Tower crane condition monitoring methods, devices, electronic equipment and storage media

Abstract

This application provides a tower crane condition detection method, device, electronic device, and storage medium. It extracts features from tower crane images to obtain feature vector parameters for multiple tower crane conditions; optimizes the speed update method, pulse emission rate update method, and fitness function of the search algorithm; searches for optimal weights and optimal biases using the search algorithm, and assigns these optimal weights and biases to a neural network; inputs the feature vector parameters of multiple tower crane conditions to the neural network with the assigned optimal weights and biases for detection, obtaining the tower crane condition result; balances the global and local search capabilities of the search algorithm, making the optimal weights and biases more suitable for the tower crane condition detection network, increasing the diversity and flexibility of the group, and thus improving the recognition accuracy of the neural network for tower crane condition detection.

Description

Technical Field

[0001] The embodiments of this application relate to, but are not limited to, the field of engineering inspection, and particularly to tower crane condition detection methods, devices, electronic equipment, and storage media. Background Technology

[0002] Monitoring the status of tower cranes during construction can effectively improve safety management and prevent accidents. Detection methods based on time-series analysis models only statistically capture the overall pattern of the time series, only able to capture linear relationships, unable to make short-term predictions, and unaffected by sudden events, making it difficult to establish a direct link with the physical characteristics of the structure itself. While neural network-based detection methods have high fault tolerance and robustness, focusing on describing the influence of local and short-term factors, directly approximating nonlinear relationships in time series using neural networks in practical applications often requires a large number of input nodes, hidden layers, and hidden layer nodes, making the network structure overly complex. This can lead to gradient explosion and vanishing gradient problems during neural network training, thus affecting the network's detection accuracy. Summary of the Invention

[0003] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0004] The purpose of this application is to at least partially solve one of the technical problems existing in the related technologies. The embodiments of this application provide a tower crane condition detection method, device, electronic device and storage medium, so as to achieve a balance between the global search capability and local search capability of the search algorithm, and make the optimal weight and optimal bias value more suitable for the detection network of tower crane condition detection.

[0005] An embodiment of the first aspect of this application provides a tower crane condition detection method, comprising:

[0006] Acquire tower crane images;

[0007] Based on the tower crane images, feature vector parameters for multiple tower crane states are obtained by feature extraction.

[0008] Construct a detection network;

[0009] Initialize the search algorithm and optimize its speed update method, pulse emission rate update method, and fitness function to coordinate the global and local search capabilities of the search algorithm;

[0010] The optimal weights and optimal biases of the neural network are searched using an optimized search algorithm, and the optimal weights and optimal biases are assigned to the neural network.

[0011] The feature vector parameters of multiple tower crane states are input into a neural network that is given the optimal weight and the optimal bias value for detection, and the tower crane state results are obtained.

[0012] In certain embodiments of the first aspect of this application, the feature vector parameters include multiple Hu invariant moments, gray-scale center coordinates, zero-order moments, and the lifting weight and amplitude corresponding to the trajectory.

[0013] In certain embodiments of the first aspect of this application, the search algorithm is the Bat algorithm; the speed update method is optimized in the following manner: v n (t+1)=wv n (t)+(x n (t)-p k (t))·f γn (t); w = μ min +(μ max -μ min )×rand()+σ×randn(); where w is the inertia weight, v n (t+1) represents the velocity of the (t+1)th iteration, v n (t) represents the velocity of the t-th iteration, x n (t) is the position of the bat in the t-th iteration, f rn (t) is the pulse frequency of the t-th iteration, p k (t) is the global optimal position of the current bat population search in the t-th iteration, μ min It is the minimum value of the random inertia weight, μ max σ is the maximum value of the random inertia weight, rand() is a random number between 0 and 1, randn() is a random number from a standard normal distribution, and σ is the standard deviation between the random inertia weight and the mean of the random inertia weight.

[0014] In some embodiments of the first aspect of this application, the search algorithm is the Bat algorithm; the pulse emission rate update method is optimized as follows: Where, r n (t) represents the pulse emission rate in the t-th iteration, r n (0) represents the original pulse emission rate, t represents the current iteration number, LG represents the total number of iterations, and β1 is the value used to adjust γ. n (t) Update the preset parameters for the curve's trend.

[0015] In some embodiments of the first aspect of this application, the search algorithm is the Bat algorithm; the fitness function is optimized in the following manner: acc0 represents the accuracy on the training set, and acc1 represents the accuracy on the test set.

[0016] In certain embodiments of the first aspect of this application, the detection network is a backpropagation neural network; the construction of the detection network includes:

[0017] Establish input layer nodes, hidden layer nodes, and output layer nodes for the detection network, wherein the number of input nodes is the same as the number of feature vector parameters, and the number of output nodes is the same as the number of tower crane status results;

[0018] The detection network is constructed by connecting the input layer nodes, the hidden layer nodes, and the output layer nodes.

[0019] In certain embodiments of the first aspect of this application, the number of hidden layer nodes is determined in the following manner:

[0020] Multiple estimated quantity values ​​are obtained based on the number of input layer nodes and the number of output layer nodes;

[0021] Construct multiple candidate detection networks, wherein the number of hidden layer nodes in each candidate detection network corresponds one-to-one with the estimated number of nodes.

[0022] The simulation results are obtained by inputting the training data set into the candidate detection network.

[0023] The estimated number of nodes corresponding to the candidate detection network with the best simulation results is determined as the number of hidden layer nodes.

[0024] A second aspect of this application provides a tower crane condition detection device, comprising:

[0025] Image acquisition unit, used to acquire images of the tower crane;

[0026] The feature extraction unit is used to extract features from the tower crane image to obtain feature vector parameters of multiple tower crane states.

[0027] Network building blocks are used to build detection networks;

[0028] The search algorithm optimization unit is used to initialize the search algorithm and optimize the speed update method, pulse emission rate update method and fitness function of the search algorithm to coordinate the global search capability and local search capability of the search algorithm.

[0029] The network optimization unit is used to search for the optimal weights and optimal biases of the neural network using an optimized search algorithm, and to assign the optimal weights and optimal biases to the neural network.

[0030] The detection unit is used to input feature vector parameters of multiple tower crane states into a neural network that is given the optimal weight and the optimal bias value for detection, so as to obtain the tower crane state result.

[0031] According to a third aspect of this application, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the tower crane status detection method as described above.

[0032] According to a fourth aspect of this application, a computer-readable storage medium stores computer-executable instructions for performing the tower crane condition detection method described above.

[0033] The above scheme has at least the following beneficial effects: It extracts feature vector parameters of multiple tower crane states by performing feature extraction on tower crane images; it optimizes the speed update method, pulse emission rate update method, and fitness function of the search algorithm to coordinate the global and local search capabilities of the search algorithm; it uses the optimized search algorithm to search for the optimal weights and optimal biases of the neural network and assigns these optimal weights and biases to the neural network; it inputs the feature vector parameters of multiple tower crane states into the neural network assigned the optimal weights and biases for detection, obtaining the tower crane state results; it balances the global and local search capabilities of the search algorithm, making the optimal weights and biases more suitable for the detection network of tower crane state detection, increasing the diversity and flexibility of the group, and thus improving the recognition accuracy of the neural network for tower crane state detection. Attached Figure Description

[0034] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0035] Figure 1 This is a flowchart illustrating the steps of the tower crane condition detection method provided in an embodiment of this application;

[0036] Figure 2 This is a structural diagram of the tower crane condition detection device provided in an embodiment of this application;

[0037] Figure 3 This is a structural diagram of the electronic device provided in the embodiments of this application;

[0038] Figure 4 This is a schematic diagram of the tower crane's deformation under normal conditions;

[0039] Figure 5 This is a schematic diagram of the loads acting on the tower base;

[0040] Figure 6 This is a schematic diagram of the displacement coordinate distribution characteristics of the top of the tower crane under normal conditions;

[0041] Figure 7 This is a schematic diagram of the displacement coordinate distribution characteristic model of the top of the tower body of a tower crane under the condition of metal structure damage;

[0042] Figure 8 This is a schematic diagram of the displacement coordinate distribution characteristics of the top of the tower crane under overload conditions;

[0043] Figure 9 This is a schematic diagram of the displacement coordinate distribution characteristics of the top of the tower crane under uneven settlement of the tower foundation. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0045] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0046] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0047] An embodiment of this application provides a method for detecting the condition of a tower crane.

[0048] Reference Figure 1 Tower crane condition detection methods include, but are not limited to, the following steps:

[0049] Step S100: Obtain an image of the tower crane;

[0050] Step S200: Extract features from the tower crane image to obtain feature vector parameters for multiple tower crane states;

[0051] Step S300: Construct the detection network;

[0052] Step S400: Initialize the search algorithm and optimize the speed update method, pulse emission rate update method and fitness function of the search algorithm to coordinate the global search capability and local search capability of the search algorithm;

[0053] Step S500: Use the optimized search algorithm to search for the optimal weights and optimal biases of the neural network, and assign the optimal weights and optimal biases to the neural network.

[0054] Step S600: Input the feature vector parameters of multiple tower crane states into a neural network with optimal weights and optimal biases to perform detection and obtain the tower crane state results.

[0055] In step S100, the tower crane image is obtained by taking pictures of the tower crane using a drone equipped with a camera.

[0056] Regarding step S200, from a condition monitoring perspective, any method of monitoring tower crane status using a single parameter is inaccurate. Therefore, in actual monitoring, only by acquiring multi-dimensional information about the tower crane can more accurate and reliable condition monitoring and diagnosis be achieved.

[0057] Reference Figure 4 In tower crane structures, the tower body, as the main load-bearing component, primarily undergoes bending deformation, while other deformations such as shear deformation are negligible. The tower crane can be simplified as an elastic rigid frame, with BC representing the jib, AB representing the tower body, and section B representing the slewing bearing. Under load, the tower body, in addition to its own weight, mainly bears the bending moment generated by the suspended load and the jib's own weight, which can be considered as a downward vertical pressure F applied at a certain point in BC. Assuming the rigid frame is divided into two parts along AB at section B, it is subjected to an axial force F. n =F, and the bending moment M = FL2 caused by the tilt, where L2 is the amplitude of the tower crane. Ignoring the compression caused by axial force on rod AB and the influence of shear force on rod AB, the deflection at any section of the tower body is a function related to its height h, i.e.: w = f(h). From the curvature expression and the deflection equation... Given EIw”=M0=FL2, after two integrations we have EIw'=EIθ=FL2L1+C and For rod AB, the rotation angle and deflection at fixed end A are both 0. Substituting into the above equation, we get C = 0 and D = 0. Therefore, the expressions for the deflection and tilt angle at section B are as follows: and Then there is In the formula, ρ represents the radius of curvature at the top of the slewing bearing, M0 is the bending moment at the top of the slewing bearing, L1 is the height from the tower crane base to the slewing bearing, and EI is the bending stiffness of the tower. Static analysis shows that changes in load, luffing, and the tower's bending stiffness all affect the deflection at the top of the tower. Considering the three main failure mechanisms that cause tower crane overturning, their impact on the tower crane structure will be reflected in the deflection at the top of the tower. Therefore, in tower crane condition monitoring, the deflection at the top of the tower can serve as an important indicator for assessing tower crane safety.

[0058] Under a certain fixed load combination, as the tower crane's slewing mechanism rotates, its cross section B will also rotate and project a trajectory onto the horizontal plane. For ease of description, the center point of the side of the slewing bearing edge of the unloaded tower crane in normal state can be selected as the origin O to establish a plane rectangular coordinate system xOy.

[0059] Reference Figure 5 From the mechanical analysis, it can be seen that under normal conditions, the bending moment M0 at the top of the slewing bearing of the tower body is affected by the load, amplitude, and rotation angle. According to the tower crane overturning stability formula specified in the tower crane design code GB / T 13752-2017, the eccentricity e of the tower base detaching from the ground should satisfy e=(M0+h×F) h ) / (F n +G k )≤b / 3; where M0 is the torque value acting on the top surface of the tower base when the tower crane boom is parallel to one side of the tower base, and G k For the self-weight load of the tower crane, F h F is the horizontal load value acting on the top surface of the tower base when the tower crane boom is parallel to one side of the tower base surface. n denoted as , where is the vertical load exerted by the tower crane on the top surface of the tower base, 'b' is the length of the bottom edge of the tower base, and 'h' is the thickness of the tower base.

[0060] Therefore, it can be deduced that the range of the bending moment at the top of the slewing bearing of the tower body during normal operation is M0≤[(F n +G k )b / 3]-h×F h The displacement range of the top of the tower crane under normal conditions is as follows:

[0061] In the xOy coordinate system, let F hx F represents the component of the horizontal load in the x-axis direction. hy Let x be the component of the horizontal load in the y-axis direction. Then, under normal conditions, the deflection ranges of the tower top in the x and y directions are as follows:

[0062] Assuming the bending moment on the tower crane remains constant in its 360° rotation direction under windless conditions, and the height L1 from the tower crane base to the slewing bearing remains constant, considering only the effect of the non-uniformity of the metal structural material on the tower's bending stiffness EI, the elastic modulus E and the moment of inertia I of the tower section caused by the non-uniformity of the tower's metal structural material exhibit a Gaussian distribution in each angular direction, denoted as (μ E ,σ E ), (μ I ,σ I ), where μ E μ I Let E and I be the average values ​​of the tower body, and σ be the average values ​​of the tower body. E σI Let E and I be the standard deviations of the tower's components. According to the 3σ principle, 99.7% of E is distributed within (μ... E -3σ E ,μ E +3σ E Within the range of ), 99.7% of I is distributed in (μ). I -3σ I ,μ I +3σ I Within the range, there are Reference Figure 6 Therefore, the projection range of the displacement coordinates of the top of the tower in the xOy coordinate system under normal conditions is a length L. wx =w xmax -w xmin (x-axis direction), width L wy =w ymax -w ymin A rectangle (in the y-axis direction).

[0063] When a tower crane is simplified to an elastic rigid frame, in the xOy coordinate system, the displacement range of the top of the tower in the x and y directions is affected by the inherent properties of the metal structure (E and I values) and the load (F). n G k F h The influence of geometric dimensions (L1, b, h).

[0064] Reference Figure 7 When a tower crane is in a state of metal structural damage, it generally manifests as cracks, irreversible deformation, or loose bolts connecting the standard sections of the tower body. This is equivalent to a reduction in the inherent properties of the metal structure. Therefore, with the load on the tower crane remaining constant, the values ​​of E and I decrease, and the projection range of the displacement coordinates of the monitoring point at the top of the tower body in the xOy plane coordinate system will increase. In the xOy plane, if each quadrant represents one side of the truss of a standard tower section, when damage occurs to the truss on one side of the standard section, the displacement coordinates of the corresponding quadrant will inevitably be projected outside the range of the tower crane model when the upper slewing mechanism rotates to the vicinity of that truss.

[0065] Reference Figure 8 When the tower crane is in an overloaded state, it is equivalent to the E and I values ​​remaining unchanged, but the load increasing. Then, after the upper slewing mechanism of the tower crane rotates 360°, the distribution characteristics of the displacement coordinates of the monitoring point at the top of the tower body are as follows.

[0066] Reference Figure 9When the tower crane foundation settles unevenly, it can be considered as the foundation undergoing a certain angle of deflection. Assuming that when the tower foundation deflects clockwise by an angle θ' (θ'≠0°), it can be considered that the entire tower crane tilts at an angle θ' towards the y-axis, while the various parts of the tower body are relatively stationary. At this time, the monitoring point at the top of the tower body will shift in the y-axis direction, and the distribution characteristics of the displacement coordinates of the monitoring point at the top of the tower body are as follows.

[0067] When the displacement coordinate distribution characteristics at the top of the tower deviate from the range of the displacement coordinate distribution characteristic model at the top of the tower under normal conditions, it indicates that the tower crane is in a dangerous state. This demonstrates that there is a mapping relationship between the trajectory of the top of the tower and the state of the tower crane, which can be effectively used to monitor the tower crane's condition and diagnose the type of tower crane fault.

[0068] Based on the tower top displacement model of the tower crane under different states, it can be seen that there is a mapping relationship between the trajectory of the tower top and the tower crane state, which can be effectively used to monitor the working state of the tower crane. Therefore, feature extraction of the tower top displacement trajectory is an important prerequisite for realizing tower crane state recognition.

[0069] Displacement signals of the top of the tower body during one revolution were collected under four conditions: normal tower crane operation, overload, uneven foundation settlement, and metal structure damage. After noise reduction using the Mallat algorithm, the axis trajectory was synthesized, and the images were set to the same size. Then, the images were binarized using the im2bw function in MATLAB to obtain trajectory images of the displacement of the top of the tower body under the four conditions.

[0070] The Hu invariant moment method is a feature extraction method that represents the geometric features of an image region as a digital matrix. It possesses invariance to image translation, rotation, and scaling, and is therefore often used to extract image features and combine them with neural networks and other recognition methods to achieve image classification and recognition. Invariant moments utilize the concept of torque, treating image pixels as point masses and pixel coordinates as lever arms, using moments of various orders to represent the shape features of the image.

[0071] Hu invariant moments are effective at describing the shape of objects, but they cannot describe the positional information of a closed trajectory in an image, nor can they describe the proportion of the trajectory in the image. Therefore, the gray-level center of the image is used to represent the positional information of the trajectory, and the zero-order moment of the image is used to represent the proportion of the trajectory in the image. Finally, the Hu invariant moment, gray-level center, zero-order moment of the trajectory at the top of the tower, and the lifting weight and amplitude corresponding to the trajectory are used as feature vector parameters.

[0072] Regarding the physical meaning of the Hu invariant moment in an image, the image can be viewed as a thin plate with uneven mass density. Then, the gray-level distribution function f(x,y) of the image is the density of the thin plate at the position (x,y).

[0073] According to the definition of a moment, the gray level of a two-dimensional image is represented by u(i,j), and the zeroth moment M...00 Represented as: The zeroth moment represents the quality of an image, which is the sum of the image's gray levels.

[0074] The first moment is expressed as: First moment M 10 and M 01 This represents the grayscale center of the image, and its coordinates are:

[0075] Based on the grayscale center coordinate x c y c We can construct central moments that are translationally invariant: Where p,q=0,1,2,..., to compensate for the effect of scale changes on the central moments, the zero-order central moment m is used. 00 Normalizing the central moments of each order yields normalized central moments, establishing scale invariance. Where r = p + q + 2, p + q = 2, 3, ...

[0076] Finally, using the normalized second and third central moments, we can derive the following seven invariant moments, namely the Hu invariant moments. These moments remain unchanged during image translation, rotation, and scaling. The calculation formulas are as follows:

[0077]

[0078] The calculated Hu invariant moment of the trajectory at the top of the tower needs to be further logarithmized for easier identification: To reduce the order of magnitude of the data.

[0079] For step S300, a detection network is constructed; specifically, the detection network is a backpropagation neural network. The BA-BP neural network is constructed using MATLAB's Neural Network Toolbox. The design of the BA-BP neural network mainly includes setting the number of network layers, the number of input and output layer nodes, the number of hidden layer nodes, and the training method for the transfer function. A BP network can contain one or more hidden layers; for most applications, a single hidden layer is sufficient.

[0080] First, establish the input layer nodes, hidden layer nodes, and output layer nodes of the detection network.

[0081] The number of input nodes is the same as the number of feature vector parameters. The feature vector parameters include 7 Hu invariant moments, grayscale center coordinates, zero-order moment, and the lifting weight and amplitude corresponding to the trajectory, totaling 12 input parameters, thus resulting in 12 input nodes. The number of output nodes is the same as the number of tower crane status results. The tower crane status results include four states: normal state, overload state, uneven foundation settlement state, and metal structure damage state, thus resulting in 4 output nodes.

[0082] The number of hidden layers and the number of neurons in each hidden layer of a backpropagation (BP) neural network have a significant impact on its performance. Different settings will affect the size, convergence speed, and generalization ability of the BP network.

[0083] Specifically, the number of hidden layer nodes is determined in the following manner:

[0084] Using empirical formulas Multiple estimated quantity values ​​are obtained based on the number of input layer nodes and the number of output layer nodes; where N i and N o These are the number of nodes in the input layer and the output layer, respectively, and 'a' is a variable chosen by the user, which is a constant between [1, 10].

[0085] Multiple candidate detection networks are constructed, and the number of hidden layer nodes in the candidate detection networks corresponds one-to-one with the estimated number of nodes; for example, if the estimated number of nodes is between 5 and 14, then neural networks with 5 to 14 hidden layer nodes are constructed sequentially.

[0086] The simulation results are obtained by inputting the training data set into the candidate detection network.

[0087] The estimated number of nodes corresponding to the candidate detection network with the best simulation results is determined as the number of hidden layer nodes.

[0088] Furthermore, the nodes in both the hidden and output layers employ the tansig function (logarithmic sigmoid function) because it has the function of a nonlinear amplification factor, which can handle and approximate nonlinear input-output relationships. Moreover, the partitioning is relatively smooth, its classification is more unique and reasonable than linear partitioning, and it has better fault tolerance. The graph and expression of the tansig function are as follows:

[0089] The detection network is constructed by connecting the input layer nodes, the hidden layer nodes, and the output layer nodes.

[0090] For step S400, the search algorithm is specifically the Bat Algorithm.

[0091] The Bat Algorithm is a heuristic search algorithm based on swarm intelligence, and it is an effective method for searching for the global optimum. This algorithm is an iterative optimization technique that initializes a set of random solutions, then iteratively searches for the optimal solution. Furthermore, it generates new local solutions around the optimal solution through random flight, thus enhancing the local search.

[0092] The Bat Algorithm (BA) is a random search algorithm that simulates the ability of bats in nature to detect prey and avoid obstacles using sonar. It simulates the basic detection and localization capabilities of bats using ultrasound to locate obstacles or prey and links this to the optimization objective. The biomimetic principle of the BA algorithm maps a population of ∑ bat individuals to NP feasible solutions in a D-dimensional problem space. The optimization and search processes are simulated as the movement and prey-hunting processes of individual bats. The fitness function value of the solution is used to measure the quality of the bat's position, and the survival of the fittest is analogous to the iterative process of replacing poorer feasible solutions with better ones during optimization and search. In the Bat Algorithm, to simulate bats detecting prey and avoiding obstacles, the following three approximate or idealized rules are assumed: all bats use echolocation to perceive distance, and they employ a clever method to distinguish prey from background obstacles; bats fly randomly at position xi with speed vi, searching for prey with a fixed frequency fmin, a variable wavelength λ, and a volume A0. Bats automatically adjust the wavelength (or frequency) of the emitted pulses and the pulse emission rate r, which belongs to [0,1], based on their proximity to the target. There are various ways to change the volume, but in the bat algorithm, it is assumed that the volume A changes from a maximum value A0 (integer) to a fixed minimum value Amin.

[0093] The initialization search algorithm involves the bats randomly distributing a set of initial solutions in a D-dimensional space. The initial parameters include: maximum pulse volume A0, maximum pulse rate R0, search pulse frequency range [fmin, fmax], volume attenuation coefficient α, search frequency enhancement coefficient γ, and search precision ε or maximum number of iterations iter_max. The bat's position x is also randomly initialized. i And find the current optimal solution x* based on the fitness value.

[0094] Calculate the fitness value for each bat, and denote p. k (t) represents the global optimal position for the current population search.

[0095] Calculate the velocity of all bats in generation t+1. The velocity update method is as follows:

[0096] v n (t+1)=wv n (t)+(x n (t)-p k (t))·f γn(t); w = μ min +(μ max -μ min )×rand()+σ×randn(); where w is the inertia weight, v n (t+1) represents the velocity of the (t+1)th iteration, v n (t) represents the velocity of the t-th iteration, x n (t) is the position of the bat in the t-th iteration, f rn (t) is the pulse frequency of the t-th iteration, p k (t) is the global optimal position of the current bat population search in the t-th iteration, μ min It is the minimum value of the random inertia weight, μ max σ is the maximum value of the random inertia weight, rand() is a random number between 0 and 1, randn() is a random number from a standard normal distribution, and σ is the standard deviation between the random inertia weight and the mean of the random inertia weight.

[0097] If the weight is fixed at 1, each bat will always have the same exploration and exploitation capabilities during prey hunting, but this also makes the search behavior inflexible. By introducing inertia weights into the velocity update formula, a larger inertia weight will give the bat a larger speed, thus a stronger exploration ability, i.e., a global search ability; a smaller inertia weight will give the bat a stronger exploitation ability, i.e., a local search ability. In this way, by adjusting the inertia weights, the influence of historical speed on current search behavior is adjusted, balancing the algorithm's global search ability with its local search ability, increasing the diversity and flexibility of the group.

[0098] Common inertia weight adjustment strategies include linear decreasing strategies, linear differential decreasing strategies, nonlinear decreasing strategies, and adaptive strategies. One option is to add a Stochastic Inertia Weight (SIW) strategy to the velocity update equation. The SIW strategy obtains inertia weights through random distribution, which avoids both insufficient local search capability in the early stages of iteration and insufficient global search capability in the later stages of iteration. This allows the algorithm to quickly escape local searches, which is beneficial for finding the global optimum and improves the algorithm's stability.

[0099] For an individual bat n, generate a random number rand0 that is uniformly distributed between [0,1]. If rand0 < γ n (t), then according to x n (t+1)=x n (t)+v n (t+1) Update bat position, otherwise follow the... Update bat locations.

[0100] The fitness of the updated bat is obtained by f(x) n (t+1)), the fitness function is as follows: acc0 represents the accuracy on the training set, and acc1 represents the accuracy on the test set.

[0101] The algorithm requires virtually no external information during operation, but it does need a fitness function to drive the location updates of the bat colony and evaluate the model's classification performance, especially the performance of the trained model on new data. The main principle for designing the fitness function is to calculate a non-complex value that can be compared. If the objective function itself is a non-negative function, it can be directly used as the fitness function for minimization problems. A smaller fitness function value indicates more accurate training. The fitness function is chosen to be 1 minus the average accuracy of the training and test sets. A smaller fitness function value indicates more accurate training while also balancing better predictive accuracy.

[0102] If f(x) is satisfied n (t+1)) <f(x n (t) and Then the position of bat n is updated to x. n (t+1), loudness updated to A n (t+1), pulse emission rate updated to γ n (t+1).

[0103] The pulse emission rate update method is as follows: Where, r n (t) represents the pulse emission rate in the t-th iteration, r n (0) represents the original pulse emission rate, t represents the current iteration number, LG represents the total number of iterations, and β1 is the value used to adjust γ. n (t) Update the preset parameters for the curve's trend.

[0104] In the early stage of the algorithm, the impulse loudness A n When the value is large, the algorithm's global search capability should be improved to find the optimal region; in the later stages of the algorithm, a strong local search capability is needed to obtain an approximate solution for the extreme point in that region. The pulse emission frequency γ of the original BA algorithm... n As the algorithm grows larger in later stages, the probability of local searches decreases. To address this, an adaptive curve reduction strategy is proposed. This strategy adjusts the rate of decrease in the pulse emission frequency according to the number of iterations. This allows more bats to perform global search operations in the early stages and more local search operations in subsequent iterations.

[0105] Repeat the above steps until the maximum number of iterations is reached or the error is less than the preset error threshold, then stop the iteration.

[0106] For step S500, the optimized search algorithm is used to search for the optimal weights and optimal biases of the neural network, and the optimal weights and optimal biases are output.

[0107] The optimal weights and the optimal biases are assigned to the neural network.

[0108] For step S600, the feature vector parameters of multiple tower crane states are input into a neural network that is given the optimal weight and the optimal bias value for detection, and the tower crane state results are obtained.

[0109] Some embodiments of this application provide a tower crane status detection device.

[0110] Reference Figure 2 The tower crane status detection device includes: an image acquisition unit 510, a feature extraction unit 520, a network construction unit 530, a search algorithm optimization unit 540, a network optimization unit 550, and a detection unit 560.

[0111] The system includes: an image acquisition unit 510 for acquiring images of the tower crane; a feature extraction unit 520 for extracting features from the tower crane images to obtain feature vector parameters for multiple tower crane states; a network construction unit 530 for constructing a detection network; a search algorithm optimization unit 540 for initializing the search algorithm and optimizing its speed update method, pulse emission rate update method, and fitness function to coordinate its global and local search capabilities; a network optimization unit 550 for using the optimized search algorithm to search for the optimal weights and optimal biases of the neural network and assigning them to the neural network; and a detection unit 560 for inputting the feature vector parameters of multiple tower crane states into the neural network assigned the optimal weights and optimal biases to obtain tower crane state results.

[0112] The content of the tower crane condition detection method embodiment is applicable to the tower crane condition detection device embodiment. The specific functions implemented by the tower crane condition detection device embodiment are the same as those of the tower crane condition detection method embodiment, and the beneficial effects achieved are also the same as those achieved by the tower crane condition detection method embodiment.

[0113] An embodiment of this application provides an electronic device. (Refer to...) Figure 3 The electronic device includes: a memory 620, a processor 610, and a computer program stored in the memory 620 and executable on the processor 610. When the processor 610 executes the computer program, it implements the tower crane status detection method as described above.

[0114] This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0115] In general, for the hardware structure of electronic devices, the processor 610 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits to execute relevant programs in order to achieve the technical solutions provided in the embodiments of this application.

[0116] The memory 620 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 620 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 620 and is called and executed by the processor 610.

[0117] Input / output interfaces are used to implement information input and output.

[0118] The communication interface is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0119] Bus 630 transmits information between various components of the device, such as processor 610, memory 620, input / output interfaces, and communication interfaces. Processor 610, memory 620, input / output interfaces, and communication interfaces communicate with each other within the device via bus 630.

[0120] Embodiments of this application provide a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions for performing the tower crane condition detection method described above.

[0121] It should be understood that the method steps in the embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can use standard programming techniques. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if necessary, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0122] Furthermore, the procedures described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by the context. The procedures described herein (or variations and / or combinations thereof) may be executed under the control of one or more computer systems configured with executable instructions, and may be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program comprises a plurality of instructions executable by one or more processors.

[0123] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, smartphones, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention also includes the computer itself.

[0124] A computer program can be applied to input data to perform the functions described herein, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0125] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0126] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A method for detecting the condition of a tower crane, characterized in that, include: Acquire tower crane images; Feature extraction is performed on the tower crane images to obtain feature vector parameters for multiple tower crane states. These feature vector parameters include multiple Hu invariant moments, gray-level center coordinates, zero-order moments, and the lifting weight and amplitude corresponding to the trajectory. The displacement signals of the top of the tower body during one revolution are collected under four states: normal tower crane, overload, uneven foundation settlement, and metal structure damage. After noise reduction using the Mallat algorithm, the axis trajectory is synthesized, and the images are set to the same size. Then, the images are binarized using the im2bw function in MATLAB to obtain trajectory images of the displacement of the top of the tower body under the four states. The Hu invariant moments are obtained by feature extraction from the trajectory images using the Hu invariant moment method. The gray-level center is used to represent the position information of the trajectory, and the zero-order moment is used to represent the proportion of the trajectory in the image. Construct a detection network, wherein the detection network is a neural network; Initialize the search algorithm and optimize its speed update method, pulse emission rate update method, and fitness function to coordinate the global and local search capabilities of the search algorithm; The optimal weights and optimal biases of the neural network are searched using an optimized search algorithm, and the optimal weights and optimal biases are assigned to the neural network. The feature vector parameters of multiple tower crane states are input into a neural network that is given the optimal weight and the optimal bias value for detection, and the tower crane state results are obtained. The search algorithm is the Bat algorithm; the pulse emission rate update method is optimized as follows: ;in, Let be the pulse emission rate in the t-th iteration. Let t be the original pulse emission rate, t be the current iteration number, and LG be the total number of iterations. For adjustment Update the preset parameters for the curve's trend; The fitness function is optimized in the following manner: ; acc0 is the accuracy of the training set, and acc1 is the accuracy of the test set.

2. The tower crane condition detection method according to claim 1, characterized in that, The search algorithm is the Bat algorithm; the speed update method is optimized as follows: ; Where w is the inertia weight, Let be the velocity of the (t+1)th iteration. Let be the velocity of the t-th iteration. It is the position of the bat in the t-th iteration. It is the pulse frequency of the t-th iteration. It is the globally optimal position of the current bat population search in the t-th iteration. It is the minimum value of the random inertia weight. It is the maximum value of the random inertia weight. It is a random number between 0 and 1. These are random numbers that follow a standard normal distribution. It is the standard deviation between the random inertia weight and the mean of the random inertia weight.

3. The tower crane condition detection method according to claim 1, characterized in that, The detection network is a backpropagation neural network; The constructed detection network includes: Establish input layer nodes, hidden layer nodes, and output layer nodes of the detection network, wherein the number of input layer nodes is the same as the number of feature vector parameters, and the number of output layer nodes is the same as the number of tower crane status results; The detection network is constructed by connecting the input layer nodes, the hidden layer nodes, and the output layer nodes.

4. The tower crane condition detection method according to claim 3, characterized in that, The number of hidden layer nodes is determined in the following manner: Multiple estimated quantity values ​​are obtained based on the number of input layer nodes and the number of output layer nodes; Construct multiple candidate detection networks, wherein the number of hidden layer nodes in each candidate detection network corresponds one-to-one with the estimated number of nodes. The simulation results are obtained by inputting the training data set into the candidate detection network. The estimated number of nodes corresponding to the candidate detection network with the best simulation results is determined as the number of hidden layer nodes.

5. A tower crane condition detection device, characterized in that, include: Image acquisition unit, used to acquire images of the tower crane; The feature extraction unit is used to extract features from the tower crane image to obtain feature vector parameters of multiple tower crane states. The feature vector parameters include multiple Hu invariant moments, gray-scale center coordinates, zero-order moments, and lifting weight and amplitude corresponding to the trajectory. A network construction unit is used to construct a detection network, wherein the detection network is a neural network; The search algorithm optimization unit is used to initialize the search algorithm and optimize the speed update method, pulse emission rate update method and fitness function of the search algorithm to coordinate the global search capability and local search capability of the search algorithm. The network optimization unit is used to search for the optimal weights and optimal biases of the neural network using an optimized search algorithm, and to assign the optimal weights and optimal biases to the neural network. The detection unit is used to input feature vector parameters of multiple tower crane states into a neural network that is given the optimal weight and the optimal bias value for detection, so as to obtain the tower crane state result; The search algorithm is the Bat algorithm; the pulse emission rate update method is optimized as follows: ;in, Let be the pulse emission rate in the t-th iteration. Let t be the original pulse emission rate, t be the current iteration number, and LG be the total number of iterations. For adjustment Update the preset parameters for the curve's trend; The fitness function is optimized in the following manner: ; acc0 is the accuracy of the training set, and acc1 is the accuracy of the test set.

6. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the tower crane condition detection method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The system stores computer-executable instructions for performing the tower crane condition detection method as described in any one of claims 1 to 4.

Images