction force between the tamper plate and the ballast shoulder track bed F is less than the preset threshold value delta , the signal collection is stopped.

[0013] Optionally, in step S3, the vertical relative displacement between the running gear axle box of the working vehicle and the sleeper and the lateral displacement of the sleeper are acquired, specifically including:

[0014] installing three laser displacement sensors and one CCD camera on the left and right sides of the running gear area of the working vehicle, wherein the positions of the laser displacement sensors on each side substantially correspond to the upper sides of the three sleepers, the bases of the laser displacement sensors are installed in movable sliding grooves, and the movement can be controlled by a servo hydraulic device;

[0015] the CCD camera is fixedly installed on the running gear frame or the detection beam of the working vehicle perpendicular to the track bed, can cover the shooting range of the three sleepers, and is calibrated by pasting a calibration plate at the end of the sleeper or using a known fastener spacing, the parameters that need to be calibrated include a camera intrinsic matrix K , a rotation matrix R , a translation vector t and a scaling coefficient α :

[0016] (1)

[0017] wherein, f x and f y are focal lengths (in pixels); c x and c y are principal point coordinates (usually the center of the image);

[0018] (2)

[0019] wherein, u v are the pixel coordinates of a certain point on the image, and the corresponding normalized camera coordinates are ( x , y , 1); ​

[0020] (3)

[0021] wherein, x c ,y c , z c are camera coordinates, and X w ,Y w , Z w are world coordinates;

[0022] (4)

[0023] wherein, D r is the actual distance between two points on the calibration plate, D p is the corresponding pixel distance in the image;

[0024] The laser displacement sensor emits laser to the upper surface of the end of the sleeper, the CCD camera captures the image and inputs to the vision measurement system to track the position of the laser point, when the laser point is not on the sleeper, the vision measurement system measures the distance between the laser point and the sleeper, and controls the servo hydraulic device to adjust the position of the laser displacement sensor, to ensure that the laser displacement sensor is in the correct measurement range;

[0025] After the position adjustment is completed, the working vehicle starts working, the tamper falls down, and when the interaction force between the bottom plate and the ballast shoulder track bed F is greater than the preset threshold delta , the laser displacement sensor starts to collect the vertical relative displacement between the axle box or the detection beam and the sleeper, the vision measurement system tracks the laser point to test the lateral displacement of the sleeper; the tamper is lifted up, and when the interaction force between the bottom plate and the ballast shoulder track bed F is less than the preset threshold delta , the signal collection is stopped.

[0026] Optionally, in step S4, the collected vibration acceleration signal, force signal and displacement signal are sliced and filtered, specifically including:

[0027] The vibration acceleration signal, force signal and displacement signal collected in each tampering operation process (falling-down-tampering-lifting-up) are resampled, and the polyphase FIR filtering method (window function weighted sinc interpolation) of “interpolation + low-pass filtering + decimation” is adopted to unify the sampling frequency of the signals f :

[0028] (5)

[0029] wherein, h [ k ] is a low-pass filter coefficient, x [·] is an original signal, L / M is a target sampling rate conversion ratio, which can be calculated by f / fs , wherein f s is an original sampling frequency;

[0030] The resampled vibration acceleration signal, force signal and displacement signal are band-stop filtered to exclude interference of the original tamper excitation frequency:

[0031] (6)

[0032] wherein, x ( t ) is an input signal, y ( t ) is a filtered signal, and HBS is a filter operator, and a corresponding second-order band-stop filter difference equation can be determined by the following formula:

[0033] (7)

[0034] wherein, b 0、 b 1、 b 2、 a 1 and a 2 are filter coefficients;

[0035] The filtered vibration acceleration signal, force signal and displacement signal are subjected to slicing processing, the window length is set to s , the overlap rate is set to p , and the number of samples after slicing N is determined by the following formula:

[0036] (8)

[0037] wherein, L x is the length of the input signal.

[0038] Optionally, in step S5, the pre-trained ballast shoulder track bed mechanical state intelligent detection model specifically comprises:

[0039] A simulation analysis model of a compactor-ballast track bed was established based on the DEM-MBD coupling method. The simulation analysis model includes components such as the compactor base plate, hydraulic connecting rod, eccentric vibrator, rails, fasteners, sleepers, and track bed. The ballasted track bed is built based on DEM, and polyhedra are used to simulate ballast particles. The compactor base plate, rails, and sleepers are simulated using rigid body elements, while the hydraulic connecting rod and fasteners are simulated using spring-damping elements. The eccentric excitation force is controlled by a loading function, as shown in the following formula:

[0040] (9)

[0041] In the formula, m1 and m2 are the masses of the two eccentric wheels, respectively;

[0042] Different initial mechanical states of the track bed are preset, including different ballast shoulder support stiffness, ballast shoulder track bed compaction and track panel lateral resistance; then, vibration acceleration signals of the compactor base plate, interaction force signals between the compactor base plate and the ballast shoulder track bed, vertical relative displacement signals between the running gear of the work vehicle and the sleeper, and sleeper lateral displacement signals are obtained under different excitation frequencies, compaction times and compaction times.

[0043] Record the mechanical state of the ballast shoulder track bed in the simulation analysis model according to the preset slice data window length and step size, obtain the track bed support stiffness, track bed compaction and track panel lateral resistance corresponding to each slice sample, and use them as labels; construct a simulation dataset based on the simulation analysis model, and divide it into training set, validation set and test set according to the proportion;

[0044] A smart detection model for the mechanical state of ballast shoulder track is established based on a hybrid expert agent. This model comprises multiple expert networks and a gating network. Each expert network is an independent deep neural network model responsible for processing different types of signals, mainly consisting of an input layer, convolutional blocks, residual blocks, fully connected layers, and an output layer. The gating network adaptively allocates weights and selects experts based on the input. The final output is the weighted sum of the expert outputs according to the gating weights, determined by the following formula:

[0045] (10)

[0046] In the formula, g k ( x ) is the first k An expert's representation of the input signal. f k ( x ) for gating network to the first k The weights assigned to each expert This is a predicted value for the mechanical state of the track bed;

[0047] The established ball shoulder track bed mechanical state intelligent detection model is trained on a simulation data set, a loss function of each task is MSE, a total loss function of multiple tasks is accumulated and summed, seen formulas (11)-(12), model parameters are updated on a test set, model parameters are verified on a verification set, and finally a test set is used to test model effects;

[0048] (11)

[0049] (12)

[0050] In the formula, T is the number of tasks, y i is a real label value, is a model prediction value, lambda t is a task allocation weight;

[0051] The ball shoulder track bed mechanical state intelligent detection model trained on the simulation data set is migrated to a field measured data set, model fine tuning is performed on limited test data, and finally a pre-trained ball shoulder track bed mechanical state intelligent detection model is obtained.

[0052] Optionally, the established ball shoulder track bed mechanical state intelligent detection model is trained on a simulation data set, and specifically includes the following steps:

[0053] The expert network 1 is responsible for processing vertical force signals and lateral force signals, and respectively performing fast Fourier transform on the vertical force and the lateral force to obtain frequency domain information, only taking real number domain results, and aligning the time domain length of the force signals, and zero padding the insufficient, so that the dimension of the input signal is 4;

[0054] The expert network 2 is responsible for processing vertical vibration acceleration signals and lateral vibration acceleration signals, and respectively performing fast Fourier transform on the vertical vibration acceleration and the lateral vibration acceleration to obtain frequency domain information, only taking real number domain results, and aligning the time domain length of the acceleration signals, and zero padding the insufficient, so that the dimension of the input signal of each measuring point is 4;

[0055] The expert network 3 is responsible for processing vertical relative displacement signals and lateral displacement signals of the track sleeper and the running gear of the working vehicle, and respectively performing fast Fourier transform on the vertical relative displacement signals and the lateral displacement signals to obtain frequency domain information, only taking real number domain results, and aligning the time domain length of the acceleration signals, and zero padding the insufficient, so that the dimension of the input signal of each measuring point is 4;

[0056] The expert network 4 is responsible for simultaneously processing vibration acceleration signals, displacement signals and force signals, and performing fast Fourier transform on the input to enhance frequency domain information;

[0057] Finally, Gaussian noise is randomly injected to all input signals before inputting into the expert network, as shown in the following formula:

[0058] (13)

[0059] where SNR is the signal-to-noise ratio, P s is the power of the signal, P n is the power of the noise.

[0060] Optionally, model fine-tuning is performed using limited test data, specifically including:

[0061] The field measured data set is processed according to the same processing steps as the simulation data set, and the track bed supporting stiffness, track bed compactness and track panel lateral resistance are measured immediately after each tamping process in the field as data tags, which correspond to the last slice sample;

[0062] The field measured data set is divided into a training set and a test set, and during model fine-tuning, all weight parameters of the expert network are frozen, i.e., the weight parameters of the expert network are not updated, and only the parameters of the gating network are updated;

[0063] Before model fine-tuning, the input signal is subjected to fast Fourier transform for frequency domain information enhancement, but random noise is not added, and the loss function is MSE, and the parameters of the gating network are updated on the training set;

[0064] Finally, the model fine-tuning effect is verified on the test set.

[0065] In a second aspect, the present application provides a ball shoulder track bed mechanical state intelligent detection system using a tamping device, comprising:

[0066] An acceleration measurement module is configured to obtain vibration acceleration data of a tamper base plate, and the vibration acceleration data of the tamper base plate is vertical and lateral acceleration data of the tamper base plate acting on the track bed during the process;

[0067] A force measurement module is configured to obtain interaction force data of the tamper base plate and the ball shoulder track bed, and the interaction force of the tamper base plate and the ball shoulder track bed is vertical and lateral force data of the tamper base plate acting on the track bed during the process;

[0068] A displacement measurement module is configured to obtain vertical relative displacement of a working vehicle running gear and a sleeper and lateral displacement of the sleeper, and the vertical relative displacement of the working vehicle running gear and the sleeper and the lateral displacement of the sleeper are displacement data of the tamper base plate acting on the track bed during the process;

[0069] A servo control module is configured to adjust the laser displacement sensor to a proper position according to the result of the vision measurement system, so as to ensure that the laser point is above the surface range of the end of the sleeper;

[0070] A data preprocessing module is configured to perform resampling, slicing and filtering on the measured data, and construct a data set;

[0071] A model pre-training module is configured to construct a hybrid expert agent model, and train the model on the data set to obtain optimal pre-training parameters;

[0072] A model fine-tuning module is configured to fine-tune the weight parameters of the gating network, and obtain a final ballast shoulder track bed mechanical state intelligent detection model.

[0073] A data detection module is configured to input the field measured data into the ballast shoulder track bed mechanical state intelligent detection model, obtain the detection values of the track bed supporting stiffness, the track bed compactness and the rail row lateral resistance, and perform visualization.

[0074] In a third aspect, the present application provides a non-transitory computer readable storage medium, which is used to store computer instructions, and the computer instructions are executed by a processor to implement the ballast shoulder track bed mechanical state intelligent detection method using the tamper device.

[0075] In a fourth aspect, the present application provides an electronic device, which comprises a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device implements the ballast shoulder track bed mechanical state intelligent detection method using the tamper device.

[0076] Compared with the prior art, the present application has the following advantages and beneficial effects: the traditional track bed detection method needs to remove fasteners and pads, take out ballast and load a jack, etc., and the present application can directly use the existing tamper device to obtain response signals in the operation process, so as to realize intelligent detection of the state of the ballast shoulder track bed after operation, and greatly improve the detection efficiency; in addition, the detection system can be directly installed on a large railway machine, and the detection result is fed back to the technical personnel in real time, so as to determine whether the track bed state meets the requirements, thereby guiding the formulation of the next maintenance plan.

[0077] The advantages of the additional aspects of the present application will be more apparent in the description part below, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0079] Figure 1 A flow chart of a ballast shoulder track bed mechanical state intelligent detection method using a tamping device is provided for the embodiments of the present application.

[0080] Figure 2 A tamping device bottom plate vibration acceleration sensor arrangement schematic diagram is provided for the embodiments of the present application.

[0081] Figure 3 A thin film pressure sensor arrangement schematic diagram is provided for the embodiments of the present application.

[0082] Figure 4 A tamping device-ballast track bed simulation analysis model schematic diagram established based on a DEM-MBD coupling method is provided for the embodiments of the present application.

[0083] Figure 5 A ballast shoulder track bed mechanical state intelligent detection model architecture diagram is provided for the embodiments of the present application.

[0084] Figure 6 A specialist network 1 architecture schematic diagram is provided for the embodiments of the present application.

[0085] Figure 7 A specialist network 2 architecture schematic diagram is provided for the embodiments of the present application.

[0086] Figure 8 A specialist network 3 architecture schematic diagram is provided for the embodiments of the present application.

[0087] Figure 9 A specialist network 4 architecture schematic diagram is provided for the embodiments of the present application.

[0088] Figure 10 A displacement measurement system arrangement diagram is provided for the embodiments of the present application.

[0089] Figure 11 A laser displacement sensor installation schematic diagram is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0090] Embodiments of the present application are described in detail below with reference to several specific embodiments, examples of which are illustrated in the accompanying drawings, wherein the same or similar elements are denoted by the same or similar reference signs throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0091] Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the use of the phrase "comprising" in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or coupling. The phrase "and / or" used herein includes any one of the associated listed items and all combinations thereof.

[0092] Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as that generally understood by those of ordinary skill in the art to which the present application belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have meanings consistent with those in the context of the prior art, and should not be interpreted with idealized or overly formal meanings unless defined as such.

[0093] For the convenience of understanding the embodiments of the present application, the following will be further explained with reference to several specific embodiments as examples in conjunction with the accompanying drawings, and each embodiment does not constitute a limitation on the embodiments of the present application.

[0094] Embodiment 1

[0095] The present application provides a technical solution: a ballast shoulder track bed mechanical state intelligent detection method using a tamping device, as shown in Figure 1 ;

[0096] Step 1, obtain the vibration acceleration signal of the tamping device bottom plate, the vibration acceleration signal includes vertical vibration acceleration signal and transverse acceleration signal, which are collected by acceleration sensor and data acquisition system;

[0097] In this embodiment, two double-axis acceleration sensors are respectively installed above the tamping device bottom plates on the left and right sides of the working vehicle, as shown in Figure 2 , the vibration acceleration signals perpendicular to the tamping device bottom plate direction and parallel to the sleeper direction can be tested respectively;

[0098] A layer of iron pad is inserted between the base plate of the compactor and the vibration device. Two-dimensional thin-film pressure sensors are embedded inside the iron pad. Figure 3 As shown, the total force is the average value of the outputs of all pressure sensors;

[0099] (1)

[0100] (2)

[0101] In the formula, N and M are the number of vertical and horizontal pressure sensors deployed, respectively. In this embodiment, they are taken as 8 and 3, respectively.

[0102] Set the threshold for the interaction force between the base plate and the ballast shoulder track bed. delta As the compactor falls, the interaction force between the base plate and the ballast shoulder track bed... F Greater than the preset threshold delta At this time, vibration acceleration signals are simultaneously acquired; the compactor is raised, and the interaction force between the base plate and the ballast shoulder track bed is measured. F Less than the preset threshold delta Stop signal acquisition when the signal is acquired.

[0103] Step 2: Obtain the interaction force signal between the compactor base plate and the ballast shoulder track bed. The force signal includes vertical force and lateral force signals, which are uniformly collected by a membrane pressure sensor and a data acquisition system.

[0104] Step 3: Obtain the vertical relative displacement between the running gear of the work vehicle and the sleeper, as well as the lateral displacement of the sleeper. The displacement signals are collected by a laser displacement sensor, a CCD camera, and a data acquisition system.

[0105] Step 4: Slice and filter the collected vibration acceleration signal, force signal and displacement signal to obtain the filtered slice vibration acceleration signal, slice force signal and slice displacement signal respectively;

[0106] Vibration acceleration, force, and displacement signals collected during each compaction operation (lowering-compacting-lifting) are resampled. A polyphase FIR filtering method (window function weighted sinc interpolation) combining interpolation, low-pass filtering, and decimation is used to unify the signal sampling frequency. f :

[0107] (3)

[0108] In the formula, h [ k [ ] represents the coefficients of the low-pass filter. x [·] represents the original signal. L / MThe conversion ratio for the target sampling rate can be calculated as f / fs wherein f s is the original sampling frequency;

[0109] The resampled vibration acceleration signal, force signal and displacement signal are band-stop filtered to exclude the interference of the original tamper excitation frequency:

[0110] (4)

[0111] wherein x ( t ) is the input signal, y ( t ) is the filtered signal, H BS is the filter operator, and the corresponding second-order band-stop filter difference equation can be determined as follows:

[0112] (5)

[0113] wherein b 0、 b 1、 b 2、 a 1 and a 2 are filter coefficients;

[0114] The filtered vibration acceleration signal, force signal and displacement signal are sliced, the window length is set as s , the overlap ratio is set as p , and the number of samples after slicing N is determined as follows:

[0115] (6)

[0116] wherein L x is the length of the input signal;

[0117] Step 5: A tamper-ballast bed simulation analysis model is established based on the DEM-MBD coupling method, which includes a tamper bottom plate, an oil pressure connecting rod, an eccentric exciter, a rail, a fastener, a sleeper and a ballast bed, wherein the ballast bed is established based on DEM, polyhedron is used to simulate ballast particles, and the material parameters are shown in Table 1; in this example, the thickness of the ballast bed is 350 mm, the width of the top surface of the ballast bed is 3600 mm, and the Hertz-MindlinNassauer Kuna (No slip) constitutive model is used for the contact force update calculation between the ballast and the ballast and between the ballast and the sleeper;

[0118] Table 1 DEM model parameters

[0119] parameter steel rail sleeper ballast Young's modulus (MPa) 2.059 x 10 5 ]]> 2.0 x 10 4 ]]> 7.5 x 10 3 ]]> Density (kg / m 3 )]]> 7860 2240 2600 Poisson's ratio 0.3 0.25 0.18 coefficient of static friction 0.4 0.85 0.78

[0120] The compactor base plate, rails, and sleepers are simulated using rigid body elements, while the hydraulic linkages and fasteners are simulated using spring-damping elements. Figure 4 As shown, the torque equation of the compactor during the entire operation is as follows:

[0121] (9)

[0122] In the formula, F 1 represents the thrust or pull force of the lifting cylinder. for t The angle between the force applied by the hydraulic cylinder and the tangent to line segment BC is given by G, where G is the weight of the compactor (1.06 kN) and m is the mass of the compactor (approximately 108 kg). a t for t The constant acceleration of the compactor For time t, the first time between the ballast and the base plate i Contact force at each contact point N for t The number of contacts between the base plate and the ballast at any given time. However, it's important to note that during the lowering and lifting phases, the number of contacts between the base plate and the ballast is zero. =0.

[0123] The eccentric excitation force is controlled by a loading function, as shown in the following formula:

[0124] (10)

[0125] In the formula, m1 and m2 are the masses of the two eccentric wheels, respectively;

[0126] The excitation force is decomposed along the X and Y axes, as shown in the following equation:

[0127] (11)

[0128] (12)

[0129] In the formula, F x The component of the excitation force along the X-axis is... F y This represents the component of the excitation force along the Y-axis.

[0130] Due to the complex mechanical behavior of the tamper and the movement process of the hydraulic cylinder cannot be simulated in the DEM module, a high-fidelity simulation model of the tamper is established in the MBD module by using the multi-body dynamics related theory and combining the movement trajectory of the hydraulic cylinder. The calculation parameters are shown in Table 2. In order to accurately simulate the pushing force and pulling force of the lifting cylinder, a translational pair is first applied between the cylinder barrel and the piston rod, and then a spring damper is used to simulate the constant force. The spring damper stiffness is 450 N / mm, and the damping is 80 N·s / mm. The specific movement settings and load application process are as follows.

[0131] (1) The descending stage: the operation time is set to 0-0.4 s. From 0 to 0.1 s, the axial force of a single spring damper is uniformly increased from 0 kN to F 1 / 2. F From 0.1 to 0.4 s, the constant axial force of a single spring damper is set to 1 / 2. Since the ballast and the bottom plate are not in contact at this stage, no excitation force is applied.

[0132] F (2) The tampering stage: the operation time is set to 0.4-1.6 s, the constant pushing force of the spring damper is consistent with the descending stage, and an eccentric excitation force e is applied at the centroid point of the tamper, and a constant downward force of 13 kN is applied at the same time.

[0133] (3) The lifting stage: the operation time is set to 1.6-2.0 s. From 1.6 to 1.7 s, the axial force of the spring damper is uniformly increased from 0 kN to F 1 / 2. From 1.7 to 2.0 s, the constant axial force of the spring damper is set to F 1 / 2. The excitation force is set in the same way as the descending stage.

[0134] Table 2 Simulation parameters of the tamper

[0135]

[0136] Step 6, preset different initial mechanical states of the track bed, the initial mechanical states of the track bed include different ballast shoulder supporting stiffness, ballast shoulder track bed compactness and track panel lateral resistance; then obtain the vibration acceleration signals of the tamper bottom plate, the interaction force signals between the tamper bottom plate and the ballast shoulder track bed, the vertical relative displacement signals between the track vehicle running gear and the sleeper and the lateral displacement signals of the sleeper under different excitation frequencies, tampering times and tampering frequencies; the data is preprocessed according to the process of step 4 to obtain the filtered response data;

[0137] The mechanical state of the ballast shoulder track bed in the simulation analysis model is recorded according to the preset slice data window length of 0.2s and step size of 0.1s. The track bed support stiffness, track bed compaction and track panel lateral resistance corresponding to each slice sample are obtained and used as labels. A simulation dataset is constructed based on the simulation analysis model and divided into training set, validation set and test set in a ratio of 8:2:1.

[0138] Step 7: Establish an intelligent detection model for the mechanical state of the ballast shoulder track bed based on a hybrid expert agent, such as... Figure 5 As shown, the intelligent detection model for the mechanical state of ballast shoulder track includes multiple expert networks and a gating network. Each expert network is an independent deep neural network model responsible for processing different types of signals, mainly composed of input layers, convolutional blocks, residual blocks, fully connected layers, and output layers. The gating network adaptively allocates weights and selects experts based on the input. The final output is the weighted sum of the outputs of each expert according to the gating weights, determined by the following formula:

[0139] (13)

[0140] In the formula, g k ( x ) is the first k An expert's representation of the input signal. f k ( x ) for gating network to the first k The weights assigned to each expert This is a predicted value for the mechanical state of the track bed;

[0141] The intelligent detection model of the mechanical state of the ballast shoulder track bed was trained on the simulation dataset. The loss function of each task was MSE. The total loss function of multiple tasks was summed, as shown in formulas (14) to (15). The model parameters were updated on the test set and the model parameters were verified on the validation set. Finally, the model effect was tested using the test set.

[0142] (14)

[0143] (15)

[0144] In the formula, T For the number of tasks, y i The actual label value. These are the model's predicted values. lambda t Assign weights to tasks;

[0145] The ballast shoulder track bed mechanical state intelligent detection model trained on the simulation data set is migrated to the field measured data set, and the model is fine-tuned using limited test data, and finally a pre-trained ballast shoulder track bed mechanical state intelligent detection model is obtained.

[0146] In this example, the expert network 1 is responsible for processing the vertical force signal and the lateral force signal, and the fast Fourier transform is performed on the vertical force and the lateral force respectively to obtain the frequency domain information, only the real number domain result is taken, and it is aligned with the time domain length of the force signal, and the insufficient is zero filled, then the dimension of the input signal is 4, the embedding dimension is 128, as shown in Figure 6 The interaction force between the one side tamper bottom plate and the track bed in this example takes the average value of the output of multiple film pressure sensors;

[0147] The expert network 2 is responsible for processing the vertical vibration acceleration signal and the lateral vibration acceleration signal, and the fast Fourier transform is performed on the vertical vibration acceleration and the lateral vibration acceleration respectively to obtain the frequency domain information, only the real number domain result is taken, and it is aligned with the time domain length of the acceleration signal, and the insufficient is zero filled, then the dimension of the input signal of each measuring point is 4, the embedding dimension is 128, as shown in Figure 7 Two vibration measuring points are arranged on the one side tamper bottom plate in this example;

[0148] The expert network 3 is responsible for processing the vertical relative displacement signal between the walking part of the working vehicle and the sleeper and the lateral displacement signal of the sleeper, and the fast Fourier transform is performed on the vertical relative displacement signal and the lateral displacement signal respectively to obtain the frequency domain information, only the real number domain result is taken, and it is aligned with the time domain length of the acceleration signal, and the insufficient is zero filled, then the dimension of the input signal of each measuring point is 4, the embedding dimension is 128, as shown in Figure 8 Three displacement measuring points are arranged on each side in this example, and the track row refers to three sleepers;

[0149] The expert network 4 is responsible for simultaneously processing the vibration acceleration signal, the displacement signal and the force signal, and performing fast Fourier transform on the input to enhance the frequency domain information, as shown in Figure 9

[0150] Finally, Gaussian noise is randomly injected into all input signals before inputting into the expert network, as shown in the following formula:

[0151] (16)

[0152] In the formula, SNR is the signal-to-noise ratio, P s is the power of the signal, P n is the power of the noise.

[0153] ​Step 8, the field measured data set is processed according to the same processing steps as the simulation data set, and the track bed supporting stiffness, track bed compactness and track panel lateral resistance are measured immediately after each tamping process is completed in the field as data labels, which correspond to the last slice sample;

[0154] The field measured data set is divided into a training set and a test set, and during model fine-tuning, the weights and parameters of the expert network are frozen, that is, the weights and parameters of the expert network are not updated, and only the parameters of the gating network are updated;

[0155] Before model fine-tuning, the input signal is subjected to fast Fourier transform for frequency domain information enhancement, but random noise is not added, and the loss function is MSE, and the parameters of the gating network are updated on the training set;

[0156] Finally, the model fine-tuning effect is verified on the test set;

[0157] Step 9, input the data to be measured into the fine-tuned ball shoulder track bed mechanical state intelligent detection model to obtain the real track bed supporting stiffness, track bed compactness and track panel lateral resistance.

[0158] Embodiment 2

[0159] Based on embodiment 1, the ball shoulder track bed mechanical state intelligent detection system using a tamping device is provided, which corresponds to the ball shoulder track bed mechanical state intelligent detection method using a tamping device, and specifically comprises:

[0160] An acceleration measurement module is configured to obtain vibration acceleration data of the tamper base plate, and the vibration acceleration data of the tamper base plate is vertical and lateral acceleration data of the tamper base plate acting on the track bed during the process;

[0161] A force measurement module is configured to obtain interaction force data of the tamper base plate and the ball shoulder track bed, and the interaction force of the tamper base plate and the ball shoulder track bed is vertical and lateral force data of the tamper base plate acting on the track bed during the process;

[0162] A displacement measurement module is configured to obtain vertical relative displacement of the working vehicle running part and the sleeper and lateral displacement of the sleeper, and the vertical relative displacement of the working vehicle running part and the sleeper and the lateral displacement of the sleeper are displacement data of the tamper base plate acting on the track bed during the process;

[0163] In this example, three laser displacement sensors and one CCD camera are respectively installed on the left and right sides of the working vehicle running part area, and the positions of the laser displacement sensors on each side substantially correspond to the top of the three sleepers, and the bases of the laser displacement sensors are installed in movable grooves, which can be controlled to move by a servo hydraulic device, as shown in Figure 10 ;

[0164] CCD camera is fixedly installed on the running part frame or detection beam of the working vehicle perpendicularly to the track bed, which can cover the shooting range of 3 sleepers. The calibration plate is pasted at the end of the sleeper or the known fastener spacing is used to calibrate the CCD camera. The parameters to be calibrated include the camera intrinsic matrix K , rotation matrix R , translation vector t and scaling factor α :

[0165] (17)

[0166] wherein, f x and f y is the focal length (in pixels); c x and c y is the principal point coordinate (usually the image center);

[0167] (18)

[0168] wherein, u v is the pixel coordinate of a point on the image, and its corresponding normalized camera coordinate is x y , 1);

[0169] (19)

[0170] wherein, x c ,y c z c is the camera coordinate, and X w ,Y w Z w is the world coordinate;

[0171] (20)

[0172] wherein, D r is the actual distance between two points on the calibration plate, D p is the corresponding pixel distance on the image;

[0173] ​​​​The laser displacement sensor emits a laser to the surface above the end of the sleeper. A CCD camera captures the image and inputs it into a vision measurement system to track the position of the laser point. When the laser point is not on the sleeper, the vision measurement system measures the distance between the laser point and the sleeper and controls the servo hydraulic device to adjust the position of the laser displacement sensor to ensure that the laser displacement sensor is within the correct measurement range.

[0174] After the position adjustment is completed, the work vehicle begins operation, the compactor is lowered, and the interaction force between the base plate and the ballast shoulder track bed is applied. F Greater than the preset threshold delta At this time, the laser displacement sensor begins to collect the vertical relative displacement between the axle box or detection beam and the sleeper, the vision measurement system tracks the laser point, and tests the lateral displacement of the sleeper; the compactor is raised, and when the interaction force between the base plate and the ballast shoulder is... F Less than the preset threshold delta When the signal acquisition stops, stop the acquisition.

[0175] The servo control module is used to adjust the laser displacement sensor to a suitable position based on the results of the vision measurement system, ensuring that the laser point is within the surface range above the sleeper end, such as... Figure 11 As shown;

[0176] The data preprocessing module is used to resample, slice, and filter the measured data to construct a dataset;

[0177] The model pre-training module is used to build a hybrid expert agent model and train it on the dataset to obtain the optimal pre-training parameters.

[0178] The model fine-tuning module is used to fine-tune the weight parameters of the gating network to obtain the final intelligent detection model of the mechanical state of the ballast shoulder track bed;

[0179] Data detection module: This module is used to input the field measured data into the intelligent detection model of the mechanical state of the ballast shoulder track bed, obtain the detection values ​​of track bed support stiffness, track bed compaction and track panel lateral resistance, and visualize them.

[0180] Example 3

[0181] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, which communicate with each other. The memory stores program instructions that can be executed by the processor. The processor calls the program instructions to execute an intelligent detection method for the mechanical state of ballast shoulder track using a compaction device. The intelligent detection method for the mechanical state of ballast shoulder track using a compaction device includes the following steps:

[0182] The vibration acceleration signal of the compactor base plate is acquired. The vibration acceleration signal includes vertical vibration acceleration signal and lateral acceleration signal, which are collected by acceleration sensor and data acquisition system.

[0183] The interaction force signals between the compactor base plate and the ballast shoulder track bed are acquired. The force signals include vertical force and lateral force signals, which are uniformly collected by a thin-film pressure sensor and a data acquisition system.

[0184] The vertical relative displacement between the running gear of the work vehicle and the sleeper, as well as the lateral displacement of the sleeper, are acquired. The displacement signals are collected by a laser displacement sensor, a CCD camera, and a data acquisition system.

[0185] The collected vibration acceleration signal, force signal and displacement signal are sliced ​​and filtered to obtain the filtered slice vibration acceleration signal, slice force signal and slice displacement signal, respectively.

[0186] The filtered slice vibration acceleration signal, slice force signal, and slice displacement signal are simultaneously input into the pre-trained intelligent detection model of the mechanical state of the ballast shoulder track bed to obtain the real mechanical state of the ballast shoulder track bed, which includes the track bed support stiffness, track bed compaction, and track panel lateral resistance.

[0187] Example 4

[0188] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements an intelligent detection method for the mechanical state of ballast shoulder track bed using a compaction device. The method includes the following steps:

[0189] The vibration acceleration signal of the compactor base plate is acquired. The vibration acceleration signal includes vertical vibration acceleration signal and lateral acceleration signal, which are collected by acceleration sensor and data acquisition system.

[0190] The interaction force signals between the compactor base plate and the ballast shoulder track bed are acquired. The force signals include vertical force and lateral force signals, which are uniformly collected by a thin-film pressure sensor and a data acquisition system.

[0191] The vertical relative displacement between the running gear of the work vehicle and the sleeper, as well as the lateral displacement of the sleeper, are acquired. The displacement signals are collected by a laser displacement sensor, a CCD camera, and a data acquisition system.

[0192] The collected vibration acceleration signal, force signal and displacement signal are sliced ​​and filtered to obtain the filtered slice vibration acceleration signal, slice force signal and slice displacement signal, respectively.

[0193] The filtered slice vibration acceleration signal, slice force signal, and slice displacement signal are simultaneously input into the pre-trained intelligent detection model of the mechanical state of the ballast shoulder track bed to obtain the real mechanical state of the ballast shoulder track bed, which includes the track bed support stiffness, track bed compaction, and track panel lateral resistance.

[0194] Example 5

[0195] Embodiment 5 of the present application provides a computer program product comprising a computer program which, when run on one or more processors, implements a method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamper device as described above. The method comprises the following process steps:

[0196] An acceleration signal of the tamper base plate is obtained, the acceleration signal comprising a vertical vibration acceleration signal and a lateral acceleration signal, and is collected by an acceleration sensor and a data acquisition system;

[0197] An interaction force signal of the tamper base plate and the ballast shoulder track bed is obtained, the force signal comprising a vertical interaction force and a lateral interaction force signal, and is collected by a thin film pressure sensor and a data acquisition system;

[0198] A vertical relative displacement of the running gear of the work vehicle and the sleeper and a lateral displacement of the sleeper are obtained, and the displacement signal is collected by a laser displacement sensor, a CCD camera and a data acquisition system;

[0199] The collected vibration acceleration signal, force signal and displacement signal are sliced and filtered to obtain a filtered sliced vibration acceleration signal, a filtered sliced force signal and a filtered sliced displacement signal, respectively;

[0200] The filtered sliced vibration acceleration signal, the filtered sliced force signal and the filtered sliced displacement signal are simultaneously input into a pre-trained intelligent detection model of the mechanical state of the ballast shoulder track bed to obtain the real mechanical state of the ballast shoulder track bed, the mechanical state of the ballast shoulder track bed comprising the track bed supporting stiffness, the track bed compactness and the rail transverse resistance.

[0201] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) having computer-usable program code embodied therein.

[0202] The present application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a means for implementing the functions specified in the flowcharts and / or block diagrams.Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0203] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flowchart Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0204] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the flowchart Figure 1 apparatuses that implement the functions specified in the flowchart Figure 1

[0205] The above description is only a specific implementation of the present application, and is not intended to limit the protection scope of the present application. Those skilled in the art should understand that various modifications or changes can be made to the disclosed technical solutions without inventive labor, and all these modifications or changes should be covered within the protection scope of the present application.​​​

Claims

1. A ballast shoulder track bed mechanical state intelligent detection method using a tamping device, characterized by, The method comprises the following steps: S1, obtaining a vibration acceleration signal of a tamper base plate, the vibration acceleration signal comprising a vertical vibration acceleration signal and a lateral acceleration signal, and the vibration acceleration signal and the lateral acceleration signal being collected by an acceleration sensor and a data collection system; S2, obtaining an interaction force signal between the tamper base plate and a ballast shoulder track bed, the force signal comprising a vertical interaction force and a lateral interaction force signal, and the force signal being collected by a thin film pressure sensor and a data collection system; S3, obtaining a vertical relative displacement between a running part of a work vehicle and a sleeper and a lateral displacement of the sleeper, and the displacement signal being collected by a laser displacement sensor, a CCD camera and a data collection system; S4, slicing and filtering the collected vibration acceleration signal, the force signal and the displacement signal to obtain a filtered sliced vibration acceleration signal, a filtered sliced force signal and a filtered sliced displacement signal; S5, inputting the filtered sliced vibration acceleration signal, the filtered sliced force signal and the filtered sliced displacement signal into a pre-trained ballast shoulder track bed mechanical state intelligent detection model to obtain a real ballast shoulder track bed mechanical state, the ballast shoulder track bed mechanical state comprising a ballast shoulder track bed supporting stiffness, a ballast shoulder track bed compactness and a track panel lateral resistance; In step S5, the pre-trained ballast shoulder track bed mechanical state intelligent detection model comprises the following steps: A tamper-having-ballast track bed simulation analysis model is established based on a DEM-MBD coupling method, the simulation analysis model comprising a tamper base plate, an oil pressure connecting rod, an eccentric exciter, a steel rail, a fastener, a sleeper and a track bed component, wherein the having-ballast track bed is established based on DEM and polyhedral particles are used to simulate ballast particles; the tamper base plate, the steel rail and the sleeper are simulated by rigid body units; the oil pressure connecting rod and the fastener are simulated by spring-damping units; and the eccentric excitation force is controlled by a loading function, as shown in the following formula: ； In the formula, m1 and m2 are the masses of the two eccentric wheels, respectively; Different initial mechanical states of the track bed are preset, the initial mechanical states of the track bed comprising different ballast shoulder supporting stiffnesses, ballast shoulder track bed compactnesses and track panel lateral resistances; then, vibration acceleration signals of the tamper base plate, interaction force signals between the tamper base plate and the ballast shoulder track bed, vertical relative displacement signals between the running part of the work vehicle and the sleeper and lateral displacement signals of the sleeper are obtained under different excitation frequencies, tampering times and tampering frequencies; The mechanical states of the ballast shoulder track bed in the simulation analysis model are recorded according to the preset slice data window length and step length to obtain the track bed supporting stiffness, the track bed compactness and the track panel lateral resistance corresponding to each slice sample, which are taken as labels; a simulation data set is constructed based on the simulation analysis model, and the simulation data set is divided into a training set, a verification set and a test set according to a proportion; A ballast shoulder track bed mechanical state intelligent detection model is established based on a hybrid expert agent, the ballast shoulder track bed mechanical state intelligent detection model comprising a plurality of expert networks and a gating network, wherein each expert network is an independent deep neural network model and is responsible for processing different types of signals, and mainly comprises an input layer, a convolution block, a residual block, a full connection layer and an output layer; the gating network adaptively allocates weights and selects experts according to the input, and the final output is the result of weighting the outputs of the experts according to the gating weights, which is determined by the following formula: ； wherein g k ( x ) is the representation of the input signal by the first k expert, f k ( x ) is the weight assigned to the first k expert by the gating network, is the predicted value of the mechanical state of the ballast. The established ball shoulder track bed mechanical state intelligent detection model is trained on the simulation data set, the loss function of each task is MSE, the total loss function of the multi-task is accumulated and summed, the model parameters are updated on the test set, the model parameters are verified on the verification set, and finally the model effect is tested on the test set; ； ； wherein T is the number of tasks, y i is the true label value, is the model prediction value, Lambda t is the weight assigned to the task; The ball shoulder track bed mechanical state intelligent detection model trained on the simulation data set is migrated to the field measured data set, the model is fine-tuned by using limited test data, and finally the pre-trained ball shoulder track bed mechanical state intelligent detection model is obtained.

2. The method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamping device according to claim 1, characterized in that, In step S1, the vibration acceleration signal of the compactor base plate is acquired, specifically including: installing dual-axis acceleration sensors on the top of the compactor base plates on both sides of the work vehicle; installing a thin-film pressure sensor inside the compactor base plate; and setting the interaction force threshold between the base plate and the ballast shoulder track bed. Delta As the compactor falls, the interaction force between the base plate and the ballast shoulder track bed... F Greater than the preset threshold Delta At this time, vibration acceleration signals are simultaneously acquired; the compactor is raised, and the interaction force between the base plate and the ballast shoulder track bed is measured. F Less than the preset threshold Delta When the signal acquisition stops, stop the acquisition.

3. The method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamping device according to claim 1, characterized in that, In step S3, the vertical relative displacement of the work vehicle running part axle box and the sleeper and the lateral displacement of the sleeper are obtained, specifically including: Three laser displacement sensors and a CCD camera are respectively installed on the left and right sides of the work vehicle running part region, wherein the positions of the laser displacement sensors on each side correspond to the top of the three sleepers, the bases of the laser displacement sensors are installed in the movable sliding groove, and the movement is controlled by a servo hydraulic device; The CCD camera is fixedly installed on the running part frame or detection beam of the working vehicle perpendicularly to the ballast bed, covers the shooting range of three sleepers, and is calibrated by pasting a calibration plate at the end of the sleeper or using a known fastener spacing. The parameters that need to be calibrated include a camera intrinsic matrix K , a rotation matrix R , a translation vector t , and a scaling coefficient α : ； wherein f x and f y is the focal length in pixels; c x and c y is the principal point coordinate; ； where (x, y) are the pixel coordinates of a point on the image, and (u, v, 1) are the corresponding normalized camera coordinates. u , v ) are the pixel coordinates of a point on the image, and (u, v, 1) are the corresponding normalized camera coordinates. x , y ,1) are the pixel coordinates of a point on the image, and (u, v, 1) are the corresponding normalized camera coordinates ； wherein, x c ,y c , z c are camera coordinates, and X w ,Y w , Z w are world coordinates; ； wherein D r is the actual distance between two points on the calibration plate, D p is the corresponding pixel distance on the image; The laser displacement sensor emits laser to the surface above the end of the sleeper, the CCD camera captures the image and inputs it to the vision measurement system to track the position of the laser point, when the laser point is not on the sleeper, the vision measurement system measures the distance between the laser point and the sleeper, and controls the servo hydraulic device to adjust the position of the laser displacement sensor to ensure that the laser displacement sensor is in the correct measurement range. After the position adjustment is completed, the work vehicle begins operation, the compactor is lowered, and the interaction force between the base plate and the ballast shoulder track bed is applied. F Greater than the preset threshold Delta At this time, the laser displacement sensor begins to collect the vertical relative displacement between the axle box or detection beam and the sleeper, the vision measurement system tracks the laser point, and tests the lateral displacement of the sleeper; the compactor is raised, and when the interaction force between the base plate and the ballast shoulder is... F Less than the preset threshold Delta When the signal acquisition stops, stop the acquisition.

4. The method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamping device according to claim 1, characterized in that, In step S4, the collected vibration acceleration signal, force signal and displacement signal are sliced and filtered, specifically including: The vibration acceleration signal, the force signal and the displacement signal collected in each tamping operation are resampled, and the polyphase FIR filtering method is used to unify the sampling frequency of the signals f : ； wherein h [·] is a low-pass filter coefficient, x [·] is an original signal, n is an output signal y [·] is an index of, k is a filter h [·] is an index of, M / L is a target sample rate conversion ratio, given by f / Fs wherein f s is an original sampling frequency; The resampled vibration acceleration signal, force signal and displacement signal are band-stop filtered to exclude the interference of the original tamper excitation frequency: ； wherein x t is the input signal, y t is the filtered signal, H BS is the filter operator, whose difference equation of the second order band-stop filter is determined by​​ ； wherein b 0, b 1, b 2, a 1 and a 2 are filter coefficients; The filtered vibration acceleration signal, the force signal and the displacement signal are subjected to slicing processing, the window length is set as s , the overlap rate is set as p , and then the number of samples after slicing N is determined by the following formula: ； In the formula, L x is the length of the input signal.

5. The method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamping device according to claim 4, characterized in that: Specifically including: The expert network 1 is responsible for processing the vertical force signal and the lateral force signal, and the fast Fourier transform is performed on the vertical force and the lateral force to obtain the frequency domain information, only the real number domain result is taken, and the time domain length of the force signal is aligned, the insufficient is zero filled, and then the dimension of the input signal is 4; The expert network 2 is responsible for processing the vertical vibration acceleration signal and the lateral vibration acceleration signal, and the fast Fourier transform is performed on the vertical vibration acceleration and the lateral vibration acceleration to obtain the frequency domain information, only the real number domain result is taken, and the time domain length of the acceleration signal is aligned, the insufficient is zero filled, and then the dimension of the input signal of each measuring point is 4; The expert network 3 is responsible for processing the vertical relative displacement signal of the work vehicle running part and the sleeper and the lateral displacement signal of the sleeper, and the fast Fourier transform is performed on the vertical relative displacement signal and the lateral displacement signal to obtain the frequency domain information, only the real number domain result is taken, and the time domain length of the acceleration signal is aligned, the insufficient is zero filled, and then the dimension of the input signal of each measuring point is 4; The expert network 4 is responsible for simultaneously processing the vibration acceleration signal, the displacement signal and the force signal, and performing fast Fourier transform on the input to enhance the frequency domain information; Finally, Gaussian noise is randomly injected into all input signals before inputting into the expert network, as follows: ； where SNR is the signal-to-noise ratio, P s is the power of the signal, P n is the power of the noise.

6. The method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamping device according to claim 4, characterized in that: Specifically including: The field measured data set is processed according to the same processing steps as the simulation data set, and the track bed supporting stiffness, the track bed compactness and the track panel lateral resistance are measured immediately after each tamping process in the field as data labels, which correspond to the last slice sample; The field measured data set is divided into a training set and a test set, and during model fine-tuning, the weights and parameters of the expert network are frozen, that is, the weights and parameters of the expert network are not updated, and only the parameters of the gating network are updated; Before model fine-tuning, the input signal is subjected to fast Fourier transform for frequency domain information enhancement, but no random noise is added, and the loss function is MSE, and the parameters of the gating network are updated on the training set; Finally, the model fine-tuning effect is verified on the test set.

7. A ballast shoulder track bed mechanical state intelligent detection system using a tamping device, characterized by, It comprises: an acceleration measurement module for obtaining vibration acceleration data of the tamper base plate, the vibration acceleration data of the tamper base plate being vertical and lateral acceleration data of the tamper base plate acting on the track bed during the process; a force measurement module for obtaining interaction force data between the tamper base plate and the ballast shoulder track bed, the interaction force between the tamper base plate and the ballast shoulder track bed being vertical and lateral force data of the tamper base plate acting on the track bed during the process; a displacement measurement module for obtaining vertical relative displacement between the running gear of the working vehicle and the sleeper and lateral displacement of the sleeper, the vertical relative displacement between the running gear of the working vehicle and the sleeper and the lateral displacement of the sleeper being displacement data of the tamper base plate acting on the track bed during the process; a servo control module for adjusting the laser displacement sensor to a suitable position according to the results of the visual measurement system to ensure that the laser point is within the surface range above the end of the sleeper; a data preprocessing module for resampling, slicing and filtering the measured data to construct a data set; a model pre-training module for constructing a hybrid expert agent model and training it on the data set to obtain optimal pre-training parameters; a model fine-tuning module for fine-tuning the weight parameters of the gating network to obtain a final intelligent detection model of the mechanical state of the ballast shoulder track bed; a data detection module for inputting the field measured data into the intelligent detection model of the mechanical state of the ballast shoulder track bed to obtain the detection values of the track bed supporting stiffness, the track bed compactness and the track panel lateral resistance, and to visualize them.

8. A non-transitory computer-readable storage medium, comprising: It comprises a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implements a method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamper device according to any one of claims 1-5.

9. An electronic device, comprising: It comprises: a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to make the electronic device execute a method for intelligently detecting the mechanical state of a ballast shoulder track bed using a tamper device according to any one of claims 1-5.

Images