Fixed lever fault detection method and system based on multi-scale feature cascade detection

Abstract

The invention discloses a fixed lever fault detection method and system based on multi-scale feature cascade detection, and belongs to the technical field of rail wagon fault detection. The problems that an existing fault detection method for related parts of a fixed lever neglects the overall semantics of the parts, the detection process is tedious, and the accuracy rate is low are solved. The method comprises the following steps: firstly, acquiring specific position coordinates of each component in a to-be-identified truck image and an image area to be subjected to state detection and entity detection in the image by utilizing a truck component positioning network; detecting the state detection image and the entity detection image by using a state detection network and an entity detection network respectively to obtain corresponding detection results; and finally, based on detection results of the state detection network and the entity detection network, performing fault detection on the fixed lever related component on the current to-be-identified truck image according to the overall semantic features of the fixed lever related component.

Description

Technical Field

[0001] The present invention belongs to the technical field of railway freight car fault detection, and in particular relates to a fixed lever fault detection method and system. Background Art

[0002] Truck brake lever components include the fixed lever, movable lever, fixed lever fulcrum, fixed lever fulcrum chain shoe ring, fixed lever fulcrum seat, and the connecting parts between these components: round pins, rivet pins, and cotter pins. These components transmit braking force to the truck and are crucial components of the truck's braking system. Failure can have unimaginable consequences.

[0003] Fault detection of fixed lever-related components faces the following difficulties:

[0004] 1. The components related to the fixed lever are usually located at the bottom of the truck, which makes inspection inconvenient;

[0005] 2. Some minor fault forms, such as broken round pins and missing rivet pin sleeves, are detected only by the human eye, resulting in a high rate of missed detection;

[0006] 3. There are many types of failures in the components related to the fixed lever, the most common of which are the following 22 types:

[0007] 1) Fixed lever broken; 2) Fixed lever lost; 3) Moving lever broken; 4) Moving lever lost; 5) Fixed lever fulcrum broken; 6) Fixed lever fulcrum lost; 7) Fixed lever fulcrum round pin broken; 8) Fixed lever fulcrum round pin lost; 9) Fixed lever fulcrum cotter pin lost; 10) Fixed lever fulcrum chain shoe broken; 11) Fixed lever fulcrum chain shoe lost; 12) Fixed lever fulcrum chain shoe round pin lost; 13) Fixed lever fulcrum The round pin of the chain shoe ring is broken; 14) The cotter pin of the chain shoe ring round pin of the fixed lever fulcrum is lost; 15) The rivet pin sleeve of the chain shoe ring of the fixed lever fulcrum is lost; 16) The fixed lever fulcrum seat is broken; 17) The fixed lever fulcrum seat is lost; 18) The round pin of the fixed lever fulcrum seat is broken; 19) The round pin of the fixed lever fulcrum seat is lost; 20) The cotter pin of the round pin of the fixed lever fulcrum seat is lost; 21) The rivet pin sleeve of the fixed lever fulcrum seat is lost; 22) The rivet of the fixed lever fulcrum seat is lost.

[0008] There are many types of faults associated with fixed lever components, and some have similar fault morphologies, making them difficult to distinguish with the naked eye. Currently, most methods rely on manual identification of faults in images captured by vehicles. This method is labor-intensive, requiring the review of nearly 10,000 images of truck operations per day. Furthermore, due to the influence of the professional level and work attitude of the relevant personnel, the missed detection rate is high. Most existing automated recognition methods are based on traditional computer vision fault detection methods, such as differential and morphological methods, which have low accuracy. Other deep learning-based methods address the problem of fixed lever-related fault detection in an overly fragmented manner, using deep learning models to individually determine whether each component is faulty. This ignores the overall semantics of the fixed lever-related components, making detection very cumbersome and time-consuming. Summary of the Invention

[0009] The present invention aims to solve the problems that the existing method for detecting faults of fixed lever related components ignores the overall semantics of the components, the detection process is cumbersome and the accuracy is low.

[0010] A method for detecting fixed lever faults based on multi-scale feature cascade detection includes the following steps:

[0011] S1. Using the truck component positioning network to obtain the specific location coordinates of each component in the truck image to be identified, and then obtain the image area to be tested for state detection and entity detection in the image;

[0012] The output of the truck component positioning network is the specific square frame position coordinates of the truck component in the current truck image to be identified. The truck components include the bolster, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, center tie rod, fixed lever fracture, and mobile lever fracture components.

[0013] S2. Use the state detection network and the entity detection network to detect the state detection image and the entity detection image respectively to obtain the detection results; the output of the state detection network is the specific square box position coordinates of the connection status of each component in the state detection area of ​​the current truck image to be identified; the output of the entity detection network is the position coordinates of the entities on each component in the entity detection area of ​​the current truck image to be identified; based on the detection results of the state detection network and the entity detection network, the overall semantics of the fixed lever related components are used to perform fixed lever fault detection on the current truck image to be identified.

[0014] Furthermore, the process of obtaining an image to be subjected to state detection and entity detection in an image by using a component extraction algorithm in step S1 includes the following steps:

[0015] Check whether the detection results of the truck parts positioning network include fixed levers;

[0016] If the fixed lever is not included: query whether the detection results of the truck component positioning network include the fixed lever fulcrum connection; if the fixed lever fulcrum connection is not included, divide the truck image into four sub-images along the center line of the width and the center line of the height of the truck image to be identified, record the coordinates of these four sub-images relative to the original truck image in a clockwise direction, and use these four sub-images as state detection images and entity detection images; if the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, and record the coordinates of the fixed lever fulcrum connection relative to the original truck image, and use the image in the fixed lever fulcrum connection area as the state detection and entity detection image;

[0017] If a fixed lever is included: record the position coordinates of the fixed lever in the truck image; query whether the detection results of the truck component positioning network include the fixed lever fulcrum connection; if the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, record the coordinates of the fixed lever fulcrum connection relative to the original truck image, and the image in the area of ​​the fixed lever fulcrum connection is used as the image for state detection and entity detection; if the fixed lever fulcrum connection is not included, the two images of the area where the fixed lever is located are used as the images for state detection and entity detection.

[0018] Furthermore, during the state detection network detection process, the connection status of each component includes normal connection status, abnormal connection status, fixed lever fulcrum tilt, fixed lever fulcrum loss, fixed lever fulcrum seat breakage, fixed lever fulcrum chain shoe ring breakage, fixed lever fulcrum breakage, fixed lever fulcrum horizontal breakage, round pin loss and wire binding, round pin loss hole, round pin breakage, cotter pin loss, connection component leakage, rivet pin sleeve loss, rivet loss;

[0019] During the entity detection network inspection process, the entities on each component include the fixed lever fulcrum, the fixed lever fulcrum chain shoe ring, the fixed lever fulcrum seat, the inclination of the fixed lever fulcrum seat, the round pin, the rivet pin, the reflective rivet pin left after the rivet pin sleeve is lost, the connecting component has protruded, the cylinder left after the cotter pin is lost, the broken end of the round pin, the broken end of the fixed lever fulcrum seat, the broken end of the fixed lever fulcrum chain shoe ring, the broken end of the fixed lever fulcrum, the broken end of the component, the hole formed by the loss of the component, the iron wire, and the inclination of the fixed lever fulcrum.

[0020] Furthermore, the specific process of using the overall semantics of the fixed lever related components to perform fixed lever fault detection on the current truck image to be identified in step S2 includes:

[0021] S21, obtaining the number of images detected in the state obtained in step S1;

[0022] S22. If the number of state detection images is equal to 4, the entire truck image to be identified is confirmed as "fixed lever fulcrum missing";

[0023] S23. If the number of state detection images is equal to 2, determine whether the state detection image is a normal image; if it is not a normal image, perform rivet missing fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection;

[0024] S24. If the number of status detection images is equal to 1, determine whether the status detection image is a normal image; if it is not a normal image, perform rivet pin sleeve missing fault detection, cotter pin missing fault detection, round pin missing fault detection, round pin breakage fault detection, entity breakage fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection.

[0025] Furthermore, the specific process of determining whether the state detection image is a normal picture includes:

[0026] A1. Obtain the number of state detection images;

[0027] A2. If the number of state detection images is equal to 4, it is not a normal image;

[0028] A3. If the number of state detection images is not equal to 4, traverse all state detection results of the state detection network:

[0029] If there is a normal connection state, obtain the detection confidence of the abnormal connection state and the normal connection state; if in addition to the abnormal connection state and the normal connection state, there are other abnormal connection states in the state detection results, then it is not a normal image; otherwise, if the detection confidence of the abnormal connection state is greater than the detection execution degree of the normal connection state, then it is not a normal image;

[0030] If there is no normal connection status, it is not a normal image;

[0031] A4: If A2 and A3 do not determine it as an abnormal image, it is determined to be a normal image.

[0032] Furthermore, the process of performing rivet missing fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection in step S23 is as follows:

[0033] B. The specific process of rivet loss fault detection includes:

[0034] If the status test result shows that the rivet is missing, it is confirmed as "the rivet of the fixed lever fulcrum seat is missing"; otherwise, if the physical test result shows that there is a hole in the fixed lever fulcrum seat, it is confirmed as "the rivet of the fixed lever fulcrum seat is missing";

[0035] C. The specific process of entity loss fault detection includes:

[0036] If the status detection result shows that the fixed lever fulcrum is missing, it is confirmed as "fixed lever fulcrum missing"; otherwise, if the physical detection result shows neither a fixed lever fulcrum nor a fixed lever fulcrum tilt, it is confirmed as "fixed lever fulcrum missing"; if the physical detection result shows that there is no fixed lever fulcrum seat, it is confirmed as "fixed lever fulcrum seat missing"; if the physical detection result shows that there is a fixed lever fulcrum seat and a fixed lever fulcrum, but no fixed lever fulcrum chain shoe ring, and the current truck is a model with a chain shoe ring, it is confirmed as "fixed lever fulcrum chain shoe ring missing";

[0037] D. The specific process of detecting the tilt fault of the fixed lever fulcrum includes:

[0038] Get the number of state detection images;

[0039] If the number of state detection images is equal to 1, continue to judge: if the state detection result shows that the fixed lever fulcrum is tilted, and the physical detection result shows that the component is broken, it is confirmed as "fixed lever fulcrum broken"; if the physical detection result does not show the fixed lever fulcrum seat, and the current truck is a model with a chain shoe ring, it is confirmed as "fixed lever fulcrum chain shoe ring missing"; if the current truck is not a model with a chain shoe ring, it is confirmed as "fixed lever fulcrum seat missing";

[0040] If the number of state detection images is equal to 2, continue to judge:

[0041] If the fixed lever fulcrum is tilted in the state detection result, further judgment is made: if the fixed lever fulcrum is broken in the state detection result, it is confirmed as "fixed lever fulcrum broken"; if the fixed lever fulcrum is broken in the physical detection result, it is confirmed as "fixed lever fulcrum broken"; if the component is broken in the physical detection result, it is confirmed as "fixed lever fulcrum broken";

[0042] If the fixed lever fulcrum seat is broken in the status detection result, it is confirmed as "fixed lever fulcrum seat broken";

[0043] If there is a broken part of the fixed lever fulcrum seat in the physical inspection results, it is confirmed as "broken fixed lever fulcrum seat";

[0044] If the status detection result shows that the chain shoe ring of the fixed lever fulcrum is broken, it is confirmed as "the chain shoe ring of the fixed lever fulcrum is broken";

[0045] If the physical inspection results show a broken link of the fixed lever fulcrum chain shoe, it is confirmed as "broken link of the fixed lever fulcrum chain shoe";

[0046] If the physical inspection result shows a hole caused by a missing round pin, and the truck is equipped with a fixed lever fulcrum chain shoe ring, it is confirmed as "Fixed lever fulcrum chain shoe ring missing round pin". Otherwise, it is confirmed as "Fixed lever fulcrum seat missing round pin";

[0047] If there is no fixed lever fulcrum seat in the physical detection result, and the current truck is a model with a fixed lever fulcrum chain shoe ring, it is confirmed as "fixed lever fulcrum chain shoe ring is lost", otherwise, it is confirmed as "fixed lever fulcrum seat is lost".

[0048] Furthermore, if the image in step S24 is not a normal image, the process of performing rivet pin sleeve missing fault detection, cotter pin missing fault detection, round pin missing fault detection, round pin breakage fault detection, and entity breakage fault detection includes:

[0049] E. The specific process of detecting the rivet pin sleeve missing fault includes:

[0050] If the status detection results show that the rivet pin sleeve is missing or the connecting part is out of place, confirm "Loss of rivet pin sleeve for fixed lever fulcrum" or "Loss of rivet pin sleeve for fixed lever fulcrum seat" according to the location of the status area;

[0051] If the physical inspection results show reflective rivet pins left by missing rivet pin sleeves, confirm "Loss of fixed lever fulcrum chain shoe ring rivet pin sleeve" or "Loss of fixed lever fulcrum seat rivet pin sleeve" according to their location.

[0052] If there is a hole on the fixed lever fulcrum seat in the physical inspection results, and the hole is on the left side of the fixed lever fulcrum seat, it is confirmed as "the fixed lever fulcrum seat rivet pin sleeve is missing";

[0053] F. The specific process of detecting the cotter pin missing fault includes:

[0054] If the status detection results show that the cotter pin is missing, the connecting part is out of place, or the round pin is broken, confirm "Loss of cotter pin at fixed lever fulcrum", "Loss of cotter pin at fixed lever fulcrum chain shoe ring", and "Loss of cotter pin at fixed lever fulcrum seat" according to the location of the area to which the status belongs;

[0055] If the physical inspection results show a cylinder formed by the loss of a round pin cotter pin, confirm the "fixed lever fulcrum cotter pin lost", "fixed lever fulcrum chain shoe ring cotter pin lost", and "fixed lever fulcrum seat cotter pin lost" according to the area where the cylinder is located.

[0056] G. The specific process of round pin missing fault detection includes:

[0057] If there is a wire tied to the round pin in the status detection result, confirm "Fixed lever fulcrum round pin missing", "Fixed lever fulcrum chain shoe ring round pin missing", "Fixed lever fulcrum seat round pin missing" according to the location of the area to which the status belongs;

[0058] If the status detection results show both holes caused by missing round pins and tilted fixed lever fulcrums, confirm "Fixed lever fulcrum chain shoe ring round pin missing" and "Fixed lever fulcrum seat round pin missing";

[0059] If there is iron wire in the physical detection result, confirm "Fixed lever fulcrum round pin missing", "Fixed lever fulcrum chain shoe ring round pin missing", "Fixed lever fulcrum seat round pin missing" according to the location of the area;

[0060] H. The specific process of round pin breakage fault detection includes:

[0061] If there is a broken round pin in the status detection result, confirm "Fixed lever fulcrum round pin broken", "Fixed lever fulcrum chain shoe ring round pin broken", "Fixed lever fulcrum seat round pin broken" according to the position of the area to which the status belongs;

[0062] If there are broken round pins in the physical inspection results, confirm "Fixed lever fulcrum round pin broken", "Fixed lever fulcrum chain shoe ring round pin broken", "Fixed lever fulcrum seat round pin broken" according to the location of the area to which it belongs;

[0063] I. The specific process of entity breakage fault detection includes:

[0064] If the fixed lever fulcrum is broken horizontally in the status detection result, confirm "Fixed lever fulcrum broken";

[0065] If the fixed lever fulcrum is broken in the status detection result, confirm "Fixed lever fulcrum broken";

[0066] If there is a fracture caused by the broken fixed lever fulcrum in the physical detection results, confirm "Fixed lever fulcrum broken";

[0067] If the status detection result shows that the fixed lever fulcrum chain shoe ring is broken, confirm "Fixed lever fulcrum chain shoe ring is broken";

[0068] If the status detection result shows that the fixed lever fulcrum seat is broken, confirm that "the fixed lever fulcrum seat is broken".

[0069] Furthermore, the process of obtaining the location of the region to which the state belongs and the location of the region to which the entity belongs includes:

[0070] According to the positioning results of the bolster, brake beam, fixed lever and wheel axle of the component positioning network, the area between the bolster and the brake beam is As the area of ​​the fixed lever fulcrum seat, the area between the brake beam and the fixed lever As the area where the chain shoe ring belongs to the fixed lever fulcrum, the area between the fixed lever and the wheel axle is fixed The area where the fixed lever fulcrum belongs;

[0071] Calculate the position coordinates of the state square box, the position coordinates of the entity square box and The IOU with the larger IOU value is determined as the state and the regional location to which the entity belongs.

[0072] Furthermore, the method further includes the step of filtering the detected fault form to generate the coordinates of a square frame of the location of the fault form in the truck image. The specific process includes:

[0073] K1. Get all current confirmation results;

[0074] K2. If both component loss and component breakage occur at the same time, delete all component loss confirmation results;

[0075] K3. If both round pin loss and round pin breakage occur at the same time, delete all round pin breakage confirmation results;

[0076] K4. If both the rivet pin sleeve and the cotter pin are lost at the same time, delete all the rivet pin sleeve loss confirmation results;

[0077] K5. If both round pin and cotter pin are lost, delete all cotter pin loss confirmation results.

[0078] K6. Delete all fault confirmation results that are not currently tested for the truck according to the predefined configuration file;

[0079] K7. Traverse the remaining confirmation results and write the coordinates of the square box where the fault form is located in the confirmation result information into the specified file path.

[0080] A fixed lever fault detection system based on multi-scale feature cascade detection, the system comprising:

[0081] Truck component positioning module: This module uses the positioning network to obtain the specific coordinates of the bolster, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, center tie rod, fixed lever fracture, and mobile lever fracture in the truck image to be identified.

[0082] Component connection status detection module: performs status detection on the fixed lever fulcrum connection to obtain normal connection status, abnormal connection status, fixed lever fulcrum tilt, fixed lever fulcrum loss, fixed lever fulcrum seat break, fixed lever fulcrum chain shoe ring break, fixed lever fulcrum break, fixed lever fulcrum horizontal break, round pin loss and wire binding, round pin loss hole, round pin break, cotter pin loss, connection component leakage, rivet pin sleeve loss, and rivet loss connection status;

[0083] Small component entity detection module: responsible for detecting the fixed lever fulcrum, fixed lever fulcrum chain shoe ring, fixed lever fulcrum seat, fixed lever fulcrum seat tilt, round pin, rivet pin, reflective rivet pin left after the rivet pin sleeve is lost, connecting component ejection, cylinder left after the cotter pin is lost, round pin breakage, fixed lever fulcrum seat breakage, fixed lever fulcrum chain shoe ring breakage, fixed lever fulcrum breakage, component breakage, holes formed by component loss, wire, and fixed lever fulcrum tilt entity included in the fixed lever fulcrum connection;

[0084] Fault detection module: Based on the detection results of the truck component positioning module, component connection status detection module, and small component entity detection module, the module uses the overall semantics of the fixed lever-related components to perform fixed lever fault detection on the current truck image to be identified.

[0085] Beneficial effects:

[0086] The method proposed in the present invention regards the fault detection of fixed lever-related components as a whole, uses deep learning methods to obtain the overall semantics of the fixed lever-related components, and jointly detects the corresponding faults based on the overall semantics of the fixed lever-related components: first, the location of each component in the current truck image is obtained, and then whether the connection status of each component is abnormal is determined. Finally, the current shape of each small component is specifically checked to determine whether a fault has occurred. Compared with previous methods, the method has higher accuracy and faster detection speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 This is the model structure diagram of the truck parts positioning network, state detection network, and entity detection network.

[0088] Figure 2 The figure is the overall flow chart for fault detection of fixed lever related components.

[0089] Figure 3 Processing diagrams for the system. DETAILED DESCRIPTION

[0090] This invention aims to address the problem that existing methods for detecting faults in fixed lever-related components ignore the overall semantics of components, resulting in a cumbersome detection process and low accuracy. First, the invention utilizes a truck component positioning network to obtain the specific location coordinates of each component in the truck image to be identified, as well as the image regions where state detection and entity detection are to be performed. The state detection image and entity detection image are then inspected using a state detection network and an entity detection network, respectively, to obtain detection results. Finally, based on the detection results of the state detection network and the entity detection network, fault detection of fixed lever-related components is performed on the truck image to be identified according to the overall semantic features of the fixed lever-related components. This is explained below with reference to specific implementation methods.

[0091] Specific implementation method 1: Combination Figures 1 to 2 To explain this embodiment,

[0092] This embodiment is a method for detecting fixed lever faults based on multi-scale feature cascade detection, comprising the following steps:

[0093] S1. Use the truck component positioning network to obtain the specific location coordinates of each component in the truck image to be identified, and obtain the image area to be detected and detected by the component extraction algorithm;

[0094] The output of the truck component positioning network is the specific square frame position coordinates of the truck components in the current truck image to be identified, including the rocker, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, center tie rod, fixed lever broken fracture, and mobile lever broken fracture components.

[0095] In some embodiments, in step S1 of "obtaining an image region to be subjected to state detection and entity detection in an image by using a component extraction algorithm," the component extraction algorithm includes the following steps:

[0096] S11. Obtain the preset image cutting parameters, including: the default state detection image area width and height, the default entity detection image area width and height;

[0097] S12. Query the truck component positioning network detection results whether it contains a fixed lever;

[0098] S13. If not including a fixed lever:

[0099] S131. Query the truck component positioning network detection results whether it contains a fixed lever fulcrum connection;

[0100] S132. If the fixed lever fulcrum connection is not included, then divide the truck image into four sub-images along the midline of the width and the midline of the height of the truck image to be identified. Record the coordinates of these four sub-images relative to the original truck image in a clockwise direction, and use these four sub-images as the state detection image and the entity detection image.

[0101] S133. If the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, and record the coordinates of the fixed lever fulcrum connection relative to the original truck image. The image within the fixed lever fulcrum connection area is used as the image for state detection and entity detection.

[0102] S14. If a fixed lever is included:

[0103] S141 records the position coordinates of the fixed lever in the truck image; query the truck component positioning network detection results include the fixed lever fulcrum connection;

[0104] S142. If the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, and record the coordinates of the fixed lever fulcrum connection relative to the original truck image. The image within the fixed lever fulcrum connection area is used as the image for state detection and entity detection.

[0105] S143. If the fixed lever fulcrum connection is not included, the two images of the area where the fixed lever is located are used as images for state detection and entity detection;

[0106] S2. Detect the state detection image and entity detection image using the state detection network and entity detection network, respectively, to obtain detection results. Then, using the fault detection algorithm and the detection results, perform fixed lever-related fault detection on the truck image to be identified.

[0107] The output of the state detection network is the specific square box position coordinates of the connection status of each component in the current truck image state detection area to be identified, including normal connection state, abnormal connection state, tilt of the fixed lever fulcrum, loss of the fixed lever fulcrum, breakage of the fixed lever fulcrum seat, breakage of the fixed lever fulcrum chain shoe ring, breakage of the fixed lever fulcrum, horizontal breakage of the fixed lever fulcrum, loss of the round pin tied with wire, loss of the round pin hole, breakage of the round pin, loss of the cotter pin, protrusion of the connection component, loss of the rivet pin sleeve, and loss of the rivet.

[0108] The output of the entity detection network is the position coordinates of the entities on each component of the current truck image entity detection area to be identified, including the fixed lever fulcrum, the fixed lever fulcrum chain shoe ring, the fixed lever fulcrum seat, the inclination of the fixed lever fulcrum seat, the round pin, the rivet pin, the reflective rivet pin left after the rivet pin sleeve is lost, the connecting component has protruded, the cylinder left after the cotter pin is lost, the broken end of the round pin, the broken end of the fixed lever fulcrum seat, the broken end of the fixed lever fulcrum chain shoe ring, the broken end of the fixed lever fulcrum, the broken end of the component, the hole formed by the loss of the component, the iron wire, and the inclination of the fixed lever fulcrum.

[0109] It should be noted that the state detection network outputs the position coordinates of the component connection state. For example, the "fixed lever fulcrum tilted" in the state detection result is the position coordinates of the tilted fixed lever in the state detection image, while the entity detection network outputs the position coordinates of the specific component entity. For example, the "fixed lever fulcrum tilted" in the entity detection result detects the tilted fixed lever entity. In simple terms, one determines whether the fixed lever fulcrum area is tilted, and the other determines which part of the fixed lever fulcrum area is tilted. The present invention comprehensively utilizes the detection results output by the state detection network and the entity detection network to complete a detection process.

[0110] In some embodiments, the specific process of the fault detection in step S2 of "using the fault detection algorithm in combination with the detection results to perform fixed lever related fault detection on the current truck image to be identified" is as follows:

[0111] S21. Get the number of images of the state detection obtained in step S1;

[0112] S22. If the number of state detection images is equal to 4, the entire truck image to be identified alarm "fixed lever fulcrum lost";

[0113] S23. If the number of state detection images is equal to 2:

[0114] S231 determines whether the state detection image is a normal picture;

[0115] S232. If the image is normal, no fault detection is performed;

[0116] S233. If the image is not normal, perform rivet missing fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection;

[0117] S24. If the number of state detection images is equal to 1:

[0118] S241 determines whether the state detection image is a normal picture, the judgment process is the same as step S231;

[0119] S242. If the image is normal, no fault detection is performed;

[0120] S243. This is not a normal picture. Perform rivet missing fault detection, rivet pin sleeve missing fault detection, cotter pin missing fault detection, round pin missing fault detection, round pin breakage fault detection, entity breakage fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection.

[0121] Preferably, the specific process of "determining whether the state detection image is a normal picture" in step S231 is as follows:

[0122] A1. Obtain the number of state detection images;

[0123] A2. If the number of state detection images is equal to 4, it is not a normal image;

[0124] A3. If the number of state detection images is not equal to 4, traverse all state detection results of the state detection network:

[0125] A31. If the connection status is normal;

[0126] A311. Get the detection confidence of abnormal connection status and normal connection status;

[0127] A312. If, in addition to the abnormal connection state and the normal connection state, there are other abnormal connection states in the status detection results, it is not a normal image;

[0128] A313. Otherwise, if the detection confidence of the abnormal connection state is greater than the detection execution of the normal connection state, then it is not a normal image;

[0129] A32. If there is no normal connection status, it is not a normal image;

[0130] A4. If A2 and A3 do not determine it as an abnormal image, it is determined to be a normal image.

[0131] Preferably, the process of "not a normal picture, performing rivet missing fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection" in step S233 includes:

[0132] B. The specific process of rivet loss fault detection is as follows:

[0133] B1. If the status detection results show that a rivet is missing, an alarm will be issued: "Rivet missing on the fixed lever fulcrum seat"

[0134] B2. Otherwise, if the physical inspection results show a hole in the fixed lever fulcrum seat (a hole caused by a missing component), an alarm will be issued: "Rivets on the fixed lever fulcrum seat are missing."

[0135] C. The specific process of entity loss fault detection is as follows:

[0136] C1. If the status detection result shows that the fixed lever fulcrum is lost, the alarm "Fixed lever fulcrum lost" will be issued;

[0137] C2. Otherwise, if the physical detection result shows neither a fixed lever fulcrum nor a tilted fixed lever fulcrum, an alarm "Fixed lever fulcrum missing" is generated.

[0138] C3. If the fixed lever fulcrum seat is missing in the physical detection result, an alarm "Fixed lever fulcrum seat missing" will be issued;

[0139] C4. If the physical inspection results show a fixed lever fulcrum seat and a fixed lever fulcrum, but no fixed lever fulcrum chain shoe ring, and the current truck is a model with a chain shoe ring, an alarm "Fixed lever fulcrum chain shoe ring missing" will be triggered.

[0140] D. The specific process of detecting the tilt fault of the fixed lever fulcrum is as follows:

[0141] D1. Get the number of state detection images;

[0142] D2. If the number of state detection images is equal to 1:

[0143] D21. If the status detection results show that the fixed lever fulcrum is tilted, or the physical detection results show that a component is broken, an alarm "Fixed lever fulcrum broken" will be issued.

[0144] D22. If the physical inspection result shows no fixed lever fulcrum seat, and the truck is equipped with a chain shoe ring, the alarm "Fixed lever fulcrum chain shoe ring missing" will be triggered. If the truck is not equipped with a chain shoe ring, the alarm "Fixed lever fulcrum seat missing" will be triggered.

[0145] D23. If the current truck image does not detect a missing entity fault, broken entity fault, missing round pin fault, or missing rivet fault, a self-alarm will be issued based on the severity level of the fixed lever fault.

[0146] D3. If the number of state detection images is equal to 2:

[0147] D31. If the status detection result shows that the fixed lever fulcrum is tilted:

[0148] D311. If the status detection result shows that the fixed lever fulcrum is broken, the alarm "Fixed lever fulcrum broken" will be displayed;

[0149] D312. If the physical inspection results show a broken fixed lever fulcrum, an alarm "broken fixed lever fulcrum" will be issued;

[0150] D313. If there is a broken part in the physical inspection results, the alarm "Fixed lever fulcrum broken" will be issued;

[0151] D32. If the fixed lever fulcrum seat is broken in the status detection results, the alarm "Fixed lever fulcrum seat broken" will be issued;

[0152] D33. If the physical inspection results show a broken fixed lever fulcrum seat, an alarm will be issued: "Fixed lever fulcrum seat broken";

[0153] D34. If the status detection result shows that the chain shoe ring of the fixed lever fulcrum is broken, the alarm "Fixed lever fulcrum chain shoe ring broken" will be triggered;

[0154] D35. If the physical inspection results show a broken link in the chain shoe of the fixed lever fulcrum, the alarm "Broken link in the chain shoe of the fixed lever fulcrum" will be triggered.

[0155] D36. If the physical inspection results show a hole caused by a missing round pin, and the truck is equipped with a fixed lever fulcrum chain shoe ring, an alarm will be issued: "Fixed lever fulcrum chain shoe ring round pin missing." Otherwise, an alarm will be issued: "Fixed lever fulcrum seat round pin missing."

[0156] D37. If the physical inspection result shows no fixed lever fulcrum seat, and the truck is equipped with a fixed lever fulcrum chain shoe ring, an alarm is generated: "Fixed lever fulcrum chain shoe ring missing." Otherwise, an alarm is generated: "Fixed lever fulcrum seat missing."

[0157] D38. If, after the above fault detection, none of the truck image entity loss fault, entity breakage fault, or round pin loss fault has occurred, an alarm will be automatically issued according to the severity level of the fixed lever related fault.

[0158] Preferably, the process of "not being a normal picture, performing rivet missing fault detection, rivet pin sleeve missing fault detection, cotter pin missing fault detection, round pin missing fault detection, round pin breakage fault detection, entity breakage fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection" in step S243 includes:

[0159] E. The specific process of detecting the rivet pin sleeve missing fault is as follows:

[0160] E1. If the status detection results show that the rivet pin sleeve is missing or the connecting component is loose, an alarm will be issued based on the location of the area to which the status belongs: "Loss of rivet pin sleeve for fixed lever fulcrum" or "Loss of rivet pin sleeve for fixed lever fulcrum seat";

[0161] E2. If the physical inspection results show reflective rivet pins left behind by missing rivet pin sleeves, an alarm will be triggered based on the area where the rivet pin sleeve is missing or the rivet pin sleeve is missing for the fixed lever fulcrum chain shoe ring.

[0162] E3. If there is a hole on the fixed lever fulcrum seat in the physical inspection results, and the hole is on the left side of the fixed lever fulcrum seat, the alarm "Fixed lever fulcrum seat rivet pin sleeve is missing" will be displayed.

[0163] F. The specific process of detecting the cotter pin missing fault is as follows:

[0164] F1. If the status detection results show a missing cotter pin, a loose connection part, or a broken round pin, an alarm will be issued for "missing cotter pin at fixed lever fulcrum," "missing cotter pin at fixed lever fulcrum chain shoe ring," or "missing cotter pin at fixed lever fulcrum seat" based on the location of the status area.

[0165] F2. If the physical inspection results show a cylinder formed by a missing round pin, an alarm will be triggered based on the area where the cylinder is located: "Fixed lever fulcrum cotter pin missing", "Fixed lever fulcrum chain shoe ring cotter pin missing", and "Fixed lever fulcrum seat cotter pin missing".

[0166] G. The specific process of round pin missing fault detection is as follows:

[0167] G1. If a wire is tied to a round pin in the status detection results, an alarm will be issued based on the location of the area to which the status belongs: "Fixed lever fulcrum round pin lost", "Fixed lever fulcrum chain shoe ring round pin lost", or "Fixed lever fulcrum seat round pin lost";

[0168] G2. If the status detection results show both a hole caused by a missing round pin and a tilted fixed lever fulcrum, an alarm will be issued: "Fixed lever fulcrum chain shoe ring round pin missing" or "Fixed lever fulcrum seat round pin missing";

[0169] G3. If there is iron wire in the physical detection result, the "fixed lever fulcrum round pin is missing", "fixed lever fulcrum chain shoe ring round pin is missing", and "fixed lever fulcrum seat round pin is missing" alarms will be issued according to the location of the area to which they belong.

[0170] H. The specific process of round pin breakage fault detection is as follows:

[0171] H1. If a broken round pin is found in the status detection results, an alarm will be issued for "fixed lever fulcrum round pin broken", "fixed lever fulcrum chain shoe ring round pin broken", or "fixed lever fulcrum seat round pin broken" according to the location of the status area.

[0172] H2. If there are broken round pins in the physical detection results, the "Fixed Lever Pivot Pin Broken", "Fixed Lever Pivot Chain Shoe Ring Pin Broken", and "Fixed Lever Pivot Seat Pin Broken" alarms will be triggered based on the location of the area to which they belong.

[0173] I. The specific process of entity breakage fault detection is as follows:

[0174] I1. If the status detection result shows that the fixed lever fulcrum is broken horizontally, an alarm "Fixed lever fulcrum broken" will be issued;

[0175] I2. If the status detection result shows that the fixed lever fulcrum is broken, the alarm "Fixed lever fulcrum broken" will be displayed;

[0176] I3. If the physical inspection results show a fracture caused by a broken fixed lever fulcrum, an alarm "Fixed lever fulcrum broken" will be issued;

[0177] I4. If the status detection results show that the fixed lever fulcrum chain shoe ring is broken, the alarm "Fixed lever fulcrum chain shoe ring broken" will be triggered;

[0178] I5. If the status detection result shows that the fixed lever fulcrum seat is broken, an alarm "Fixed lever fulcrum seat broken" will be issued.

[0179] C'. The specific process of entity loss fault detection is the same as step C in step S233.

[0180] D'. The specific process of detecting the tilt fault of the fixed lever fulcrum is the same as step D in step S233.

[0181] Furthermore, the process of obtaining the location of the state region and the location of the entity region is as follows:

[0182] J1. According to the positioning results of the bolster, brake beam, fixed lever, and wheel axle of the component positioning network, the area between the bolster and the brake beam is As the area of ​​the fixed lever fulcrum seat, the area between the brake beam and the fixed lever As the area where the chain shoe ring belongs to the fixed lever fulcrum, the area between the fixed lever and the wheel axle is fixed The area where the fixed lever fulcrum belongs;

[0183] J2. Calculate the position coordinates of the state square box, the position coordinates of the entity square box and The IOU with the larger IOU value is determined as the state and the regional location to which the entity belongs. The IOU calculation formula is as follows:

[0184]

[0185] in is the area enclosed by the state and entity square frame position coordinates, for Regional area.

[0186] S3. Filter the detected fault shape and generate the square frame coordinates of the location of the fault shape in the truck image.

[0187] In some embodiments, the specific process of "filtering the detected fault form to generate the square coordinates of the position of the fault form in the truck image" in step S3 is as follows:

[0188] K1. Get all current alarm results;

[0189] K2. If both component loss and component breakage occur simultaneously, delete all component loss alarms;

[0190] K3. If both pin loss and pin breakage occur at the same time, delete all pin breakage alarms.

[0191] K4. If both the rivet pin sleeve and the cotter pin are lost, delete all rivet pin sleeve loss alarms.

[0192] K5. If both round pin and cotter pin are lost at the same time, delete all cotter pin loss alarms;

[0193] K6. Delete all fault alarms that are not currently detected for the truck according to the pre-defined configuration file;

[0194] K7. Traverse the remaining alarm results and write the coordinates of the square box where the fault form is located in the alarm information into the specified file path.

[0195] In this implementation, the truck component localization network, state detection network, and entity detection network all employ the FasterRCNN architecture. The image input sizes are (1024, 1024), (800, 256), and (256, 256), corresponding to the truck image to be identified, the state detection image region, and the entity detection image region, respectively. The detection results of the state detection network and the entity detection network are expressed in the form of [x1, y1, x2, y2, confidence], where x1, y1, x2, y2 represent the coordinates of the current detection result in the detection image, and the confidence value is the probability that the network considers the detection result at that coordinate. A larger probability indicates a higher probability that the coordinate region is the current detection result. As the input image scale decreases, the image features gradually refine from overall truck component features, component connection state features, to component entity features.

[0196] All networks are trained on the corresponding training datasets.

[0197] Specific implementation method 2: Combination Figure 3 To explain this embodiment,

[0198] This embodiment is a fixed lever fault detection system based on multi-scale feature cascade detection, which is actually a system corresponding to the fixed lever fault detection method based on multi-scale feature cascade detection.

[0199] The system described in this embodiment includes a truck component positioning module, a component connection status detection module, a small component entity detection module, a fault detection module, and a fault message generation module;

[0200] The truck component positioning module obtains the specific location coordinates of the bolster, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, middle tie rod, fixed lever fracture, and mobile lever fracture in the truck image to be identified through the positioning network;

[0201] The component connection status detection module detects the status of the fixed lever fulcrum connection and obtains the normal connection status, abnormal connection status, fixed lever fulcrum tilt, fixed lever fulcrum loss, fixed lever fulcrum seat breakage, fixed lever fulcrum chain shoe ring breakage, fixed lever fulcrum breakage, fixed lever fulcrum horizontal breakage, round pin loss and wire binding, round pin loss hole, round pin breakage, cotter pin loss, connection component leakage, rivet pin sleeve loss, and rivet loss connection status;

[0202] The small component entity detection module is responsible for detecting the fixed lever fulcrum, the fixed lever fulcrum chain shoe ring, the fixed lever fulcrum seat, the fixed lever fulcrum seat tilt, the round pin, the rivet pin, the reflective rivet pin left after the rivet pin sleeve is lost, the connecting component is ejected, the cylinder left after the cotter pin is lost, the broken round pin, the broken fixed lever fulcrum seat, the broken fixed lever fulcrum chain shoe ring, the broken fixed lever fulcrum, the broken component, the hole formed by the loss of the component, the iron wire, and the fixed lever fulcrum tilt entity contained in the fixed lever fulcrum connection;

[0203] The fault detection module integrates the detection results of the truck component positioning module, the component connection status detection module, and the small component entity detection module to identify the fault form occurring in the current truck image;

[0204] The fault message generation module generates the square frame position coordinates of the fault form in the current truck image and filters out unreasonable fault forms.

[0205] The above examples are merely illustrative of the calculation model and process of the present invention and are not intended to limit the embodiments of the present invention. Persons skilled in the art will readily appreciate that other variations or modifications based on the above description are possible. This list of embodiments is not exhaustive; however, any obvious variations or modifications derived from the technical solution of the present invention remain within the scope of protection of the present invention.

Claims

1. A fixed lever fault detection method based on multi-scale feature cascade detection, characterized in that: The following steps are involved: S1. Using the truck component positioning network to obtain the specific location coordinates of each component in the truck image to be identified, and then obtain the image area to be tested for state detection and entity detection in the image; The output of the truck component positioning network is the specific square frame position coordinates of the truck component in the current truck image to be identified. The truck components include the bolster, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, center tie rod, fixed lever fracture, and mobile lever fracture components. S2. Use the state detection network and the entity detection network to detect the state detection image and the entity detection image respectively to obtain the detection results; the output of the state detection network is the specific square box position coordinates of the connection status of each component in the state detection area of ​​the current truck image to be identified; the output of the entity detection network is the position coordinates of the entities on each component in the entity detection area of ​​the current truck image to be identified; based on the detection results of the state detection network and the entity detection network, the overall semantics of the fixed lever related components are used to perform fixed lever fault detection on the current truck image to be identified.

2. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 1, characterized in that: The process of "obtaining an image to be subjected to state detection and entity detection in an image by a component extraction algorithm" in step S1 includes the following steps: Check whether the detection results of the truck parts positioning network include fixed levers; If the fixed lever is not included: query whether the detection results of the truck component positioning network include the fixed lever fulcrum connection; if the fixed lever fulcrum connection is not included, divide the truck image into four sub-images along the center line of the width and the center line of the height of the truck image to be identified, record the coordinates of these four sub-images relative to the original truck image in a clockwise direction, and use these four sub-images as state detection images and entity detection images; if the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, and record the coordinates of the fixed lever fulcrum connection relative to the original truck image, and use the image in the fixed lever fulcrum connection area as the state detection and entity detection image; If a fixed lever is included: record the position coordinates of the fixed lever in the truck image; query whether the detection results of the truck component positioning network include the fixed lever fulcrum connection; if the fixed lever fulcrum connection is included, record the position coordinates of the fixed lever fulcrum connection relative to the fixed lever, record the coordinates of the fixed lever fulcrum connection relative to the original truck image, and the image in the area of ​​the fixed lever fulcrum connection is used as the image for state detection and entity detection; if the fixed lever fulcrum connection is not included, the two images of the area where the fixed lever is located are used as the images for state detection and entity detection.

3. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 2, characterized in that: During the status detection network detection process, the connection status of each component includes normal connection status, abnormal connection status, fixed lever fulcrum tilt, fixed lever fulcrum loss, fixed lever fulcrum seat breakage, fixed lever fulcrum chain shoe ring breakage, fixed lever fulcrum breakage, fixed lever fulcrum horizontal breakage, round pin loss and wire binding, round pin loss hole, round pin breakage, cotter pin loss, connection component leakage, rivet pin sleeve loss, and rivet loss; During the entity detection network inspection process, the entities on each component include the fixed lever fulcrum, the fixed lever fulcrum chain shoe ring, the fixed lever fulcrum seat, the inclination of the fixed lever fulcrum seat, the round pin, the rivet pin, the reflective rivet pin left after the rivet pin sleeve is lost, the connecting component has protruded, the cylinder left after the cotter pin is lost, the broken end of the round pin, the broken end of the fixed lever fulcrum seat, the broken end of the fixed lever fulcrum chain shoe ring, the broken end of the fixed lever fulcrum, the broken end of the component, the hole formed by the loss of the component, the iron wire, and the inclination of the fixed lever fulcrum.

4. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 3, characterized in that: The specific process of using the overall semantics of the fixed lever related components to perform fixed lever fault detection on the current truck image to be identified in step S2 includes: S21, obtaining the number of images detected in the state obtained in step S1; S22. If the number of state detection images is equal to 4, the entire truck image to be identified is confirmed as "fixed lever fulcrum missing"; S23. If the number of state detection images is equal to 2, determine whether the state detection image is a normal image; if it is not a normal image, perform rivet missing fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection; S24. If the number of status detection images is equal to 1, determine whether the status detection image is a normal image; if it is not a normal image, perform rivet pin sleeve missing fault detection, cotter pin missing fault detection, round pin missing fault detection, round pin breakage fault detection, entity breakage fault detection, entity missing fault detection, and fixed lever fulcrum tilt fault detection.

5. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 4, characterized in that: The specific process of determining whether the state detection image is a normal image includes: A1. Obtain the number of state detection images; A2. If the number of state detection images is equal to 4, it is not a normal image; A3. If the number of state detection images is not equal to 4, traverse all state detection results of the state detection network: If there is a normal connection state, obtain the detection confidence of the abnormal connection state and the normal connection state; if in addition to the abnormal connection state and the normal connection state, there are other abnormal connection states in the state detection results, then it is not a normal image; otherwise, if the detection confidence of the abnormal connection state is greater than the detection execution degree of the normal connection state, then it is not a normal image; If there is no normal connection status, it is not a normal image; A4: If A2 and A3 do not determine it as an abnormal image, it is determined to be a normal image.

6. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 5, characterized in that: The process of detecting rivet loss fault, entity loss fault, and fixed lever fulcrum tilt fault in step S23 is as follows: B. The specific process of rivet loss fault detection includes: If the status test results show that the rivet is missing, it is confirmed as "Loss of rivet on fixed lever fulcrum seat"; otherwise, if the physical test results show that there is a hole on the fixed lever fulcrum seat, it is confirmed as "Loss of rivet on fixed lever fulcrum seat"; C. The specific process of entity loss fault detection includes: If the status detection result shows that the fixed lever fulcrum is missing, it is confirmed as "Fixed lever fulcrum missing"; otherwise, if the physical detection result shows neither a fixed lever fulcrum nor a fixed lever fulcrum tilt, it is confirmed as "Fixed lever fulcrum missing"; if the physical detection result shows no fixed lever fulcrum seat, it is confirmed as "Fixed lever fulcrum seat missing"; if the physical detection result shows a fixed lever fulcrum seat and a fixed lever fulcrum, but no fixed lever fulcrum chain shoe ring, and the current truck is a model with a chain shoe ring, it is confirmed as "Fixed lever fulcrum chain shoe ring missing"; D. The specific process of detecting the tilt fault of the fixed lever fulcrum includes: Get the number of state detection images; If the number of state detection images is equal to 1, continue the judgment: if the state detection result shows that the fixed lever fulcrum is tilted, and the physical detection result shows that the component is broken, it is confirmed as "fixed lever fulcrum broken"; if the physical detection result does not show the fixed lever fulcrum seat, and the current truck is a model with a chain shoe ring, it is confirmed as "fixed lever fulcrum chain shoe ring missing"; if the current truck is not a model with a chain shoe ring, it is confirmed as "fixed lever fulcrum seat missing"; If the number of state detection images is equal to 2, continue to judge: If the fixed lever fulcrum is tilted in the state detection result, further judgment is made: if the fixed lever fulcrum is broken in the state detection result, it is confirmed as "fixed lever fulcrum broken"; if the fixed lever fulcrum is broken in the physical detection result, it is confirmed as "fixed lever fulcrum broken"; if the component is broken in the physical detection result, it is confirmed as "fixed lever fulcrum broken"; If the fixed lever fulcrum seat is broken in the status detection results, it is confirmed as "fixed lever fulcrum seat broken"; If the physical inspection results show a broken fixed lever fulcrum seat, confirm it as "broken fixed lever fulcrum seat"; If the status detection result shows that the fixed lever fulcrum chain shoe ring is broken, it is confirmed as "fixed lever fulcrum chain shoe ring is broken"; If the physical inspection results show a broken link of the fixed lever fulcrum chain shoe, it is confirmed as "broken link of the fixed lever fulcrum chain shoe"; If the physical inspection result shows a hole caused by a missing round pin, and the truck is equipped with a fixed lever fulcrum chain shoe ring, the problem is confirmed as "Fixed lever fulcrum chain shoe ring missing round pin". Otherwise, the problem is confirmed as "Fixed lever fulcrum seat missing round pin". If there is no fixed lever fulcrum seat in the physical detection result, and the current truck is a model with a fixed lever fulcrum chain shoe ring, it is confirmed as "fixed lever fulcrum chain shoe ring is missing", otherwise, it is confirmed as "fixed lever fulcrum seat is missing".

7. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 6, characterized in that: The image in step S24 is not a normal image. The process of detecting the rivet pin sleeve missing fault, the split pin missing fault, the round pin missing fault, the round pin breakage fault, and the entity breakage fault includes: E. The specific process of detecting the rivet pin sleeve missing fault includes: If the status detection results show that the rivet pin sleeve is missing or the connecting component is out of place, confirm "Loss of fixed lever fulcrum rivet pin sleeve" or "Loss of fixed lever fulcrum seat rivet pin sleeve" according to the location of the status area; If the physical inspection results show reflective rivet pins left by missing rivet pin sleeves, confirm "Loss of fixed lever fulcrum chain shoe ring rivet pin sleeve" or "Loss of fixed lever fulcrum seat rivet pin sleeve" based on their location. If there is a hole on the fixed lever fulcrum seat in the physical inspection results, and the hole is on the left side of the fixed lever fulcrum seat, it is confirmed that the "fixed lever fulcrum seat rivet pin sleeve is missing"; F. The specific process of detecting the cotter pin missing fault includes: If the status detection results show a missing cotter pin, a protruding connecting component, or a broken round pin, confirm "missing cotter pin at fixed lever fulcrum," "missing cotter pin at fixed lever fulcrum chain shoe ring," or "missing cotter pin at fixed lever fulcrum seat" based on the location of the status area. If the physical inspection results show a cylinder formed by a missing round pin, confirm the "missing fixed lever fulcrum cotter pin", "missing fixed lever fulcrum chain shoe ring cotter pin", or "missing fixed lever fulcrum seat cotter pin" based on the area where the cylinder is located. G. The specific process of round pin missing fault detection includes: If there is a wire tied to the round pin in the status detection result, confirm "Fixed lever fulcrum round pin missing", "Fixed lever fulcrum chain shoe ring round pin missing", "Fixed lever fulcrum seat round pin missing" according to the location of the status area; If the status test results show both a hole caused by a missing round pin and a tilted fixed lever fulcrum, confirm that "Fixed lever fulcrum chain shoe ring round pin is missing" and "Fixed lever fulcrum seat round pin is missing"; If there is iron wire in the physical inspection results, confirm "Fixed Lever Pivot Pin Missing", "Fixed Lever Pivot Chain Shoe Ring Pin Missing", and "Fixed Lever Pivot Seat Pin Missing" according to the location of the area. H. The specific process of round pin breakage fault detection includes: If a round pin is broken in the status detection result, confirm "Fixed lever fulcrum round pin broken", "Fixed lever fulcrum chain shoe ring round pin broken", "Fixed lever fulcrum seat round pin broken" according to the location of the status area; If there are broken round pins in the physical inspection results, confirm "Fixed lever fulcrum round pin broken", "Fixed lever fulcrum chain shoe ring round pin broken", and "Fixed lever fulcrum seat round pin broken" according to the location of the area to which they belong; I. The specific process of entity breakage fault detection includes: If the fixed lever fulcrum is broken horizontally in the status detection results, confirm "Fixed lever fulcrum broken"; If the fixed lever fulcrum is broken in the status detection results, confirm "Fixed lever fulcrum broken"; If there is a fracture caused by the broken fixed lever fulcrum in the physical detection results, confirm "Fixed lever fulcrum broken"; If the status detection result shows that the fixed lever fulcrum chain shoe ring is broken, confirm "Fixed lever fulcrum chain shoe ring is broken"; If the status detection result shows that the fixed lever fulcrum seat is broken, confirm "Fixed lever fulcrum seat is broken".

8. The method for detecting fixed lever faults based on multi-scale feature cascade detection according to claim 7, characterized in that: The process of obtaining the location of the state region and the location of the entity region includes: According to the positioning results of the bolster, brake beam, fixed lever and wheel axle of the component positioning network, the area between the bolster and the brake beam is As the area of ​​the fixed lever fulcrum seat, the area between the brake beam and the fixed lever As the area where the chain shoe ring belongs to the fixed lever fulcrum, the area between the fixed lever and the wheel axle is fixed The area where the fixed lever fulcrum belongs; Calculate the position coordinates of the state square box, the position coordinates of the entity square box and The IOU with the larger IOU value is determined as the state and the regional location to which the entity belongs.

9. A method for detecting fixed lever faults based on multi-scale feature cascade detection according to any one of claims 1 to 8, characterized in that: The method further includes the step of filtering the detected fault form and generating the coordinates of a square frame of the position where the fault form is located in the truck image. The specific process includes: K1. Get all current confirmation results; K2. If both component loss and component breakage occur at the same time, delete all component loss confirmation results; K3. If both round pin loss and round pin breakage occur at the same time, delete all round pin breakage confirmation results; K4. If both the rivet pin sleeve and the cotter pin are lost at the same time, delete all the rivet pin sleeve loss confirmation results; K5. If both round pin and cotter pin are lost, delete all cotter pin loss confirmation results. K6. Delete all fault confirmation results that are not currently tested for the truck according to the predefined configuration file; K7. Traverse the remaining confirmation results and write the coordinates of the square box where the fault form is located in the confirmation result information into the specified file path.

10. A fixed lever fault detection system based on multi-scale feature cascade detection, characterized in that: The system comprises: Truck component positioning module: This module uses the positioning network to obtain the specific coordinates of the bolster, brake beam, wheel axle, fixed lever, mobile lever, fixed lever fulcrum connection, mobile lever fulcrum connection, center tie rod, fixed lever fracture, and mobile lever fracture in the truck image to be identified. Component connection status detection module: performs status detection on the fixed lever fulcrum connection to obtain normal connection status, abnormal connection status, fixed lever fulcrum tilt, fixed lever fulcrum loss, fixed lever fulcrum seat break, fixed lever fulcrum chain shoe ring break, fixed lever fulcrum break, fixed lever fulcrum horizontal break, round pin loss and wire binding, round pin loss hole, round pin break, cotter pin loss, connection component leakage, rivet pin sleeve loss, and rivet loss connection status; Small component entity detection module: responsible for detecting the fixed lever fulcrum, fixed lever fulcrum chain shoe ring, fixed lever fulcrum seat, fixed lever fulcrum seat tilt, round pin, rivet pin, reflective rivet pin left after the rivet pin sleeve is lost, connecting component ejection, cylinder left after the cotter pin is lost, round pin breakage, fixed lever fulcrum seat breakage, fixed lever fulcrum chain shoe ring breakage, fixed lever fulcrum breakage, component breakage, holes formed by component loss, wire, and fixed lever fulcrum tilt entity included in the fixed lever fulcrum connection; Fault detection module: Based on the detection results of the truck component positioning module, component connection status detection module, and small component entity detection module, the module uses the overall semantics of the fixed lever-related components to perform fixed lever fault detection on the current truck image to be identified.

Images