Magnetic force sensor-based track condition holographic detection system and method of use thereof

Abstract

The application is a track state holographic detection system based on magnetic force sensing and a method thereof, comprising a magnetic force sensing detection module and a magnetic marker module; the magnetic force sensing detection module comprises a sensing array adjusting mechanism, a sensor mounting seat symmetrically arranged at the two sides of the end of the steel rail and capable of moving in multiple directions, a closed-loop linear Hall sensing array arranged on the sensor mounting seat and distributed along the sectional profile of the steel rail, and a multi-channel sensor including a rail head area, a rail waist area and a rail bottom area, the sensing array adjusting mechanism adjusting the spatial position and attitude of the sensor array by moving in multiple directions; the magnetic marker module comprises a magnetic marker unit including a permanent magnet arranged at the rail bottom and the sleeper end of the track, and the magnetic marker unit and the magnetic force sensing detection module have a signal detection reaction; the application has the advantages of improving the multi-dimensional holographic detection capability of the track state monitoring system and solving the magnetic interference problem in complex environments.

Description

Technical Field

[0001] This invention relates to the field of track engineering technology based on magnetic force detection, specifically to a track state holographic detection system based on magnetic force sensing and its usage method. Background Technology

[0002] In the field of rail transit operation and maintenance, the accuracy of the geometric parameters of the track structure directly affects driving safety and passenger comfort. Currently, track condition detection technologies mainly include optical inspection, manual inspection, and automated inspection equipment. However, these traditional methods have many limitations. Optical inspection equipment, such as laser measurement systems, suffers significant accuracy degradation in complex environments such as tunnel dust and water mist, with a success rate generally below 60%, making it difficult to meet the needs of all-weather operation. Manual inspection is inefficient, with a detection speed of only 50 meters per hour, and is greatly affected by human factors, resulting in poor data reliability and an inability to achieve holographic detection of all track parameters. Existing automated inspection equipment mostly uses a single sensor solution, resulting in blind spots and poor track type adaptability. Specialized equipment is usually required for different rail specifications, increasing maintenance costs. Furthermore, under extreme environmental conditions, such as high temperature (≥60℃), low temperature (≤-20℃), or high vibration environments, the stability of existing equipment decreases significantly, especially in magnetic field interference environments, lacking effective magnetic shielding and compensation mechanisms. Some automated inspection equipment still relies on internal combustion engine drives, which not only does not conform to modern green construction concepts but also causes noise pollution. Although geomagnetic sensing technology has begun to be applied in the field of railway attitude measurement, existing technologies can only achieve simple detection of basic angle parameters and have not yet formed multi-dimensional holographic detection capabilities. Furthermore, the problem of magnetic interference in complex environments remains unresolved. These issues severely restrict the accuracy, efficiency, and applicability of track inspection, making it difficult to meet the urgent needs of modern rail transit for high-precision, high-efficiency, and all-weather inspection. Therefore, there is an urgent need to develop a new type of track condition detection system to overcome these technical bottlenecks. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a holographic detection system for track status based on magnetic sensing and its usage method. This system aims to improve the multi-dimensional holographic detection capability of track status monitoring systems and solve the problem of magnetic interference in complex environments. It overcomes the technical defects of existing technologies, such as the susceptibility of detection accuracy to interference from complex environments like dust and water mist, low efficiency and blind spots in manual and single-sensor detection, insufficient track type adaptability requiring dedicated configuration, poor stability in extreme temperature and magnetic field environments, and environmental and noise problems caused by reliance on internal combustion engine. This invention achieves high reliability, high efficiency, full-parameter, all-weather detection of track geometric parameters and meets the requirements of green operation.

[0004] To achieve the above objectives, a holographic track condition detection system based on magnetic sensing is designed, comprising a magnetic sensing detection module and a magnetic marking module. The magnetic sensing detection module includes: a sensor array adjustment mechanism with sensor mounting seats symmetrically arranged along both sides of the rail at its end, capable of multi-directional movement; a closed-loop linear Hall effect sensor array, mounted on the sensor mounting seats and distributed along the rail cross-sectional contour, including multi-channel sensors in the rail head region, rail web region, and rail bottom region; the sensor array adjustment mechanism adjusts the spatial position and orientation of the sensor array through multi-directional movement. The magnetic marking module includes: magnetic marking units, comprising permanent magnets disposed at the rail bottom and sleeper ends; the magnetic marking units serve as positioning references for magnetic sensing, reacting with the magnetic sensing detection module; the marking interval of the magnetic marking units matches the sleeper spacing.

[0005] Furthermore, the present invention also includes: the closed-loop linear Hall sensor array includes a 24-channel sensor, wherein 10 channels are arranged in the rail head region, 8 channels are arranged in the rail waist region, and 4 channels are arranged in the rail bottom region.

[0006] Furthermore, the present invention also includes: the sensor array adjustment mechanism comprising: a lateral movement component, including a lateral movement bracket, the lateral movement bracket being mounted on a track via two legs, and a laterally arranged lateral ball screw on the lateral movement bracket, the lateral ball screw being driven by a servo motor at one end; a vertical compensation mechanism, including a vertical movement bracket, the vertical movement bracket being threaded onto the lateral ball screw and moving laterally with the lateral ball screw, the vertical movement bracket being provided with a vertical ball screw, the vertical ball screw being driven by a vertical motor at one end; and a sensor mounting plate, threaded onto the vertical ball screw; attitude The attitude adjustment joint, a universal joint structure, is located at both ends of the sensor mounting plate, with the other end connected to the sensor mounting base. A vertical torque motor is mounted on the sensor mounting plate along the horizontal direction, and its rotation shaft passes through the attitude adjustment joint, driving it to swing vertically. A horizontal torque motor is mounted on the attitude adjustment joint along the vertical direction, and its rotation shaft passes through the sensor mounting base, driving it to swing horizontally. The attitude adjustment joint, in conjunction with the horizontal and vertical torque motors, enables real-time adjustment of the pitch and yaw angles of the sensor array.

[0007] Furthermore, the present invention also includes: the sensor mounting base is further provided with: an inertial sensor, which provides feedback data to realize closed-loop control and detects the parallelism between the reference plane of the sensor array and the center line of the track; and a grating ruler, whose detection head is set towards the track to monitor the displacement of the sensor array adjustment mechanism in real time.

[0008] Furthermore, the present invention also includes: the magnetic marking module further includes a magnetic marking fixing device, which consists of a rail marking holder and a sleeper marking embedded part; the rail marking holder is made of austenitic stainless steel and fits against the bottom of the rail through a dovetail groove structure; the sleeper marking embedded part adopts a T-shaped structure design, with one side of the T-shaped structure embedded in the sleeper and the other end exposed to the outside, and integrates a permalloy magnetic shielding sleeve.

[0009] Furthermore, the present invention also includes: the sensor array adjustment mechanism is further provided with a high-precision lateral compensation component, which is disposed between the sensor mounting plate and the vertical moving bracket, including: a guide rail base that is threadedly engaged with a vertical ball screw on the vertical moving bracket, a guide rail that is horizontally disposed on the guide rail base, a slider that is slidably engaged with the guide rail, one side of the slider being fixedly connected to the sensor mounting plate, and the telescopic end of a telescopic cylinder disposed on the guide rail base being fixedly connected to the slider. The telescopic end of the telescopic cylinder drives the slider to reciprocate on the slide rail, thereby causing the sensor mounting plate to reciprocate in the horizontal direction, so that the symmetry axis of the sensor array on the sensor mounting plate is completely coincident with the symmetry axis of the single track.

[0010] Furthermore, the present invention also includes: a disc spring assembly disposed between the sensor mounting plate and the vertical motor, the disc spring assembly being sleeved on the vertical ball screw, thereby enabling the sensor array to float elastically within a certain range and providing limiting guidance for the vertical displacement of the sensor mounting plate.

[0011] Furthermore, the present invention also includes: a system hardware module comprising multi-source data acquisition hardware, motion control hardware, and environmental monitoring hardware; the multi-source data acquisition hardware adopts an FPGA+ARM architecture to realize synchronous acquisition and preprocessing of sensor array signals; the motion control hardware adopts a PLC main controller, in conjunction with a servo driver, to achieve precise control of the adjustment mechanism; the environmental monitoring hardware integrates a temperature sensor, a humidity sensor, and a dust concentration sensor; a protection module comprising a permalloy magnetic shielding layer, a micro-positive pressure maintaining device, and a temperature control device, used to suppress external magnetic field interference, prevent dust intrusion, and maintain the system operating temperature; and a robot connection mechanism comprising a quick-change interface assembly and a buffer shock absorption unit; the quick-change interface assembly adopts a tenon-and-mortise positioning structure, in conjunction with a pneumatic locking device; the buffer shock absorption unit includes two sets of metal rubber shock absorbers to achieve vibration attenuation in the XYZ axes.

[0012] This invention also provides a method for using a track state holographic detection system based on magnetic sensing. The method employs the aforementioned track state holographic detection system based on magnetic sensing and includes the following steps: acquiring magnetic field distribution data collected by multi-channel closed-loop linear Hall sensors distributed along the rail cross-sectional profile; preprocessing the magnetic field distribution data, including using wavelet threshold filtering to eliminate pulse interference, adaptive Kalman filtering to compensate for environmental magnetic field drift, temperature compensation to correct sensor temperature drift, and wavelet transform to remove low-frequency drift; inputting the preprocessed magnetic field distribution data into an improved U-Net network, wherein the improved U-Net network uses a MobileNetV2-style inverted residual block reconstruction basic feature extraction unit, through a 1×1 The process of convolutional dimensionality upscaling, depthwise separable convolutional feature extraction, and 1×1 convolutional dimensionality reduction reduces the number of parameters. An ECA attention module and a gated attention mechanism are integrated between the encoder and decoder. The ECA attention module dynamically learns channel dependencies through adaptive average pooling and lightweight 1D convolution, while the gated attention mechanism generates an adaptive weight map for refined feature selection. Bilinear interpolation is used instead of transposed convolution as an upsampling method, combined with 1×1 convolution for channel matching and 3×3 convolution for feature refinement. The improved U-Net network processes magnetic field images to predict key rail dimensions. The predicted key rail dimensions and geometric parameters are fused to generate a digital holographic model of the track section, enabling holographic detection of track conditions.

[0013] Furthermore, the present invention also includes: the parameter increment of the ECA attention module in the improved U-Net network is controlled within 1%, and the dual attention mechanism works synergistically to improve the model's ability to capture features of target boundaries and small-sized defects; the improved U-Net network uses bilinear interpolation as an upsampling method to avoid the checkerboard artifact problem caused by transposed convolution, ensuring the smoothness and spatial consistency of the output segmentation map; the improved U-Net network adopts a lightweight architecture design, significantly improving inference speed, and can be deployed on edge devices or embedded systems to process track detection data in real time in resource-constrained engineering scenarios.

[0014] Compared with the prior art, the advantages of this invention are:

[0015] The track condition detection system provided by this invention demonstrates comprehensive and significant technical advantages in practical applications. Regarding detection accuracy, the system effectively overcomes the interference of complex environments such as tunnel dust and water mist on measurement results, increasing the success rate from less than 60% with traditional optical methods to over 95%. This ensures high reliability and repeatability of track geometric parameters such as gauge, level, elevation, and alignment, providing accurate data for condition assessment. In terms of detection efficiency, the system achieves fully automated operation with a detection speed exceeding 5 kilometers per hour, nearly 100 times more efficient than manual inspection. It simultaneously acquires all track parameters, completely eliminating blind spots caused by single sensors, significantly improving detection completeness and coverage, and effectively avoiding the risks of missed and false detections. Regarding environmental adaptability, the system maintains stable operation in a wide temperature range of -30℃ to 70℃, under harsh conditions such as high vibration and strong magnetic field interference, effectively suppressing external magnetic field interference in the measurement process, ensuring data accuracy and system robustness, and significantly expanding the range of applicable scenarios. The system boasts excellent rail type compatibility, eliminating the need for specialized equipment replacement or parameter reconfiguration for different rail specifications. This significantly reduces maintenance complexity and equipment configuration costs, while improving resource utilization efficiency. In terms of environmental performance, the pure electric drive system operates with zero emissions and low noise, fully aligning with green construction and sustainable development principles, and improving the on-site working environment. Furthermore, the system's high-precision, multi-dimensional detection data can be directly integrated with an intelligent track condition analysis platform, providing reliable data support for safe operation, scientific maintenance, and predictive maintenance. This effectively prevents safety hazards caused by abnormal track conditions and extends the service life of infrastructure. In summary, this system achieves breakthrough improvements in accuracy, efficiency, adaptability, economy, and environmental friendliness, effectively addressing the core shortcomings of existing technologies. It provides an efficient, reliable, and practical technical solution for the rail transit inspection field, possessing outstanding practical value and significant potential for wider application. Attached Figure Description

[0016] Figure 1 This is an isometric view of a holographic detection device for track status based on magnetic force sensing;

[0017] Figure 2 This is a front view of the sensor array adjustment mechanism;

[0018] Figure 3 This is an isometric view of the sensor array adjustment mechanism;

[0019] Figure 4 It is a software architecture diagram showing how it is used;

[0020] Figure 5 This is a diagram of the traditional U-Net network structure;

[0021] Figure 6 This is a diagram of the improved U-Net network structure;

[0022] In the diagram: 1. Magnetic sensing detection module; 2. Magnetic marking module; 3. Track; 11. Sensor array; 12. Sensor array adjustment mechanism;

[0023] 121 Lateral movement assembly; 1211 High-precision lateral compensation assembly; 1212 Lateral movement bracket; 1213 Lateral ball screw; 1214 Servo motor; 1215 Support foot;

[0024] 122 Vertical compensation mechanism; 1221 Disc spring assembly; 1222 grating ruler; 1223 Vertical moving bracket; 1224 Vertical ball screw;

[0025] 123 Attitude adjustment joint; 1231 Horizontal torque motor; 1232 Vertical torque motor; 1233 Inertial sensor;

[0026] 124 Vertical motor; 125 Sensor mounting plate; 126 Sensor mounting base. Detailed Implementation

[0027] To make the purpose, principle and structure of the present invention clearer, the following description is provided in conjunction with the accompanying drawings and specific embodiments.

[0028] See Figures 1-6 This invention provides a holographic detection system for track status based on magnetic sensing and its usage method, comprising: a magnetic sensing detection module 1 and a magnetic marking module 2. The magnetic sensing detection module 1 consists of a closed-loop linear Hall sensor array 11 and a sensor array adjustment mechanism 12; the magnetic marking module 2 consists of magnetic marking units and magnetic marking fixing devices; the system also includes a system hardware module, a protection module, and a robot connection mechanism (not shown in the figure).

[0029] In the magnetic force sensing detection module 1, the closed-loop linear Hall sensor array 11 contains twenty-four channels of sensors, distributed along the rail cross-sectional profile, with ten channels arranged in the rail head region, eight channels in the rail web region, and four channels in the rail bottom region. The sensor array adjustment mechanism 12 includes a lateral movement component 121, a vertical compensation mechanism 122, and an attitude adjustment joint 123.

[0030] The lateral movement assembly 121 includes a lateral movement bracket 1212, which is mounted on the track 3 by two legs 1215. The lateral movement bracket 1212 is provided with a lateral ball screw 1213, which is driven by a servo motor 1214 located at one end.

[0031] The vertical compensation mechanism 122 includes a vertical moving bracket 1223, which is threaded onto a horizontal ball screw 1213. Preferably, an ear plate with an internally threaded hole is provided on the back of the vertical moving bracket 1223, through which the internally threaded hole passes onto the horizontal ball screw 1213. The ear plate moves laterally with the horizontal ball screw 1213. When the horizontal ball screw 1213 rotates, the internally threaded hole of the ear plate will advance or retract with the rotation of the horizontal ball screw 1213, thus forming a motion relationship where the vertical moving bracket 1223 is driven by the horizontal ball screw 1213 to make lateral displacement. A vertical ball screw 1224 is provided on the vertical moving bracket 1223, and the vertical ball screw 1224 is driven by a vertical motor 124 located at one end.

[0032] The sensor mounting plate 125 is threaded onto the vertical ball screw 1224. Similarly, a lug with an internally threaded hole on the back of the sensor mounting plate 125 can be threaded onto the vertical ball screw 1224 for follow-along movement. A disc spring assembly 1221 is installed between the end face of the lug of the sensor mounting plate 125 and the end face of the fixed end of the vertical motor 124. The disc spring assembly 1221 is a disc spring assembly with a through hole in the center (also known as a Belleville spring assembly), an elastic component composed of multiple standard conical disc springs stacked together. Each spring has a through hole (inner diameter) in the center and an outer diameter at the outer edge; the conical structure gives it non-linear elastic characteristics. During assembly, the springs can be stacked in the same direction to increase stiffness and load-bearing capacity, stacked top to top to increase the total deformation and reduce stiffness, or a hybrid method can be used to customize the load-deformation curve. All through holes are strictly aligned to form a continuous, through-channel. To ensure stable compression, a vertical ball screw 1224 is installed in the channel to prevent skew and instability. An external guide sleeve constrains radial displacement, and support washers at both ends ensure uniform load distribution. Preferably, flat washers can be added between the discs for fine-tuning. The through-hole not only enables precise guidance and coaxial compression but also facilitates the passage of the vertical ball screw 1224, simplifying installation and maintenance. In the event of vibration in the working environment, the device structure creates a certain degree of elastic floating between the end face of the ear plate of the sensor mounting plate 125 and the end face of the fixed end of the vertical motor 124, maximizing the filtering of high-frequency, low-amplitude vibrations in the working environment. A disc spring assembly 1221 is fitted onto the vertical ball screw 1224, enabling elastic floating of the sensor array within a certain range and providing limiting guidance for the vertical displacement of the sensor mounting plate 125.

[0033] The attitude adjustment joint 123 is a universal joint structure, located at both ends of the sensor mounting plate 125, with the other end connected to the sensor mounting base 126. Structurally, it includes a first joint plate parallel to the sensor mounting plate 125 and a second joint plate parallel to the sensor mounting base 126. The first and second joint plates are fixedly connected or integrally formed, completing the directional movable connection between the non-parallel sensor mounting plate 125 and the sensor mounting base 126.

[0034] The fixed end of the vertical torque motor 1232 is set on the sensor mounting plate 125 along the horizontal longitudinal direction. Specifically, the "horizontal longitudinal direction" is perpendicular to the vertically set sensor mounting plate 125. Its rotation axis passes through the attitude adjustment joint 123 (the first joint plate parallel to the sensor mounting plate 125) and is fixedly connected to the attitude adjustment joint 123, driving the attitude adjustment joint 123 to swing in the vertical direction (vertical plane).

[0035] The fixed end of the horizontal torque motor 1231 is vertically mounted on the attitude adjustment joint 123 (the second joint plate parallel to the sensor mounting base 126), and its rotation axis passes through the sensor mounting base 126 and is fixedly connected to the sensor mounting base 126, driving the sensor mounting base 126 to swing horizontally; the attitude adjustment joint 123, together with the horizontal torque motor 1231 and the vertical torque motor 1232, realizes real-time adjustment of the pitch and yaw angles of the sensor array.

[0036] An inertial sensor 1233 is also provided on the sensor mounting base 126 for feedback data to achieve closed-loop control and to detect the parallelism between the reference plane of the sensor array and the center line of the track. The inertial sensor 1233 is based on Newton's law of inertia. It detects the displacement, force or vibration changes caused by inertia when the sensor mounting base 126 moves, and uses physical effects such as capacitance and piezoresistive force to convert the changes into electrical signals, thereby measuring the linear acceleration, angular velocity and other motion parameters of the sensor mounting base 126 in real time.

[0037] The sensor mounting base 126 is also equipped with a grating ruler 1222, whose detection head is set towards the track panel to monitor the displacement of the sensor array adjustment mechanism 12 in real time. The grating ruler 1222 is based on the optical moiré fringe principle. When the scale grating and the indicator grating are relatively displaced, alternating bright and dark moiré fringes are generated. The photoelectric receiving element converts the light intensity change of the fringes into electrical pulse signals. By counting the number of pulses and judging the phase difference, the displacement and direction of movement are measured in real time with high precision.

[0038] In magnetic marking module 2, the magnetic marking unit includes permanent magnets set at the bottom of the rail and the ends of the sleepers of track 3, serving as positioning references for magnetic force sensing. These magnets interact with the magnetic force sensing detection module 1, and the marking intervals of the magnetic marking units match the sleeper spacing. The magnetic marking fixing device consists of a rail marking holder and a sleeper marking embedded part. The rail marking holder is made of austenitic stainless steel and fits against the bottom of the rail through a dovetail groove structure. The sleeper marking embedded part adopts a T-shaped structure design, with one side of the T-shaped structure embedded in the sleeper and the other end exposed externally, and it integrates a permalloy magnetic shielding sleeve.

[0039] The system hardware modules include multi-source data acquisition hardware, motion control hardware, and environmental monitoring hardware. The multi-source data acquisition hardware adopts an FPGA+ARM architecture to achieve synchronous acquisition and preprocessing of sensor array signals. The motion control hardware uses a PLC main controller, in conjunction with a servo driver, to achieve precise control of the adjustment mechanism. The environmental monitoring hardware integrates temperature, humidity, and dust concentration sensors. The protection module includes a permalloy magnetic shielding layer, a micro-positive pressure maintenance device, and a temperature control device to suppress external magnetic field interference, prevent dust intrusion, and maintain the system's operating temperature. The robot connection mechanism consists of a quick-change interface assembly and a shock-absorbing unit. The quick-change interface assembly adopts a tenon-and-mortise positioning structure, in conjunction with a pneumatic locking device. The shock-absorbing unit contains two sets of metal-rubber shock absorbers to achieve vibration attenuation in the XYZ axes.

[0040] Example 1: In addition, as a preferred technical solution, the sensor array adjustment mechanism 12 is also provided with a high-precision lateral compensation component 1211, which is disposed between the sensor mounting plate 125 and the vertical moving bracket 1223, and is connected to the vertical moving bracket 1223 in place of the ear plate of the sensor mounting plate 125.

[0041] The high-precision lateral compensation component 1211 includes a guide rail base that is threadedly engaged with a vertical ball screw 1224 on a vertical moving bracket 1223. The rear side of the guide rail base also has an ear plate with an internally threaded hole that is threadedly engaged with the vertical ball screw 1224. It also includes a guide rail horizontally mounted on the guide rail base, a slider that slides along the guide rail, one side of the slider being fixedly connected to a sensor mounting plate 125, and the telescopic end of a telescopic cylinder mounted on the guide rail base being fixedly connected to the slider. The telescopic end of the telescopic cylinder drives the slider to reciprocate on the guide rail, causing the sensor mounting plate 125 to reciprocate horizontally. This precisely controls the position of the sensor array mounted on the sensor mounting plate 125, ensuring that the axis of symmetry of the sensor array 11 on the sensor mounting plate 125 completely coincides with the axis of symmetry of the single-strand track 3, resulting in the sensor array 11 being highly symmetrically positioned on both sides of the single-strand track 3, thus improving detection accuracy.

[0042] Preferably, the alignment of the symmetry axis of the sensor array 11 with the symmetry axis of the single track 3 can be achieved by reflecting the symmetry of the device with respect to the single track 3 through the detection parameters of the combined sensor array 11, the grating ruler 1222 and the inertial sensor 1233, thereby controlling the movement stroke of the sensor array adjustment mechanism 12 to achieve alignment.

[0043] Example 2: This invention also provides a method for using a track state holographic detection system based on magnetic sensing. The method employs the aforementioned track state holographic detection system based on magnetic sensing and includes the following steps: acquiring magnetic field distribution data collected by multi-channel closed-loop linear Hall sensors distributed along the rail cross-section contour; preprocessing the magnetic field distribution data, including using wavelet threshold filtering to eliminate pulse interference, adaptive Kalman filtering to compensate for environmental magnetic field drift, temperature compensation to correct sensor temperature drift, and wavelet transform to remove low-frequency drift; inputting the preprocessed magnetic field distribution data into an improved U-Net network, which uses MobileNetV2-style inverted residual block reconstruction of basic feature extraction units, and performs 1×1 convolution to enhance... The process of using dimensionality- and depth-separable convolutional feature extraction and 1×1 convolutional dimensionality reduction reduces the number of parameters. An ECA attention module and a gated attention mechanism are integrated between the encoder and decoder. The ECA attention module dynamically learns channel dependencies through adaptive average pooling and lightweight 1D convolution, while the gated attention mechanism generates an adaptive weight map for refined feature selection. Bilinear interpolation is used instead of transposed convolution as an upsampling method, combined with 1×1 convolution for channel matching and 3×3 convolution for feature refinement. An improved U-Net network is used to process magnetic field images and predict key rail dimensions. The predicted key rail dimensions and geometric parameters are fused to generate a digital holographic model of the track section, achieving holographic detection of three track states. The improved U-Net network has a parameter increment of less than 1% in the ECA attention module, and the dual attention mechanism works together to improve the model's ability to capture features of target boundaries and small defects. The improved U-Net network uses bilinear interpolation as an upsampling method to avoid the checkerboard artifact problem caused by transposed convolution, ensuring the smoothness and spatial consistency of the output segmentation map. The improved U-Net network adopts a lightweight architecture design, which significantly improves the inference speed and can be deployed on edge devices or embedded systems to process track 3 detection data in real time in resource-constrained engineering scenarios.

[0044] In the actual operation, the system first initializes, performing dynamic magnetic field calibration, sensor calibration, and mechanism reset. The operator selects the track type through the human-machine interface. Then, the robot is fine-tuned and moved to the inspection station, and the support device unfolds and locks. The sensor array adjustment mechanism 12 adjusts according to the built-in technical parameters of the current track type. The lateral movement component 121 moves the vertical mechanism above the initial inspection position, and the vertical motor 124 drives the vertical mechanism downward until the closed-loop linear Hall sensor array 11 is in the preset position. Under the real-time detection and control of the disc spring assembly 1221 and the grating ruler 1222, the sensor array floats elastically. The attitude adjustment joint 123 adjusts the pitch and yaw angles of the sensor array in real time through the horizontal torque motor 1231 and the vertical torque motor 1232. Closed-loop control is achieved through feedback data from the inertial sensor 1233 to ensure the parallelism between the inspection reference plane and the center line of the track. After all sensors are in position, the magnetic sensor array automatically starts, and the robot drives the inspection system to move along the track for continuous sampling, synchronously collecting magnetic data and transmitting it to the software system in real time. The software system performs noise reduction on the raw data, generates a 3D model of the track, calculates the deviations of key geometric parameters from the design values, and generates a deviation report. The fine-tuning decision layer generates an adjustment plan based on the deviation data, and the control execution layer drives the automatic track-setting robot to perform adjustments, conducting an intermediate check after each certain displacement. After the track adjustment is completed, the system automatically performs a re-check. If all parameter deviations are within the allowable range, the fine-tuning is deemed qualified; otherwise, a second adjustment is performed. Upon successful completion, the system stores the inspection report and adjustment record, including the raw data, holographic model, deviation curve, and adjustment parameters, providing data support for subsequent analysis.

[0045] Example 3: As a solution that can be used directly, the present invention provides a specific implementation scheme.

[0046] The overall mechanical architecture of the system consists of a magnetic sensing and detection module 1, a magnetic marking module 2, a system hardware module, and a protection module.

[0047] The magnetic force sensing detection module 1 consists of a closed-loop linear Hall sensor array 11 and a sensor array 11 adjustment mechanism. The 24-channel closed-loop linear Hall sensor array 11 is the core component of the entire device, and it is distributed along the rail cross-section profile, with 10 channels in the rail head region, 8 channels in the rail web region, and 4 channels in the rail bottom region (the accuracy is determined according to the detection accuracy, with the highest accuracy requirement in the rail head region).

[0048] The sensor array 11 adjustment mechanism consists of a lateral movement component 121, a vertical compensation mechanism 122, and an attitude adjustment joint 123. The lateral movement component 121 employs a ball screw drive structure, using a servo motor to drive the sensor mounting seats on both sides to move symmetrically along a high-precision linear guide rail, adapting to the gauge requirements of different rail specifications. The vertical compensation mechanism 122 integrates a disc spring assembly and a grating ruler 1222, enabling the sensor array to elastically float within a certain range, compensating for detection height deviations caused by rail surface unevenness, and ensuring a constant distance between the sensor and the rail surface. The attitude adjustment joint 123 uses an abstract universal joint structure, working with horizontal and vertical torque motors to achieve real-time adjustment of the pitch and yaw angles of the sensor array 11. Closed-loop control using feedback data from the inertial sensor 1233 ensures that the parallelism between the detection reference plane and the rail centerline is below a certain value.

[0049] Magnetic marking module 2 consists of magnetic marking units and magnetic marking fixing devices. The magnetic marking units have pre-installed high-remanence neodymium iron boron permanent magnets at the rail base and sleeper ends as positioning references for magnetic force sensing, with marking intervals matching the sleeper spacing. The magnetic marking fixing device consists of rail marking holders and sleeper marking embedded parts. The rail marking holders are forged from austenitic stainless steel and fit snugly against the bottom of the rail via a dovetail groove structure, housing powerful permanent magnets. The sleeper marking embedded parts adopt a T-shaped structure design, made of tempered 45# steel, with an integrated permalloy magnetic shielding sleeve at the exposed end to ensure the marking magnetic field direction is perpendicular to the sleeper surface, with a perpendicularity error below a certain value.

[0050] The system hardware module consists of multi-source data acquisition hardware, motion control hardware, and environmental monitoring hardware. The multi-source data acquisition hardware adopts an "FPGA+ARM" architecture. The positioning module integrates a UWB ultra-wideband positioning unit, combining the absolute position information from the sleeper magnetic markers to achieve real-time coordinate calibration of the detection system along the track panel. The data acquisition card uses a high-speed AD converter to simultaneously acquire magnetic sensor data, attitude data from the 1233 inertial sensor, and vertical distance data from the laser displacement sensor. The FPGA chip is responsible for the synchronous acquisition and preprocessing of sensor signals, with a built-in digital filtering algorithm to eliminate high-frequency noise. The ARM processor is responsible for data caching, protocol conversion, and communication with the host computer. The communication interface uses Profinet industrial Ethernet to achieve low-latency data interaction with the automatic fine-tuning robot control cabinet. The motion control hardware uses a PLC main controller as its core, working with servo drives to achieve precise control of the adjustment mechanism and fine-tuning actuator. The system is equipped with communication interfaces such as EtherCAT, supporting real-time control command transmission and status feedback. Environmental monitoring hardware integrates temperature sensors, humidity sensors, dust concentration sensors, etc., to trigger corresponding protective measures, such as activating heating / heat dissipation, dust removal, and other functions.

[0051] In terms of anti-interference design, the main components of the protection module employ a permalloy magnetic shielding layer for sensor encapsulation to suppress interference from the Earth's magnetic field and train electromagnetic radiation. For sealing protection, a micro-positive pressure maintenance device is installed internally to prevent dust intrusion. Regarding temperature control, the device integrates heating and heat sinks, using a temperature sensor to monitor the internal temperature in real time, automatically heating at low temperatures and automatically cooling at high temperatures to ensure the ideal operating temperature range for electronic components. For extreme working environments, the following additional structures are also included:

[0052] In high-temperature environments, the transmission components utilize PTFE self-lubricating bearings to prevent high-temperature grease failure. The outer shell features a honeycomb heat dissipation structure, coupled with a miniature axial flow fan for forced cooling. In low-temperature environments, the moving joints use low-temperature grease, with heating elements maintaining the joint temperature above a certain level. Connecting bolts are made of low-temperature steel to prevent low-temperature brittle fracture. The inner layer of the protective shell features an aerogel insulation layer to reduce condensation caused by temperature differences between the inside and outside. In dusty / water mist environments, the moving parts employ a labyrinthine sealing structure, combined with a V-shaped dustproof ring, raising the protection level to IP69K. The optical window uses sapphire glass with a hydrophobic coating to reduce water mist adhesion. An automatic cleaning mechanism is incorporated, using a miniature air pump to periodically spray compressed air to remove surface dust. The air nozzle uses a fan-shaped nozzle to cover all detection windows.

[0053] The robot connection mechanism consists of a quick-change interface assembly and a shock-absorbing unit. The quick-change interface assembly adopts a mortise and tenon positioning structure, combined with a pneumatic locking device, to achieve rapid docking between the detection system and the fine-tuning robot. The shock-absorbing unit connects two sets of metal-rubber shock absorbers in series to attenuate vibrations along the XYZ axes. The mounting base is equipped with stress relief grooves to prevent structural deformation caused by impacts during robot movement.

[0054] It also includes the system software architecture.

[0055] The overall software architecture is as follows: The system software adopts a layered architecture design, including the overall software architecture, data acquisition and preprocessing layer, feature extraction and parameter calculation layer, holographic modeling layer, fine-tuning decision layer, control execution layer, and data storage and interaction layer. Each layer interacts with data through standardized interfaces.

[0056] The data acquisition and preprocessing layer consists of: acquisition parameters that can be configured via software, including sampling frequency, filtering parameters, triggering conditions, etc., to perform noise suppression and feature enhancement on the raw sensor data; interference suppression algorithms: wavelet threshold filtering is used to eliminate pulse interference, combined with adaptive Kalman filtering to compensate for environmental magnetic field drift; temperature compensation algorithm is used to correct the influence of sensor temperature drift; wavelet transform is applied to remove low-frequency drift; and Kalman filtering algorithm is used to eliminate random noise.

[0057] The data synchronization mechanism is as follows: based on the UWB positioning timestamp, a linear interpolation method is used to achieve millisecond-level synchronization of magnetic, IMU, and laser data.

[0058] The preprocessed data is stored in binary format with timestamps for easy subsequent analysis.

[0059] The feature extraction and parameter calculation layer consists of the following: Track gauge calculation: The magnetic field peak spacing of the magnetic markers is detected by laterally symmetrically arranged Hall sensors, combined with a temperature compensation model; Elevation / track alignment calculation: Based on the pitch / yaw angle output by inertial sensor 1233, combined with vertical / lateral distance data, the track centerline is fitted using the least squares method; Twist degree calculation: The twist degree per unit length is calculated by detecting the rail top height difference through two sets of sensors with a longitudinal spacing of 200mm, combined with track gauge data; Fastener position detection: The center coordinates of the fastener are located by identifying the magnetic field characteristics (phase difference, amplitude ratio) of the sleeper magnetic markers.

[0060] The holographic modeling layer employs a hybrid modeling method that integrates physical constraints and deep learning. It fuses geometric parameters and fastener position data to generate a digital model of the rail section, supporting visualization and historical data comparison. Based on the 3D MagneticInverse Routine algorithm framework (based on a three-dimensional magnetic inversion algorithm framework), it uses magnetic field distribution data to invert the three-dimensional contour of the rail. A CNN model (an improved U-Net structure) is introduced to process the magnetic field image and predict key rail dimensional parameters. Harmonic analysis methods from MPI imaging technology are combined to extract higher-order harmonic features of the magnetic field, improving the accuracy of 3D reconstruction. A structure-aware temporal bilateral filtering algorithm is used to optimize the dynamic detection sequence. The modeling output includes key parameters such as track gauge and rail top elevation, with a data update rate ≥10Hz.

[0061] The fine-tuning decision layer consists of: based on the deviation analysis between the test data and the design standards, PID feedback control is used to generate adjustment commands, realizing closed-loop control of test-fine-tuning, and generating a fine-tuning scheme: Deviation calculation: compared with the design BIM model, the deviation values ​​of each degree of freedom in three-dimensional space are calculated; Path planning: the particle swarm optimization algorithm is used to plan the fine-tuning execution path to avoid interference during the adjustment process; Safety verification: the adjustment process is simulated to verify whether the equipment travel and load limits are exceeded; Decision output: the target position parameters of the six-degree-of-freedom platform are generated.

[0062] The control execution layer converts the target parameters output by the decision layer into control commands; it uses a PID + feedforward control algorithm to achieve closed-loop position control; it supports three control modes: position mode, speed mode, and torque mode; it achieves synchronous control of multi-axis motion; and it has safety functions such as limit protection, overload protection, and emergency stop.

[0063] The data storage and interaction layer provides a visual operation interface and data display function: real-time display of the orbit 3 holographic model and deviation data; supports switching between manual and automatic operation modes; records detection data and adjustment history, supports data export and report generation; and provides a remote communication interface to support cloud data synchronization and remote monitoring.

[0064] Compared with the prior art, the present invention has the following main advantages:

[0065] 1. High Detection Accuracy: Utilizing a magnetic sensor array 11 and multi-source data fusion technology, sub-millimeter level detection of multiple parameters such as track gauge and alignment is achieved (key parameter accuracy reaches ±0.1mm); 2. Strong Environmental Adaptability: Employing magnetic shielding and dynamic compensation technology, the detection success rate is ≥99.5% in dusty and water mist environments, an improvement of over 60% compared to traditional optical equipment; its extreme environment design allows it to operate stably in temperature ranges of -60-80℃ and high-vibration environments; 3. Full Track Adaptability: Through a 1300-1600mm wide-range adjustment mechanism and modular sensor head design, it adapts to various track types, including slotted tracks, improving equipment versatility by 60%; 4. Intelligent and Efficient: Detection efficiency reaches 200m / h, achieving closed-loop control with a fine-tuning robot (response time ≤1s), significantly reducing labor costs and operational risks; 5. Green and Environmentally Friendly: Adopting a wide-voltage redundant power supply system to replace traditional internal combustion engine drive, resulting in low noise and zero emissions, in line with green construction concepts.

[0066] The overall mechanical architecture of the system consists of a magnetic sensing and detection module 1, a magnetic marking module 2, system hardware modules, and a protection module. The magnetic sensing and detection module 1 comprises a closed-loop linear Hall effect sensor array 11 and a sensor array adjustment mechanism 12. The 24-channel closed-loop linear Hall effect sensor array 11 is the core component of the entire device, distributed along the rail cross-sectional profile, with ten channels in the rail head region, eight channels in the rail web region, and four channels in the rail bottom region (the specific channels are determined by the detection accuracy, with the highest accuracy requirement in the rail head region).

[0067] The sensor array adjustment mechanism 12 consists of a lateral movement component 121, a vertical compensation mechanism 122, and an attitude adjustment joint 123. The lateral movement component 121 employs a ball screw drive structure, using a servo motor to drive the sensor mounting seats on both sides to move symmetrically along a high-precision linear guide rail, adapting to the gauge requirements of different rail specifications. The vertical compensation mechanism 122 integrates a disc spring assembly 1221 and a grating ruler 1222, enabling the sensor array to elastically float within a certain range, compensating for detection height deviations caused by rail surface unevenness, and ensuring a constant distance between the sensor and the rail surface. The attitude adjustment joint 123 adopts an abstract universal joint structure, working with a horizontal torque motor 1231 and a vertical torque motor 1232 to achieve real-time adjustment of the pitch and yaw angles of the sensor array 11. Closed-loop control using feedback data from the inertial sensor 1233 ensures that the parallelism between the detection reference plane and the rail centerline is below a certain value. In addition, a high-precision lateral compensation component 1211 is added, which can precisely adjust the final position of the sensor array so that its axis of symmetry is completely coincident with the axis of symmetry of the single rail.

[0068] Magnetic marking module 2 consists of magnetic marking units and magnetic marking fixing devices. The magnetic marking units have pre-installed high-remanence neodymium iron boron permanent magnets at the rail base and sleeper ends as positioning references for magnetic force sensing, with marking intervals matching the sleeper spacing. The magnetic marking fixing device consists of rail marking holders and sleeper marking embedded parts. The rail marking holders are forged from austenitic stainless steel and fit snugly against the bottom of the rail via a dovetail groove structure, housing powerful permanent magnets. The sleeper marking embedded parts adopt a T-shaped structure design, made of tempered 45# steel, with an integrated permalloy magnetic shielding sleeve at the exposed end to ensure the marking magnetic field direction is perpendicular to the sleeper surface, with a perpendicularity error below a certain value.

[0069] The system hardware module consists of multi-source data acquisition hardware, motion control hardware, and environmental monitoring hardware. The multi-source data acquisition hardware adopts an "FPGA+ARM" architecture. The positioning module integrates a UWB ultra-wideband positioning unit, combining the absolute position information of the sleeper magnetic markers to achieve real-time coordinate calibration of the detection system along the track panel. The data acquisition card uses a high-speed AD converter to simultaneously acquire magnetic sensor data, inertial sensor attitude data, and laser displacement sensor vertical distance data. The FPGA chip is responsible for the synchronous acquisition and preprocessing of sensor signals, with a built-in digital filtering algorithm to eliminate high-frequency noise. The ARM processor is responsible for data caching, protocol conversion, and communication with the host computer. The communication interface uses Profinet industrial Ethernet to achieve low-latency data interaction with the automatic fine-tuning robot control cabinet. The motion control hardware uses a PLC main controller as its core, working with servo drivers to achieve precise control of the adjustment and fine-tuning actuators. The system is equipped with communication interfaces such as EtherCAT to support real-time control command transmission and status feedback. The environmental monitoring hardware integrates temperature sensors, humidity sensors, and dust concentration sensors to trigger corresponding protective measures, such as activating heating / heat dissipation and dust removal functions.

[0070] In terms of anti-interference design, the main components of the protection module employ a permalloy magnetic shielding layer for sensor encapsulation to suppress interference from the Earth's magnetic field and train electromagnetic radiation. For sealing protection, a micro-positive pressure maintenance device is installed internally to prevent dust intrusion. Regarding temperature control, the device integrates heating and heat sinks, using a temperature sensor to monitor the internal temperature in real time, automatically heating at low temperatures and automatically cooling at high temperatures to ensure the ideal operating temperature range for electronic components. For extreme working environments, the following additional structures are also included:

[0071] In high-temperature environments, the transmission components utilize PTFE self-lubricating bearings to prevent high-temperature grease failure. The outer shell features a honeycomb heat dissipation structure, coupled with a miniature axial flow fan for forced cooling. In low-temperature environments, the moving joints use low-temperature grease, with heating elements maintaining the joint temperature above a certain level. Connecting bolts are made of low-temperature steel to prevent low-temperature brittle fracture. The inner layer of the protective shell features an aerogel insulation layer to reduce condensation caused by temperature differences between the inside and outside. In dusty / water mist environments, the moving parts employ a labyrinthine sealing structure, combined with a V-shaped dustproof ring, raising the protection level to IP69K. The optical window uses sapphire glass with a hydrophobic coating to reduce water mist adhesion. An automatic cleaning mechanism is incorporated, using a miniature air pump to periodically spray compressed air to remove surface dust. The air nozzle uses a fan-shaped nozzle to cover all detection windows.

[0072] The robot connection mechanism consists of a quick-change interface assembly and a shock-absorbing unit. The quick-change interface assembly adopts a mortise and tenon positioning structure, combined with a pneumatic locking device, to achieve rapid docking between the detection system and the fine-tuning robot. The shock-absorbing unit connects two sets of metal-rubber shock absorbers in series to attenuate vibrations along the XYZ axes. The mounting base is equipped with stress relief grooves to prevent structural deformation caused by impacts during robot movement.

[0073] How to use a magnetic sensing-based holographic orbital state detection system:

[0074] 1. Initialization Phase: The system powers on and performs a self-test, then executes dynamic magnetic field calibration (collecting the magnetic field distribution of the tunnel environment and establishing a magnetic field interference model), sensor calibration (zero-point calibration, linearity calibration), and mechanism reset. The calibration time is ≤5 minutes. The operator selects the track type through the human-machine interface. 2. Positioning Phase: The robot is fine-tuned and moved to the inspection station. The support device is deployed and locked. The robot is initially aligned with the track using a laser positioning system.

[0075] 2. Adaptation and Adjustment: Based on the technical parameters of the current rail type built into the system, the lateral movement component 121 of the sensor array adjustment mechanism 12 moves the vertical mechanism along the linear guide rail to above the initial detection position, and the sensor confirms its relative position with the rail. The vertical motor 124 drives the vertical mechanism downward until the eight-channel portion of the closed-loop linear Hall sensor array 11 is exactly in the rail web area. Under the real-time detection and control of the disc spring assembly 1221 and the grating ruler 1222, the sensor array can elastically float within a certain range to compensate for the detection height deviation caused by the unevenness of the rail surface, ensuring that the distance between the sensor and the rail surface remains constant. The attitude adjustment joint 123 realizes the real-time adjustment of the pitch and yaw angles of the sensor array 11 through the horizontal torque motor 1231 and the vertical torque motor 1232, and the closed-loop control through the feedback data of the inertial sensor 1233 ensures that the parallelism between the detection reference plane and the center line of the rail panel is below a certain value.

[0076] 3. Detection phase: After each sensor is in place, the magnetic sensor array starts automatically. The robot drives the detection system to move along the track to continuously sample and collect magnetic force synchronously, which is then transmitted to the software system in real time.

[0077] 4. Holographic Modeling: The preprocessing layer performs noise reduction on the original data, and the holographic modeling layer generates a 3D model of the track in real time, calculates the deviation of key geometric parameters from the design values, and generates a deviation report. The deviation data is displayed intuitively on the holographic model through color coding.

[0078] 5. Fine-tuning execution: The fine-tuning decision layer generates an adjustment plan based on the deviation data, and the control execution layer drives the automatic fine-tuning robot of the track panel to adjust according to the planned path. An intermediate check is performed every 50mm of displacement to ensure that the adjustment direction is correct.

[0079] 6. Re-inspection and confirmation: After the track panel adjustment is completed, the system will automatically perform a re-inspection. If all parameter deviations are within the allowable range and meet the specification requirements, the fine adjustment is deemed qualified; otherwise, a second adjustment will be performed.

[0080] 7. Data storage: After passing the test, the test report and adjustment records are stored, including the original data, holographic model, deviation curve and adjustment parameters, to provide data support for subsequent analysis.

[0081] Example 4: To adapt to the dual requirements of deployment efficiency and segmentation accuracy in actual engineering detection scenarios, this solution has been specifically optimized for the traditional U-Net. The specific improvements are as follows:

[0082] 1. Lightweight architecture reconstruction for resource-constrained deployments: By introducing MobileNetV2-style inverted residual blocks, the network's basic feature extraction unit is reconstructed. The inverted residual block employs an efficient workflow of "1×1 convolution dimensionality upscaling - depthwise separable convolution feature extraction - 1×1 convolution dimensionality reduction," significantly reducing model parameter size and floating-point computation while preserving feature representation capabilities. Compared to traditional convolutional blocks, it effectively reduces the number of parameters, significantly improves inference speed, and can be flexibly deployed in resource-constrained engineering scenarios such as edge devices and embedded systems.

[0083] 2. Dual Attention Fusion for Enhanced Key Feature Capture: A dual-layer feature optimization system, combining the ECA attention module and a gated attention mechanism, is constructed: "channel enhancement - semantic filtering." The ECA attention module dynamically learns inter-channel dependencies through adaptive average pooling and lightweight 1D convolution, precisely enhancing key channel features with parameter increments controlled within 1%. The gated attention mechanism embeds the feature interaction path between the encoder and decoder, generating an adaptive weight map for refined feature filtering. The synergistic effect of these dual attention mechanisms effectively improves the model's ability to capture features from key regions such as target boundaries and small defects, optimizing feature representation accuracy in complex scenarios.

[0084] 3. Upgraded upsampling strategy to ensure output stability: Bilinear interpolation is used instead of traditional transposed convolution as the core upsampling method. This restores the feature map size while avoiding the "checkerboard artifact" problem that transposed convolution is prone to. Combined with subsequent 1×1 convolution channel matching and 3×3 convolution feature refinement, the smoothness and spatial consistency of the output segmentation map are guaranteed. Furthermore, the lightweight design further reduces computational overhead, making the segmentation results more in line with the actual requirements of engineering detection for accuracy and stability.

[0085] The above description is merely a specific embodiment of the invention, but the scope of protection of the invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the invention, based on the technical solution and concept of the invention, should be covered within the scope of protection of the invention.

Claims

1. A holographic detection system for track state based on magnetic force sensing, characterized in that, Includes a magnetic force sensing and detection module and a magnetic marking module; The magnetic force sensing detection module includes: The sensor array adjustment mechanism has sensor mounting bases symmetrically arranged along both sides of the rail at its end, and can move in multiple directions. A closed-loop linear Hall effect sensor array, mounted on a sensor mounting base and distributed along the rail cross-sectional profile, includes multi-channel sensors in the rail head, rail web, and rail bottom regions. The sensor array adjustment mechanism adjusts the spatial position and orientation of the sensor array by performing multi-directional movements; The magnetic marking module includes: The magnetic marking unit includes permanent magnets set on the bottom of the track and the ends of the sleepers. The magnetic marking unit serves as a positioning reference for magnetic force sensing and reacts with the magnetic force sensing detection module. The marking interval of the magnetic marking unit matches the sleeper spacing. The sensor array adjustment mechanism includes: The lateral movement assembly includes a lateral movement bracket, which is mounted on a track via two legs. The lateral movement bracket is also equipped with a lateral ball screw, which is driven by a servo motor located at one end. The vertical compensation mechanism includes a vertical moving bracket, which is threaded onto a horizontal ball screw and moves laterally with the horizontal ball screw. The vertical moving bracket also has a vertical ball screw, which is driven by a vertical motor located at one end. The sensor mounting plate is threaded onto the vertical ball screw; The attitude adjustment joint is a cross-shaped universal joint structure, located at both ends of the sensor mounting plate, with the other end connected to the sensor mounting base; A vertical torque motor is mounted on the sensor mounting plate along the horizontal longitudinal direction. The rotation shaft of the vertical torque motor is mounted through the attitude adjustment joint, driving the attitude adjustment joint to swing in the vertical direction. A horizontal torque motor is mounted vertically on the attitude adjustment joint. The rotation shaft of the horizontal torque motor is mounted through the sensor mounting base, driving the sensor mounting base to swing horizontally. The attitude adjustment joint, in conjunction with the horizontal torque motor and the vertical torque motor, enables real-time adjustment of the pitch and yaw angles of the sensor array; The sensor array adjustment mechanism also includes a high-precision lateral compensation component, which is positioned between the sensor mounting plate and the vertical moving bracket, and includes: The guide rail base is threaded with the vertical ball screw on the vertical moving bracket. The guide rail is horizontally set on the guide rail base. The slider is slidably engaged with the guide rail. One side of the slider is fixedly connected to the sensor mounting plate. The telescopic end of the telescopic cylinder set on the guide rail base is fixedly connected to the slider. The telescopic end of the telescopic cylinder drives the slider to reciprocate on the slide rail, which drives the sensor mounting plate to reciprocate in the horizontal direction, so that the symmetry axis of the sensor array on the sensor mounting plate is completely coincident with the symmetry axis of the single track. It also includes a disc spring assembly disposed between the sensor mounting plate and the vertical motor. The disc spring assembly is sleeved on the vertical ball screw and is composed of multiple standard conical disc springs stacked together, which realizes the elastic floating of the sensor array within a certain range and provides limiting guidance for the vertical displacement of the sensor mounting plate.

2. The orbital state holographic detection system based on magnetic sensing as described in claim 1, characterized in that, The closed-loop linear Hall sensor array includes twenty-four channels, with ten channels arranged in the rail head region, eight channels arranged in the rail waist region, and four channels arranged in the rail bottom region.

3. The orbital state holographic detection system based on magnetic sensing as described in claim 1, characterized in that, The sensor mounting base is also equipped with: Inertial sensors provide feedback data to enable closed-loop control and detect the parallelism between the reference plane of the sensor array and the center line of the track. The grating ruler, with its detection head facing the track, monitors the displacement of the sensor array adjustment mechanism in real time.

4. The orbital state holographic detection system based on magnetic sensing as described in claim 1, characterized in that, The magnetic marking module also includes a magnetic marking fixing device, which consists of a rail marking holder and a sleeper marking embedded part. The rail marking holder is made of austenitic stainless steel and fits against the bottom of the rail through a dovetail groove structure. The sleeper marking embedded part adopts a T-shaped structure design, with one side of the T-shaped structure embedded in the sleeper and the other end exposed to the outside, and integrates a permalloy magnetic shielding sleeve.

5. The orbital state holographic detection system based on magnetic sensing as described in claim 1, characterized in that, It also includes system hardware modules, which include multi-source data acquisition hardware, motion control hardware, and environmental monitoring hardware; The multi-source data acquisition hardware adopts an FPGA+ARM architecture to realize the synchronous acquisition and preprocessing of sensor array signals. The motion control hardware uses a PLC main controller, in conjunction with a servo driver, to achieve precise control of the adjustment mechanism; the environmental monitoring hardware integrates a temperature sensor, a humidity sensor, and a dust concentration sensor. It also includes a protection module, which includes a permalloy magnetic shielding layer, a micro-positive pressure maintaining device, and a temperature control device, used to suppress external magnetic field interference, prevent dust intrusion, and maintain the system operating temperature. It also includes a robot connection mechanism, which consists of a quick-change interface assembly and a buffer and shock absorption unit; the quick-change interface assembly adopts a tenon and mortise positioning structure and is equipped with a pneumatic locking device; the buffer and shock absorption unit includes two sets of metal rubber shock absorbers to achieve vibration attenuation in the XYZ three axes.

6. A method of using a magnetic sensing-based holographic detection system for orbital state, comprising the magnetic sensing-based holographic detection system for orbital state as described in any one of claims 1-5, characterized in that, The method includes the following steps: Acquire magnetic field distribution data collected by multi-channel closed-loop linear Hall sensors distributed along the profile of the rail section; The magnetic field distribution data is preprocessed, including using wavelet threshold filtering to eliminate pulse interference, adaptive Kalman filtering to compensate for environmental magnetic field drift, temperature compensation to correct sensor temperature drift, and wavelet transform to remove low-frequency drift. The preprocessed magnetic field distribution data is input into an improved U-Net network. The improved U-Net network uses inverted residual blocks in the style of MobileNetV2 to reconstruct the basic feature extraction unit. The number of parameters is reduced through a process of 1×1 convolution dimensionality upscaling, depthwise separable convolution feature extraction, and 1×1 convolution dimensionality reduction. An ECA attention module and a gated attention mechanism are fused between the encoder and decoder. The ECA attention module dynamically learns the channel dependencies through adaptive average pooling and lightweight 1D convolution. The gated attention mechanism generates an adaptive weight map to refine the features. Bilinear interpolation is used to replace transposed convolution as an upsampling method, combined with 1×1 convolution for channel matching and 3×3 convolution for feature refinement. The improved U-Net network is used to process magnetic field images and predict key dimensional parameters of the rail. By integrating the predicted key dimensional and geometric parameters of the rail, a digital holographic model of the rail section is generated, enabling holographic detection of the track condition.

7. The method of using the orbital state holographic detection system based on magnetic sensing as described in claim 6, characterized in that, The parameter increment of the ECA attention module in the improved U-Net network is controlled within 1%, and the dual attention mechanism works together to improve the model's ability to capture features of target boundaries and small-sized defects. The improved U-Net network uses bilinear interpolation as an upsampling method, which avoids the checkerboard artifact problem caused by transposed convolution and ensures the smoothness and spatial consistency of the output segmentation map. The improved U-Net network adopts a lightweight architecture design, which can be deployed on edge devices or embedded systems to process track detection data in real time in resource-constrained engineering scenarios.

Images