A method and system for monitoring fatigue risk of in-service steel crane girder

Abstract

The present application belongs to the technical field of building component crack identification and monitoring, and aims at the technical problem that it is difficult to monitor the fatigue crack of the in-service steel crane beam at present, and provides a fatigue risk monitoring method for the in-service steel crane beam, which comprises the following steps: S1, obtaining the data of the steel crane beam to be monitored and obtaining the classification result of the high-risk area of the steel crane beam; S2, obtaining the crack image; S3, obtaining the crack length estimation value at the current moment according to the obtained crack image, and obtaining the crack length trend graph; S4, obtaining the average development rate of the crack at each stage according to the crack length trend graph, and using the preset crack rate-risk classification deep neural network to classify the average development rate of the crack at each stage, and obtaining the risk level of the fatigue crack risk position of the steel crane beam; S5, when the risk level is the high risk level, the early warning information is generated. The present application can realize the accurate positioning of the high-risk position of the fatigue crack and the accurate risk assessment.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of building component crack identification and monitoring, and particularly relates to a method and system for monitoring fatigue risk of a steel crane beam in service. BACKGROUND

[0002] In an industrial building steel structure system, a steel crane beam, as a key load transfer component, is subjected to dynamic cyclic loading for a long time. Its service performance directly affects the structural safety and production continuity of the industrial plant. However, engineering practice shows that, due to the coupling of complex factors (including but not limited to: eccentric effect caused by asymmetric track arrangement, brake force of the overhead crane, geometric form difference of the crane beam, etc.), the crane beam system is prone to fatigue cracks in the stress concentration area. Such cracks have significant nonlinear propagation characteristics. When the crack propagation rate exceeds the critical threshold, it may lead to catastrophic fracture accidents, seriously threatening industrial safety production. Therefore, it is necessary to monitor the fatigue cracks. However, the current fatigue monitoring method has the following technical problems:

[0003] 1. Difficulty in predicting crack initiation position: The traditional specification formula is based on ideal boundary conditions, which is difficult to accurately represent the complex stress field formed by the coupling of multiple factors under actual working conditions. Research shows that, for crane beams with the same structure, the maximum principal stress region shifts significantly under different factor coupling, resulting in significant dispersion of the crack initiation position. The traditional stress checking method has a mismatch between the theoretical model and the actual working conditions.

[0004] 2. Limitations of manual inspection efficiency: Factors such as dust pollution in industrial environments, high-altitude operation restrictions, and equipment obstructions make manual inspection difficult and costly. Manual inspection has strong dependence on subjective experience and low crack quantification accuracy, making it difficult to meet the accuracy requirements for crack propagation rate monitoring.

[0005] 3. Bottlenecks of intelligent monitoring technology: Existing visual monitoring equipment has three technical obstacles: (1) dependence on external power supply leads to low deployment feasibility, as stable power cannot be provided in 90% of industrial sites; (2) battery endurance bottleneck, with an average continuous working time of less than 6 months for commercially available devices, which cannot meet the 5-year monitoring period required by ASTM E647; (3) false positive rate of image recognition algorithms under complex field noise is difficult to avoid, and is significantly affected by vibration noise, oil contamination, and other interferences, making it difficult to meet the use requirements in real service environment.

[0006] 4. Lack of safety evaluation system: The existing specification lacks a quantitative correlation model between crack propagation rate and structure residual life, and the false alarm rate of manual experience judgment is more than 40%, which cannot establish a grading warning mechanism based on fracture mechanics. SUMMARY

[0007] The purpose of the present application is to overcome the shortcomings of the prior art, provide a steel crane beam fatigue risk monitoring method and system in service, which can realize accurate positioning, long-term monitoring, risk accurate evaluation and noise interference elimination of high-risk parts of fatigue cracks.

[0008] To achieve the above purpose, the technical scheme adopted by the present application is:

[0009] A steel crane beam fatigue risk monitoring method in service, comprising the following steps:

[0010] S1, obtaining steel crane beam data to be monitored, and classifying the obtained crane beam data by using a pre-trained fatigue crack positioning deep neural network to obtain a high-risk area classification result of the steel crane beam;

[0011] S2, deploying a crack image acquisition terminal in the high-risk area according to the high-risk area classification result to obtain a crack image;

[0012] S3, obtaining a crack length estimation value at the current moment according to the obtained crack image, and summarizing the length estimation value data to obtain a crack length trend graph;

[0013] S4, correcting the crack length on the crack length trend graph, obtaining the average crack development rate of each stage by using the corrected crack length trend graph, and classifying the average crack development rate of each stage by using a pre-set crack rate-risk classification deep neural network to obtain the risk level of the fatigue crack risk part of the steel crane beam;

[0014] S5, when the risk level is a high risk level, generating an early warning information.

[0015] Preferably, in step S1, the pre-trained fatigue crack positioning deep neural network is obtained by the following method:

[0016] S11, establishing a steel crane beam fatigue cracking database;

[0017] S12, constructing a first deep neural network, training by using the data in the steel crane beam fatigue cracking database to obtain a fatigue crack positioning deep neural network model, then correcting the model to obtain the pre-trained fatigue crack positioning deep neural network.

[0018] Preferably, in step S12, the model correction method is as follows:

[0019] S121, obtaining prior knowledge in the field;

[0020] S122, correcting the model according to the prior knowledge in the field to obtain the pre-trained fatigue crack positioning deep neural network.

[0021] Preferably, in step S2, the crack image acquisition terminal comprises:

[0022] a main processor, configured to be responsible for scheduling and execution of terminal core business logic, configure image sensor parameters, perform crack image shooting tasks, perform preliminary processing on original images, and communicate with the micro control unit;

[0023] a communication module, configured to upload the collected high-definition crack images and equipment state information to a remote server;

[0024] an image sensor, configured to shoot high-definition crack images under the cooperation of the fill light;

[0025] a fill light, configured to provide a controllable light source for light compensation in dark or night environment;

[0026] a FLASH memory, configured to temporarily store high-definition crack images shot by the image sensor;

[0027] a micro control unit, configured to control the power on / off of each module and perform power-on self-test and task termination protocol.

[0028] Preferably, in step S4, the calculation formula of the average crack development rate of each stage is:

[0029]

[0030] wherein, L d (t1) is the final display length of the crack in the crack length trend graph of a certain stage, L d (t0) is the initial display length of a certain stage in the crack length trend graph, Δ s t is the total time interval of the stage.

[0031] Preferably, in step S4, the preset crack rate-risk classification deep neural network is constructed by the following method:

[0032] S41, a fatigue crack rate-risk database is established;

[0033] S42, a second deep neural network is constructed, and the data in the fatigue crack rate-risk database is used for training to obtain the preset crack rate-risk classification deep neural network.

[0034] Preferably, the first deep neural network and the second deep neural network have the same structure, and both include an input layer, a plurality of hidden layers, a plurality of corresponding Sigmoid activation layers and an output layer connected in sequence.

[0035] Preferably, in step S4, the display length of the crack length trend graph after correction at time t is:

[0036]

[0037] L d (t) is the display length of the crack length trend graph at time t, L c (t) is the identification length at time t.

[0038] The application also provides an in-service steel crane beam fatigue risk monitoring system, comprising:

[0039] A data acquisition module is configured to acquire crack images through a crack image acquisition terminal and send the acquired crack images to a data processing module;

[0040] The data processing module is configured to acquire a current crack length estimation value according to the acquired crack images and aggregate the length estimation value data to obtain a crack length trend graph.

[0041] The data post-processing module is configured to acquire the average crack development rate of each stage according to the crack length trend graph and perform risk classification on the average crack development rate of each stage by using a preset crack rate-risk classification deep neural network to obtain the risk level of the fatigue crack risk position of the steel crane beam, and generate an early warning information when the risk level is a high risk level.

[0042] Preferably, the in-service steel crane beam fatigue risk monitoring system further comprises:

[0043] A data display module is configured to display monitoring point distribution information, fatigue crack images, crack length identification results and early warning information.

[0044] An equipment control module is configured to adjust and control the shooting interval and the light intensity of the crack image acquisition terminal in real time.

[0045] A system management module is configured to manage users, roles, organizations and resource menus.

[0046] Compared with the prior art, the application has the following beneficial effects:

[0047] (1) Compared with the traditional manual experience judgment, the monitoring method proposed by the application can quickly and accurately predict the high-risk area of fatigue crack initiation by using the pre-trained fatigue crack positioning deep neural network in combination with the steel crane beam body data and the headstock data; and the dynamic classification and grading early warning of the fatigue crack risk are realized by using the crack rate-risk classification deep neural network in combination with the crack length trend graph to calculate the average development rate of each stage, so that the early warning accuracy of the application is greatly improved and the misjudgment rate is significantly reduced.

[0048] (2) The application fuses prior knowledge in the field, introduces prior knowledge in the field (the influence of track eccentricity on stress distribution) in model training, calibrates prediction probability through a correction function, and improves the applicability and accuracy of deep neural networks under actual working conditions;

[0049] (3) The crack image acquisition terminal provided by the application has dynamic power management and abnormal interruption processing mechanism, solves the problem of insufficient battery endurance of existing visual monitoring equipment relying on external power supply, and through software optimization, significantly shortens the device response time, further reduces the device running power consumption, and meets the long-term monitoring demand;

[0050] (4) The application can effectively exclude the interference of industrial environment noise on the crack recognition algorithm by correcting the crack length and determining the crack length data that does not conform to the statistical law as an outlier, so as to ensure the accuracy and reliability of the crack length data; since the correction method is based on statistical law and physical constraint, the false positive rate is significantly reduced;

[0051] (5) The fatigue risk monitoring system provided by the application integrates data acquisition, processing, post-processing, display, device control and management modules, realizes full-process automatic management from crack detection to risk assessment, and the device control module in the application can dynamically adjust the shooting interval and the intensity of the light supplement according to the light conditions and the risk level, so as to further optimize the monitoring effect. In summary, the application is significantly superior to the prior art in crack positioning, risk assessment, monitoring efficiency, anti-interference ability and system integration, and provides an efficient, accurate and reliable solution for fatigue risk monitoring of in-service steel crane beam. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 The plug-in plate type steel crane beam model provided by the application;

[0053] Figure 2 The plug-in plate type crane beam finite element model established by the application; (a) is a non-eccentric model; (b) is an eccentric model when the track eccentricity is 30mm;

[0054] Figure 3 It is the overall stress nephogram of working condition 1;

[0055] Figure 4 It is the stress nephogram at the support of working condition 1;

[0056] Figure 5 It is the overall stress nephogram of working condition 2;

[0057] Figure 6 It is the stress nephogram at the support of working condition 2;

[0058] Figure 7 It is the framework diagram of the in-service steel crane beam fatigue risk monitoring system provided by the application;

[0059] Figure 8 is a crack length curve and a crack development rate curve chart for a high-risk area;

[0060] Figure 9 is a crack growth trend chart in the present application;

[0061] Figure 10 is an actual application chart of the crack image acquisition terminal in the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described in detail below. Figures 1 to 10 The technical solutions in the embodiments of the present application will be described in detail below.

[0063] Embodiment 1

[0064] The embodiments of the present application provide a method for monitoring fatigue risk of a steel crane beam in service, comprising the following steps:

[0065] S1, obtaining steel crane beam data to be monitored, and classifying the obtained crane beam data by using a pre-trained fatigue crack positioning depth neural network to obtain a high-risk area classification result of the steel crane beam;

[0066] In the present embodiment, the obtained steel crane beam data to be monitored includes steel crane beam body data, crane data borne by the steel crane beam, and track eccentricity data of the steel crane beam. Based on the original design drawing of the steel crane beam, the steel crane beam body data and the crane data borne by the steel crane beam can be easily obtained. The track eccentricity data of the steel crane beam needs to be obtained through on-site detection, and the detection method is as follows:

[0067] The distance h from the upper flange edge to the web edge of the steel crane beam is measured on site by using a tape measure f 1, the distance h from the upper flange edge to the crane track edge f 2, consult the original design drawing to obtain the steel crane beam web width t w 1, the track width t w 2, then the track eccentricity value O e is:

[0068]

[0069] If the track eccentricity value O e is greater than the steel crane beam web width t w 1, it is judged that there is condition A: track eccentricity.

[0070] In this embodiment, the pre-trained fatigue crack positioning deep neural network is constructed by the following method:

[0071] S11, a steel crane beam fatigue cracking database is established;

[0072] Based on years of experience in crane beam detection projects, a large amount of steel crane beam fatigue cracking data is obtained, and the steel crane beam fatigue cracking data is taken as the output label. The specific division of each cracking part of the steel crane beam is as follows: upper flange area, support insertion end area, support sealing plate area, and lower flange midspan area. The body data and crane load data of the crane beam with fatigue cracking are obtained by consulting the corresponding design drawings. The input features are the body data of the steel crane beam and the crane data carried by the steel crane beam. The steel crane beam fatigue cracking database is established.

[0073] S12, a first deep neural network is constructed, which specifically adopts an existing neural network architecture. The embodiment does not involve improvement of the architecture. The first deep neural network is trained through the data in the steel crane beam fatigue cracking database to obtain a fatigue crack positioning deep neural network model. Then the fatigue crack positioning deep neural network model is corrected to obtain a pre-trained fatigue crack positioning deep neural network. The first deep neural network in the present application comprises an input layer, a plurality of hidden layers, a plurality of Sigmoid activation layers and an output layer connected in sequence. The hidden layers and the Sigmoid activation layers correspond one by one.

[0074] The present application utilizes prior knowledge in the field (effect of track eccentricity on plug-in plate type steel crane beam support area) to correct the fatigue crack positioning deep neural network model. The stress concentration position of the plug-in plate type steel crane beam support area is greatly affected by the track eccentricity. Under the action of the crane load, when there is no track eccentricity, the stress concentration position of the support is the sealing plate. When there is eccentricity, the stress concentration position of the support appears in the sealing plate and the plug-in plate. The eccentricity can cause the stress of the stress concentration position of the plug-in plate to increase by more than 30%. Therefore, the prior knowledge in the field is used to correct the prediction probability of the model to form a pre-trained fatigue crack positioning deep neural network.

[0075] In the embodiment of the present application, the method for correcting the model is as follows:

[0076] S121, the prior knowledge in the field is obtained, and the specific acquisition method is as follows:

[0077] The shape information of the plug-in plate type steel crane beam is obtained to obtain a steel crane beam model, as shown in Figure 1 Then a finite element model is established based on the steel crane beam model, and the model is divided into a no-track eccentricity model and a track eccentricity model, as shown in Figure 2As shown, the specific two models of comparative working conditions are set as shown in Table 1, then the finite element analysis is carried out on the two models, the stress results are obtained as shown in Table 2, and the changes of the stress concentration parts of the plug-in plate type steel crane beam support area under the two working conditions are compared, and the results are as shown in Figures 3 to 6 As shown.

[0078] Table 1 Specific setting of two models of comparative working conditions

[0079] Condition No eccentricity of track Eccentricity 1.5 times web thickness Condition 1 √ Condition 2 √

[0080] Table 2 Changes of stress concentration parts of plug-in plate type steel crane beam support area under two working conditions

[0081] Condition Stress at closure MPa Stress at insertion MPa Condition 1 144.3 86.7 Condition 2 158.9 135.6

[0082] As shown in Table 2, the existence of track eccentricity significantly increases the stress level at the sealing plate and the plug-in plate, and the stress at the sealing plate increases from 144.3 MPa to 158.9 MPa in working condition 2 (with track eccentricity) compared with working condition 1 (without track eccentricity), with an increase of about 10.1%; the stress at the plug-in plate increases from 86.7 MPa to 135.6 MPa, with an increase of 56.4%. And through Figures 3 to 6 It can be seen that under the action of the crown block load, when there is no track eccentricity, the stress concentration part of the support is the sealing plate, and when there is eccentricity, the stress concentration part of the support appears at the sealing plate and the plug-in plate. Through the above results, it is shown that the track eccentricity has a significant influence on the stress distribution of the steel crane beam support area, especially the stress sensitivity of the plug-in plate part is higher, and the stress increase is much larger than that of the sealing plate. The eccentricity can cause the stress at the stress concentration part of the plug-in plate to increase by more than 30%.

[0083] S122, according to the prior knowledge in the field, the model is modified to obtain a pre-trained fatigue crack positioning depth neural network.

[0084] The embodiment of the application calibrates the data probability of the model by modifying the model according to the prior knowledge in the field, so that the probability of positioning to the support area is increased when the condition A "track eccentricity" exists.

[0085] Specifically, by training a calibration function, inputting condition A and the original prediction probability p, and then outputting the modified probability:

[0086]

[0087] Wherein, delta is an increment set according to prior knowledge.

[0088] S2, according to the high-risk area classification result, deploying a crack image acquisition terminal in the high-risk area to obtain a crack image, as shown in Figure 10 As shown in

[0089] In this embodiment, the crack image acquisition terminal comprises:

[0090] a main processor, configured to be responsible for scheduling and execution of terminal core business logic, configure image sensor parameters, execute crack image shooting task, and perform preliminary processing on original images, and communicate with a micro control unit; specifically, the main processor in this embodiment is configured to:

[0091] (1) system control, running a simplified Linux system, responsible for scheduling and execution of terminal core business logic, including image acquisition, data processing, and communication coordination.

[0092] (2) image processing, configured to configure image sensor parameters, execute crack image shooting task, and perform preliminary processing on original images.

[0093] (3) collaborative management, communicating with the micro control unit, receiving instructions, and starting or closing high-power modules as needed.

[0094] a communication module, configured to upload the collected high-definition crack images and device state information to a remote server;

[0095] an image sensor, configured to shoot high-definition crack images under the cooperation of a fill light;

[0096] a fill light, configured to provide a controllable light source for light compensation in dark or night environment;

[0097] a FLASH memory, configured to temporarily store high-definition crack images shot by the image sensor;

[0098] a micro control unit, configured to control the power on / off of each module (such as activating / closing the main processor and the communication module in sequence), and perform power-on self-test (ADC detects power supply stability) and task termination protocol (deep sleep to 10 muA).

[0099] Further, the crack image acquisition terminal in the present application has the following innovative design functions:

[0100] 1. Low-power control unit MCU design method

[0101] (1) independent power management architecture

[0102] An STM32F030 micro control unit (MCU) physically isolated from the main processor (RV1106) and the communication module (4G) is adopted to realize hardware-level power control, and the MCU controls the working time of all hardware components to ensure that different functional modules are powered on at the correct and required time.

[0103] The MCU has:

[0104] a) Ultra-low sleep current: typical value ≤ 5 μA (including real-time clock (RTC) running power consumption);

[0105] b) Multi-wake-up source management: integrated RTC timing wake-up and GPIO external interrupt wake-up dual-channel mechanism;

[0106] (2) Dynamic power management strategy

[0107] Power-on self-test process:

[0108] a) The MCU is first started, and the power supply state is detected by the analog-to-digital converter (ADC) in cooperation with the power management unit (PMU). The ADC and PMU collect external power and battery voltage data and transmit them to the MCU;

[0109] b) After the voltage stability verification is passed, the PMU is activated in sequence → RV1106 → 4G module power supply;

[0110] Task termination protocol:

[0111] a) After the service period ends, the reverse power-off sequence (4G → RV1106 → PMU) is executed;

[0112] b) Enter deep sleep mode autonomously (standby current ≤ 10 μA), start RTC timing before sleep, and the MCU is started again after reaching the set wake-up time.

[0113] (3) Abnormal interrupt handling mechanism

[0114] Establish a communication state monitoring channel, when the following abnormal conditions are detected:

[0115] a) Network interruption duration ≥ 120 seconds;

[0116] b) Data flow overrun alarm;

[0117] c) Server connection failure number > 3 times;

[0118] Execute emergency handling program, as follows:

[0119] a) Forcefully terminate the current process and turn off the high-power module power supply (RV1106 and 4G module);

[0120] b) Automatically start the intermittent sleep mode (work / sleep cycle = 15 minutes), wake up every 15 minutes and detect whether the abnormal condition is resolved. If the abnormal condition disappears, proceed normally according to the established work flow.

[0121] 2. Software optimization method of main processor (RV1106)

[0122] In a conventional design, the main processor (RV1106), communication module (4G), image sensor, light supplement lamp and FLASH memory of the crack image acquisition terminal consume a large amount of electric energy, and 95% of the electric energy is consumed by these devices in a single shooting task.

[0123] Moreover, a set of simplified Linux systems runs in the main processor (RV1106), and it takes 30 seconds from starting the system kernel to loading the business program, which includes communication with the MCU, image sensor configuration, light supplement lamp starting, shooting business execution and image uploading through 4G. The 4G module takes about 20 seconds from starting to configuring the AT instruction and then completing network registration, and the time for completing network registration by the 4G module needs to be longer in an area with weak wireless signal due to the influence of the 4G base station signal.

[0124] Based on the above problems, the crack image acquisition terminal in the application optimizes the software inside it, and the software optimization method specifically includes startup process optimization and communication protocol optimization.

[0125] a) The image sensor parameter configuration and 4G network registration are executed synchronously;

[0126] b) The light supplement lamp preheating is completed in the data transmission stage;

[0127] c) The AT instruction pre-caching technology is adopted, so that the network registration time of the 4G module is shortened by 40% (from the original 20 seconds to 12 seconds).

[0128] S3, an estimated value of the crack length at the current moment is obtained based on the obtained crack image, the estimated value data of the crack length are summarized, and a crack length trend graph is obtained.

[0129] The estimated value of the crack length at the current moment in the embodiment is specifically obtained by the steel structure fatigue crack identification method disclosed in the application patent with the publication number CN117523470A.

[0130] S4, the crack length on the crack length trend graph is rectified, the average crack development rate of each stage is obtained by using the rectified crack length trend graph, and the risk classification of the average crack development rate of each stage is performed by using a preset crack rate-risk classification depth neural network, so that the risk level of the fatigue crack risk position of the steel crane beam is obtained.

[0131] In the embodiment of the application, the calculation formula of the average crack development rate of each stage is:

[0132]

[0133] L d (t1) is the final display length of the crack in a certain stage of the crack length trend graph, Ld (t0) is the initial display length of a certain stage in the crack length trend graph, Δ s t is the total time interval of this stage.

[0134] In the embodiment of the present application, the preset crack rate-risk classification deep neural network is constructed by the following method:

[0135] S41, a fatigue crack rate-risk database is established;

[0136] Using relevant project experience, the development of cracks in the upper flange area of the steel crane beam, the development of cracks in the support sealing plate area, and the development of cracks in the support insertion end area are tracked respectively. The lower flange area is a pure tension area, and once cracking occurs, the crane beam will be quickly damaged, so the crack risk level of the lower flange area is directly divided into high risk, and the development rate is not calculated. From the discovery of the crack, the crack length data changes with time are tracked, and the crack risk level is calculated as low risk. The crack length acquisition technology disclosed in the invention patent with publication number CN117523470A is used to record the crack length every day, calculate the crack development rate at each stage, and combine the crack location, crack length, and crack development rate to artificially divide the risk level corresponding to each rate, and then obtain the fatigue crack rate-risk database.

[0137] The characteristics of the established fatigue crack rate-risk database are the fatigue crack locations, lengths, and development rates of the steel crane beam, and the labels are the artificially labeled risk levels;

[0138] S42, a second deep neural network is constructed, which specifically uses an existing neural network architecture, and the present embodiment does not involve improvement of the architecture; the data in the fatigue crack rate-risk database is used to train the second deep neural network, and a preset crack rate-risk classification deep neural network is obtained. The second deep neural network in the present embodiment includes an input layer, a plurality of hidden layers, a plurality of corresponding Sigmoid activation layers, and an output layer connected in sequence.

[0139] In the present embodiment, the crack length on the crack length trend graph is corrected based on statistics, and the crack length data that does not conform to the statistical law is determined as an outlier to avoid producing unreasonable early warning information. The specific correction method is as follows:

[0140] (1) Basic definition

[0141] L c (t) - the identified length at time t;

[0142] Ave c (t-1) - the latest average value at time t, the length data L cNot involved in the calculation of the latest average value of this stage;

[0143] L d (t) - the display length of the crack length trend at time t;

[0144] L d (t1) - the final display length of the crack length trend in this stage;

[0145] L d (t0) - the initial display length of the crack length trend in this stage;

[0146] Δ(L) - length difference;

[0147] Δ s t - the total time interval of this stage;

[0148] S(t) ∈ {1, 2, 3,...} - the current stage;

[0149] T k - the kth turning point time;

[0150] I(t) - the set of times that meet the following conditions:

[0151] Condition 1: belongs to the current stage S(τ) = S(t);

[0152] Condition 2: the time is not later than the current point τ < t;

[0153] Condition 3: the data is accepted as a non-outlier;

[0154] v t - the crack growth rate per stage;

[0155] (2) Calculation of the average value per stage: Where:

[0156] I (t-1) = {τ | S(τ) = S(t), τ < t, L c (τ) is accepted as a non-outlier}, when the secondary data is not involved in the calculation of the latest average value, because the current point may be an outlier;

[0157] (3) Length difference calculation: Δ(L) = L c (t) - Ave c (t-1);

[0158] (4) Determine the starting point condition: the starting point of the stage is the first data point in the monitoring stage, which is manually confirmed and reviewed as the data reference.

[0159] (5) Determine the normal point condition: |Δ(L)| ≤ 10 ∩ Lc (t) ≥ L d (t-1);

[0160] The formula contains two conditions that must be met simultaneously: ① deviation condition: |Δ(L)|≤10, the absolute deviation of the current identified length from the current stage average is not more than 10 units, which is used to filter the significant abnormal values identified by the algorithm due to industrial site noise interference; ② monotonicity condition: L c (t) ≥ L d (t-1), the current identified length should be greater than or equal to the previous displayed length, ensuring that the crack length trend chart always presents a non-decreasing trend (the crack will not "shorten"); when the two conditions are met, the data is displayed in red in the length trend chart and participates in the trend chart connection, and when only the previous condition is met, the data is displayed in gray in the trend chart and does not participate in the trend chart connection;

[0161] (6) determined as unreasonable point condition: |Δ(L)|≤10∩L c (t) < L d (t-1), the unreasonable point participates in subsequent average value calculation.

[0162] The formula contains two conditions that must be met simultaneously: ① deviation condition: |Δ(L)|≤10, the absolute deviation of the current identified length from the current stage average is not more than 10 units, which is used to filter the significant abnormal values identified by the algorithm due to industrial site noise interference; ② monotonicity condition: L c (t) < L d (t-1), the current identified length should be less than the previous displayed length, and the crack length trend chart presents a decreasing trend (the crack "shortens"); the data is displayed in gray in the trend chart and does not participate in the trend chart connection.

[0163] (7) determined as outlier condition: Δ(L) < -10∩L c (t) < L d (t-1), the outlier does not participate in subsequent average value calculation.

[0164] The formula contains two conditions that must be met simultaneously: ① negative deviation condition: Δ(L) < -10, the current identified length is 10 units lower than the stage average, which means that the crack length suddenly abnormally decreases, and the data may be disturbed by noise; ② monotonicity condition: L c (t) < L d (t-1), the current identified length is less than the previous valid display value, which means that the crack "shortens", violating the physical law.

[0165] (8) determined as turning point condition: Δ(L) > 10∩L c (t) > L d (t-1);

[0166] The formula contains two conditions that must be met simultaneously: ① positive deviation condition: Δ(L) > 10, the current identified length exceeds the average value of the current stage by 10 units, which means that the crack length increases rapidly, and the crack may develop rapidly; ② monotonicity condition: L c (t) ≥ L d (t-1), the current identified length should be greater than or equal to the previous displayed length, to ensure that the crack length trend chart always shows a non-decreasing trend (the crack will not "shorten"); when the two conditions are met, the data point is determined as a turning point, which means that the previous stage ends, and the turning point data does not participate in the calculation of the average value of the previous stage and the development rate of the previous stage.

[0167] (9) The displayed length of the crack length trend chart at time t:

[0168]

[0169] (10) Stage transition function

[0170] S(t) = S(t-1) + 1 - when t meets the turning point condition

[0171] (11) Calculate the crack development rate of each stage

[0172]

[0173] S5, when the risk level is a high risk level, generate an early warning information. The crack length curve and the crack development rate curve of the high risk area are as shown in Figure 8 . Specifically, when the cumulative change of the crack average development rate of each stage is 5 times or the crack average development rate of the current stage enters a high risk level, an early warning information is generated.

[0174] The risk level of the fatigue crack risk position of the steel crane beam in the embodiment of the application is specifically divided as shown in the following table 3.

[0175] Table 3: Risk level division of fatigue crack risk position of steel crane beam

[0176]

[0177]

[0178] Example 2

[0179] As shown in Figure 7 , the embodiment of the application further provides a fatigue risk monitoring system for a steel crane beam in service, which specifically comprises:

[0180] The data acquisition module is used for acquiring a crack image through a crack image acquisition terminal and sending the acquired crack image to the data processing module; the data acquisition module in the embodiment of the application and the crack image acquisition terminal use an HTTP protocol to collect, transmit and store image data.

[0181] The data processing module is used for acquiring a crack length estimation value at a current moment according to the acquired crack image and collecting length estimation value data to obtain a crack length trend graph; the data processing module in the embodiment of the application is embedded with a fatigue crack recognition algorithm (for example, a crack recognition method disclosed in the invention patent with the publication number CN117523470A), and the crack image is subjected to the data processing module to obtain the crack length estimation value at the current moment.

[0182] The data post-processing module is used for performing a deviation correction on the crack length on the crack length trend graph, acquiring an average crack development rate of each stage by using the deviation-corrected crack length trend graph, performing risk classification on the average crack development rate of each stage by using a preset crack rate-risk classification deep neural network, obtaining a risk level of a fatigue crack risk position of the steel crane girder, and generating an early warning information when the risk level is a high risk level.

[0183] The in-service steel crane girder fatigue risk monitoring system provided in the embodiment of the application further comprises:

[0184] The data display module is used for displaying monitoring point distribution information, a fatigue crack image, a crack length recognition result and early warning information.

[0185] The device control module is used for automatically adjusting and controlling a shooting interval of the crack image acquisition terminal in real time according to a current monitoring point reminder information and a number of early warning information, realizing dynamic tracking of the risk; and is used for automatically adjusting a light supplement lamp intensity in real time according to a current monitoring point light and a steel structure paint film condition, avoiding overexposure from affecting crack recognition effect.

[0186] The system management module is used for managing users, roles, organizations and resource menus inside the system.

[0187] In the embodiment of the application, the preset crack rate-risk classification deep neural network is obtained by the following method:

[0188] Step 1, establishing a fatigue crack rate-risk database;

[0189] Step 2, constructing a second deep neural network, training by using data in the fatigue crack rate-risk database, and obtaining the preset crack rate-risk classification deep neural network.

[0190] The performance of the deviation correction method proposed by the data post-processing module in the embodiment 2 of the application is researched as follows

[0191] The crack images taken in time sequence are recognized by the algorithm of the data processing module, and the following recognition lengths are obtained. The data is corrected by the correction method, the crack development rate of each stage is calculated, and the results are shown in Table 4.

[0192] Table 4 crack development rate of each stage

[0193]

[0194]

[0195]

[0196] The crack growth trend chart is shown in Figure 9 , and it can be seen from Figure 9 and Table 4 that the correction method proposed in the embodiment of the application can well eliminate the algorithm recognition error caused by industrial environment noise and truly reflect the crack growth trend.

[0197] Although the embodiments of the application have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, replacements and variations can be made to these embodiments without departing from the principles and purposes of the application, and the scope of the application is defined by the claims and their equivalents.

Claims

1. A method of in-service steel crane girder fatigue risk monitoring, characterized by, The method comprises the following steps: S1, obtaining steel crane beam data to be monitored, and classifying the obtained crane beam data by using a pre-trained fatigue crack positioning deep neural network to obtain a high-risk area classification result of the steel crane beam; S2, deploying a crack image acquisition terminal in the high-risk area according to the high-risk area classification result to obtain a crack image; S3, obtaining a crack length estimation value at the current moment according to the obtained crack image, and collecting the length estimation value data to obtain a crack length trend graph; S4, performing rectification processing on the crack length on the crack length trend graph, obtaining an average crack development rate of each stage by using the rectified crack length trend graph, and performing risk classification on the average crack development rate of each stage by using a pre-set crack rate-risk classification deep neural network to obtain a risk level of a fatigue crack risk position of the steel crane beam; The display length of the rectified crack length trend graph at time t is: ； The normal point condition is: ; The unreasonable point condition is: ; The outlier condition is: ; The turning point condition is: ; wherein, is a display length of the crack length trend graph at time t, is a recognition length at time t; is a length difference value; is a display length of the crack length trend graph at time t-1; S5, when the risk level is a high risk level, generating a warning information.

2. The method of fatigue risk monitoring of in-service steel crane girder according to claim 1, characterized in that, In step S1, the pre-trained fatigue crack positioning deep neural network is obtained by the following method: S11, establishing a steel crane beam fatigue cracking database; S12, constructing a first deep neural network, training the first deep neural network by using data in the steel crane beam fatigue cracking database to obtain a fatigue crack positioning deep neural network model, and then correcting the model to obtain the pre-trained fatigue crack positioning deep neural network.

3. The method of fatigue risk monitoring of in-service steel crane girder according to claim 2, characterized in that, In step S12, the model correction method is as follows: S121, obtaining domain prior knowledge; S122, correcting the model according to the domain prior knowledge to obtain the pre-trained fatigue crack positioning deep neural network.

4. The method of in-service steel crane girder fatigue risk monitoring of claim 1, wherein, In step S2, the crack image acquisition terminal comprises: a main processor configured to schedule and execute terminal core business logic, configure image sensor parameters, perform crack image shooting tasks, perform preliminary processing on original images, and communicate with a micro control unit; a communication module configured to upload collected high-definition crack images and device state information to a remote server; an image sensor configured to shoot high-definition crack images under the cooperation of a fill light; a fill light configured to provide a controllable light source for light compensation in a dark or night environment; a FLASH memory configured to temporarily store high-definition crack images shot by the image sensor; a micro control unit configured to control the power on / off of each module and perform power-on self-test and task termination protocols.

5. The method of in-service steel crane girder fatigue risk monitoring of claim 1, wherein, In step S4, the calculation formula of the average crack development rate of each stage is: ； wherein, is the final displayed length of the crack at a stage in the crack length profile, is the initial displayed length of the crack at a stage in the crack length profile, is the total time interval for the stage.

6. The method of in-service steel crane girder fatigue risk monitoring of claim 2, wherein, In step S4, the pre-set crack rate-risk classification deep neural network is obtained by the following method: S41, establishing a fatigue crack rate-risk database; S42, constructing a second deep neural network, training the second deep neural network by using data in the fatigue crack rate-risk database to obtain the pre-set crack rate-risk classification deep neural network.

7. The method of in-service steel crane girder fatigue risk monitoring of claim 6, wherein, The first deep neural network and the second deep neural network have the same structure, and each comprises an input layer, a plurality of hidden layers, a plurality of corresponding Sigmoid activation layers and an output layer connected in sequence.

8. An in-service steel crane girder fatigue risk monitoring system, characterized by, The method for monitoring fatigue risk of in-service steel crane girder according to any one of claims 1-7 is implemented, comprising: a data acquisition module, configured to acquire a crack image by a crack image acquisition terminal and send the acquired crack image to a data processing module; the data processing module is used for obtaining the crack length estimation value at the current time according to the acquired crack image, and the length estimation value data is summarized to obtain the crack length trend chart; the data post-processing module is used for correcting the crack length on the crack length trend chart, obtaining the average development rate of the crack at each stage by using the corrected crack length trend chart, and classifying the risk of the average development rate of the crack at each stage by using the preset crack rate-risk classification deep neural network, obtaining the risk level of the fatigue crack risk position of the steel crane girder, and generating the early warning information when the risk level is the high risk level.

9. A fatigue risk monitoring system for in-service steel crane girder according to claim 8, characterized in that, Also includes: a data display module for displaying monitoring point distribution information, fatigue crack image, crack length identification result and early warning information; an equipment control module for real-time adjustment and control of the shooting interval and the light intensity of the crack image acquisition terminal; a system management module for managing users, roles, organizations and resource menus.

Images