Magnetic suspension gap monitoring method and system based on image processing

Abstract

The invention discloses a magnetic suspension gap monitoring method and system based on image processing, and belongs to the field of image analysis, and the method comprises the steps: collecting an image sequence of a magnetic suspension bearing gap, and an air pressure data sequence and a temperature data sequence in a vacuum pipeline; calculating an optical distortion field through an image segmentation model according to the air pressure data and the temperature data at the same moment, carrying out geometric correction on the feature map of the image frame at the same moment based on the optical distortion field, and outputting a gap region segmentation sequence and a confidence coefficient sequence; calculating an optical flow field in a corresponding segmentation area in the image sequence to obtain a total displacement field, and subtracting the rigid motion displacement field from the total displacement field to obtain a gap micro-vibration displacement signal; and performing order analysis on the gap micro-vibration displacement signal to extract vibration characteristics, and weighting the vibration characteristics based on the confidence sequence to generate a monitoring result. According to the invention, accurate monitoring of the magnetic suspension gap can be realized under the extreme working condition of magnetic suspension of the high-speed vacuum pipeline.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image analysis, in particular to a magnetic suspension gap monitoring method and system based on image processing. BACKGROUND

[0002] The magnetic suspension train is an important development direction of future rail transportation, and the accurate monitoring of the suspension gap is the key to ensuring the safe and stable operation of the system. At present, the gap detection method based on image processing is widely studied due to its non-contact and high precision characteristics, such as using edge detection algorithm or convolutional neural network to extract the gap region from the image captured by the industrial camera. This kind of method performs well in the conventional environment, but faces serious challenges in the special application scenario of high-speed magnetic suspension train in vacuum tube.

[0003] Firstly, the dynamic changes of air pressure and temperature inside the vacuum tube will cause the change of air refractive index, resulting in nonlinear geometric distortion of the collected gap image. The existing image segmentation algorithm does not consider this physical environmental factor, which leads to a serious decline in the positioning accuracy of the gap edge and makes it difficult to meet the micron-level monitoring requirements. Secondly, the small gap changes caused by high-speed train running are coupled with the macroscopic motion of the train body in the image sequence, and the traditional image analysis method is difficult to effectively separate and extract the vibration characteristics representing the periodic excitation of the track. In addition, the inaccessibility of the vacuum tube environment makes camera window pollution unavoidable, and the existing technology lacks self-awareness of image quality degradation, which will lead to the performance degradation of the entire monitoring system once the image quality decreases. SUMMARY

[0004] The present application provides a magnetic suspension gap monitoring method and system based on image processing, which can realize accurate monitoring of the magnetic suspension gap under extreme working conditions of high-speed vacuum tube magnetic suspension.

[0005] The present application provides a magnetic suspension gap monitoring method based on image processing, comprising: Collecting an image sequence of the magnetic suspension bearing gap through the vacuum tube wall window, and synchronously acquiring an air pressure data sequence and a temperature data sequence in the vacuum tube; Inputting the image sequence, the air pressure data sequence and the temperature data sequence into a pre-trained image segmentation model according to the time alignment relationship, so that the image segmentation model calculates the optical distortion field according to the air pressure data and the temperature data at the same time, and performs geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence; Based on the gap region segmentation sequence, an optical flow field is calculated in a corresponding region in the image sequence to obtain a total displacement field of the magnetic suspension bearing gap, and a rigid motion displacement field is subtracted from the total displacement field to obtain a gap micro-vibration displacement signal; wherein the rigid motion displacement field is calculated by an inertial measurement unit data of the magnetic suspension train; An order analysis is performed on the gap micro-vibration displacement signal to extract vibration characteristics, an average of the confidence sequence is taken as a weight, the vibration characteristics are weighted, and a monitoring result of the magnetic suspension bearing gap is generated according to a weighting result; wherein the vibration characteristics include vibration energy of a magnetic suspension bearing rotation frequency and first vibration energy of a track welded joint through frequency.

[0006] The embodiment of the application can provide a spatio-temporal alignment multi-modal data source for subsequent image processing by synchronously collecting gap images and physical environment data in a vacuum pipeline; can dynamically compensate optical distortion caused by changes in air pressure and temperature in the vacuum pipeline by driving an image segmentation model with physical environment data for feature-level geometric correction; can extract micro-vibration signals representing dynamic characteristics of the bearing itself by separating the rigid motion of the vehicle body from the image optical flow to provide pure input for subsequent state recognition; and can generate intelligent monitoring results with state judgment and reliability evaluation by performing order analysis on the micro-vibration signals and weighting fusion with model confidence. Compared with the prior art which can only measure the static gap in a conventional environment, the application can realize accurate monitoring of the magnetic suspension gap under the extreme working conditions of high-speed vacuum pipeline magnetic suspension.

[0007] Further, the image sequence of the magnetic suspension bearing gap is collected through a vacuum pipeline wall window, and air pressure data sequence and temperature data sequence in the vacuum pipeline are synchronously acquired, including: A pulsed infrared laser light source and a high-speed camera equipped with a narrow-band filter are used to collect the image sequence through the vacuum pipeline wall window, and a timestamp is attached to each frame of the image sequence; wherein the filter wavelength of the narrow-band filter matches the wavelength of the pulsed infrared laser light source; Air pressure data and temperature data in the vacuum pipeline are acquired by a sensor, and air pressure data sequence and temperature data sequence that are time-synchronized with the image sequence are generated based on the timestamp.

[0008] The embodiment of the application can acquire high-contrast gap images in a vacuum pipeline environment and establish accurate corresponding relationships between multiple data sources by using an anti-interference optical scheme and a time sequence synchronization method.

[0009] Further, the image sequence, the air pressure data sequence and the temperature data sequence are input to a pre-trained image segmentation model in a time alignment relationship, so that the image segmentation model calculates an optical distortion field according to air pressure data and temperature data at the same time, and performs geometric correction on a feature map of an image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence, comprising: feature extraction is performed on the image sequence by an encoder to obtain each first feature map; wherein one image frame of the image sequence corresponds to one first feature map; Each optical environment parameter vector is obtained by mapping the air pressure data sequence and the temperature data sequence through a fully connected neural network layer; wherein the air pressure data and the temperature data at the same time in the air pressure data sequence and the temperature data sequence correspond to one optical environment parameter vector; Each optical environment parameter vector is respectively deconvoluted to generate each optical distortion field corresponding to the scale of each first feature map; Each second feature map is obtained by respectively multiplying each optical distortion field and the corresponding first feature map element by element; Each second feature map is upsampled by a decoder to extract fusion features, and the fusion features are processed through a double-branch output structure to obtain a gap region segmentation sequence and a confidence sequence; wherein at each upsample stage, the optical distortion field is upsampled to the corresponding scale and multiplied by the feature map at the corresponding stage element by element.

[0010] The embodiment of the application can realize adaptive geometric correction at the image feature level by constructing a coding and decoding network structure that fuses physical priors, and improve the accuracy of segmentation results in complex optical environments.

[0011] Further, based on the gap region segmentation sequence, an optical flow field is calculated in the corresponding region of the image sequence to obtain a total displacement field of the magnetic bearing gap, and the total displacement field is subtracted by a rigid motion displacement field to obtain a gap micro-vibration displacement signal, comprising: Based on the gap region segmentation sequence, corresponding gap region image blocks in the image sequence are intercepted, and a dense optical flow between the gap region image blocks of consecutive frames is calculated to generate a total displacement field; The residual displacement field is obtained by subtracting the rigid motion displacement field from the total displacement field pixel by pixel; The residual displacement field is spatially averaged in the magnetic bearing gap to obtain a gap micro-vibration displacement signal.

[0012] The embodiment of the present application can effectively inhibit the influence of background interference and macroscopic motion of the vehicle body, and accurately capture the microscopic dynamic change characteristics of the gap region, by calculating the optical flow in the accurately segmented gap region and decoupling the fused inertial data.

[0013] Further, the rigid motion displacement field is calculated by the inertial measurement unit data of the maglev train, and includes: Obtaining the inertial measurement unit data of the maglev train; Performing coordinate transformation operation from the train coordinate system to the image coordinate system on the inertial measurement unit data; Performing integral operation on the transformed inertial measurement unit data to generate the rigid motion displacement field in the image plane.

[0014] The embodiment of the present application can reconstruct the rigid displacement field in the image plane by converting the inertial data to the image coordinate system for motion solving, and provide compensation reference for micro-vibration separation.

[0015] Further, the order analysis on the gap micro-vibration displacement signal to extract the vibration characteristics includes: Obtaining the real-time speed signal of the maglev train; Resampling the gap micro-vibration displacement signal according to the real-time speed signal, and converting the signal from time domain to angle domain; Performing Fourier transform on the resampled gap micro-vibration displacement signal to obtain the order spectrum; Extracting the vibration energy of the maglev bearing rotation frequency and the first vibration energy of the track welded joint pass frequency from the order spectrum, and composing the vibration energy and the first vibration energy into a feature vector as the vibration characteristics.

[0016] The embodiment of the present application can eliminate the influence of variable speed working conditions and extract the vibration characteristics directly related to mechanical physical characteristics by converting the vibration signal to the angle domain related to the speed for analysis.

[0017] Further, the mean value of the confidence sequence is taken as the weight to weight the vibration characteristics, and the monitoring result of the maglev bearing gap is generated according to the weighting result, including: Calculating the mean value of the confidence sequence, and mapping the vibration characteristics to a first state score; Multiplying the mean value and the first state score to obtain a state score; Comparing the state score with a preset abnormal threshold, and generating the monitoring result of the maglev bearing gap according to the comparison result.

[0018] The embodiment of the present application can automatically reduce the reliability of the abnormal early warning and improve the reliability of the monitoring result when the image quality is reduced to cause the model to be unreliable by weighting and fusing the model confidence and the vibration feature.

[0019] Another embodiment of the present application also provides a magnetic suspension gap monitoring system based on image processing, comprising a data acquisition module, an image segmentation module, a vibration signal module and an abnormal monitoring module. The data acquisition module is used for collecting image sequences of the magnetic suspension bearing gap through a vacuum pipeline wall window, and synchronously acquiring air pressure data sequences and temperature data sequences in the vacuum pipeline. The image segmentation module is used for inputting the image sequences, the air pressure data sequences and the temperature data sequences into a pre-trained image segmentation model in a time alignment relationship, so that the image segmentation model calculates an optical distortion field according to the air pressure data and the temperature data at the same time, performs geometric correction on a feature map of an image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence. The vibration signal module is used for calculating an optical flow field in a corresponding region in the image sequences based on the gap region segmentation sequence, obtaining a total displacement field of the magnetic suspension bearing gap, and obtaining a gap micro-vibration displacement signal by subtracting a rigid motion displacement field from the total displacement field, wherein the rigid motion displacement field is calculated through inertial measurement unit data of the magnetic suspension train. The abnormal monitoring module is used for performing order analysis on the gap micro-vibration displacement signal to extract a vibration feature, weighting the vibration feature by taking a mean value of the confidence sequence as a weight, and generating a monitoring result of the magnetic suspension bearing gap according to the weighting result, wherein the vibration feature includes vibration energy of a magnetic suspension bearing rotation frequency and first vibration energy of a track welded joint through frequency.

[0020] Another embodiment of the present application also provides a terminal device, comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, when the processor executes the computer program, the steps of a magnetic suspension gap monitoring method based on image processing are implemented.

[0021] Another embodiment of the present application also provides a computer readable storage medium item, comprising a stored computer program, when the computer program runs, the device where the computer readable storage medium is located executes the steps of a magnetic suspension gap monitoring method based on image processing. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of an embodiment of a magnetic suspension gap monitoring method based on image processing provided by the present application is shown. Figure 2 A structural schematic diagram of an embodiment of the image processing-based magnetic suspension gap monitoring system provided by the present application. DETAILED DESCRIPTION

[0023] For the purposes of the present application, the technical solutions and advantages, the technical solutions in the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms “include” and “have” and any variations thereof in the specification and claims of the present application and the above description of drawings are intended to cover non-exclusive inclusion.

[0025] In the description of the embodiments of the present application, the technical terms “first”, “second”, etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of “a plurality of” is two or more, unless otherwise explicitly and specifically limited.

[0026] Reference herein to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0027] In the description of the embodiments of the present application, the term “and / or” is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character “ / ” herein generally represents an “or” relationship between the front and rear associated objects.

[0028] In the description of the embodiments of the present application, the term “a plurality of” refers to two or more (including two), and similarly, “a plurality of groups” refers to two or more groups (including two groups), and “a plurality of pieces” refers to two or more pieces (including two pieces).

[0029] Referring toFigure 1 To solve the problem that the prior art can only measure the static gap in a conventional environment, an embodiment of the present application provides a magnetic suspension gap monitoring method based on image processing, comprising steps S101 to S104: In step S101, an image sequence of the magnetic suspension bearing gap is collected through a vacuum pipeline wall window, and a pressure data sequence and a temperature data sequence in the vacuum pipeline are synchronously acquired.

[0030] It should be noted that collecting the image sequence of the magnetic suspension bearing gap through the vacuum pipeline wall window refers to a process of installing an optical window at a specific position of the vacuum pipeline wall in the vacuum pipeline environment of the high-speed operation of the magnetic suspension train, and continuously acquiring an image sequence that can clearly reflect the relative position change between the bearing and the track by using a high-speed imaging system arranged towards the gap region.

[0031] It should be noted that synchronously acquiring the pressure data sequence and the temperature data sequence in the vacuum pipeline refers to that, while collecting each frame of gap image, the pressure data and the temperature data in the vacuum pipeline at the image collection time are measured and recorded in a time alignment relationship, so as to form an environment parameter sequence corresponding to the image sequence one by one in time.

[0032] Preferably, the collecting of the image sequence of the magnetic suspension bearing gap through the vacuum pipeline wall window and the synchronous acquisition of the pressure data sequence and the temperature data sequence in the vacuum pipeline comprise: An image sequence is collected through the vacuum pipeline wall window by using a pulsed infrared laser light source and a high-speed camera equipped with a narrow-band filter, and a time stamp is attached to each frame of the image sequence; wherein the filtering wavelength of the narrow-band filter matches the wavelength of the pulsed infrared laser light source; The pressure data and the temperature data in the vacuum pipeline are acquired by a sensor, and based on the time stamp, a pressure data sequence and a temperature data sequence that are time-synchronized with the image sequence are generated.

[0033] Specifically, the light source of the high-speed imaging system can be a pulsed infrared laser; wherein the pulse width of the pulsed infrared laser needs to be extremely narrow, so as to ensure that the duration of the laser pulse is much smaller than the characteristic time of the relative motion of the train bearing during the exposure time of the high-speed camera, so as to freeze the high-speed moving bearing parts in the image. The high-speed camera needs to be equipped with a narrow-band filter to allow the laser reflection light of a specific wavelength to enter the high-speed camera, while filtering out all other wavelengths of stray light in the vacuum pipeline, thereby improving the signal-to-noise ratio and contrast of the image. While shooting the image, a time stamp is attached to each frame of image as a time reference for subsequent fusion processing of multiple source data.

[0034] Specifically, since the magnetic levitation train will cause the change of air refractive index when it runs at high speed in the vacuum tube, and further cause the optical distortion of the photographed gap image, it is necessary to collect the air pressure data and temperature data affecting the air refractive index through the sensor. The air pressure sensor can be directly installed on the measuring point of the pipe wall surface and communicated with the inside of the pipe through the vacuum flange; the temperature sensor can be distributedly arranged at multiple key positions near the bearing gap to monitor the possible temperature gradient. A timestamp derived from the same high-precision clock source is attached to each frame of image and each set of collected air pressure data and temperature data to realize time alignment.

[0035] In step S102, the image sequence, the air pressure data sequence and the temperature data sequence are input into the pre-trained image segmentation model in a time alignment relationship, so that the image segmentation model calculates an optical distortion field according to the air pressure data and the temperature data at the same time, and performs geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence.

[0036] It should be noted that the image segmentation model calculates the optical distortion field according to the air pressure data and the temperature data at the same time means that the synchronously acquired air pressure data and temperature data are injected into the image segmentation model as prior knowledge, so that the image segmentation model can dynamically generate an optical distortion field matching the current optical environment in the forward inference process.

[0037] It should be noted that the image segmentation model calculates the optical distortion field according to the air pressure data and the temperature data at the same time means that the synchronously acquired air pressure data and temperature data are injected into the image segmentation model as prior knowledge, so that the image segmentation model can dynamically generate an optical distortion field matching the current optical environment in the forward inference process.

[0038] Preferably, the image sequence, the air pressure data sequence and the temperature data sequence are input into the pre-trained image segmentation model in a time alignment relationship, so that the image segmentation model calculates an optical distortion field according to the air pressure data and the temperature data at the same time, and performs geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence, comprising: extracting features of the image sequence through an encoder to obtain each first feature map; wherein one image frame of the image sequence corresponds to one first feature map; mapping the air pressure data sequence and the temperature data sequence through a fully connected neural network layer to obtain each optical environment parameter vector; wherein the air pressure data and the temperature data at the same time in the air pressure data sequence and the temperature data sequence correspond to one optical environment parameter vector; performing an inverse convolution operation on each of the optical environment parameter vectors to generate each optical distortion field corresponding to each of the first feature maps; performing element-wise multiplication of each of the optical distortion fields and the corresponding first feature maps to obtain each second feature map; performing upsampling operations on each of the second feature maps through the decoder to extract fusion features, and processing the fusion features through a double-branch output structure to obtain a gap region segmentation sequence and a confidence sequence; wherein, at each upsampling stage, the optical distortion field is upsampled to a corresponding scale and multiplied element-wise with the feature map of the corresponding stage.

[0039] Specifically, the encoder performs multi-scale feature extraction on the input image frame to generate a first feature map. At the same time, the air pressure and temperature data at the same time are fused and transformed in dimension through a fully connected layer to generate a parameter vector representing the current optical environment state. The parameter vector is upsampled to the same spatial size as the first feature map through an inverse convolution operation to form an optical distortion field. It is worth noting that the optical environment parameter vector does not directly correspond to a certain physical formula, but is an implicit feature learned by the model from a large amount of data during the training process, which can represent the degree of influence of the current environment on the optical path.

[0040] Specifically, the image segmentation model element-wise fuses the optical distortion field with the first feature map to realize geometric correction at the feature level through a differentiable spatial transformation network. It is worth noting that at each upsampling stage of the decoder, the original optical distortion field is upsampled to the corresponding scale and then fused with the feature map of the current stage to form a multi-level optical distortion correction mechanism.

[0041] Specifically, at the output stage of the decoder, the fusion features are input into two independent 1x1 convolution layer branch processing channels; wherein, one branch processing channel outputs the probability of each pixel belonging to the gap region through a sigmoid activation function to form a gap region segmentation sequence; the other branch processing channel outputs a confidence score corresponding to each gap region segmentation result through the same structure but independent learning of parameter weights to form a confidence sequence; the two branch processing channels share the same input features, but learn the segmentation task and reliability evaluation task respectively.

[0042] Specifically, the pre-trained image segmentation model is obtained by supervised training on an image dataset containing images collected under different air pressure and temperature conditions and labeled gap regions. It is worth noting that the confidence branch is optimized through a joint loss function during the training process. The joint loss function guides the confidence branch to learn to evaluate the uncertainty of the model's segmentation results, so that the model outputs low confidence in areas with poor image quality and high confidence in areas with high segmentation confidence.

[0043] In step S103, based on the gap region segmentation sequence, optical flow fields are calculated in the corresponding regions in the image sequence to obtain a total displacement field of the magnetic suspension bearing gap, and a rigid motion displacement field is subtracted from the total displacement field to obtain a gap micro-vibration displacement signal.

[0044] It should be noted that, based on the gap region segmentation sequence, optical flow fields are calculated in the corresponding regions in the image sequence to obtain a total displacement field of the magnetic suspension bearing gap, which means that the gap region segmentation sequence output by the image segmentation model is used as a spatial mask to extract an image block sequence containing only the magnetic suspension bearing gap region from the original image sequence, and then dense optical flow between consecutive frames is calculated in this region to obtain a total displacement field representing the overall motion trend of the gap region.

[0045] It should be noted that, by subtracting the rigid motion displacement field from the total displacement field, a gap micro-vibration displacement signal is obtained, which means that the displacement component caused by the rigid motion of the vehicle body is removed from the total displacement field calculated from the image sequence, so that the gap micro-vibration displacement signal caused purely by the vibration of the magnetic suspension bearing itself is extracted.

[0046] Preferably, based on the gap region segmentation sequence, optical flow fields are calculated in the corresponding regions in the image sequence to obtain a total displacement field of the magnetic suspension bearing gap, and a rigid motion displacement field is subtracted from the total displacement field to obtain a gap micro-vibration displacement signal, comprising: Based on the gap region segmentation sequence, corresponding gap region image blocks are intercepted in the image sequence, and dense optical flow between the gap region image blocks of consecutive frames is calculated to generate a total displacement field; From the total displacement field, the rigid motion displacement field is subtracted pixel by pixel to obtain a residual displacement field; The residual displacement field is spatially averaged in the magnetic suspension bearing gap to obtain a gap micro-vibration displacement signal.

[0047] Specifically, according to the gap region segmentation sequence output by the image segmentation model, a minimum circumscribed rectangular region containing the magnetic suspension bearing gap is located and intercepted in each frame of the original image sequence. After obtaining the gap region image block sequence, a dense optical flow algorithm can be used to calculate the motion vector of each pixel point in the image block between consecutive frames. By performing motion consistency verification and outlier filtering on the optical flow calculation results, a total displacement field is generated.

[0048] Specifically, the train motion data collected by the inertial measurement unit is mapped to the image coordinate system through coordinate transformation, and a rigid motion displacement field consistent with the spatial resolution of the total displacement field is reconstructed. The residual displacement field is obtained by subtracting the rigid motion displacement field from the total displacement field in a pixel-by-pixel vector subtraction manner.

[0049] Specifically, the residual displacement field is subjected to spatial averaging processing in the magnetic suspension bearing gap region. The gap micro-vibration displacement signal is generated by calculating the mean value of the displacement vectors of all effective pixels in the gap region.

[0050] It should be noted that the rigid motion displacement field calculated by the inertial measurement unit data of the magnetic suspension train refers to the linear acceleration and angular velocity data collected by the inertial measurement unit installed on the train body in real time, and the pixel displacement field caused by the macroscopic motion of the train body in the image plane coordinate system is reconstructed through coordinate transformation and kinematic calculation.

[0051] Preferably, the rigid motion displacement field is calculated by the inertial measurement unit data of the magnetic suspension train, comprising: acquiring inertial measurement unit data of a magnetic suspension train; performing coordinate transformation operation from train coordinate system to image coordinate system on the inertial measurement unit data; performing integral operation on the transformed inertial measurement unit data to generate rigid motion displacement field in the image plane.

[0052] Specifically, the train motion data is collected in real time by the inertial measurement unit installed on the train body, including three-axis angular velocity and three-axis linear acceleration. Based on the intrinsic matrix obtained by camera calibration and the extrinsic transformation matrix between the camera and the inertial measurement unit, the train motion data in the train coordinate system is converted to the camera coordinate system. The converted angular velocity is time-integrated to obtain the rotation angle, and the displacement component of each pixel point caused by rotation is calculated according to the perspective projection relationship in the camera imaging model. At the same time, the converted linear acceleration is twice-integrated to obtain the translation displacement, which is mapped to the translation component in the image plane in combination with the camera intrinsic parameter. Finally, the rotation component and the translation component vector are superimposed to generate the rigid motion displacement field.

[0053] In step S104, order analysis is performed on the gap micro-vibration displacement signal to extract vibration characteristics, the mean value of the confidence sequence is taken as a weight, the vibration characteristics are weighted, and a monitoring result of the magnetic suspension bearing gap is generated according to the weighting result; wherein the vibration characteristics include the vibration energy of the magnetic suspension bearing rotation frequency and the first vibration energy of the track welded joint through frequency.

[0054] It should be noted that the order analysis of the gap micro-vibration displacement signal to extract the vibration feature means that the vibration signal in the time domain is converted into an angle domain signal synchronized with the rotation speed, and the vibration feature directly related to the physical characteristics of the mechanical rotating component is extracted by analyzing the order spectrum.

[0055] Preferably, the order analysis of the gap micro-vibration displacement signal to extract the vibration feature comprises: obtaining a real-time rotation speed signal of the maglev train; resampling the gap micro-vibration displacement signal according to the real-time rotation speed signal to convert the signal from the time domain to the angle domain; performing Fourier transform on the resampled gap micro-vibration displacement signal to obtain an order spectrum; extracting the vibration energy of the maglev bearing rotation frequency and the first vibration energy of the track welded joint passing frequency from the order spectrum, and grouping the vibration energy and the first vibration energy into a feature vector as the vibration feature.

[0056] Specifically, the real-time rotation speed signal of the maglev train is obtained through a rotation speed sensor. The gap micro-vibration displacement signal is resampled at equal angle intervals according to the rotation speed signal to realize the conversion from the time domain to the angle domain. Fourier transform is performed on the angle domain signal to obtain an order spectrum. The abscissa of the order spectrum is the order, i.e. the ratio of the frequency to the rotation frequency, and the ordinate represents the vibration energy. Two features are extracted from the order spectrum: one is the vibration energy corresponding to the maglev bearing rotation frequency, which reflects the unbalance state of the rotor system; the other is the first vibration energy of the track welded joint passing frequency, which is related to the train speed and the track welded joint spacing, and directly represents the impact excitation strength of the track on the joint to the bearing.

[0057] It should be noted that the mean value of the confidence sequence is used as the weight to weight the vibration feature, and the monitoring result of the maglev bearing gap is generated according to the weighting result, which means that the reliability of the vibration feature is quantified by using the reliability evaluation result output by the image segmentation model, and the intelligent monitoring result with state judgment and reliability evaluation is generated through the weighting fusion mechanism.

[0058] Preferably, the mean value of the confidence sequence is used as the weight to weight the vibration feature, and the monitoring result of the maglev bearing gap is generated according to the weighting result, which comprises: calculating the mean value of the confidence sequence and mapping the vibration feature to a first state score; multiplying the mean value and the first state score to obtain a state score; comparing the state score with a preset abnormal threshold to generate the monitoring result of the maglev bearing gap according to the comparison result.

[0059] Specifically, the mean value of the confidence sequence in the current analysis time window is calculated to obtain a weight factor representing the overall reliability of the image segmentation in this period. At the same time, the extracted vibration feature vector is mapped to a first state score between 0 and 1 through a preset scoring model. The first state score is multiplied by the weight factor to obtain the final state score. The final state score is compared with a preset abnormal threshold value. If the state score is less than the abnormal threshold value, an abnormal alarm is generated. It is worth noting that the abnormal threshold value can be set according to the state score distribution under historical normal data through statistical methods or based on specific safety margin requirements.

[0060] The embodiment of the present application can provide spatio-temporal aligned multi-modal data sources for subsequent image processing by synchronously collecting gap images and physical environment data in the vacuum pipeline; can dynamically compensate optical distortion caused by changes in air pressure and temperature in the vacuum pipeline by driving the image segmentation model with physical environment data for feature-level geometric correction; can extract micro-vibration signals representing the dynamic characteristics of the bearing itself by separating the rigid motion of the vehicle body from the image optical flow, providing pure input for subsequent state recognition; and can generate intelligent monitoring results with state judgment and reliability evaluation by performing order analysis on the micro-vibration signals and weighting fusion with model confidence.

[0061] Optionally, in the embodiment of the present application, the image sequence of the magnetic suspension bearing gap is collected through the vacuum pipeline wall window, and the air pressure data sequence and the temperature data sequence in the vacuum pipeline are synchronously acquired, comprising: An image sequence is collected through the vacuum pipeline wall window by using a pulsed infrared laser light source and a high-speed camera equipped with a narrow-band filter, and a time stamp is attached to each frame of the image sequence; wherein the filtering wavelength of the narrow-band filter matches the wavelength of the pulsed infrared laser light source; The air pressure data and the temperature data in the vacuum pipeline are acquired by a sensor, and based on the time stamp, the air pressure data sequence and the temperature data sequence that are time-synchronized with the image sequence are generated.

[0062] The embodiment of the present application can acquire high-contrast gap images in a vacuum pipeline environment and establish accurate correspondence between multiple data sources by using an anti-interference optical scheme and a time sequence synchronization method.

[0063] Optionally, in the embodiment of the present application, the image sequence, the air pressure data sequence and the temperature data sequence are input to a pre-trained image segmentation model in a time alignment relationship, so that the image segmentation model calculates an optical distortion field according to the air pressure data and the temperature data at the same time, and performs geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence, comprising: extracting features of the image sequence by an encoder to obtain each first feature map, wherein one image frame of the image sequence corresponds to one first feature map; mapping the air pressure data sequence and the temperature data sequence through a fully connected neural network layer to obtain each optical environment parameter vector, wherein air pressure data and temperature data at the same time in the air pressure data sequence and the temperature data sequence correspond to one optical environment parameter vector; performing an inverse convolution operation on each optical environment parameter vector to generate each optical distortion field corresponding to the scale of each first feature map; performing element-wise multiplication of each optical distortion field and each corresponding first feature map to obtain each second feature map; performing an upsampling operation on each second feature map through a decoder to extract fusion features, and processing the fusion features through a double-branch output structure to obtain a gap region segmentation sequence and a confidence sequence, wherein at each upsampling stage, the optical distortion field is upsampled to the corresponding scale and multiplied element-wise with the feature map at the corresponding stage.

[0064] The embodiment of the present application can realize adaptive geometric correction at the image feature level by constructing a coding and decoding network structure that fuses physical priors, and improve the accuracy of segmentation results in complex optical environments.

[0065] Optionally, in the embodiment of the present application, based on the gap region segmentation sequence, an optical flow field is calculated in the corresponding region in the image sequence to obtain a total displacement field of the magnetic bearing gap, and the total displacement field is subtracted by a rigid motion displacement field to obtain a gap micro-vibration displacement signal, comprising: Based on the gap region segmentation sequence, corresponding gap region image blocks in the image sequence are intercepted, and dense optical flow between the gap region image blocks of consecutive frames is calculated to generate a total displacement field; From the total displacement field, a rigid motion displacement field is subtracted pixel by pixel to obtain a residual displacement field; The residual displacement field is spatially averaged in the magnetic bearing gap to obtain a gap micro-vibration displacement signal.

[0066] The embodiment of the present application can effectively suppress the influence of background interference and macroscopic motion of the vehicle body and accurately capture the microscopic dynamic change characteristics of the gap region by calculating the optical flow in the accurately segmented gap region and decoupling the inertial data.

[0067] Optionally, in the embodiment of the present application, the rigid motion displacement field is calculated from the inertial measurement unit data of the maglev train, comprising: Obtaining the inertial measurement unit data of the maglev train; performing a coordinate transformation operation from a train coordinate system to an image coordinate system on the inertial measurement unit data; performing an integral operation on the transformed inertial measurement unit data to generate a rigid motion displacement field in the image plane.

[0068] The embodiment of the present application can reconstruct the rigid displacement field in the image plane by converting the inertial data to the image coordinate system for motion solving, and provide a compensation reference for micro-vibration separation.

[0069] Optionally, in the embodiment of the present application, the order analysis on the gap micro-vibration displacement signal to extract the vibration feature comprises: acquiring a real-time rotating speed signal of the maglev train; resampling the gap micro-vibration displacement signal according to the real-time rotating speed signal, and converting the signal from the time domain to the angle domain; performing a Fourier transform on the resampled gap micro-vibration displacement signal to obtain an order spectrum; extracting the vibration energy of the rotating frequency of the maglev bearing and the first vibration energy of the track welding joint pass frequency from the order spectrum, and taking the vibration energy and the first vibration energy as a feature vector as the vibration feature.

[0070] The embodiment of the present application can eliminate the influence of variable speed working conditions and extract the vibration feature directly related to the mechanical physical characteristics by converting the vibration signal to the angle domain related to the rotating speed for analysis.

[0071] Optionally, in the embodiment of the present application, the mean value of the confidence sequence is taken as a weight to weight the vibration feature, and the monitoring result of the maglev bearing gap is generated according to the weighting result, comprising: calculating the mean value of the confidence sequence, and mapping the vibration feature to a first state score; multiplying the mean value and the first state score to obtain a state score; comparing the state score with a preset abnormal threshold, and generating the monitoring result of the maglev bearing gap according to the comparison result.

[0072] The embodiment of the present application can automatically reduce the reliability of the abnormal early warning when the model confidence and the vibration feature are weighted and fused, and improve the reliability of the monitoring result when the image quality decreases to cause the model to be unreliable.

[0073] As shown in the above method embodiment, the corresponding system embodiment is provided; Figure 2 ​An embodiment of the present application provides a magnetic suspension gap monitoring system based on image processing, comprising a data acquisition module 201, an image segmentation module 202, a vibration signal module 203 and an abnormality monitoring module 204; The data acquisition module 201 is configured to collect an image sequence of a magnetic suspension bearing gap through a vacuum pipeline wall window, and synchronously acquire a pressure data sequence and a temperature data sequence in the vacuum pipeline; The image segmentation module 202 is configured to input the image sequence, the pressure data sequence and the temperature data sequence in a time alignment relationship to a pre-trained image segmentation model, so that the image segmentation model calculates an optical distortion field according to pressure data and temperature data at the same time, and performs geometric correction on a feature map of an image frame at the same time based on the optical distortion field, and outputs a gap region segmentation sequence and a confidence sequence; The vibration signal module 203 is configured to calculate an optical flow field in a corresponding region in the image sequence based on the gap region segmentation sequence, obtain a total displacement field of the magnetic suspension bearing gap, and obtain a gap micro-vibration displacement signal by subtracting a rigid motion displacement field from the total displacement field, wherein the rigid motion displacement field is calculated by using inertial measurement unit data of a magnetic suspension train; The abnormality monitoring module 204 is configured to perform order analysis on the gap micro-vibration displacement signal to extract vibration features, take a mean value of the confidence sequence as a weight, weight the vibration features, and generate a monitoring result of the magnetic suspension bearing gap according to a weighting result, wherein the vibration features include vibration energy of a magnetic suspension bearing rotation frequency and first vibration energy of a track welded joint through frequency.

[0074] Optionally, in the embodiment of the present application, the data acquisition module 201 comprises an image data submodule and an environment data submodule. The image data submodule is configured to collect an image sequence through a vacuum pipeline wall window by using a pulsed infrared laser light source and a high-speed camera equipped with a narrow-band filter, and attach a time stamp to each frame of the image sequence, wherein a filtering wavelength of the narrow-band filter matches a wavelength of the pulsed infrared laser light source. The environment data submodule is configured to acquire pressure data and temperature data in the vacuum pipeline through a sensor, and generate a pressure data sequence and a temperature data sequence that are time-synchronized with the image sequence based on the time stamp.

[0075] The embodiment of the present application can acquire high-contrast gap images in a vacuum pipeline environment and establish an accurate corresponding relationship between multiple source data by using an anti-interference optical scheme and a time sequence synchronization method.

[0076] Optionally, in the embodiment of the present application, the image segmentation module 202 comprises a feature extraction submodule, a data mapping submodule, an optical distortion field submodule, a geometric correction submodule and a result output submodule. The feature extraction submodule is configured to extract features of the image sequence by an encoder to obtain each first feature map; wherein one image frame of the image sequence corresponds to one first feature map. The data mapping submodule is configured to map the air pressure data sequence and the temperature data sequence by a fully connected neural network layer to obtain each optical environment parameter vector; wherein the air pressure data and the temperature data at the same time in the air pressure data sequence and the temperature data sequence correspond to one optical environment parameter vector. The optical distortion field submodule is configured to perform an inverse convolution operation on each optical environment parameter vector to generate each optical distortion field corresponding to the scale of each first feature map. The geometric correction submodule is configured to multiply each optical distortion field and the corresponding first feature map element by element to obtain each second feature map. The result output submodule is configured to perform an up-sampling operation on each second feature map by a decoder to extract fusion features, and process the fusion features by a double-branch output structure to obtain a gap region segmentation sequence and a confidence sequence; wherein at each up-sampling stage, the optical distortion field is up-sampled to the corresponding scale and multiplied with the feature map at the corresponding stage element by element.

[0077] The embodiment of the present application can realize adaptive geometric correction at the image feature level by constructing a coding and decoding network structure that fuses physical priors, and improve the accuracy of segmentation results in complex optical environments.

[0078] Optionally, in the embodiment of the present application, the vibration signal module 203 comprises an optical flow calculation submodule, a data decoupling submodule and a vibration signal submodule. The optical flow calculation submodule is configured to intercept corresponding gap region image blocks in the image sequence based on the gap region segmentation sequence, and calculate the dense optical flow between the gap region image blocks of consecutive frames to generate a total displacement field. The data decoupling submodule is configured to subtract a rigid motion displacement field from the total displacement field pixel by pixel to obtain a residual displacement field. The vibration signal submodule is configured to spatially average the residual displacement field in the magnetic bearing gap to obtain a gap micro-vibration displacement signal.

[0079] The embodiment of the present application can effectively inhibit the influence of background interference and macroscopic motion of the vehicle body, and accurately capture the microscopic dynamic change characteristics of the gap region by calculating the optical flow in the precisely segmented gap region and decoupling the fused inertial data.

[0080] Optionally, in the embodiment of the present application, the vibration signal module 203 further comprises: an inertial measurement submodule, a coordinate transformation submodule and a rigid motion displacement field submodule; The inertial measurement submodule is configured to acquire inertial measurement unit data of the maglev train. The coordinate transformation submodule is configured to perform a coordinate transformation operation from a train coordinate system to an image coordinate system on the inertial measurement unit data. The rigid motion displacement field submodule is configured to perform an integral operation on the transformed inertial measurement unit data to generate a rigid motion displacement field in the image plane.

[0081] The embodiment of the present application can reconstruct the rigid displacement field in the image plane by converting the inertial data to the image coordinate system for motion solving, thereby providing a compensation reference for micro-vibration separation.

[0082] Optionally, in the embodiment of the present application, the anomaly monitoring module 204 comprises: a rotational speed signal submodule, a resampling submodule, a order spectrum submodule and a vibration feature submodule. The rotational speed signal submodule is configured to acquire real-time rotational speed signals of the maglev train. The resampling submodule is configured to resample the gap micro-vibration displacement signals according to the real-time rotational speed signals, and convert the signals from the time domain to the angle domain. The order spectrum submodule is configured to perform Fourier transform on the resampled gap micro-vibration displacement signals to obtain an order spectrum. The vibration feature submodule is configured to extract vibration energy of a rotational frequency of the maglev bearing and first vibration energy of a through frequency of the rail welded joint from the order spectrum, and group the vibration energy and the first vibration energy into a feature vector as the vibration feature.

[0083] The embodiment of the present application can eliminate the influence of variable speed working conditions and extract vibration features directly related to mechanical physical characteristics by converting the vibration signals to the angle domain related to the rotational speed for analysis.

[0084] Optionally, in the embodiment of the present application, the anomaly monitoring module 204 further comprises: a state score submodule, a score weighting submodule and a monitoring result submodule. The state score submodule is configured to calculate a mean value of the confidence sequence, and map the vibration feature to a first state score. The score weighting submodule is configured to multiply the mean value by the first state score to obtain a state score. The monitoring result submodule is configured to compare the state score with a preset abnormal threshold, and generate a monitoring result of the magnetic suspension bearing gap according to a comparison result.

[0085] The embodiment of the present application can automatically reduce the reliability of the abnormal early warning when the image quality decreases to cause the model to be unreliable, and improve the reliability of the monitoring result by weighting and fusing the model confidence and the vibration feature.

[0086] It can be understood that the above-mentioned device item embodiment corresponds to the method item embodiment of the present application, and can realize the image processing-based magnetic suspension gap monitoring method provided by any one of the method item embodiments of the present application.

[0087] The embodiment of the present application can provide spatiotemporal alignment multi-modal data sources for subsequent image processing by the data acquisition module 201 synchronously collecting gap images and physical environment data in the vacuum pipeline; can dynamically compensate optical distortion caused by changes in air pressure and temperature in the vacuum pipeline by the image segmentation module 202 driving the image segmentation model to perform feature-level geometric correction using the physical environment data; can extract micro-vibration signals representing the dynamic characteristics of the bearing itself by the vibration signal module 203 separating the rigid motion of the vehicle body from the image optical flow; and can generate intelligent monitoring results with state judgment and reliability evaluation by the abnormal monitoring module 204 performing order analysis on the micro-vibration signals and weighting and fusing the model confidence.

[0088] It should be noted that the device embodiments described above are only schematic, and part or all of the modules can be selected to achieve the purpose of the embodiment of the present application according to actual needs. In addition, in the device embodiment provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which can be realized as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0089] On the basis of the above-mentioned embodiment of the image processing-based magnetic suspension gap monitoring method, another embodiment of the present application provides a terminal device, which comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the image processing-based magnetic suspension gap monitoring method of any one of the embodiments of the present application is realized.

[0090] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0091] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The terminal device can include, but is not limited to, a processor and a memory.

[0092] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and connects all parts of the terminal device through various interfaces and lines.

[0093] On the basis of the above-mentioned method embodiment, another embodiment of the present application provides a computer readable storage medium, including a stored computer program, wherein when the computer program runs, the device where the computer readable storage medium is located executes the image processing-based magnetic suspension gap monitoring method in any one of the above-mentioned method embodiments of the present application.

[0094] The modules / units integrated in the device / terminal equipment, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each method embodiment can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0095] The above is the preferred embodiment of the present application. It should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which are also considered within the scope of protection of the present application.

Claims

1. A method for monitoring the gap of magnetic levitation based on image processing, characterized in that, include: Image sequences of the gap between the magnetic levitation bearings were acquired through viewing windows on the wall of the vacuum pipe, and pressure and temperature data sequences inside the vacuum pipe were acquired simultaneously. The image sequence, air pressure data sequence, and temperature data sequence are input into a pre-trained image segmentation model according to the time alignment relationship. The image segmentation model calculates the optical distortion field based on the air pressure data and temperature data at the same time, and performs geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and outputs the gap region segmentation sequence and confidence sequence. Based on the gap region segmentation sequence, the optical flow field is calculated in the corresponding region of the image sequence to obtain the total displacement field of the magnetic levitation bearing gap. The rigid motion displacement field is then subtracted from the total displacement field to obtain the gap micro-vibration displacement signal. The rigid motion displacement field is calculated using data from the inertial measurement unit of the magnetic levitation train. The gap micro-vibration displacement signal is subjected to order analysis to extract vibration features. The mean of the confidence sequence is used as a weight to weight the vibration features, and the monitoring results of the magnetic levitation bearing gap are generated based on the weighting results. The vibration features include the vibration energy of the magnetic levitation bearing rotation frequency and the primary vibration energy of the rail welding joint passing frequency.

2. The magnetic levitation gap monitoring method based on image processing as described in claim 1, characterized in that, The process of acquiring image sequences of the magnetic levitation bearing gap through a viewing window on the wall of the vacuum pipe, and simultaneously acquiring air pressure and temperature data sequences within the vacuum pipe, includes: A pulsed infrared laser source and a high-speed camera equipped with a narrowband filter are used to acquire image sequences through a window on the wall of a vacuum tube, and a timestamp is added to each frame of the image sequence; wherein the filtering wavelength of the narrowband filter is matched with the wavelength of the pulsed infrared laser source; The air pressure and temperature data inside the vacuum pipe are acquired by sensors, and based on the timestamp, an air pressure data sequence and a temperature data sequence synchronized with the image sequence are generated.

3. The magnetic levitation gap monitoring method based on image processing as described in claim 1, characterized in that, The process involves inputting the image sequence, air pressure data sequence, and temperature data sequence into a pre-trained image segmentation model according to a time-aligned relationship. This allows the image segmentation model to calculate the optical distortion field based on the air pressure and temperature data at the same time, and to perform geometric correction on the feature maps of image frames at the same time based on the optical distortion field. The model then outputs a gap region segmentation sequence and a confidence sequence, including: The image sequence is processed by an encoder to extract features, resulting in first feature maps; wherein, one image frame of the image sequence corresponds to one first feature map; The pressure data sequence and temperature data sequence are mapped by a fully connected neural network layer to obtain each optical environment parameter vector; wherein, the pressure data and temperature data at the same moment in the pressure data sequence and temperature data sequence correspond to one optical environment parameter vector; Perform deconvolution operations on each of the optical environment parameter vectors to generate optical distortion fields corresponding to each of the first feature map scales; Each optical distortion field is multiplied element-wise with its corresponding first feature map to obtain each second feature map; The decoder performs upsampling operations on each of the second feature maps to extract fused features, and processes the fused features through a dual-branch output structure to obtain a gap region segmentation sequence and a confidence sequence; wherein, in each upsampling stage, the optical distortion field is upsampled at the corresponding scale and multiplied element-wise with the feature map of the corresponding stage.

4. The magnetic levitation gap monitoring method based on image processing as described in claim 1, characterized in that, Based on the gap region segmentation sequence, the optical flow field is calculated in the corresponding region of the image sequence to obtain the total displacement field of the magnetic levitation bearing gap. The gap micro-vibration displacement signal is obtained by subtracting the rigid motion displacement field from the total displacement field, including: Based on the gap region segmentation sequence, corresponding gap region image blocks are extracted from the image sequence, and dense optical flow between the gap region image blocks in consecutive frames is calculated to generate a total displacement field. The rigid motion displacement field is subtracted pixel by pixel from the total displacement field to obtain the residual displacement field; The residual displacement field is spatially averaged within the gap of the magnetic levitation bearing to obtain the gap micro-vibration displacement signal.

5. The magnetic levitation gap monitoring method based on image processing as described in claim 4, characterized in that, The rigid motion displacement field is calculated using data from the inertial measurement unit of the magnetic levitation train, and includes: Acquire inertial measurement unit data of the maglev train; Perform a coordinate transformation operation from the train coordinate system to the image coordinate system on the inertial measurement unit data; The transformed inertial measurement unit data is integrated to generate a rigid motion displacement field in the image plane.

6. The magnetic levitation gap monitoring method based on image processing as described in claim 1, characterized in that, The step of performing order analysis on the gap micro-vibration displacement signal to extract vibration features includes: Obtain the real-time rotational speed signal of the maglev train; The gap micro-vibration displacement signal is resampled based on the real-time rotation speed signal, and the signal is converted from the time domain to the angular domain. The order spectrum is obtained by performing a Fourier transform on the resampled gap micro-vibration displacement signal. The vibration energy of the magnetic levitation bearing rotation frequency and the primary vibration energy of the track welding joint passing frequency are extracted from the order spectrum, and the vibration energy and the primary vibration energy are combined to form a feature vector as the vibration feature.

7. The magnetic levitation gap monitoring method based on image processing as described in claim 6, characterized in that, The step of using the mean of the confidence sequence as a weight to weight the vibration characteristics, and generating the monitoring result of the magnetic levitation bearing clearance based on the weighting result, includes: Calculate the mean of the confidence sequence and map the vibration features to a first state score; Multiply the mean by the first state score to obtain the state score; The state score is compared with a preset anomaly threshold, and the monitoring result of the magnetic levitation bearing gap is generated based on the comparison result.

8. A magnetic levitation gap monitoring system based on image processing, characterized in that, include: Data acquisition module, image segmentation module, vibration signal module, and anomaly monitoring module; The data acquisition module is used to acquire image sequences of the gap between the magnetic levitation bearings through the viewing window on the wall of the vacuum pipe, and simultaneously acquire air pressure data sequences and temperature data sequences inside the vacuum pipe. The image segmentation module is used to input the image sequence, air pressure data sequence, and temperature data sequence into a pre-trained image segmentation model according to the time alignment relationship, so that the image segmentation model can calculate the optical distortion field based on the air pressure data and temperature data at the same time, and perform geometric correction on the feature map of the image frame at the same time based on the optical distortion field, and output the gap region segmentation sequence and confidence sequence. The vibration signal module is used to calculate the optical flow field in the corresponding region of the image sequence based on the gap region segmentation sequence, to obtain the total displacement field of the magnetic levitation bearing gap, and to obtain the gap micro-vibration displacement signal by subtracting the rigid motion displacement field from the total displacement field; wherein, the rigid motion displacement field is calculated from the inertial measurement unit data of the magnetic levitation train; The anomaly monitoring module is used to perform order analysis on the gap micro-vibration displacement signal to extract vibration features, use the mean of the confidence sequence as a weight to weight the vibration features, and generate the monitoring result of the magnetic levitation bearing gap based on the weighting result; wherein, the vibration features include the vibration energy of the magnetic levitation bearing rotation frequency and the primary vibration energy of the rail welding joint passing frequency.

9. A terminal device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a magnetic levitation gap monitoring method based on image processing as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform a magnetic levitation gap monitoring method based on image processing as described in any one of claims 1-7.

Images