Safety intelligent monitoring system for portal crane

Abstract

The invention relates to the technical field of portal cranes, in particular to an intelligent safety monitoring system for a portal crane, which comprises a rigidity calculation module used for generating a real-time rigidity attenuation rate of a speed reducing mechanism from a vibration component, load distribution, burst energy and an impact counting rate through a rigidity calculation model based on a gear tooth collision theory; the crack prediction module is used for generating a prediction crack type and a crack degree grade according to the real-time rigidity of the speed reducing mechanism, the crack vibration component and a crack prediction model based on the improved CNN; and the safety management module is used for adjusting the main beam load distribution safety coefficient of the portal crane according to the predicted crack type and crack degree grade. According to the method, the real-time stiffness attenuation rate of multi-sensor fusion is calculated, the data processing amount of the improved CNN model is reduced, predictive monitoring of major faults of the speed reducing mechanism of the gantry crane is achieved, the safety of the gantry crane is remarkably improved, and the service life of the gantry crane is remarkably prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of gantry cranes, and particularly to a safety intelligent monitoring system for gantry cranes. Background Art

[0002] At present, according to the regulations of "General Bridge Cranes", the rated speed of the trolley running can reach up to 100 m / min at most, and in special cases, it can even exceed 112 m / min. With the increase of the trolley speed, the accident risk also increases accordingly.

[0003] In this process, the deceleration mechanism is responsible for controlling the lateral movement of the trolley carrying the goods on the main beam. The movement of the trolley will cause a change in the load distribution on the main beam. Therefore, the stable operation of the deceleration mechanism is crucial for the smoothness of the lifting process. Its failure will lead to the paralysis of the entire crane operation, and may even cause serious safety accidents, threatening the lives of the lifting operation personnel. Therefore, if the faults of key components can be detected in time during the operation of the reducer and timely and accurate maintenance measures are taken, the incidence of safety accidents can be greatly reduced, the downtime of mechanical equipment due to faults can be shortened, the normal operation of the equipment can be ensured, and the service life of the crane can be extended.

[0004] Therefore, how to realize the full-process closed-loop safety intelligent predictive monitoring of gantry cranes from signal acquisition to crack grading through multi-source data fusion is a technical problem to be solved at present. Summary of the Invention

[0005] For this reason, the present invention provides a safety intelligent monitoring system for gantry cranes. Through multi-sensor fusion, intelligent algorithms and dynamic safety regulation based on vibration sensors, torque sensors and acoustic emission sensors, especially by calculating the real-time stiffness attenuation rate of multi-sensor fusion, the data processing volume of the improved CNN model is reduced, and the predictive monitoring of major faults of the deceleration mechanism of gantry cranes is realized, significantly improving the safety and service life of gantry cranes.

[0006] To achieve the above object, the present invention proposes a safety intelligent monitoring system for gantry cranes. A plurality of vibration sensors, torque sensors and acoustic emission sensors are arranged on the gear pair of the deceleration mechanism of the gantry crane. The safety intelligent monitoring system includes:

[0007] A monitoring value processing module, configured to extract vibration components and crack vibration components from the vibration values monitored by the plurality of vibration sensors, extract the load distribution from the torque values monitored by the torque sensors, and extract burst energy, impact count rate and crack energy from the energy signals monitored by the acoustic emission sensors;

[0008] A stiffness calculation module for generating a real-time stiffness decay rate of a reduction mechanism by inputting the vibration components, load distribution, burst energy, and impact count rate into a stiffness calculation model based on gear tooth collision theory;

[0009] A crack prediction module for generating a predicted crack type and crack degree level by inputting the real-time stiffness of the reduction mechanism, crack vibration components into a crack prediction model based on an improved CNN;

[0010] A safety management module for adjusting the safety factor of the main girder load distribution of a gantry crane according to the predicted crack type and crack degree level.

[0011] Furthermore, the stiffness calculation module includes a theoretical stiffness calculation unit, a real-time stiffness calculation unit, and a decay rate calculation unit;

[0012] The theoretical stiffness calculation unit is used to calculate the theoretical stiffness based on a stiffness displacement decomposition model;

[0013] The real-time stiffness calculation unit is used to determine a displacement error, a load distribution correction factor, and a damping coefficient correction factor according to the vibration components, load distribution, burst energy, and impact count rate, and generate the real-time stiffness based on the displacement error, load distribution correction factor, and damping coefficient correction factor;

[0014] The decay rate calculation unit is used to calculate the real-time stiffness decay rate according to the theoretical stiffness and the real-time stiffness.

[0015] Furthermore, the vibration components include a bearing housing vibration component, a gear tooth meshing vibration component, and a gear tooth meshing rotational frequency component, the displacement error includes a bearing housing displacement error and a gear tooth meshing displacement error, and the real-time stiffness calculation unit includes a bearing housing displacement error correction sub-unit and a meshing displacement correction sub-unit;

[0016] The bearing housing displacement error correction sub-unit is used to calculate the actual bearing housing displacement error based on the time integral of the bearing housing vibration component and the bearing housing natural frequency, and calculate the bearing housing displacement error according to the actual bearing housing displacement error and the maximum displacement error;

[0017] The meshing displacement correction sub-unit is used to obtain the actual displacement error value by multiplying the ratio of the gear tooth meshing vibration component to the gear tooth meshing rotational frequency component by a first weight coefficient, and determine the gear tooth meshing displacement error according to the ratio of the actual displacement error value to the gear design backlash.

[0018] Furthermore, the real-time stiffness calculation unit further includes a meshing stiffness correction sub-unit;

[0019] The meshing stiffness correction sub-unit is used to determine the load distribution correction factor by multiplying the second weight coefficient and the single-tooth meshing stiffness based on the proportional relationship between the load distribution and the rated load distribution.

[0020] Furthermore, the real-time stiffness calculation unit further includes a burst energy correction term calculation sub-unit, an impact count rate correction term calculation sub-unit, and a weighted calculation sub-unit;

[0021] The burst energy correction term calculation sub-unit is used to calculate the first correction term by multiplying the ratio of the burst energy to the acoustic emission threshold by the third weight coefficient;

[0022] The impact count rate correction term calculation sub-unit is used to calculate the second correction term by multiplying the ratio of the impact count rate to the acoustic emission reference value by the fourth weight coefficient;

[0023] The weighted calculation sub-unit is connected to the burst energy correction term calculation sub-unit and the impact count rate correction term calculation sub-unit, and is used to calculate the damping coefficient correction factor based on the first correction term, the second correction term, and the meshing damping coefficient.

[0024] Furthermore, the theoretical stiffness calculation unit is used to calculate the theoretical stiffness based on Hooke's law and Hertz contact theory.

[0025] In the above solution, precise compensation of multi-source errors is achieved through fine decomposition of vibration components, the calculation accuracy of stiffness is improved through dynamic correction of displacement errors, the adaptability to complex working conditions is strengthened through the fusion of multiple weight coefficients, and through the hierarchical correction and dynamic weights of the first to fourth weights, it is more in line with the on-site situation under complex working conditions, improving the accuracy and reliability of the stiffness decay rate calculation.

[0026] Furthermore, the crack prediction model includes a data parallel input network, a multi-branch feature extraction network, a multi-scale feature fusion network, and a multi-task classification network;

[0027] The data parallel input network is used to sample and standardize the real-time stiffness decay rate, crack vibration components, and crack energy respectively, and generate stiffness decay time series features, vibration component time-frequency domain features, and crack energy statistical features;

[0028] The multi-branch feature extraction network is used to extract the short-term correlation of stiffness decay from the stiffness decay time series features to generate the first branch feature, extract the time domain features, frequency domain features, and weighted fusion after time-frequency domain branching from the vibration component time-frequency domain features to generate the second branch feature, and perform convolutional mapping on the crack energy statistical features to generate the third branch feature;

[0029] The multi-scale feature fusion network is used to generate fused features from the first branch feature, the second branch feature, and the third branch feature through feature splicing and SCSA;

[0030] The multi-task classification network is used to generate a predicted crack type from the fused features through a classification branch, and generate a crack degree level from the fused features through a regression branch.

[0031] Furthermore, the crack prediction module includes a crack type classification training unit;

[0032] The crack type classification training unit is used to optimize and train the parallel input network, the multi-branch feature extraction network, the multi-scale feature fusion network, and the classification branch through an improved focal loss function based on difficult-to-classify samples.

[0033] Furthermore, the crack prediction module further includes a crack degree level regression training unit;

[0034] The crack degree level regression training unit is used to optimize and train the parallel input network, the multi-branch feature extraction network, the multi-scale feature fusion network, and the regression branch through a regression loss function based on the number of degree levels and the Sigmoid function.

[0035] Furthermore, the safety factor is used to control the shortest distance of the trolley of the gantry crane from the end point, and the monitoring value processing module, the stiffness calculation module, the crack prediction module, and the safety management module are embedded in the industrial control computer for deployment.

[0036] In the above solution, through the data parallel input network, the multi-branch feature extraction network, the multi-scale feature fusion network, and the multi-task classification network, the processes of signal preprocessing, feature extraction, and multi-source feature fusion of the CNN model are carried out, realizing the multi-branch feature fusion and lightweight design of the improved CNN, solving the problems of insufficient feature utilization, high computational resource occupancy, and weak anti-noise ability existing in the traditional model in the crack prediction of gantry cranes, providing a core algorithm support with high precision and low latency for predictive maintenance, and overcoming the problems of class imbalance and feature similarity through the classification branch, the improved focal loss function, the regression branch, and the regression loss function, providing a reliable end-to-end learning framework for the safety monitoring of the reduction mechanism of gantry cranes.

[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0038] 1. Through multi-sensor fusion based on vibration sensors, torque sensors and acoustic emission sensors, intelligent algorithms and dynamic safety regulation, especially by calculating the real-time stiffness decay rate of multi-sensor fusion, the data processing volume of the improved CNN model is reduced, realizing the predictive monitoring of major faults in the reduction mechanism of gantry cranes, and significantly improving the safety and service life of gantry cranes.

[0039] 2. Through the fine decomposition of vibration components to achieve precise compensation of multi-source errors, through the dynamic correction of displacement errors to improve the stiffness calculation accuracy, through the fusion of multi-weight coefficients to strengthen the adaptability to complex working conditions, and through the hierarchical correction and dynamic weights of the first to fourth weights, it is more in line with the on-site situation under complex working conditions, improving the accuracy and reliability of the stiffness decay rate calculation.

[0040] 3. Through the data parallel input network, multi-branch feature extraction network, multi-scale feature fusion network and multi-task classification network, the processes of signal preprocessing, feature extraction and multi-source feature fusion of the CNN model are carried out, realizing the multi-branch feature fusion and lightweight design of the improved CNN, solving the problems of insufficient feature utilization, high computational resource occupancy and weak anti-noise ability existing in the traditional model for crack prediction in gantry cranes, providing a core algorithm support with high precision and low latency for predictive maintenance, and through the classification branch, improved focal loss function, regression branch and regression loss function, realizing the problem of overcoming class imbalance and feature similarity, providing a reliable end-to-end learning framework for the safety monitoring of the reduction mechanism of gantry cranes. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic structural diagram of the safety intelligent monitoring system for gantry cranes according to an embodiment of the present invention;

[0042] Figure 2 It is a schematic flow diagram of the safety intelligent monitoring system for gantry cranes according to an embodiment of the present invention;

[0043] Figure 3 It is a schematic flow diagram of calculating the real-time stiffness of the safety intelligent monitoring system for gantry cranes according to an embodiment of the present invention;

[0044] Figure 4 It is a schematic structural diagram of the crack prediction model based on the improved CNN of the subway vehicle state evaluation system based on trackside data fusion according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] In order to make the objectives and advantages of the present invention clearer and more understandable, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0046] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principle of the present invention and do not limit the protection scope of the present invention.

[0047] It should be noted that in the description of the present invention, the terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation of the present invention.

[0048] In addition, it should also be noted that in the description of the present invention, unless otherwise clearly specified and defined, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0049] As Figures 1 to 4 shown, the present invention provides a safety intelligent monitoring system for gantry cranes. Through multi-sensor fusion based on vibration sensors, torque sensors, and acoustic emission sensors, intelligent algorithms, and dynamic safety regulation, especially by calculating the real-time stiffness decay rate of multi-sensor fusion, the data processing volume of the improved CNN model is reduced, and predictive monitoring of major faults in the reduction mechanism of gantry cranes is achieved, significantly improving the safety and service life of gantry cranes.

[0050] As Figures 1 to 4 shown, this embodiment proposes a safety intelligent monitoring system for gantry cranes. A plurality of vibration sensors, torque sensors, and acoustic emission sensors are arranged on the gear pair of the reduction mechanism of the gantry crane. The safety intelligent monitoring system includes:

[0051] A monitoring value processing module for extracting vibration components and crack vibration components from the vibration values monitored by the plurality of vibration sensors, extracting load distributions from the torque values monitored by the torque sensors, and extracting burst energy, impact count rate, and crack energy from the energy signals monitored by the acoustic emission sensors;

[0052] A stiffness calculation module for generating the real-time stiffness decay rate of the reduction mechanism through a stiffness calculation model based on the tooth collision theory using the vibration components, load distributions, burst energy, and impact count rate;

[0053] A crack prediction module, which is used to generate predicted crack types and crack severity levels based on the real-time stiffness of the reduction mechanism, crack vibration components, and an improved CNN-based crack prediction model;

[0054] A safety management module, which is used to adjust the safety factor of the main girder load distribution of the gantry crane according to the predicted crack types and crack severity levels.

[0055] It should be noted that the reduction mechanism is used to control the movement of the trolley on the main girder.

[0056] It can be understood that the vibration sensor monitors the crack vibration components, combined with the torque sensor monitoring the load distribution, the acoustic emission sensor monitoring the burst energy and the impact count rate, can capture the early damage characteristics such as the micro-crack initiation of the gear pair and the bearing housing and the pitting of the tooth surface. Compared with the traditional single vibration / acoustic emission monitoring, the fault identification sensitivity is significantly improved. Furthermore, the complex working conditions are simplified into the real-time stiffness attenuation rate, and the multi-source data is fused again in the improved CNN through the real-time stiffness attenuation rate, reducing the experimental dependence on the predicted crack types and crack severity levels, improving the calculation efficiency of the model, and especially suitable for the safety monitoring of non-standard reduction mechanisms.

[0057] Such as Figures 1 to 3 As shown, further, the stiffness calculation module includes a theoretical stiffness calculation unit, a real-time stiffness calculation unit, and an attenuation rate calculation unit;

[0058] The theoretical stiffness calculation unit is used to calculate the theoretical stiffness based on the stiffness displacement decomposition model;

[0059] The real-time stiffness calculation unit is used to determine the displacement error, load distribution correction factor, and damping coefficient correction factor according to the vibration component, load distribution, burst energy, and impact count rate, and generate the real-time stiffness based on the displacement error, load distribution correction factor, and damping coefficient correction factor;

[0060] The attenuation rate calculation unit is used to calculate the real-time stiffness attenuation rate according to the theoretical stiffness and the real-time stiffness.

[0061] Specifically, the process of the attenuation rate calculation unit calculating the real-time stiffness attenuation rate is as follows:

[0062]

[0063] In the formula, η is the real-time stiffness attenuation rate, K t represents the real-time stiffness, and K0 represents the theoretical stiffness.

[0064] Further, the vibration components include the bearing housing vibration component, the gear tooth meshing vibration component, and the gear tooth meshing rotation frequency component, the displacement errors include the bearing housing displacement error and the gear tooth meshing displacement error, and the real-time stiffness calculation unit includes a bearing housing displacement error correction subunit and a meshing displacement correction subunit;

[0065] The bearing housing displacement error correction subunit is used to calculate the actual displacement error of the bearing housing based on the time integral of the bearing housing vibration component and the natural frequency of the bearing housing, and calculate the bearing housing displacement error according to the actual displacement error of the bearing housing and the maximum displacement error;

[0066] The meshing displacement correction subunit is used to obtain the actual value of the displacement error by multiplying the ratio of the gear tooth meshing vibration component to the gear tooth meshing rotation frequency component by the first weight coefficient, and determine the gear tooth meshing displacement error according to the ratio of the actual value of the displacement error to the designed backlash of the gear.

[0067] Specifically, the process of the bearing housing vibration component subunit calculating the bearing housing displacement error is as follows:

[0068]

[0069] In the formula, δ1, δ0, δ1′, and Δδ1 respectively represent the bearing housing displacement error, the bearing housing theoretical error, the maximum displacement error allowed by the bearing housing design (for example, the standard of gantry and bridge cranes stipulates that the maximum radial runout of the bearing housing is 0.1 mm), and the actual displacement error of the bearing housing deduced from the vibration signal, A vib (t) is the time variation function of the bearing housing vibration component, ω n is the natural frequency of the bearing housing, which is obtained through experiments or acquisition.

[0070] Specifically, the meshing displacement correction subunit is:

[0071]

[0072] In the formula, δ2, δ 02 , Δδ2, and δ3 respectively represent the gear tooth meshing displacement error, the gear tooth meshing displacement error theoretical value, the actual value of the displacement error corrected by extracting the meshing frequency component from the vibration spectrum, and the designed backlash of the gear (the standard tooth side clearance of the gantry and bridge crane reducer is 0.08 to 0.15 mm, preferably 0.1 mm), A f+kfr , A f respectively represent the kth-order sideband amplitude of the meshing frequency (the amplitude of the gear tooth meshing frequency plus the rotation frequency of kfr), the amplitude of the gear meshing frequency, λ k represents the first weight coefficient, that is, the modulation sensitivity coefficient, which is related to the tooth surface roughness and is calibrated through the finite element experiment of the coefficient correction unit.

[0073] Furthermore, the real-time stiffness calculation unit further includes a meshing stiffness correction subunit;

[0074] The meshing stiffness correction subunit is used to determine the load distribution correction factor by multiplying the ratio of the load distribution to the rated load distribution by the second weight coefficient and the single-tooth meshing stiffness.

[0075] Specifically, the meshing stiffness correction subunit is:

[0076]

[0077] In the formula, k i , k i ′ respectively represent the load distribution correction factor at the i-th position and the single-tooth meshing stiffness at the i-th position (determined according to the material and tooth profile of the tooth), T1 and T0 respectively represent the real-time load distribution and the rated load distribution, and α represents the second weight coefficient, that is, the load-stiffness sensitivity coefficient, which is calibrated through the finite element experiment of the coefficient correction unit.

[0078] Furthermore, the real-time stiffness calculation unit includes a sudden energy correction term calculation subunit, an impact count rate correction term calculation subunit, and a weighted calculation subunit;

[0079] The sudden energy correction term calculation subunit is used to calculate the first correction term by multiplying the ratio of the sudden energy to the acoustic emission threshold by the third weight coefficient;

[0080] The impact count rate correction term calculation subunit is used to calculate the second correction term by multiplying the ratio of the impact count rate to the acoustic emission reference value by the fourth weight coefficient;

[0081] The weighted calculation subunit is connected to the sudden energy correction term calculation subunit and the impact count rate correction term calculation subunit, and is used to calculate the damping coefficient correction factor based on the first correction term, the second correction term, and the meshing damping coefficient.

[0082] Specifically, the calculation process of the damping coefficient correction factor is:

[0083]

[0084] In the formula, c i , c i ′ respectively represent the damping coefficient correction factor at the i-th position and the meshing damping coefficient at the i-th position, E h , E t respectively represent the sudden energy and the acoustic emission threshold in the healthy state, N h , N0 respectively represent the impact count rate and the acoustic emission reference value in the healthy state, and γ, η respectively represent the third weight coefficient and the fourth weight coefficient.

[0085] Furthermore, the real-time stiffness calculation unit further includes a model integration subunit;

[0086] The model integration subunit is used to determine the real-time stiffness by multiplying the sum of the time derivatives of the bearing seat displacement error and the time derivative of the tooth engagement displacement error by the damping coefficient correction factor, and adding the product of the load distribution correction factor and the pressure angle coefficient.

[0087] Specifically, the calculation process of the real-time stiffness is as follows:

[0088]

[0089] In the formula, K t represents the real-time stiffness, δ1 represents the bearing seat displacement error, represents the time derivative of the bearing seat displacement error, δ2 represents the tooth engagement displacement error, represents the time derivative of the tooth engagement displacement error, c i represents the damping coefficient correction factor at the i-th position, k i represents the load distribution correction factor at the i-th position, cosβ i represents the pressure angle coefficient.

[0090] Furthermore, the theoretical stiffness calculation unit is used to calculate the theoretical stiffness based on Hooke's law and Hertz contact theory.

[0091] Specifically, the process of the theoretical stiffness calculation unit calculating the theoretical stiffness is as follows:

[0092] Formula 1:

[0093] Formula 2:

[0094] Formula 3:

[0095] In the formula, P represents the contact normal force, δ represents the distance by which the corresponding points of the object approach, R represents the comprehensive radius of curvature, E represents the comprehensive elastic modulus, and K represents the theoretical stiffness. Among them, Formula 1 and Formula 2 are the extensions of Hooke's law in the gear structure. It can be understood that within the elastic limit, stress is proportional to strain, and the relationship between the deformation parameter δ and the force P of the gear under specific contact problems is refined. Formula 3 is constructed based on the relationship between contact deformation and contact force in Hertz contact theory, and the contact state is determined through geometric and material characteristics.

[0096] In the above solution, precise compensation for multi-source errors is achieved through fine decomposition of vibration components, the calculation accuracy of stiffness is improved through dynamic correction of displacement errors, the adaptability to complex working conditions is enhanced through fusion of multiple weight coefficients, and through hierarchical correction and dynamic weights of the first to fourth weights, it is more in line with the on-site situation under complex working conditions, improving the accuracy and reliability of the calculation of the stiffness decay rate.

[0097] As Figure 4 shown, further, the crack prediction model includes a data parallel input network, a multi-branch feature extraction network, a multi-scale feature fusion network, and a multi-task classification network;

[0098] The data parallel input network is used to sample and standardize the real-time stiffness decay rate, crack vibration components, and crack energy respectively, generating stiffness decay time series features, vibration component time-frequency domain features, and crack energy statistical features;

[0099] The multi-branch feature extraction network is used to extract the short-term correlation of stiffness decay from the stiffness decay time series features, generating the first branch feature, extracting time domain features, frequency domain features, and weighted fusion after time-frequency domain branches from the vibration component time-frequency domain features, generating the second branch feature, and performing convolutional mapping on the crack energy statistical features, generating the third branch feature;

[0100] The multi-scale feature fusion network is used to generate fused features from the first branch feature, the second branch feature, and the third branch feature through feature splicing and SCSA;

[0101] The multi-task classification network is used to generate the predicted crack type from the fused features through the classification branch and generate the crack degree level from the fused features through the regression branch.

[0102] As Figure 4 shown, specifically, the parallel input network is as follows: Input feature dimension: For the real-time stiffness decay rate, 1D time series data is used, with a length of 256 points and a sampling rate of 1 kHz; for the crack vibration components, 3 channels are used: time domain waveform, frequency domain FFT amplitude, and wavelet time-frequency diagram, that is, the vibration component time-frequency domain features include vibration time domain features, vibration frequency domain features, and vibration time-frequency domain features; for the crack energy features, statistical values of acoustic emission burst energy and impact count rate are used, in a 2D matrix. Input processing: Normalization layer: Perform z-score standardization on the stiffness features and energy features, and perform amplitude normalization on the vibration features. Sliding window segmentation: For the time series stiffness data, overlapping windows are used to extract local features. The overlapping window preferably has a window length of 64 and an overlap rate of 50%.

[0103] As Figure 4As shown, specifically, the time-frequency domain characteristics of the vibration component are as follows: Branch 1 processes the time series characteristics of stiffness attenuation: extraction of the time series characteristics of stiffness attenuation, 1D depthwise separable convolutional layer (Depth wise Separable Conv1D), number of filters: 32, kernel size 5, stride 1, activation function ReLU, to reduce the number of parameters and capture the short-term correlation of stiffness changes.

[0104] As Figure 4 shown, Branch 2 processes the time series characteristics of stiffness attenuation, including: the time domain branch uses: Conv2D (convolution size 16@3×3, ReLU activation function), MaxPool2D (pooling size 2×2), Conv2D (convolution size 32@3×3, ReLU activation function); the frequency domain branch uses: Conv2D (convolution size 16@5×1, ReLU activation function), GlobalAveragePooling1D (for the FFT amplitude spectrum), and the time-frequency domain branch uses: Conv2D (convolution size 16@3×3, ReLU activation function) for the wavelet map, Inception module for the wavelet map; Channel Attention (SE Block) is used to weight and fuse the time domain branch, frequency domain branch, and time-frequency domain branch to suppress noise interference.

[0105] As Figure 4 shown, Branch 3 processes the crack energy statistical characteristics: maps the acoustic emission energy statistical values to a 32-dimensional vector through a fully connected embedding layer, and the energy statistical values are preferably the mean, variance, and peak value.

[0106] As Figure 4 shown, specifically, the multi-scale feature fusion network includes: cross-branch feature splicing: splicing the first branch feature of 64 dimensions, the second branch feature of 128 dimensions, and the third branch feature of 32 dimensions into a 224-dimensional vector. Spatial-Channel Co-Attention (SCSA module): focuses on the crack-sensitive area through the spatial attention map, and strengthens the key frequency band features through channel attention.

[0107] As Figure 4 shown, specifically, the classification branch of the multi-task classification network is: fully connected layer (FC-128), Dropout(0.5), Softmax function outputs 4 categories: fatigue / overload / corrosion / no crack.

[0108] As Figure 4 shown, specifically, the regression branch of the multi-task classification network is: fully connected layer (FC-64), Gaussian process layer: outputs the continuous degree value and discretizes it into levels I-IV.

[0109] Furthermore, the crack prediction module includes a crack type classification training unit;

[0110] The crack type classification training unit is used to optimize and train the parallel input network, multi-branch feature extraction network, multi-scale feature fusion network, and classification branch through an improved focal loss function based on difficult-to-classify samples.

[0111] Specifically, the improved focal loss function is as follows:

[0112]

[0113] In the formula, Loss class represents the improved focal loss function, c represents 4 types of crack types: fatigue / overload / corrosion / no crack, α c , y c , p c , γc respectively represent the class weight, true label encoding, predicted class probability, and focusing parameter. The focusing parameter is determined according to whether the crack type is a difficult-to-classify sample. For example, if the overload crack and the fatigue crack are easily confused, the focusing parameter is set to 2 to increase the loss contribution of difficult samples.

[0114] Furthermore, the crack prediction module further includes a crack degree level regression training unit;

[0115] The crack degree level regression training unit is used to optimize and train the parallel input network, multi-branch feature extraction network, multi-scale feature fusion network, and regression branch through a regression loss function based on the degree level number and the Sigmoid function.

[0116] Specifically, the regression loss function is as follows:

[0117]

[0118] In the formula, Loss degree represents the regression loss function, k represents 4 types of crack degree levels, y k , σ, s k respectively represent the binary indicator variable, the Sigmoid function, and the score of the model for the k-th level threshold.

[0119] Furthermore, the safety factor is used to control the shortest distance of the trolley of the gantry crane from the end point. The monitoring value processing module, stiffness calculation module, crack prediction module, and safety management module are embedded in the industrial control computer for deployment. The lightweight deployment of the improved CNN model with a volume less than 8MB is realized.

[0120] In the above solution, through the data parallel input network, multi-branch feature extraction network, multi-scale feature fusion network and multi-task classification network, the processes of signal preprocessing, feature extraction and multi-source feature fusion of the CNN model are carried out, realizing the multi-branch feature fusion and lightweight design of the improved CNN, solving the problems of insufficient feature utilization, high computational resource occupancy and weak anti-noise ability existing in the traditional model for crack prediction of gantry cranes, providing a core algorithm support with high precision and low latency for predictive maintenance, and overcoming the problems of class imbalance and feature similarity through the classification branch, improved focal loss function, regression branch and regression loss function, providing a reliable end-to-end learning framework for the safety monitoring of the reduction mechanism of gantry cranes.

[0121] In this embodiment, through multi-sensor fusion, intelligent algorithms and dynamic safety regulation based on vibration sensors, torque sensors and acoustic emission sensors, especially by calculating the real-time stiffness decay rate of multi-sensor fusion, the data processing volume of the improved CNN model is reduced, realizing the predictive monitoring of major faults of the reduction mechanism of gantry cranes, and significantly improving the safety and service life of gantry cranes. Precise compensation of multi-source errors is achieved through fine decomposition of vibration components, the calculation accuracy of stiffness is improved through dynamic correction of displacement errors, the adaptability to complex working conditions is strengthened through fusion of multiple weight coefficients, and through hierarchical correction and dynamic weights of the first to fourth weights, it is more in line with the on-site situation under complex working conditions, improving the accuracy and reliability of the calculation of the stiffness decay rate. Through the data parallel input network, multi-branch feature extraction network, multi-scale feature fusion network and multi-task classification network, the processes of signal preprocessing, feature extraction and multi-source feature fusion of the CNN model are carried out, realizing the multi-branch feature fusion and lightweight design of the improved CNN, solving the problems of insufficient feature utilization, high computational resource occupancy and weak anti-noise ability existing in the traditional model for crack prediction of gantry cranes, providing a core algorithm support with high precision and low latency for predictive maintenance, and overcoming the problems of class imbalance and feature similarity through the classification branch, improved focal loss function, regression branch and regression loss function, providing a reliable end-to-end learning framework for the safety monitoring of the reduction mechanism of gantry cranes.

[0122] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

[0123] The foregoing are only preferred embodiments of the present invention and are not intended to limit the present invention; for those skilled in the art, various modifications and variations can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A safety intelligent monitoring system for gantry cranes, characterized in that, A gear pair of a reduction mechanism of a bridge and gantry crane is provided with a plurality of vibration sensors, torque sensors and acoustic emission sensors, and the safety intelligent monitoring system includes: A monitoring value processing module, configured to extract a vibration component and a crack vibration component from vibration values monitored by the plurality of vibration sensors, extract a load distribution from a torque value monitored by the torque sensor, and extract a burst energy, an impact count rate and a crack energy from an energy signal monitored by the acoustic emission sensor; A stiffness calculation module, configured to generate a real-time stiffness attenuation rate of the reduction mechanism through a stiffness calculation model based on the tooth collision theory by using the vibration component, the load distribution, the burst energy and the impact count rate; A crack prediction module, configured to generate a predicted crack type and a crack degree level through a crack prediction model based on an improved CNN by using the real-time stiffness attenuation rate, the crack vibration component and the crack energy; A safety management module, configured to adjust a safety factor of the main beam load distribution of the bridge and gantry crane according to the predicted crack type and the crack degree level.

2. The safety intelligent monitoring system for gantry cranes according to claim 1, wherein, The stiffness calculation module includes a theoretical stiffness calculation unit, a real-time stiffness calculation unit and an attenuation rate calculation unit; The theoretical stiffness calculation unit is configured to calculate a theoretical stiffness based on a stiffness displacement decomposition model; The real-time stiffness calculation unit is configured to determine a displacement error, a load distribution correction factor and a damping coefficient correction factor according to the vibration component, the load distribution, the burst energy and the impact count rate, and generate a real-time stiffness based on the displacement error, the load distribution correction factor and the damping coefficient correction factor; The attenuation rate calculation unit is configured to calculate the real-time stiffness attenuation rate according to the theoretical stiffness and the real-time stiffness.

3. The safety intelligent monitoring system for gantry cranes according to claim 2, characterized in that The vibration component includes a bearing seat vibration component, a tooth meshing vibration component and a tooth meshing rotation frequency component, the displacement error includes a bearing seat displacement error and a tooth meshing displacement error, and the real-time stiffness calculation unit includes a bearing seat displacement error correction sub-unit and a meshing displacement correction sub-unit; The bearing seat displacement error correction sub-unit is configured to calculate an actual bearing seat displacement error based on a time integral of the bearing seat vibration component and the bearing seat natural frequency, and calculate the bearing seat displacement error according to the actual bearing seat displacement error and the maximum displacement error; The meshing displacement correction sub-unit is configured to obtain an actual displacement error value by multiplying a ratio of the tooth meshing vibration component to the tooth meshing rotation frequency component by a first weight coefficient, and determine the tooth meshing displacement error according to a ratio of the actual displacement error value to the gear design backlash.

4. The safety intelligent monitoring system for gantry cranes according to claim 3, characterized in that, The real-time stiffness calculation unit further includes a meshing stiffness correction sub-unit; The meshing stiffness correction sub-unit is configured to determine the load distribution correction factor based on a proportional relationship between the load distribution and the rated load distribution multiplied by a second weight coefficient and the single tooth meshing stiffness.

5. The safety intelligent monitoring system for gantry cranes according to claim 4, characterized in that The real-time stiffness calculation unit further includes a burst energy correction term calculation sub-unit, an impact count rate correction term calculation sub-unit and a weighted calculation sub-unit; The burst energy correction term calculation sub-unit is configured to calculate a first correction term by multiplying a ratio of the burst energy to the acoustic emission threshold by a third weight coefficient; The impact count rate correction term calculation subunit is used to calculate the second correction term by multiplying the ratio of the impact count rate to the acoustic emission reference value by the fourth weight coefficient; The weighted calculation subunit is connected to the burst energy correction term calculation subunit and the impact count rate correction term calculation subunit, and is used to calculate the damping coefficient correction factor based on the first correction term, the second correction term, and the meshing damping coefficient.

6. The safety intelligent monitoring system for gantry cranes according to claim 2, characterized in that, The theoretical stiffness calculation unit is used to calculate the theoretical stiffness based on Hooke's law and Hertz contact theory.

7. The safety intelligent monitoring system for gantry cranes according to claim 1, wherein The crack prediction model includes a data parallel input network, a multi-branch feature extraction network, a multi-scale feature fusion network, and a multi-task classification network; The data parallel input network is used to sample and standardize the real-time stiffness decay rate, the crack vibration component, and the crack energy respectively, and generate the stiffness decay time series feature, the vibration component time-frequency domain feature, and the crack energy statistical feature; The multi-branch feature extraction network is used to extract the short-term correlation of stiffness decay from the stiffness decay time series feature to generate the first branch feature, extract the time domain feature, the frequency domain feature, and perform weighted fusion after the time-frequency domain branch on the vibration component time-frequency domain feature to generate the second branch feature, and perform convolutional mapping on the crack energy statistical feature to generate the third branch feature; The multi-scale feature fusion network is used to generate the fusion feature by feature splicing and SCSA for the first branch feature, the second branch feature, and the third branch feature; The multi-task classification network is used to generate the predicted crack type through the classification branch for the fusion feature, and generate the crack degree level through the regression branch for the fusion feature.

8. The safety intelligent monitoring system for gantry cranes according to claim 7, characterized in that The crack prediction module includes a crack type classification training unit; The crack type classification training unit is used to optimize and train the parallel input network, the multi-branch feature extraction network, the multi-scale feature fusion network, and the classification branch through an improved focal loss function based on difficult-to-classify samples.

9. The safety intelligent monitoring system for gantry cranes according to claim 7, characterized in that, The crack prediction module further includes a crack degree level regression training unit; The crack degree level regression training unit is used to optimize and train the parallel input network, the multi-branch feature extraction network, the multi-scale feature fusion network, and the regression branch through a regression loss function based on the degree level number and the Sigmoid function.

10. The safety intelligent monitoring system for gantry cranes according to any one of claims 1 to 9, characterized in that, The safety factor is used to control the shortest distance of the trolley of the bridge and gantry crane from the end point, and the monitoring value processing module, the stiffness calculation module, the crack prediction module, and the safety management module are embedded in the industrial control computer for deployment.

Images