Intelligent management system for maintenance and fault early warning of tower crane

Abstract

The invention relates to the technical field of crane fault early warning, and discloses a tower crane maintenance and fault early warning intelligent management system, which comprises a meteorological score calculation module for calculating to obtain a meteorological score; the aging score calculation module is used for calculating and obtaining an aging score; the data acquisition module is used for acquiring image data and vibration data; the load eccentric value calculation module is used for calculating to obtain a load eccentric value; the vibration feature extraction module is used for calculating to obtain a vibration feature value; the braking fault score calculation module is used for calculating to obtain a braking fault score; the dynamic threshold value generation module is used for obtaining a fault prediction score threshold value through a prediction model; the braking fault judgment module judges that the braking fault score exceeds the fault prediction score threshold value, and the fault risk exists; according to the method, the meteorological factors, the aging factors of the tower crane, the stress factors of the lifting hook and the vibration factors of multiple positions of the braking component are quantified, and the braking fault score and the dynamic threshold value are generated, so that the accuracy of fault early warning of the tower crane is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of crane fault warning, and more specifically, it relates to an intelligent management system for tower crane maintenance and fault warning. Background Art

[0002] Tower cranes are key mechanical equipment widely used in high-rise building construction, and their main function is to vertically transport building materials and equipment to designated positions. The fault monitoring of traditional tower cranes usually adopts the method of fixed threshold judgment. For example, the time duration from recording the gear position zero signal, the hoisting brake signal of the frequency converter, and the feedback signal of the brake to the actual completion of the hook brake is recorded, and it is judged that when the hook speed is not zero within the corresponding preset braking time duration, it indicates that the tower crane has a braking fault.

[0003] However, the fixed threshold judgment does not take into account the aging condition of the tower crane, and meteorological factors during operation, such as temperature, wind speed, and rainfall, etc., will all affect its braking performance, resulting in low accuracy of fault monitoring and lack of a fault warning mechanism. With the development of artificial intelligence, currently, the vibration data of the braking components are collected, and time-domain and frequency-domain analysis of the vibration data is carried out through wavelet transform combined with machine learning to achieve fault warning.

[0004] However, the above-mentioned scheme does not consider the associated influence of different positions of the braking components. The vibration data of a single position may be biased, which may lead to deviation in fault warning. In addition, the weight, shape, and hanging method of the object hung on the hook will also affect the force distribution of the hook, thereby affecting the accuracy of fault prediction. Summary of the Invention

[0005] The present invention provides an intelligent management system for tower crane maintenance and fault warning to solve the technical problems in the above background art.

[0006] The present invention provides an intelligent management system for tower crane maintenance and fault warning, including: A meteorological score calculation module, which is used to calculate a meteorological score based on meteorological data; The meteorological data includes: temperature, wind speed, and rainfall; An aging score calculation module, which is used to calculate an aging score based on the operation data of the tower crane; The operation data includes: cumulative working duration, average daily usage times, and historical fault times; A data acquisition module, which is used to simultaneously acquire the image data of the object and the vibration data at N preset points after the object is completely lifted by the tower crane; The vibration data is represented by the vibration acceleration at M time points; The load eccentricity value calculation module is used to calculate the horizontal deviation distance between the center of gravity of an object and the center of the hook of a tower crane based on the image data of the object, and mark it as the load eccentricity value; The vibration feature extraction module is used to extract vibration feature values from the vibration data at N preset points; The braking fault score calculation module is used to calculate the braking fault score based on the meteorological score, aging score, load eccentricity value, and vibration feature value; The dynamic threshold generation module is used to input the meteorological score, aging score, load eccentricity value, and vibration feature value into a prediction model to obtain a fault prediction score threshold; The braking fault judgment module is used to judge that if the braking fault score is greater than or equal to the fault prediction score threshold, it immediately notifies the tower crane to immediately lower the lifted object.

[0007] Furthermore, both the number N of preset points and the number M of time points of the vibration data are user-defined parameters.

[0008] Furthermore, the larger the value of the meteorological score, the worse the meteorological conditions. Among them, the meteorological score The calculation formula is as follows: ; Among them, T, Wind, and Rain respectively represent temperature, wind speed, and rainfall, , and respectively represent user-defined temperature reference value, wind speed reference value, and rainfall reference value, , and respectively represent user-defined first weight coefficient, second weight coefficient, and third weight coefficient, and , and The sum value of is 1.

[0009] Furthermore, the larger the value of the aging score, the greater the aging degree of the tower crane. Among them, the aging score The calculation formula is as follows: ; Among them, Work, Freq, and Break respectively represent the cumulative working hours, average daily usage times, and historical fault times, , and respectively represent user-defined fourth weight coefficient, fifth weight coefficient, and sixth weight coefficient, and , and The sum value of is 1, and e represents the natural constant.

[0010] Further, calculating the load eccentricity value includes the following steps: Step S201, identifying the center of gravity of the object and the hook center of the tower crane through the target detection model; Step S202, calculating the horizontal pixel difference between the center of gravity of the object and the hook center of the tower crane; Step S203, calculating the horizontal deviation distance between the center of gravity of the object and the hook center of the tower crane based on the horizontal pixel difference, and marking it as the load eccentricity value; Load eccentricity value The calculation formula is as follows: ; Where represents the horizontal pixel difference, Width represents the physical width of the camera sensor, represents the actual distance between the camera and the object, f represents the camera focal length, represents the number of pixels of the image data in the horizontal direction.

[0011] Further, extracting vibration feature values from the vibration data of N preset points includes the following steps: Step S301, taking the average value of the vibration accelerations at M time points of the vibration data of N preset points as the basic vibration data; Step S302, calculating the average value, standard deviation, root mean square value, peak value, waveform factor, impulse factor and margin factor of the basic vibration data; Among them, the waveform factor is the ratio of the peak value to the root mean square value, the impulse factor is the ratio of the peak value to the average value, and the margin factor is the ratio of the peak value to the standard deviation; Step S303, converting the basic vibration data into frequency domain data through fast Fourier transform, and calculating the main frequency, center frequency and spectral energy of the frequency domain data; Step S304, taking the peak value, waveform factor, impulse factor and margin factor of the basic vibration data and the main frequency, center frequency and spectral energy of the frequency domain data as vibration feature values.

[0012] Further, the larger the value of the braking fault score, the greater the braking fault risk. The braking fault score is obtained by weighted summation of the meteorological score, aging score, load eccentricity value and vibration feature value, and the corresponding weight coefficients are all custom parameters.

[0013] Further, the prediction model consists of a hidden layer and a classifier; The hidden layer is used to update the concatenated vector composed of the meteorological score, aging score, load eccentricity value and vibration feature value, and output a hidden vector. The dimension number of the hidden vector is a custom parameter; The hidden vector output by the classifier input hidden layer, and the category space of the classifier represents the fault prediction score threshold.

[0014] Further, the calculation formula of the hidden layer includes: ; where represents the hidden vector output by the hidden layer, H represents the concatenated vector input to the hidden layer, , and respectively represent the first weight matrix, the second weight matrix, and the third weight matrix, and are all user-defined hyperparameters, , and respectively represent the first bias vector, the second bias vector, and the third bias vector, and are all user-defined hyperparameters, e represents the natural constant, T represents the transpose operation, PReLU represents the PReLU activation function, and Swish represents the Swish activation function.

[0015] Further, constructing the sample label of the training sample of the prediction model includes the following steps: Step S401, calculate the braking fault score after the object is completely lifted by the tower crane; Step S402, after the tower crane completes the preset transfer distance, perform braking. If it is determined that there is a fault, gradually reduce the weight and return to step S401 to execute until no fault occurs, and record 90% of the corresponding braking fault score as the sample label of the training sample, otherwise enter step S403; Step S403, if it is determined that there is no fault, gradually increase the weight until a fault occurs, and record 90% of the corresponding braking fault score as the sample label of the training sample; Step S404, repeatedly execute step S401 to step S403 until U training samples are obtained; where the number U of training samples is a user-defined parameter.

[0016] The beneficial effects of the present invention are as follows: The present invention comprehensively analyzes meteorological factors, the aging factors of tower cranes, the stress factors of hooks, and the vibration factors at multiple positions of braking components, and quantifies the above factors to generate braking fault scores and prediction dynamic thresholds, thereby improving the accuracy of fault warning for tower cranes. Description of the Drawings

[0017] Figure 1 is a schematic diagram of the intelligent management system for tower crane maintenance and fault warning of the present invention; Figure 2 is a flowchart of calculating the load eccentricity value of the present invention; Figure 3It is the flowchart for the feature extraction of the present invention to obtain vibration eigenvalues; Figure 4 It is the flowchart for the sample labels of the training samples for constructing the prediction model of the present invention.

[0018] In the figure: meteorological score calculation module 101, aging score calculation module 102, data acquisition module 103, load eccentricity value calculation module 104, vibration feature extraction module 105, braking fault score calculation module 106, dynamic threshold generation module 107, braking fault judgment module 108. Detailed implementation manners

[0019] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in relation to some examples can also be combined in other examples.

[0020] It should be noted that unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by those of ordinary skill in the art to which the present invention pertains. The "first", "second", and similar terms used in one or more embodiments of the present invention do not denote any order, quantity, or importance, but are only used to distinguish different components. The terms such as "including" or "comprising" mean that the elements or items appearing before this word cover the elements or items listed after this word and their equivalents, without excluding other elements or items. The terms such as "connected" or "coupled" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms such as "upper", "lower", "left", "right", etc. are only used to represent relative positional relationships, and when the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0021] As Figures 1 to 4 shown, the intelligent management system for tower crane maintenance and fault warning includes: A meteorological score calculation module 101, which is used to calculate the meteorological score according to meteorological data; The meteorological data includes: temperature, wind speed, and rainfall; An aging score calculation module 102, which is used to calculate the aging score according to the operation data of the tower crane; The operation data includes: cumulative working hours, average daily usage times, and historical fault times; The data acquisition module 103 is used to collect the image data of the object and the vibration data of N preset points simultaneously after the object is completely lifted by the tower crane; The vibration data is represented by the vibration acceleration at M time points; The load eccentricity value calculation module 104 is used to calculate the horizontal deviation distance between the center of gravity of the object and the center of the hook of the tower crane based on the image data of the object, and mark it as the load eccentricity value; The vibration feature extraction module 105 is used to extract the vibration feature values from the vibration data of N preset points; The braking fault score calculation module 106 is used to calculate the braking fault score based on the meteorological score, aging score, load eccentricity value and vibration feature value; The dynamic threshold generation module 107 is used to input the meteorological score, aging score, load eccentricity value and vibration feature value into the prediction model to obtain the fault prediction score threshold; The braking fault judgment module 108 is used to judge that if the braking fault score is greater than or equal to the fault prediction score threshold, it immediately notifies the tower crane to immediately lower the lifted object.

[0022] In an embodiment of the present invention, both the number N of preset points and the number M of time points of the vibration data are user-defined parameters; for example, the preset points can be set on the brake housing (which can monitor the wear of the brake), the reducer housing (which can monitor the wear of the internal gears of the reducer), the hoisting motor mounting base (which can monitor the linkage vibration between the motor and the transmission shaft), etc. Multiple vibration sensors can be installed at each preset point to eliminate errors; the number M of time points of the vibration data is related to the acquisition duration and acquisition frequency, where both the acquisition duration and the acquisition frequency are user-defined parameters. For example, if the acquisition duration is set to 2 seconds and the acquisition frequency is set to 500 Hz, then the number M of time points of the vibration data = 2×500 = 1000.

[0023] In an embodiment of the present invention, the larger the value of the meteorological score, the worse the meteorological conditions, where the meteorological score The calculation formula is as follows: ; Where T, Wind and Rain respectively represent temperature, wind speed and rainfall, , and respectively represent the user-defined temperature reference value, wind speed reference value and rainfall reference value, , and respectively represent the user-defined first weight coefficient, second weight coefficient and third weight coefficient, and , and The sum value of is 1.

[0024] It should be noted that the calculation of the meteorological score does not involve dimensional calculation, and the default values of the custom temperature reference value, wind speed reference value, and rainfall reference value are 25°C, 5 m / s, and 10 mm / h respectively, and the default values of the custom first weight coefficient, second weight coefficient, and third weight coefficient are 0.2, 0.4, and 0.4 respectively.

[0025] In an embodiment of the present invention, the larger the value of the aging score, the greater the aging degree of the tower crane, where the aging score The calculation formula is as follows: ; where Work, Freq, and Break respectively represent the cumulative working hours, average daily usage times, and historical failure times, , and respectively represent the custom fourth weight coefficient, fifth weight coefficient, and sixth weight coefficient, and , and The sum value of is 1, and e represents the natural constant.

[0026] It should be noted that the calculation of the aging score does not involve dimensional calculation, and the default values of the custom fourth weight coefficient, fifth weight coefficient, and sixth weight coefficient are 0.2, 0.3, and 0.5 respectively.

[0027] In an embodiment of the present invention, as Figure 2 shown, calculating the load eccentricity value includes the following steps: Step S201, identifying the center of gravity of the object and the hook center of the tower crane through the target detection model; The target detection model can be YOLOv8, Mask R-CNN, etc. Training samples for training the target detection model are constructed through the annotation tool, and the annotation tool can be LabelImg, COCO Annotator, etc., which will not be elaborated here; Step S202, calculating the horizontal pixel difference between the center of gravity of the object and the hook center of the tower crane; That is, the difference between the pixel values corresponding to the center of gravity of the object and the hook center of the tower crane in the horizontal direction; Step S203, calculating the horizontal deviation distance between the center of gravity of the object and the hook center of the tower crane based on the horizontal pixel difference and marking it as the load eccentricity value; The load eccentricity value The calculation formula is as follows: ; where represents the horizontal pixel difference, Width represents the physical width of the camera sensor, represents the actual distance between the camera and the object, f represents the camera focal length, represents the number of pixels of the image data in the horizontal direction.

[0028] For example, the horizontal pixel difference is 100 pixels, the physical width of the camera sensor is 6.17 mm = 0.00617 m, the camera focal length = 5.1 mm = 0.0051 m, the actual distance between the camera and the object is 10 m, and the number of pixels of the image data in the horizontal direction (image width) is 1920 pixels. Then, according to the above calculation formula, the load eccentricity value obtained is approximately 63.01 cm ≈ 0.63 m.

[0029] In an embodiment of the present invention, as Figure 3 shown, feature extraction is performed on the vibration data of N preset points to obtain vibration characteristic values, including the following steps: Step S301, taking the average value of the vibration accelerations at M time points of the vibration data of N preset points as the basic vibration data; That is, the basic vibration data is represented by the average value of the vibration accelerations at M time points; Step S302, calculating the average value, standard deviation, root mean square value, peak value, waveform factor, impulse factor, and margin factor of the basic vibration data; Among them, the waveform factor is the ratio of the peak value to the root mean square value, the impulse factor is the ratio of the peak value to the average value, and the margin factor is the ratio of the peak value to the standard deviation; Step S303, converting the basic vibration data into frequency domain data through fast Fourier transform, and calculating the main frequency, center frequency, and spectral energy of the frequency domain data; The frequency domain data is represented by the amplitude spectrum, that is, the amplitudes of different frequency components. Among them, the main frequency represents the frequency component corresponding to the largest amplitude, the center frequency represents the frequency component corresponding to the "center of gravity" of the amplitude spectrum, which is used to reflect the overall frequency distribution, and the spectral energy represents the sum of the amplitudes of all frequency components, which is used to reflect the vibration intensity; Step S304, taking the peak value, waveform factor, impulse factor, and margin factor of the basic vibration data, and the main frequency, center frequency, and spectral energy of the frequency domain data as the vibration characteristic values.

[0030] It should be noted that the calculation formula of the center frequency is as follows: , where represents the i-th frequency component, represents the amplitude of the i-th frequency component, and the calculation formula of the spectral energy E is as follows: , In addition, the spectral entropy, harmonic ratio, spectral bandwidth, etc. of the frequency-domain data can be calculated, which will not be elaborated here.

[0031] In one embodiment of the present invention, the larger the value of the braking fault score, the greater the braking fault risk. The braking fault score is obtained by calculating the weighted sum of the meteorological score, aging score, load eccentricity value, and vibration characteristic value, and the corresponding weight coefficients are all custom parameters, which will not be elaborated here.

[0032] In one embodiment of the present invention, the prediction model consists of a hidden layer and a classifier; The hidden layer is used to update the concatenated vector composed of the meteorological score, aging score, load eccentricity value, and vibration characteristic value, and output a hidden vector. The number of dimensions of the hidden vector is a custom parameter; That is, the number of dimensions of the concatenated vector is 10, which are the meteorological score, aging score, load eccentricity value, peak value of the basic vibration data, waveform factor, impulse factor, and margin factor, as well as the main frequency, center frequency, and spectral energy of the frequency-domain data; The classifier inputs the hidden vector output by the hidden layer, and the category space of the classifier represents the fault prediction score threshold.

[0033] In one embodiment of the present invention, the calculation formula of the hidden layer includes: ; Where represents the hidden vector output by the hidden layer, H represents the concatenated vector input by the hidden layer, , and represent the first weight matrix, the second weight matrix, and the third weight matrix respectively, and are all custom hyperparameters, , and represent the first bias vector, the second bias vector, and the third bias vector respectively, and are all custom hyperparameters. e represents the natural constant, T represents the transpose operation, PReLU represents the PReLU activation function, and Swish represents the Swish activation function.

[0034] It should be noted that the first weight matrix can be defined as 64×10 in size, then the dimension number of the first bias vector is 64, the second weight matrix can be defined as 32×64 in size, then the dimension number of the second bias vector is 32, and the third weight matrix can be defined as 16×32 in size, then the dimension number of the third bias vector is 16, that is, the dimension number of the hidden vector output by the final hidden layer is 16; in addition, compared with the fixed slope of Leaky ReLU, PReLU can adaptively adjust the slope, and Swish is smoother than ReLU, can retain the contribution of negative numbers, and improve the expression ability of the deep learning model. Similarly, the activation function of the classifier can also use the Swish activation function to map the hidden vector to the fault prediction score threshold, which will not be elaborated here.

[0035] In an embodiment of the present invention, as Figure 4 shown, constructing the sample label of the training sample of the prediction model includes the following steps: Step S401, calculate the braking fault score after the object is completely lifted by the tower crane; Step S402, after the tower crane completes the preset transfer distance, perform braking. If it is determined that there is a fault, gradually reduce the weight and return to step S401 to execute until no fault occurs, and record 90% of the corresponding braking fault score as the sample label of the training sample. Otherwise, enter step S403; Step S403, if it is determined that there is no fault, gradually increase the weight until a fault occurs, and record 90% of the corresponding braking fault score as the sample label of the training sample; Step S404, repeatedly execute steps S401 to S403 until U training samples are obtained; where the number U of training samples is a user-defined parameter, and the default value of the number U of training samples is 2000.

[0036] It should be noted that the sample label of the training sample is the fault prediction score threshold. During the training process of the prediction model, the mean square error between the value output by the prediction model at each iteration and the sample label of the training sample is specified as the loss function, and the parameters of the prediction model are updated backward using an adaptive gradient optimizer (such as Adam, AMSGrad, etc.) to minimize the loss value calculated by the loss function; in addition, through the prediction model, the dynamic threshold can be set, thereby improving the accuracy of braking fault prediction. And in order to reduce the experimental cost of constructing the training sample of the prediction model, the tower crane can be modeled using a simulation model, and the simulation model can be Adams, Simscape Multibody, etc., which will not be elaborated here.

[0037] It should be noted that the setting of the interval and the threshold value is for the convenience of comparison. The size of the threshold value depends on the amount of sample data and the base quantity set by those skilled in the art for each group of sample data, as long as it does not affect the proportional relationship between the parameters and the quantified values. And the above formulas are all calculations of taking the numerical values after dimensionless processing. The formulas are all obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0038] The above has described the embodiments of this example, but this example is not limited to the above specific implementation manners. The above specific implementation manners are only illustrative and not restrictive. Under the inspiration of this example, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this example.

Claims

1. The tower crane maintenance and fault warning intelligent management system is characterized by: include: A meteorological score calculation module, which is used to calculate the meteorological score based on the meteorological data; Meteorological data include: temperature, wind speed and rainfall; An aging score calculation module, which is used to calculate an aging score based on the operation data of the tower crane; Operation data includes: cumulative working hours, average daily usage times and historical failure times; A data acquisition module, which is used to simultaneously collect image data of the object and vibration data of N preset points after the object is fully lifted by the tower crane; The vibration data is represented by the vibration acceleration at M time points; A load eccentricity value calculation module, which is used to calculate the horizontal deviation distance between the center of gravity of the object and the hook center of the tower crane according to the image data of the object, and mark it as the load eccentricity value; A vibration feature extraction module is used to extract features from vibration data of N preset points to obtain vibration feature values; A brake fault score calculation module, which is used to calculate the brake fault score according to the meteorological score, the aging score, the load eccentricity value and the vibration characteristic value; A dynamic threshold generation module, which is used to input the meteorological score, aging score, load eccentricity value and vibration characteristic value into the prediction model to obtain the fault prediction score threshold; The brake fault judgment module is used to judge that the brake fault score is greater than or equal to the fault prediction score threshold, and then immediately notify the tower crane to immediately put down the hoisted object.

2. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The number of preset points N and the number of time points of vibration data M are both custom parameters.

3. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The larger the weather score value, the worse the weather conditions. The calculation formula is as follows: ； T, Wind and Rain represent temperature, wind speed and rainfall respectively. , and Respectively represent the custom temperature reference value, wind speed reference value and rainfall reference value, , and represent the first, second and third custom weight coefficients, respectively, and , and The sum of is 1.

4. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The larger the aging score value is, the greater the aging degree of the tower crane is. The calculation formula is as follows: ； Work, Freq and Break represent the cumulative working time, average daily usage times and historical failure times respectively. , and represent the fourth, fifth and sixth custom weight coefficients, respectively, and , and The sum of is 1, and e represents a natural constant.

5. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The calculation of load eccentricity value includes the following steps: Step S201, identifying the center of gravity of the object and the hook center of the tower crane through the target detection model; Step S202, calculating the horizontal pixel difference between the center of gravity of the object and the center of the hook of the tower crane; Step S203, calculating the horizontal deviation distance between the center of gravity of the object and the hook center of the tower crane according to the horizontal pixel difference, and marking it as the load eccentricity value; Load eccentricity The calculation formula is as follows: ; in represents the horizontal pixel difference, Width represents the physical width of the camera sensor, represents the real distance between the camera and the object, f represents the focal length of the camera, Indicates the number of pixels in the horizontal direction of the image data.

6. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: Feature extraction is performed on the vibration data of N preset points to obtain vibration feature values, including the following steps: Step S301, taking the average values ​​of the vibration accelerations at M time points of the vibration data of N preset points as basic vibration data; Step S302, calculating the average value, standard deviation, root mean square value, peak value, waveform factor, pulse factor and margin factor of the basic vibration data; The waveform factor is the ratio of the peak value to the RMS value, the pulse factor is the ratio of the peak value to the average value, and the margin factor is the ratio of the peak value to the standard deviation; Step S303, converting the basic vibration data into frequency domain data by fast Fourier transform, and calculating the main frequency, center frequency and spectrum energy of the frequency domain data; Step S304: taking the peak value, waveform factor, pulse factor and margin factor of the basic vibration data and the main frequency, center frequency and spectrum energy of the frequency domain data as vibration characteristic values.

7. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The larger the value of the brake fault score is, the greater the risk of brake failure is. The brake fault score is calculated by weighted summation of the meteorological score, aging score, load eccentricity value and vibration characteristic value. The corresponding weight coefficients are all custom parameters.

8. The tower crane maintenance and fault warning intelligent management system according to claim 1 is characterized in that: The prediction model consists of hidden layers and classifiers; The hidden layer is used to update the concatenated vector composed of the meteorological score, aging score, load eccentricity value and vibration eigenvalue, and output the hidden vector. The number of dimensions of the hidden vector is a custom parameter. The classifier inputs the hidden vector output by the hidden layer, and the category space of the classifier represents the fault prediction score threshold.

9. The tower crane maintenance and fault warning intelligent management system according to claim 8, characterized in that: The calculation formula for the hidden layer includes: ； in represents the hidden vector of the hidden layer output, H represents the concatenated vector of the hidden layer input, , and Respectively represent the first weight matrix, the second weight matrix and the third weight matrix, and they are all custom hyperparameters. , and They represent the first bias vector, the second bias vector and the third bias vector respectively, and they are all custom hyperparameters, e represents a natural constant, T represents a transpose operation, PReLU represents a PReLU activation function, and Swish represents a Swish activation function.

10. The tower crane maintenance and fault warning intelligent management system according to claim 1, characterized in that: Constructing sample labels for training samples of the prediction model includes the following steps: Step S401, calculating a brake fault score after the object is completely lifted by the tower crane; Step S402: After the tower crane has completed the preset transfer distance, it brakes. If it is judged that there is a fault, it gradually reduces the weight and returns to step S401 until there is no fault, and records 90% of the corresponding brake fault score as the sample label of the training sample, otherwise it goes to step S403; Step S403, if it is determined that there is no fault, the weight is gradually increased until a fault occurs, and 90% of the corresponding brake fault score is recorded as the sample label of the training sample; Step S404, repeating steps S401 to S403 until U training samples are obtained; The number of training samples U is a custom parameter.

Images