A vehicle door state recognition method and system based on time series data image processing

Abstract

The application discloses a kind of based on time series data image processing vehicle door state recognition method and system, the method includes: collecting time series data in the movement process of vehicle door and carrying out data preprocessing, and is divided into calibration dataset and to be diagnosed dataset;Respectively to the calibration dataset and to be diagnosed dataset after pre-processing are carried out distribution density gray scale chart processing, and difference density chart is generated;Change area recognition and change degree evaluation are carried out based on difference density chart, and the degree of deviation of vehicle door state is obtained;Establish the partition map of vehicle door functional component, and carry out functional component association based on the degree of deviation of vehicle door state, and identify abnormal functional component.The application can quantitatively evaluate the abnormal degree of vehicle door, and associate abnormal area with moving component, with strong versatility, good interpretability, and facilitate accurate maintenance on site.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of abnormal diagnosis of rail transit vehicle door system, in particular to a door state recognition method and system based on time series data image processing. BACKGROUND

[0002] Due to the complex structure, high utilization rate, and various operation environments of the rail vehicle door system, the number of faults accounts for more than 30% of the total number of faults of all vehicle parts. At present, the vehicle door maintenance mainly relies on regular system repair, which is time-consuming and costly. Therefore, it is urgent to monitor the state of the vehicle door by means of intelligent means, to convert regular maintenance into predictive state repair, and to improve the operation reliability of the vehicle door and reduce the maintenance cost.

[0003] At present, the recognition of the state of the rail vehicle door mainly relies on the time series data of the motor, extracts the data features to establish a machine learning diagnosis model, and recognizes the data trend. However, the trend analysis cannot quantitatively evaluate the abnormal degree of the door operation, which is not conducive to determining the push cycle of the early warning information. In addition, in practical application, it is found that due to the influence of factors such as door position, wind pressure, and electric control, the data trend of different doors is not consistent when the state deviates. When the diagnosis algorithm is applied, it often needs to be retested and modeled, which is time-consuming and costly. The model has poor universality and high false alarm and missed alarm rates. At the same time, the abnormal state diagnosis model of the door is separated from the operation mechanism of the door, and the diagnosis result cannot be directly associated with the key components, which lacks effective guidance for on-site maintenance. SUMMARY

[0004] The present application aims to provide a door state recognition method and system based on time series data image processing, which can quantitatively evaluate the abnormal degree of the door and associate the abnormal area with the moving parts.

[0005] Technical solution: The door state recognition method based on time series data image processing comprises the following steps:

[0006] (1) Collecting the time series data in the door movement process and performing data preprocessing, and dividing into a calibration data set and a to-be-diagnosed data set;

[0007] (2) Processing the preprocessed calibration data set and to-be-diagnosed data set respectively by distribution density gray scale processing, and generating a difference density map;

[0008] (3) Based on the difference density map, the change area is recognized and the change degree is evaluated, and the state deviation degree of the door is obtained;

[0009] (4) Establishing a door functional component partition map, and associating the functional components based on the state deviation degree of the door to identify the functional components that occur abnormally.

[0010] Preferably, the time series data in the door movement process in step (1) includes motor current, motor speed, motor angle parameter data in the door opening and closing process.

[0011] Preferably, the data preprocessing in step (1) includes abnormal data elimination and / or displacement domain conversion of time series data.

[0012] Preferably, the displacement domain conversion includes: for a single set of motor current, motor speed, motor angle data, a displacement domain curve image of a single set of data is drawn with the motor angle or the converted displacement as the horizontal coordinate and the motor speed or the motor current as the vertical coordinate, the row number of the lowest point of each column of pixels in the image is read as the speed or current data corresponding to the angle or displacement value, and the displacement domain data is obtained.

[0013] Preferably, the calibration data set includes a time series data set or a displacement domain data set of a single door or a group of doors, and the to-be-diagnosed data set includes a time series data set or a displacement domain data set of a single door; wherein the group of doors refers to doors with similar operating conditions in the same time period.

[0014] Preferably, the distribution density grayscale image processing in step (2) includes:

[0015] For a single door time series data set or a displacement domain data set, quantitative data is selected from the data set as the image processing data set; a full zero array is constructed with the maximum value of a certain parameter data in the data set as the array row number and the maximum length of the parameter data as the array column number, each value in each data of the parameter data set is added by 1 at the corresponding position of the array to form a data density array related to the data distribution quantity; and the data density array values are scaled to the interval of 0-255 to form the time series or displacement domain distribution density grayscale image of the parameter data.

[0016] For a group of door time series data set or displacement domain data set, quantitative data is selected from the data set as the image processing data set; a full zero array is constructed with the maximum value of a certain parameter data in the data set as the array row number and the maximum length of the parameter data as the array column number, each value in each data of the parameter data set is added by 1 at the corresponding position of the array to form a data density array related to the data distribution quantity; then the sparse pixel values are calculated according to a specified proportion, the sparse pixel values in the array are set to zero to obtain a common area density array of the group time series or displacement domain data; and the density array values are scaled to the interval of 0-255 to form the time series or displacement domain distribution density grayscale image of the parameter data.

[0017] Preferably, the image data alignment is further included before the difference density image is generated in step (2).

[0018] Preferably, the image data alignment comprises selecting the maximum size of the maximum length and width values in the distribution density gray scale maps of the calibration data set and the data set to be diagnosed, and filling zeros in the corresponding positions of other images according to the maximum size.

[0019] Preferably, the generating of the difference density image in step (2) comprises removing the same area as the distribution density gray scale map of the calibration data after alignment from the distribution density gray scale map of the data set to be diagnosed after alignment to generate the difference density image.

[0020] Preferably, the change area identification in step (3) comprises identifying the area where the data points are more concentrated in the difference density image based on an intelligent algorithm, and the intelligent algorithm comprises a pixel and sliding window identification method and an effective pixel column identification method.

[0021] Preferably, the pixel and sliding window identification method comprises:

[0022] 1) calculating and sorting the area pixels of the difference density image using a sliding window;

[0023] 2) obtaining a large change area and a refined change area based on a change window identification rule;

[0024] 3) dividing the data change into a whole change or a local change according to the large change area and the change window identification rule;

[0025] wherein the change window identification rule is:

[0026] (i) if the pixel sum of the first sorted window is greater than 2 times the pixel sum of the second sorted window, it is considered that the image has a single local area change, and the change area is the time sequence or displacement area corresponding to the first sorted window;

[0027] (ii) if the pixel sum of more than half of the windows is greater than 0.5 times the pixel sum of the first sorted window, it is considered that the image has a whole area change, and the change area is the whole change;

[0028] (iii) if the pixel sum of the first sorted window is less than 2 times the pixel sum of the second sorted window, and the pixel sum of less than half of the windows is greater than 0.5 times the pixel sum of the first sorted window, it is considered that the image has a multi-segment local area change, and the change area is the time sequence or displacement area corresponding to each window whose pixel sum is greater than 0.5 times the pixel sum of the first sorted window.

[0029] Preferably, the effective pixel column identification method comprises:

[0030] 1) presetting an effective pixel gray scale threshold and an effective pixel column continuous minimum length threshold;

[0031] 2) obtaining all effective pixel columns in the image based on the difference density image;

[0032] 3) judging the continuity of the effective pixel column, and obtaining the effective pixel column continuous region;

[0033] 4) judging the length of the effective pixel column continuous region, if the maximum continuous region length is less than the minimum length threshold of the effective pixel column continuity, the region is not output, otherwise all sub-regions are output as change regions.

[0034] Preferably, the region change degree evaluation in step (3) is performed by a feature analysis evaluation method, and the door state offset degree is obtained by feature analysis; the feature analysis evaluation method includes a region gravity center offset evaluation method, a co-occurrence matrix evaluation method, a mean square error evaluation method, and an effective pixel point evaluation method.

[0035] Preferably, the gravity center offset evaluation method includes:

[0036] 1) respectively calculating the gravity center coordinates of the change regions of the time sequence or displacement domain distribution density images of the calibration data and the data to be diagnosed, for a discrete two-dimensional image, assuming that the pixel value is I(i,j), wherein i represents the horizontal coordinate of the pixel point, j represents the vertical coordinate of the pixel point, and the calculation formula of the gravity center (G x ,G y ) of the image is:

[0037] G x =(∑(i*I(i,j))) / M

[0038] G y =(∑(j*I(i,j))) / M

[0039] wherein M is the sum of all pixel values of the image: M=∑∑(I(i,j))

[0040] 2) calculating the gravity center of the calibration data calculating the absolute value ΔG x , ΔG y and the relative value Rate(G x ), Rate(G y ) of the gravity center offset of the data to be diagnosed relative to the calibration data, and the calculation formula is as follows:

[0041]

[0042]

[0043]

[0044]

[0045] 3) comparing with the set gravity center alarm threshold value, and considering that the change region vehicle door state deviates if exceeding the threshold value.

[0046] Preferably, the co-occurrence matrix evaluation method comprises:

[0047] 1) calculating the calibration data co-occurrence matrix correlation feature: randomly selecting two groups of quantitative data in the calibration data set as the to-be-diagnosed data and the calibration data, using the pixel and sliding window identification method to identify the change region, obtaining the large change region of each difference density image, and calculating the calibration data co-occurrence matrix correlation feature p std ;

[0048] 2) calculating the to-be-diagnosed data co-occurrence matrix correlation feature: calculating the to-be-diagnosed data co-occurrence matrix correlation feature p diag according to the large change region of the difference density image;

[0049] 3) comparing p std , p diag to make a judgment, and considering that the change region vehicle door state deviates if p diag > p std .

[0050] Preferably, the step (4) comprises establishing a vehicle door functional component partition map according to the motor angle data, the partition interval is greater than the minimum angle that can be recognized by the motor position sensor, and the types of functional components in the same partition are the least.

[0051] Preferably, the step (4) comprises converting the vehicle door state deviation data into motor angle data, and based on the angle data, combining the vehicle door functional component partition map to find the corresponding functional component partition and the functional components in the partition range.

[0052] Preferably, it further comprises model iterative updating, and the model iterative updating comprises obtaining the main line operation label data based on the main line operation feedback situation, taking the main line operation label data as the model input, and automatically updating the model parameters.

[0053] The vehicle door state recognition system based on time series data image processing comprises:

[0054] A data acquisition and preprocessing module is configured to acquire time series data in a vehicle door movement process and perform data preprocessing, and divide the data into a calibration data set and a to-be-diagnosed data set.

[0055] A difference density image generation module is configured to perform distribution density grayscale processing on the preprocessed calibration data set and to-be-diagnosed data set respectively, and generate a difference density image.

[0056] A change area and degree identification module is configured to perform change area identification and change degree evaluation based on the difference density map, and obtain a door state offset degree.

[0057] An abnormal function component association module is configured to establish a door function component partition map, and perform function component association based on the door state offset degree, and identify an abnormal function component.

[0058] Preferably, the door state identification system based on time sequence data image processing is used to implement the door state identification method based on time sequence data image processing.

[0059] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program implements the door state identification method based on time sequence data image processing when loaded into the processor.

[0060] A computer readable storage medium stores a computer program, and the computer program implements the door state identification method based on time sequence data image processing when executed by a processor.

[0061] Advantages: Compared with the prior art, the present application has the following significant advantages: the present application forms a gray-scale image related to data distribution density by image processing of motor data, identifies data change areas through multiple intelligent algorithms, and quantitatively evaluates abnormal degrees; obtains a door with a state offset and an offset area, establishes a door function component partition map, and associates the door state offset area with door system function components according to the door function component partition map, and combines and pushes as early warning information for easy on-site precise maintenance; the method is closely combined with the door movement mechanism, associates data areas with moving components, has good interpretability, and is beneficial to guiding on-site maintenance operation; and avoids the long time-consuming and high cost caused by traditional door state diagnosis requiring test modeling, saves deployment time, and has strong universality; and can automatically iterate and optimize according to the application situation, continuously improve the algorithm model performance, and improve the door system operation reliability. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A flowchart of the method of the present application;

[0063] Figure 2 A data curve schematic diagram before door preprocessing in the embodiment of the present application; wherein, (a) is a motor opening door angle curve, (b) is a motor closing door angle curve, (c) is a motor opening door speed curve, (d) is a motor closing door speed curve, (e) is a motor opening door current curve, and (f) is a motor closing door current curve;

[0064] Figure 3Fig. 1 is a schematic diagram of data curves after pretreatment of a vehicle door in an embodiment of the present application; wherein (a) is a motor opening door angle curve, (b) is a motor closing door angle curve, (c) is a motor opening door speed curve, (d) is a motor closing door speed curve, (e) is a motor opening door current curve, and (f) is a motor closing door current curve;

[0065] Figure 4 Fig. 6 is a distribution density gray scale diagram of calibration data opening door speed displacement domain in an embodiment of the present application;

[0066] Figure 5 Fig. 7 is a distribution density gray scale diagram of calibration data opening door current displacement domain in an embodiment of the present application;

[0067] Figure 6 Fig. 8 is a distribution density gray scale diagram of calibration data closing door speed displacement domain in an embodiment of the present application;

[0068] Figure 7 Fig. 9 is a distribution density gray scale diagram of calibration data closing door current displacement domain in an embodiment of the present application;

[0069] Figure 8 Fig. 10 is a distribution density gray scale diagram of to-be-diagnosed data opening door speed displacement domain in an embodiment of the present application;

[0070] Figure 9 Fig. 11 is a distribution density gray scale diagram of to-be-diagnosed data opening door current displacement domain in an embodiment of the present application;

[0071] Figure 10 Fig. 12 is a distribution density gray scale diagram of to-be-diagnosed data closing door speed displacement domain in an embodiment of the present application;

[0072] Figure 11 Fig. 13 is a distribution density gray scale diagram of to-be-diagnosed data closing door current displacement domain in an embodiment of the present application;

[0073] Figure 12 Fig. 14 is a difference density diagram of to-be-diagnosed data opening door speed in an embodiment of the present application;

[0074] Figure 13 Fig. 15 is a difference density diagram of to-be-diagnosed data opening door current in an embodiment of the present application;

[0075] Figure 14 Fig. 16 is a difference density diagram of to-be-diagnosed data closing door speed in an embodiment of the present application;

[0076] Figure 15 Fig. 17 is a difference density diagram of to-be-diagnosed data closing door current in an embodiment of the present application;

[0077] Figure 16 Fig. 18 is a partition diagram of a vehicle door functional component in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The technical solutions of the application will be further described below with reference to the drawings.

[0079] As shown in the drawings, the door state recognition method based on time series data image processing of the application comprises the following steps: Figure 1

[0080] (1) Collecting time series data in the door movement process and performing data preprocessing. Specifically, it includes:

[0081] (1.1) Collecting time series data in the door movement process includes motor current, motor speed, motor angle and other parameter data in the opening and closing process.

[0082] (1.2) Remove the motor data after locked-rotor and eliminate obvious abnormal data.

[0083] (1.3) Displacement domain data conversion.

[0084] For single group of motor current, speed and angle data, the displacement domain curve image of single group data is drawn with angle or converted displacement as abscissa and speed or current as ordinate, the row number of the lowest point of each column of image is read as the speed or current data corresponding to the displacement value, and the displacement domain data is obtained.

[0085] Wherein, the motor angle data is converted into displacement by using the following angle-displacement conversion formula:

[0086] dis = angle * d / 360

[0087] In the formula, dis is the displacement of the door, unit mm, angle is the motor angle, unit °, d is the lead of the screw, and the lead of the sliding door used in this embodiment is 64.

[0088] (1.4) Dividing the calibration data set into the diagnosis data set

[0089] The calibration data set includes single door or group door time series data set or displacement domain data set; the diagnosis data set includes single door time series data set or displacement domain data set. The group door refers to the door with similar operating conditions in the same time period, and the time series data set or displacement domain data set of the group door can balance the abnormal situation of single door data.

[0090] (2) Distributing density gray scale processing to the preprocessed calibration data set and diagnosis data set respectively, and generating difference density map. Specifically, it includes:

[0091] ​(2.1) The distribution density gray scale map processing of the calibration data set. It mainly includes four ways: single door time sequence data distribution density map processing, single door displacement domain data distribution density map processing, group door time sequence data distribution density map processing, and group door displacement domain data distribution density map processing.

[0092] (a) For the single door time sequence data set of the calibration data, select quantitative data from the single door time sequence data set as the data set used for subsequent image processing. A full zero array is constructed with the maximum value of a certain parameter data in the data set, such as opening door speed, opening door current, closing door speed, or closing door current, as the number of rows, and the maximum value of the length of the parameter data as the number of columns. For each value in each data set of the parameter data, add 1 to the corresponding position of the array to form a data density array related to the number of data distribution. Scale the data density array value to the interval of 0-255 to form the time sequence distribution density gray scale map of the parameter data, and obtain the single door opening door speed, opening door current, closing door speed, and closing door current time sequence distribution density gray scale map of the calibration data.

[0093] (b) For the single door displacement domain data set of the calibration data, select quantitative data from the single door displacement domain data set as the data set used for subsequent image processing. A full zero array is constructed with the maximum value of a certain parameter data in the data set, such as opening door speed, opening door current, closing door speed, or closing door current, as the number of rows, and the maximum value of the length of the parameter data as the number of columns. For each value in each data set of the parameter data, add 1 to the corresponding position of the array to form a data density array related to the number of data distribution. Scale the data density array value to the interval of 0-255 to form the displacement domain distribution density gray scale map of the parameter data, and obtain the single door calibration data opening door speed, opening door current, closing door speed, and closing door current displacement domain distribution density gray scale map.

[0094] (c) For the group door time sequence data set of the calibration data, select quantitative data from each door time sequence data set of the group door as the data set used for subsequent image processing. A full zero array is constructed with the maximum value of a certain parameter data in the data set, such as opening door speed, opening door current, closing door speed, or closing door current, as the number of rows, and the maximum value of the length of the parameter data as the number of columns. For each value in each data set of the parameter data, add 1 to the corresponding position of the array to form a data density array related to the number of data distribution. Calculate the sparse pixel value of the array according to the specified proportion, and set the sparse pixel value of the array to zero to obtain the common area density array of the group time sequence domain data. Scale the density array value to the interval of 0-255 to form the group time sequence distribution density gray scale map of the parameter data, and obtain the group door calibration data opening door speed, opening door current, closing door speed, and closing door current time sequence distribution density gray scale map.

[0095] (d) For the population vehicle door displacement domain data set of the calibration data, select quantitative data from each door displacement domain data set of the population vehicle door as the data set for subsequent image processing, construct a zero array with the maximum value of a parameter data such as opening door rotation speed, opening door current, closing door rotation speed or closing door current in the data set as the number of rows and the maximum length of the parameter data as the number of columns, add 1 to each value in each data of the parameter data set at the corresponding position of the array to form a data density array related to the number of data distribution. Calculate the sparse pixel value of the array according to the specified ratio, set the sparse pixel value in the array to zero to obtain the population displacement domain common area density array. Scale the density array value to the interval of 0-255 to form the population displacement domain distribution density gray scale image of the parameter data, and obtain the population vehicle door calibration data opening door rotation speed, opening door current, closing door rotation speed and closing door current displacement domain distribution density gray scale image.

[0096] (2.2) Distribution density gray scale image processing of the data set to be diagnosed. It mainly includes two ways: single door time series data distribution density image processing and single door displacement domain data distribution density image processing. The specific steps are as in (a) and (b) of step (2.1).

[0097] (2.3) Image data alignment.

[0098] Compare the length and width of the calibration data distribution density gray scale image obtained in step (2.1) and the distribution density gray scale image of the data to be diagnosed obtained in step (2.2), and take the maximum length and width value as the maximum size, and fill zero in the corresponding position of the other image array according to the maximum size.

[0099] (2.4) Generation of difference density image.

[0100] Based on the aligned distribution density gray scale image of the data to be diagnosed, remove the same area of the aligned calibration data distribution density gray scale image; the removal methods include but are not limited to: difference of pixels, zeroing of overlapping area pixels, to generate a difference density image.

[0101] Among them, the method of generating a difference density image by zeroing the pixels in the overlapping area includes: for each pixel (i, j) of the distribution density gray scale image of the data to be diagnosed, if the pixel value at the same position of the calibration data distribution density gray scale image is not 0, then directly set the value of the pixel (i, j) of the distribution density gray scale image of the data to be diagnosed to 0, and respectively obtain the opening door rotation speed, opening door current, closing door rotation speed or closing door current difference density image.

[0102] (3) Change area recognition and change degree evaluation based on the difference density image to obtain the door state deviation degree. Specifically includes:

[0103] (3.1) Change area recognition.

[0104] The intelligent algorithm is used to identify the area where data points are concentrated in the difference density image, that is, the concentrated change area of the data to be diagnosed relative to the calibration data. The intelligent algorithm includes pixel and sliding window identification method and effective pixel column identification method.

[0105] The main steps of the pixel and window identification method include:

[0106] 1) Initial area identification: the first sliding window is used to calculate and sort the area pixels of the difference density image; and the first sliding window corresponding large change area is obtained based on the change window identification rule;

[0107] 2) The large change area identified is calculated and sorted by the second sliding window; the second sliding window corresponding refined change area is obtained based on the change window identification rule; and the window of the second sliding window is smaller than that of the first sliding window;

[0108] 3) According to the large change area and the change window identification rule, the data change is divided into full range or local;

[0109] The above change window identification rule is:

[0110] (i) If the pixel sum of the first sorted window is greater than 2 times the pixel sum of the second sorted window, it is considered that the image is a single local area change, and the change area is the time sequence or displacement area corresponding to the first sorted window;

[0111] (ii) If the pixel sum of more than half of the windows is greater than 0.5 times the pixel sum of the first sorted window, it is considered that the image has full range change, and the change area is the full range;

[0112] (iii) If the pixel sum of the first sorted window is less than 2 times the pixel sum of the second sorted window, and the pixel sum of less than half of the windows is greater than 0.5 times the pixel sum of the first sorted window, it is considered that the image has multiple local area changes, and the change area is the time sequence or displacement area corresponding to each window whose window pixel sum is greater than 0.5 times the pixel sum of the first sorted window.

[0113] The main steps of the effective pixel column identification method include:

[0114] 1) The effective pixel gray threshold and the effective column continuous minimum length threshold are set in advance;

[0115] 2) Based on the difference density image, all effective pixel columns of the image are obtained;

[0116] 3) The continuity of the effective pixel column is judged, and the effective pixel column continuous area is obtained;

[0117] 4) judging the length of the continuous region of the effective pixel column, if the maximum continuous region length is less than the minimum length threshold of the continuous effective pixel column, the region is not output, otherwise all sub-regions are output as change regions.

[0118] (3.2) Region change degree evaluation.

[0119] The change degree of the change region identified in step (3.2) is evaluated by a feature analysis evaluation method, and a data region with a change degree meeting the evaluation rule is obtained, i.e., the door state offset degree is obtained. The feature analysis evaluation method includes a region center of gravity offset evaluation method, a co-occurrence matrix evaluation method, a mean square error evaluation method, and an effective pixel point evaluation method.

[0120] The main steps of the center of gravity offset evaluation method include:

[0121] 1) Calculate the center of gravity coordinates of the change region of the time series or displacement domain distribution density image of the calibration data and the data to be diagnosed, respectively. For a discrete two-dimensional image, let the pixel value be I(i,j), where i represents the horizontal coordinate of the pixel point and j represents the vertical coordinate of the pixel point. The calculation formula of the center of gravity (Gx, Gy) of the image is:

[0122] G x =(∑(i*I(i,j))) / M

[0123] G y =(∑(j*I(i,j))) / M

[0124] Where M is the sum of all pixel values of the image: M = ∑∑(I(i,j))

[0125] 2) Calculate the center of gravity of the calibration data the center of gravity of the data to be diagnosed Calculate the absolute amount of center of gravity offset ΔG x , ΔG y and the relative amount Rate(G x ), Rate(G y ) of the data to be diagnosed relative to the calibration data, and the calculation formula is as follows:

[0126]

[0127]

[0128]

[0129]

[0130] 3) Compare with the set center of gravity alarm threshold, and consider that the center of gravity offset degree meets the degree evaluation condition if it exceeds the threshold.

[0131] The main steps of the co-occurrence matrix evaluation method include:

[0132] 1) Calculate the calibration data co-occurrence matrix correlation feature: randomly select two groups of quantitative data in the calibration data set as the to-be-diagnosed data and the calibration data, use the pixel and sliding window recognition method to identify the change area, obtain the large change area of each difference density image, and calculate the calibration data co-occurrence matrix correlation feature p std ;

[0133] 2) Calculate the co-occurrence matrix correlation feature of the to-be-diagnosed data: calculate the co-occurrence matrix correlation feature p diag of the to-be-diagnosed data according to the large change area of the difference density image.

[0134] 3) Compare p std , p diag , and make a judgment. If p diag > p std , it is considered that the region change degree meets the co-occurrence matrix correlation evaluation condition. The co-occurrence matrix evaluation method can be used alone, but because of the large amount of calculation, it can be used for secondary evaluation of the change degree when the barycenter offset meets the evaluation requirements.

[0135] (3.3) Region integration.

[0136] The data regions that meet the evaluation rules are integrated and processed, and the main methods include merging adjacent regions and removing small regions.

[0137] (4) Establish a vehicle door functional component partition map, and associate the functional components based on the vehicle door state offset degree to identify the abnormal functional components.

[0138] The establishment of the vehicle door functional component partition map includes partitioning the vehicle door movement process according to the motor angle data, and the partition interval is greater than the minimum angle that can be recognized by the motor position sensor, and the types of functional components in the same partition are minimized, as shown in Figure 16 .

[0139] When using time series data, the image change region obtained by diagnosis is the data time offset data. It is necessary to convert the time region to a displacement region based on the original angle data. For the time region t1-t2, the corresponding displacement region d1-d2 is:

[0140] d1 = angle(t1)

[0141] d2 = angle(t2)

[0142] When using displacement domain data, the image change region obtained by diagnosis is the data angle or displacement offset data.

[0143] The displacement offset is converted into motor rotation angle. The functional partition corresponding to the data change area is obtained according to the door functional component partition diagram. The associated functional components of the corresponding partition are obtained from the functional partition.

[0144] The formula for converting angular displacement is as follows.

[0145] angle = dis * 360 / d

[0146] In the formula, angle is the motor rotation angle in degrees, dis is the door displacement in mm, and d is the lead screw pitch.

[0147] (5) Model Iteration. Based on the feedback from the mainline operation and maintenance, obtain the operation tag data, input the mainline operation data and data tags into the parameter iterative model, and automatically update the model parameters iteratively.

[0148] To further illustrate the method of the present invention, a door condition assessment is performed using a door on a mainline train as an example. The specific steps are as follows:

[0149] (1) Take 30 sets of door opening and closing data from the door calibration and motor timing data to be diagnosed, and then perform preprocessing operations to form the calibration dataset and the dataset to be diagnosed. Specifically, this includes:

[0150] (1.1) Collect timing data during the movement of the car door, including parameters such as motor current, motor speed, and motor angle during the opening and closing process.

[0151] (1.2) Remove stalled motor data and eliminate obviously abnormal data. The data curves of the door motor timing data before and after preprocessing are shown below. Figures 2-3 As shown.

[0152] (1.3) Displacement domain data conversion.

[0153] For a single set of motor current, speed, and angle data, this embodiment uses the displacement after angle conversion as the abscissa and the speed or current as the ordinate to plot a displacement domain curve image of the single set of data. The row number of the lowest pixel in each column of the image is read as the speed or current data corresponding to that displacement value, thus obtaining the displacement domain data. The absolute value of the door closing displacement data is taken and then negative, while the absolute value of other data is taken.

[0154] The following angular displacement conversion formula is used to convert the motor angular data into displacement:

[0155] dis = angle * d / 360

[0156] In the formula, dis is the door displacement in mm, angle is the motor rotation angle in °, and d is the lead screw. In this embodiment, the lead screw of the sliding door is 64.

[0157] (1.4) Divide the calibration data set into the data set to be diagnosed

[0158] The single gate displacement field data after preprocessing and displacement field data conversion is used as the calibration data set and the data set to be diagnosed.

[0159] (2) The preprocessed calibration data set and the data set to be diagnosed are respectively subjected to distribution density gray scale processing, and a difference density map is generated.

[0160] (2.1) Calibration data set distribution density gray scale processing. Select quantitative data from the single gate displacement field data set as the data set used for subsequent image processing. A full zero array is constructed with the maximum value of a parameter data such as the opening door speed, the opening door current, the closing door speed or the closing door current in the data set as the number of rows and the maximum value of the length of the parameter data as the number of columns. For each value in each data of the parameter data set, add 1 to the corresponding position of the array to form a data density array related to the distribution quantity of the data. Scale the data density array value to the interval of 0-255 to form the displacement field distribution density gray scale of the parameter data, and obtain the single gate calibration data opening door speed, opening door current, closing door speed and closing door current displacement field distribution density gray scale, as shown in FIG. 2. Figures 4-7

[0161] (2.2) Distribution density gray scale processing of the data set to be diagnosed

[0162] The processing process is as in step (2.1), and finally the opening door speed, opening door current, closing door speed and closing door current displacement field distribution density gray scale are obtained, as shown in FIG. 3. Figures 8-11

[0163] (2.3) Image data alignment.

[0164] The maximum size of the calibration data distribution density gray scale and the maximum size of the distribution density gray scale of the data to be diagnosed are used as the maximum size, and zero is filled in the upper side and the right side of the opening door data image array according to the maximum size, and zero is filled in the upper side and the left side of the closing door data image array according to the maximum size.

[0165] (2.4) The difference density image is generated by using the zeroing method of the overlapping area pixels, as shown in FIG. 4. Figures 12-15

[0166] (3) Based on the difference density map, the change area is identified and the change degree is evaluated, and the door state deviation degree is obtained. Specifically, it includes:

[0167] (3.1) Change area identification.

[0168] In this embodiment, the change area is identified by using the pixel and sliding window identification method.

[0169] ​​​1) Initial region identification: set the sliding window length 50, step 50, calculate the pixel sum of each window of the displacement domain difference density image, sort the pixel sum, and identify the large change region according to the pixel sum region identification rule;

[0170] 2) Set the sliding window length 3, step 3, calculate the pixel sum of the identified large change region, and identify the refined change region according to the pixel sum region identification rule.

[0171] 3) Identify the door data change region as the opening door speed 703-718mm, 650-698mm, the opening door current 712-718mm, the closing door speed 63-75mm, and the closing door current 51-63mm.

[0172] (3.2) Region change degree evaluation.

[0173] This embodiment uses the region barycenter offset feature and the symbiotic matrix correlation feature to evaluate the region change degree.

[0174] (3.2.1) Use the barycenter offset feature evaluation method to evaluate the region change degree

[0175] a1) Calculate the barycenter coordinates of the change region of the displacement domain distribution density image of the calibration data and the data to be diagnosed, respectively.

[0176] Calculate the barycenter coordinates of the change region of the time series or displacement domain distribution density image of the calibration data and the data to be diagnosed. For a discrete two-dimensional image, let its pixel value be I(i,j), where i represents the horizontal coordinate of the pixel point, j represents the vertical coordinate of the pixel point, and the calculation formula of the barycenter (G x ,G y ) of the image is as follows:

[0177] G x =(∑(i*I(i,j))) / M

[0178] G y =(∑(j*I(i,j))) / M

[0179] Where M is the sum of all pixel values of the image: M=∑∑(I(i,j))

[0180] 2) According to the barycenter of the calibration data Calculate the absolute amount of barycenter offset ΔG x , ΔG y and the relative amount Rate(G x ), Rate(G y ) of the data to be diagnosed relative to the calibration data, and the calculation formula is as follows:

[0181]

[0182]

[0183]

[0184]

[0185] wherein, respectively are the horizontal and vertical coordinates of the center of gravity of the calibration data, respectively are the horizontal and vertical coordinates of the center of gravity of the data to be diagnosed, and x , ΔG y respectively are the absolute values of the horizontal and vertical center of gravity offsets of the data to be diagnosed relative to the calibration data, and x Rate(G y ) respectively are the relative values of the horizontal and vertical center of gravity offsets of the data to be diagnosed relative to the calibration data.

[0186] a3) comparing with the set center of gravity alarm threshold value, and if the threshold value is exceeded, it is considered that the center of gravity offset degree meets the degree evaluation condition.

[0187] In this example, the center of gravity alarm threshold value is set as shown in Table 1, which is set according to the application situation of the previous model tracking.

[0188] Table 1: Center of gravity alarm threshold value setting value

[0189] Image type AG x ]] AG y ]] Rate(G x )]]> Rate(G y )]]> Opening door speed 0.01 32 0.002 0.001 Opening door current 0.009 34 0.008 0.002 Closing door speed 0.004 15 0.0008 0.001 Closing door current 0.007 12 0.001 0.008

[0190] For the change area obtained in step (3.1): opening door speed 703-718 mm, 650-698 mm, opening door current 712-718 mm, closing door speed 63-75 mm, closing door current 51-63 mm, based on the above table, the center of gravity offset of each area is calculated, and the calculation results are shown in Table 2.

[0191] Table 2: Center of gravity offset calculation value of each area

[0192]

[0193] a4) comparing the offset values calculated in Table 2 with the threshold value setting values in Table 1, it is considered that the change degree meets the center of gravity evaluation condition. It can be seen that the area that meets the center of gravity offset degree evaluation condition is related to the door speed 63-75 mm and the closing door current 51-63 mm, wherein 63-75 mm and 51-63 mm represent the distance from the single door to the closing position.

[0194] (3.2.2) Secondary evaluation using symbiotic matrix correlation feature evaluation method

[0195] For the abnormal region meeting the barycenter offset evaluation, the corresponding large change region is evaluated for the symbiotic matrix correlation feature, and the main steps are as follows.

[0196] b1) Calculate the symbiotic matrix correlation feature of the calibration data: randomly select two groups of quantitative data in the calibration data set as the to-be-diagnosed data and the calibration data, perform change region identification, obtain the large change region of the displacement domain difference density image, and calculate the symbiotic matrix correlation feature p std .

[0197] b2) Calculate the symbiotic matrix correlation feature of the to-be-diagnosed data: the abnormal region obtained after the barycenter evaluation is 63-75 mm of the door closing speed and 51-63 mm of the door closing current, and it can be known that the large change region of the displacement domain difference density image is 50-100 mm of the door closing speed and 50-100 mm of the door closing current. The symbiotic matrix correlation features p diag .

[0198] b3) Judge the corresponding displacement domain difference density images p std , p diag , if p diag >p std , it is considered that the symbiotic matrix correlation evaluation condition is met.

[0199] According to the above process, the door closing speed p std is 0.45, the door closing current p diag is 0.68, p diag >p std , and the door closing speed 63-75 mm meets the correlation evaluation condition for the large region change degree; the door closing current p std is 0.50, the door closing current p diag is 0.71, p diag >p std , and the door closing current 51-63 mm meets the correlation evaluation condition for the large region change degree.

[0200] After evaluation, the door closing speed 63-75 mm and the door closing current 51-63 mm corresponding to the large region meet the symbiotic matrix correlation evaluation condition.

[0201] (3.3) Region integration.

[0202] The data regions meeting the evaluation rules for the change degree are integrated and processed. In this embodiment, the integrated alarm region is 51-75 mm of the door closing.

[0203] (4) Establish a vehicle door functional component partition map, and associate the functional components based on the door state offset degree to identify the functional components that occur abnormities.

[0204] The establishing of the car door function component partition map includes partitioning the car door movement process according to the motor angle data, the partition interval is greater than the minimum angle that can be recognized by the motor position sensor, and the types of function components in the same partition are minimized, such as Figure 16 As shown.

[0205] The door closing displacement offset is converted to the motor angle, the data change area corresponding to the function partition is obtained according to the car door function component partition map, and the associated function components of the corresponding partition are obtained from the function partition. The conversion formula of the angle displacement is as follows: angle = dis * 360 / d, wherein d is 64,

[0206] angle = dis * 360 / d

[0207] The corresponding function partition is 10-12 according to the car door function component partition map, and the associated components include the belt drive system, the motor, the motor fixed seat bearing, the middle support of the screw rod, the middle support of the end unlocking, the nylon nut of the screw nut pair, the long guide column slide cylinder pair, the short guide column slide cylinder pair, the block pair, and the screw nut pair. The screw segment movement. Therefore, the abnormal parts of the car door state in the embodiment can include the belt drive system, the motor, the motor fixed seat bearing, the middle support of the screw rod, the middle support of the end unlocking, the nylon nut of the screw nut pair, the long guide column slide cylinder pair, the short guide column slide cylinder pair, the block pair, and the screw nut pair. The screw segment movement.

[0208] (5) Model iteration. According to the feedback of the main line operation and maintenance, the operation label data is obtained, the main line operation data and the data label are input into the parameter iteration model, and the model parameters are automatically iterated and updated.

[0209] The present application forms a gray scale image related to data distribution density by image processing of motor data, identifies the car door with state deviation and the deviation area through multiple intelligent algorithms, associates the change area with the door system function components according to the car door function component partition map, combines the car door state deviation area with the associated components as early warning information for pushing, and facilitates accurate maintenance on site. At the same time, according to the application situation, the model is continuously iterated and optimized.

[0210] The car door state recognition system based on time series data image processing provided by the present application comprises:

[0211] A data acquisition and preprocessing module is used for acquiring time series data in the car door movement process and performing data preprocessing, and is divided into a calibration data set and a to-be-diagnosed data set;

[0212] A difference density image generation module is used for respectively performing distribution density gray scale image processing on the preprocessed calibration data set and to-be-diagnosed data set, and generating a difference density image;

[0213] a change area and degree identification module, configured to perform change area identification and change degree evaluation based on the difference density map, to obtain a door state offset degree;

[0214] an abnormal function component association module, configured to establish a door function component partition map, and perform function component association based on the door state offset degree, to identify an abnormal function component.

[0215] The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is loaded into the processor, the above-mentioned door state recognition method based on time-series data image processing is implemented.

[0216] The computer readable storage medium stores a computer program, and when the computer program is executed by the processor, the above-mentioned door state recognition method based on time-series data image processing is implemented.

[0217] The computer readable storage medium includes RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory or any other medium that can be used to store instructions or data structures in the form of a program code and can be accessed by a computer.

[0218] The processor is configured to execute the computer program stored in the memory to implement each step in the method involved in the above-mentioned embodiments.

Claims

1. A vehicle door state recognition method based on time-series data image processing, characterized by, The method comprises the following steps: (1) collecting time series data in the process of door movement and performing data preprocessing, and dividing into calibration data set and data set to be diagnosed; The data preprocessing comprises displacement domain conversion on the time series data; The displacement domain conversion comprises: for single group motor current, motor speed, motor angle data, drawing displacement domain curve image of single group data with motor angle or converted displacement as horizontal coordinate and motor speed or motor current as vertical coordinate, reading the row number of the lowest point of each column of pixels in the image as the speed or current data corresponding to the angle or displacement value, and obtaining displacement domain data; (2) performing distribution density gray scale processing on the preprocessed calibration data set and data set to be diagnosed respectively, removing the same area as the aligned calibration data distribution density gray scale based on the aligned data set to be diagnosed distribution density gray scale, and generating difference density map; (3) identifying change area and evaluating change degree based on the difference density map, and obtaining door state deviation degree; comprising: identifying the area where data points are more concentrated in the difference density map based on intelligent algorithm, that is, the concentrated change area of the data to be diagnosed relative to the calibration data; evaluating the change degree of the identified change area by feature analysis evaluation method, and obtaining the door state deviation degree, the feature analysis evaluation method comprises: firstly, using the barycentric offset method for preliminary evaluation, and when the barycentric offset meets the evaluation requirement, using the co-occurrence matrix evaluation method for secondary evaluation of the change degree; the co-occurrence matrix evaluation method comprises: identifying large change area by pixel and sliding window identification method, obtaining large change area of each difference density image, calculating calibration data co-occurrence matrix correlation feature and data to be diagnosed co-occurrence matrix correlation feature; if the co-occurrence matrix correlation feature of the data to be diagnosed is greater than that of the calibration data, it is considered that the change area has deviation of door state; (4) establishing door function component partition map, and associating function components based on door state deviation degree to identify abnormal function components; The establishment of the door function component partition map comprises partitioning the door movement process according to the motor angle data, and the partition interval is greater than the minimum angle that can be recognized by the motor position sensor, and the types of function components in the same partition are the least; The function component association comprises converting the door state deviation data into motor angle data, and finding the corresponding function component partition and the function components in the partition range based on the angle data and the door function component partition map.

2. The method of claim 1, wherein, The data preprocessing in step (1) comprises abnormal data elimination.

3. The method of claim 2, wherein the method further comprises: The calibration data set comprises time series data set or displacement domain data set of single door or group door, and the data set to be diagnosed comprises time series data set or displacement domain data set of single door; wherein the group door refers to the door with similar operating conditions in the same time period.

4. The method of claim 3, wherein the method further comprises: The distribution density gray scale processing in step (2) comprises: For a single door time series data set or displacement domain data set, quantitative data in the data set is selected as an image processing data set; a full zero array is constructed with the maximum value of a certain parameter data in the data set as the number of array rows and the maximum value of the length of the parameter data as the number of array columns, and each value in each data of the parameter data set is added by 1 at the corresponding position of the array to form a data density array related to the number of data distribution; the data density array values are scaled to the interval of 0-255 to form a time series or displacement domain distribution density grayscale image of the parameter data. For a group door time series data set or displacement domain data set, quantitative data in the data set is selected as an image processing data set; a full zero array is constructed with the maximum value of a certain parameter data in the data set as the number of array rows and the maximum value of the length of the parameter data as the number of array columns, and each value in each data of the parameter data set is added by 1 at the corresponding position of the array to form a data density array related to the number of data distribution; then the array sparse pixel values are calculated according to a specified proportion, the array sparse pixel values are set to zero, and a group time series or displacement domain data common area density array is obtained; the density array values are scaled to the interval of 0-255 to form a time series or displacement domain distribution density grayscale image of the parameter data.

5. The method of claim 4, wherein the method further comprises: The image data alignment before step (2) generating the difference density image further includes selecting the maximum length and width values in the distribution density grayscale images of the calibration data set and the data set to be diagnosed as the maximum size, and performing zero padding in the corresponding positions of other images according to the maximum size.

6. The method of claim 5, wherein the method further comprises: The image data alignment before step (2) generating the difference density image further includes selecting the maximum length and width values in the distribution density grayscale images of the calibration data set and the data set to be diagnosed as the maximum size, and performing zero padding in the corresponding positions of other images according to the maximum size.

7. The method of claim 1, wherein the method further comprises: The intelligent algorithm used in step (3) includes a pixel and sliding window recognition method and an effective pixel column recognition method.

8. The method of claim 1, wherein, The pixel and sliding window recognition method includes: 1) calculating and sorting the region pixel sums of the difference density image using a sliding window; 2) obtaining a large change region and a refined change region based on a change window recognition rule; 3) dividing the data change into a whole change or a local change according to the large change region and the change window recognition rule; wherein the change window recognition rule is: (i) if the pixel sum of the first ranked window is greater than 2 times the pixel sum of the second ranked window, the image is considered to have a single local region change, and the change region is the time series or displacement region corresponding to the first ranked window; (ii) if the pixel sum of more than half of the windows is greater than 0.5 times the pixel sum of the first ranked window, the image is considered to have a whole region change, and the change region is the whole region; (iii) if the pixel sum of the first ranked window is less than 2 times the pixel sum of the second ranked window and the pixel sum of less than half of the windows is greater than 0.5 times the pixel sum of the first ranked window, the image is considered to have multiple local region changes, and the change region is the time series or displacement region corresponding to each window with a pixel sum greater than 0.5 times the pixel sum of the first ranked window. 9.The method of claim 7, wherein, The effective pixel column recognition method includes: 1) presetting an effective pixel grayscale threshold and an effective pixel column continuous minimum length threshold; 2) obtaining all effective pixel columns in the difference density image based on the difference density image; 3) judging the continuity of the effective pixel columns to obtain an effective pixel column continuous region; 4) judging the length of the continuous region of the effective pixel column, if the maximum continuous region length is less than the minimum length threshold of the effective pixel column, the region is not output, otherwise all sub-regions are output as change regions.

10. The method of claim 1, wherein, The feature analysis evaluation method of step (3) further includes a mean square error evaluation method and an effective pixel point evaluation method.

11. The method of claim 1, wherein, The barycenter offset evaluation method includes: 1) calculate the gravity coordinates of the variation region of the time series or displacement domain distribution density image of the calibration data and the data to be diagnosed respectively, for a discrete two-dimensional image, let its pixel value be , where i represents the horizontal coordinate of the pixel point, j represents the vertical coordinate of the pixel point, and the calculation formula of the gravity (G x , G y ) of the image is: wherein M is the sum of all pixel values of the image: 2) According to the calibration data gravity center ( , ), the data gravity center ( , ) to be diagnosed, the absolute amount of gravity center offset of the data to be diagnosed relative to the calibration data 、 and the relative amount 、 are calculated, and the calculation formula is as follows:​​ 3) comparing with the set barycenter alarm threshold, if exceeding the threshold, it is considered that the change region of the vehicle door state has offset.

12. The method of claim 1, wherein, It also includes model iterative updating, which includes obtaining positive line operation label data based on positive line operation feedback, taking the positive line operation label data as model input, and automatically updating the model parameters.

13. A system for recognizing a state of a vehicle door based on time-series data image processing, for implementing the method for recognizing a state of a vehicle door based on time-series data image processing according to any one of claims 1 to 12, characterized by, It includes: A data acquisition and preprocessing module for acquiring time series data during vehicle door movement and preprocessing data, which is divided into a calibration dataset and a dataset to be diagnosed; A difference density image generation module for processing the distribution density gray scale of the preprocessed calibration dataset and the dataset to be diagnosed, and generating a difference density map; A change region and degree identification module for identifying change regions and evaluating change degrees based on the difference density map to obtain the degree of vehicle door state offset; An abnormal function component association module for establishing a vehicle door function component partition map and associating function components based on the degree of vehicle door state offset to identify abnormal function components.

14. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is loaded into the processor to realize the vehicle door state recognition method based on time series data image processing according to any one of claims 1-12.

15. A computer-readable storage medium, the computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to realize the vehicle door state recognition method based on time series data image processing according to any one of claims 1-12.

Images