Intelligent sensing system for hoisting risk of large equipment

Abstract

The invention discloses a large equipment hoisting risk intelligent sensing system, and belongs to the technical field of hoisting risk identification. The method is used for solving the technical problem that an overload or out-of-control risk cannot be early warned in time due to the fact that online state monitoring of key devices such as a torque limiter and an anti-falling safety device is not integrated in an existing scheme. Through fusion of a fiber bragg grating, a millimeter wave radar and a torque sensor, full-scale monitoring from material damage to overall dynamic behaviors is achieved, an improved Transform model is combined with a space-time self-attention mechanism, dynamic updating of the eccentric distance is implemented based on torque-strain combined calculation, and the real-time control requirement of hoisting operation is met; according to the method, a three-dimensional risk tensor fusing structure defect-load offset-environment interference is constructed, the three-dimensional risk tensor is converted into a Lorenz system initial condition through state mapping, a future trajectory is predicted through a pre-training model, and end-to-end modeling of dynamic risk and nonlinear evolution is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lifting risk identification, and particularly relates to a large equipment lifting risk intelligent perception system. BACKGROUND

[0002] The large equipment lifting risk perception refers to the ability and process of identifying, monitoring, evaluating and warning potential safety risks in real time or quasi-real time during the lifting operation of large and heavy equipment by using technical means and management methods.

[0003] In the implementation of the prior art scheme, for the detection of hidden defects such as internal wire breakage and corrosion of the steel wire rope, manual visual inspection or simple tools are mostly relied on, real-time dynamic monitoring is difficult to achieve, and some systems do not integrate online state monitoring of key devices such as torque limiters and anti-falling safety devices, so that overload or out-of-control risks cannot be timely warned. SUMMARY

[0004] The present application relates to the technical field of lifting risk identification, and particularly relates to a large equipment lifting risk intelligent perception system.

[0005] The object of the present application can be achieved by the following technical solutions: A large equipment lifting risk intelligent perception system, comprising: A lifting dynamic characteristic data acquisition and processing module acquires and processes data through a multi-modal sensor network to obtain a strain distribution sequence of a steel wire rope body, a relative displacement sequence of a lifting hook-heavy object, a real-time output torque sequence of a torque limiter, synchronously acquires environmental interference parameters, and pre-processes the acquired sequence data to generate a multi-dimensional fusion data set; A multi-dimensional risk data processing and fusion module extracts features of the strain distribution sequence of the steel wire rope body based on the generated multi-dimensional fusion data set through an improved Transformer model, identifies the position and number density of internal wire breakage defects, simultaneously calculates a dynamic load eccentricity distance by fusing the relative displacement sequence of the lifting hook-heavy object and the torque sequence, and constructs a three-dimensional risk tensor in combination with wind speed turbulence intensity; A risk prediction and evaluation dynamic control module analyzes data of the three-dimensional risk tensor through a pre-trained chaotic prediction model, obtains a deviation degree of a predicted trajectory and a reference trajectory, analyzes data of the deviation degree, dynamically triggers an electromagnetic locking mechanism of an anti-falling safety device according to the analysis result, and synchronously pushes to a lifting monitoring platform.

[0006] Preferably, when the strain distribution sequence of the steel wire rope body is acquired by the fiber Bragg grating sensor array, strain data of the rope body under tensile and bending combined loads during the lifting process is acquired, and the strain distribution sequence is output. The connection point of the hook and the weight is continuously tracked by the millimeter wave radar, and a relative displacement polar coordinate sequence is output; The polar coordinates are converted into three-dimensional displacement in the Cartesian coordinate system through coordinate conversion, and the obtained three-dimensional displacement is processed and combined to obtain a relative displacement sequence; When collecting the torque sequence output by the torque limiter in real time through the high-precision torque sensor, the voltage signal output by the high-precision torque sensor is collected, the torque-voltage linear relationship is fitted and the actual torque is output, and the obtained actual torque is sorted and combined according to the time sequence to obtain a torque sequence; When collecting environmental interference parameters, an ultrasonic anemometer is deployed beside the radar installation position to collect wind speed and turbulence intensity; The air humidity is collected by a temperature and humidity sensor; The wind speed turbulence intensity and the air humidity constitute the environmental interference parameters.

[0007] Preferably, the collected sequence data is preprocessed, including strain distribution sequence temperature drift compensation, relative displacement sequence clutter suppression and smoothing, and wind speed interference correction to displacement.

[0008] Preferably, when generating a multi-dimensional fusion data set, the timestamps of each sensor are linearly interpolated and corrected to obtain corrected timestamps; the axial position x of the fiber grating sensor is converted into a global coordinate; the Cartesian coordinates of the millimeter wave radar are directly included in the global coordinate system to realize the spatial correlation of the strain distribution and the displacement; The preprocessed strain distribution sequence, the corrected displacement sequence, the torque sequence, and the environmental interference parameters are integrated to generate a structured multi-dimensional fusion data set.

[0009] Preferably, when extracting features from the steel wire rope body strain distribution sequence, the strain feature vector is output through the last layer of the improved Transformer encoder; the number of broken wires K is calculated, and the expression involved is, , is a ceiling function; is the number density.

[0010] Preferably, the load eccentricity is calculated and obtained according to the torque output by the torque limiter and the total hoisting weight; The load eccentricity is converted into a dimensionless load eccentricity coefficient C.

[0011] Preferably, when constructing a three-dimensional risk tensor, the defect severity S is calculated and obtained based on the number density and position of the broken wires; The environmental interference weight W is calculated and obtained based on the wind speed turbulence intensity; The defect severity S, the load eccentricity coefficient C, and the environmental interference weight W are processed and obtained as three dimensions to construct a three-dimensional risk tensor: ;in, This is an element-wise product operation.

[0012] Preferably, the three-dimensional risk tensor R is mapped to the typical initial interval of the Lorenz system; the evolution of the risk trajectory is described using the standard Lorenz equation.

[0013] Preferably, the future risk trajectory output by the chaotic model is obtained and converted into a risk tensor dimension: ; in, The predicted trajectories at the mixed time steps are respectively Defect severity, load eccentricity coefficient, and environmental interference weight; These are the predicted values ​​of the Lorenz system state variables output by the chaotic prediction model.

[0014] Preferably, a weighted Euclidean distance is used to measure the deviation between the predicted trajectory and the reference trajectory, the calculated deviation is analyzed, and the electromagnetic locking mechanism is dynamically triggered based on the analysis results. If the deviation is greater than Y1 and the duration is greater than or equal to Y2, the locking signal of the fall arrestor is triggered; Y1 and Y2 are both real numbers greater than 0.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention achieves full-scale monitoring from material damage to overall dynamic behavior by integrating fiber optic gratings, millimeter-wave radar, and torque sensors. Compared to existing single-sensor systems, it can effectively improve the coverage of risk features. Through temperature compensation, wind speed correction, and Kalman filtering, it can effectively reduce strain measurement errors and displacement measurement errors, ensuring the correlation of multi-source data across three dimensions of time, space, and physical quantities, providing high-quality input for subsequent risk perception models. It also solves the data drift problem caused by temperature, humidity, and wind interference in outdoor hoisting, enhancing environmental robustness. By employing mature fiber optic grating and millimeter-wave radar technologies, it avoids complex processes such as nano-coatings, improving sensor deployment efficiency.

[0016] This invention combines an improved Transformer model with a spatiotemporal self-attention mechanism, which can effectively improve the accuracy of wire breakage location identification and reduce quantity density error. Based on torque-strain joint calculation, dynamic updates of eccentricity are implemented to meet the real-time control requirements of hoisting operations. A three-dimensional risk tensor integrating structural defects, load offset, and environmental interference is constructed to achieve multi-dimensional risk assessment. Compared with the existing single overload warning, this invention can effectively reduce the false alarm rate and achieve full-process risk management including pre-event prevention, in-event monitoring, and post-event traceability.

[0017] This invention transforms a three-dimensional risk tensor into initial conditions for a Lorenz system through state mapping, and predicts future trajectories using a pre-trained model, achieving end-to-end modeling of dynamic risks and nonlinear evolution. By calculating the weighted deviation between the baseline trajectory and the predicted trajectory, the degree of risk anomaly can be quantified, solving the problem of missed detection of sudden nonlinear risks by the traditional threshold method. By calculating the deviation and analyzing whether it exceeds the limit, a superconducting magnetic locking mechanism is dynamically triggered to quickly achieve physical braking, enabling early warning of nonlinear risks. Compared with traditional vibration monitoring, this invention can effectively improve the warning time window, avoid accidents caused by insufficient braking, and improve adaptive robustness. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the operation of a large equipment hoisting risk intelligent perception system according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, the present invention is an intelligent risk perception system for hoisting large equipment, comprising: The hoisting dynamic characteristic data acquisition and processing module deploys a multimodal sensor network to acquire and process data, obtaining the strain distribution sequence of the wire rope, the relative displacement sequence of the hook and the load, and the torque sequence output in real time by the torque limiter. It also simultaneously acquires environmental interference parameters and preprocesses the acquired sequence data to generate a multidimensional fusion dataset. Specific steps include: Deploying a multimodal sensor network includes deploying fiber Bragg grating sensor arrays, setting up millimeter-wave radar systems, and integrating torque sensors. During the deployment of the fiber Bragg grating sensor array, three sets of fiber Bragg grating sensor arrays are laid at 1 / 4, 1 / 2, and 3 / 4 of the length of the steel wire rope, with each set containing four sensors. They are evenly distributed along the circumference of the rope and spaced at 90° intervals. The sensors are attached to the surface of the rope with epoxy resin, forming a 45° angle with the axis of the rope, and the direction of the sensitive axis is consistent with the direction of the principal stress of the rope. When setting up a millimeter-wave radar system, a 24GHz FMCW millimeter-wave radar is installed on the top of the hoisting arm, with the radar antenna facing the hook-load connection point. The pitch angle can be adjusted to ±30° through the universal joint bracket. When integrating torque sensors, a high-precision strain gauge torque sensor is installed on the input shaft of the torque limiter via a bushing, and the sensor signal is transmitted to the data acquisition module via a slip ring. When acquiring the strain distribution sequence of a steel wire rope using a fiber optic grating sensor array, strain data of the rope under combined tensile and bending loads during hoisting is collected, and the strain distribution sequence is output. n is the number of sensors, which takes the value of 12; x is the axial position coordinate of the cable body; t is the timestamp, which is in ms. The connection point between the hook and the load is continuously tracked using millimeter-wave radar, outputting a relative displacement polar coordinate sequence. , The straight-line distance from the hook-weight connection point to the radar, measured by millimeter-wave radar. Let X be the angle between the target's projection onto the horizontal plane and the X-axis of the radar coordinate system, where the X-axis corresponds to the direction of the lifting arm. Let the angle between the line connecting the target and the radar and the horizontal plane be positive (upward) and negative (downward). The polar coordinates are then converted to a three-dimensional displacement in Cartesian coordinates through coordinate transformation. x(t)= r(t)×cosθ(t)×cosφ(t); y(t)= r(t)×sinθ(t)×cosφ(t); z(t) = r(t) × sinφ(t); The obtained three-dimensional displacements are combined to obtain a relative displacement sequence (x(t), y(t), z(t)); where x(t), y(t), and z(t) are the horizontal displacement, horizontal lateral displacement, and vertical displacement, respectively; the horizontal direction corresponds to the axis of the lifting arm, the horizontal lateral displacement corresponds to the axis of the lifting arm, and the vertical direction corresponds to the height. When acquiring the real-time torque sequence output by the torque limiter using a high-precision torque sensor, the voltage signal U output by the high-precision torque sensor is also acquired. A linear relationship between torque and voltage is fitted, and the actual torque T is output. The relevant expression is: ;in, This is the sensitivity coefficient, with a default value of 250. This is the zero-drift compensation value; the specific value can be obtained through preliminary simulation tests. The actual torques T acquired are sorted and combined according to the acquisition time sequence to obtain the torque sequence. f represents the number of data points; When collecting environmental interference parameters, an ultrasonic anemometer is deployed next to the radar installation location to collect wind speed and turbulence intensity. These are the standard deviation of wind speed and the average wind speed, respectively. Air humidity H(t) is collected using a temperature and humidity sensor; Wind speed, turbulence intensity, and air humidity constitute environmental disturbance parameters; The acquired sequence data is preprocessed, including temperature drift compensation for strain distribution sequence, clutter suppression and smoothing of relative displacement sequence, and correction of displacement due to wind speed interference. When performing temperature drift compensation for the strain distribution sequence, the fiber grating strain data is corrected using a temperature effect correction formula. The relevant expression is as follows: ;in, The corrected strain value; The original strain value; is the temperature coefficient of the fiber grating, which corresponds to the strain drift caused by a unit temperature change; This refers to the change in ambient temperature. This represents the wavelength shift at the center of the grating. The initial center wavelength of the grating; This is the thermo-optic coefficient, with a default value of [value missing]. ; It needs to be explained that, through , Separate temperature strain from load strain to ensure that strain data reflects the actual stress on the structure. When performing clutter suppression and smoothing on relative displacement sequences, a constant false alarm rate (CFAR) detection algorithm is used to remove background clutter from millimeter-wave radar displacement data, with the false alarm probability set to... Preserve the target point cloud; The constant false alarm rate (CFAR) detection algorithm is an adaptive threshold detection technology widely used in signal processing fields such as radar, sonar, communication, and remote sensing. Its core objective is to maintain a constant false alarm probability in the detection system in an unknown and potentially changing background noise or clutter environment. The false alarm probability is the probability of mistakenly identifying noise / interference as a target. It is an existing conventional technical means, and the specific implementation steps will not be elaborated here. Furthermore, a Kalman filter is used to smooth the three-dimensional displacement sequence, and the state equation is: ;in, This is the predicted value for time t based on the state at time t-1; Let A be the optimal estimated state vector at time t-1; let B be the state transition matrix; and let C be the control matrix. This is the process noise vector; It should be explained that by balancing prediction and measurement through the state transition matrix and control matrix, radar noise can be suppressed and the relative displacement sequence can be smoothed. When correcting for displacement due to wind speed disturbance, the displacement data is dynamically corrected using wind speed turbulence intensity. The relevant expression is as follows: ;in, For the corrected displacement; This is the filtered displacement; This is the wind speed influence coefficient, an interference weight calibrated through wind tunnel testing, with a default value of 0.05. It needs to be explained that, through and Quantifying the impact of wind speed fluctuations on displacement can improve the accuracy of hoisting trajectory measurement; the displacement specifically refers to the real-time swing displacement of the hoisting equipment during hoisting. Through the combined effect of these three factors, high-fidelity dynamic characteristic data can be provided for subsequent risk assessment of the hoisting system. When generating the multidimensional fusion dataset, the timestamps of each sensor are... Perform linear interpolation correction to obtain the corrected timestamp. The expression involved is: Where i is the number of the different sensors; The time deviation was calculated by comparing the BeiDou second pulse signal. Establish a global coordinate system for the hoisting system, with the fixed ground reference point O as the origin, the X-axis along the direction of the hoisting arm, the Y-axis perpendicular to the ground and upward, and the Z-axis horizontal. Convert the axial position x of the fiber Bragg grating sensor to global coordinates: This refers to the length of the lifting arm. The boom elevation angle is provided by an angle sensor. The Cartesian coordinates (x, y, z) of millimeter-wave radar are directly incorporated into the global coordinate system, realizing the spatial correlation between strain distribution and displacement; Integrating pre-processed strain distribution sequences Corrected displacement sequence Torque sequence T(t), environmental disturbance parameters ( Generate a structured multidimensional fusion dataset DR, H(t)). The data storage format is CSV, and each record contains a timestamp, 12 strain values, 3 displacement components, 1 torque value, wind speed and turbulence intensity, humidity, and sensor spatial coordinates.

[0022] In this embodiment of the invention, the fusion of fiber optic gratings, millimeter-wave radar, and torque sensors enables full-scale monitoring from material damage to overall dynamic behavior. Compared to existing single sensors, this effectively improves the coverage of risk features. Through temperature compensation, wind speed correction, and Kalman filtering, strain measurement errors and displacement measurement errors can be effectively reduced, ensuring the correlation of multi-source data across three dimensions of time, space, and physical quantities. This provides high-quality input for subsequent risk perception models, solves the data drift problem caused by temperature, humidity, and wind interference in outdoor hoisting, and enhances environmental robustness. The use of mature fiber optic grating and millimeter-wave radar technologies avoids complex processes such as nano-coatings, improving sensor deployment efficiency.

[0023] The multidimensional risk data processing and fusion module, based on the generated multidimensional fused dataset, extracts features from the strain distribution sequence of the wire rope using an improved Transformer model, identifies the location and quantity density of internal wire breakage defects, and simultaneously calculates the dynamic load eccentricity by fusing the relative displacement sequence of the hook and the load with the torque sequence. Finally, it constructs a three-dimensional risk tensor by combining wind speed and turbulence intensity. Specific steps include: strain distribution sequence in multidimensional fusion dataset Convert to 3D tensor For the number of time steps, For time step index; For the number of sensors; Z-score normalization was performed on the time series of each sensor, and for each time step... The sensor sequence is normalized using Min-Max; Z-score normalization and Min-Max normalization are both existing conventional technical solutions, and the specific implementation steps and contents are not described here. Output normalized spatiotemporal features The value range is [0,1]; When constructing the improved Transformer model, a CNN-Transformer hybrid architecture is used. The input layer receives normalized spatiotemporal features. It is converted into a feature vector through the embedding layer: ;in, For linear mapping layers, the output feature dimension is... Learnable location encoding captures temporal sequence information. ; Spatial local feature extraction module, i.e., CNN layer: 2D convolutional layers are used to extract the spatial correlation between sensor locations. For example, a broken wire defect can cause a sudden change in strain between adjacent sensors, and the relevant expression is: ;in, The input tensor is 4D with dimension . , Input the number of channels; The kernel size; To fill the edges, add 1 unit on each side of the sensor location and the feature dimension; This specifies the number of output channels; the default value is 4. Input Dimensions Output dimension After being flattened, it is converted to The flatten operation is a common tensor reshaping operation. Its core function is to transform a multidimensional input tensor into a one-dimensional vector. An improved Transformer encoder, incorporating a spatiotemporal self-attention mechanism and encoder stacking; The spatiotemporal self-attention mechanism treats the time step and sensor position as spatiotemporal tokens, and introduces a spatiotemporal bias matrix when calculating attention weights. ;in, The spatiotemporal bias matrix, ;in, These are the query matrix, key matrix, and value matrix, respectively, obtained from the flattened feature tensor through linear transformation. Feature dimensions for each attention head; To scale the dot product attention; The attention score is converted into a probability distribution using a normalization function. Encoder stack: 3-layer encoder, each layer containing 1 spatiotemporal self-attention module and 1 feedforward network. The spatiotemporal self-attention module contains 8 attention heads. The hidden layer has a dimension of 512 and the activation function is GELU. The output layer includes a defect location header and a quantity density header; The defect location header consists of 2D convolutional layers and a sigmoid activation function, and outputs a location probability map. ; The number density header, consisting of a fully connected layer and a ReLU activation function, outputs the defect number density. ; The model is trained and optimized using existing conventional techniques; the specific implementation steps will not be elaborated here. When extracting features from the strain distribution sequence of a wire rope, the strain feature vector is output from the last layer of the improved Transformer encoder. It includes defect-sensitive features in the spatiotemporal domain, such as local strain gradient and abrupt change frequency; For location probability maps Using a threshold decision, with a threshold value of 0.7, and optimizing using the ROC curve, a binarized defect location matrix is ​​obtained. ;in, For spatial location index; Extracting the binarized defect location matrix Sensor index with a median value of 1 The expression corresponding to the axial position of the wire rope is as follows: ;in, This refers to the total length of the wire rope. =1, 2, 3, ... ; For the location of the defect, ; The expression involved in calculating the number density is: ; The number of broken wires, K, is output based on the quantity density. , It is a rounding function; When calculating the dynamic load eccentricity, the total suspended weight F(t) is calculated using the wire rope strain distribution sequence. The relevant expression is: Where E is the elastic modulus of the wire rope, with a default value of 210. For the cross-sectional area, D is the diameter of the wire rope; Calculate the real-time lever arm length of the lifting boom using millimeter-wave radar displacement sequences. The expression involved is: ;in, Displacement along the direction of the lifting arm; This refers to horizontal lateral displacement; Based on the output torque T(t) of the torque limiter and the total lifting weight F(t), the load eccentricity e(t) is calculated using the following expression: ; The expression for converting the load eccentricity e(t) into a dimensionless load eccentricity coefficient is as follows: ;in, The default value for the safety eccentricity threshold is 0.5. A value of 1 indicates that the danger threshold has been reached; When constructing the three-dimensional risk tensor, the defect severity S is calculated based on the broken wire number density and location weighting, and the relevant expression is: ;in, The position weight can be determined based on the preliminary test data of the wire rope. And, based on wind speed turbulence intensity The calculation of the environmental disturbance weight W involves the following expression: ;in, This is the safe threshold for turbulence intensity. The obtained defect severity S, load eccentricity coefficient C, and environmental disturbance weight W are used as three dimensions to construct a three-dimensional risk tensor: ;in, For element-wise multiplication, the tensor dimension is (1×1×1); the physical meaning of the three-dimensional risk tensor is the defect-load-environment coupling risk value, with a value range of [0, 1]. When classifying risk levels, if R < 0.3, it is considered low risk; If 0.3 ≤ R < 0.7, it is classified as medium risk; If R ≥ 0.7, it is judged as high risk and a real-time warning is triggered.

[0024] In this embodiment of the invention, the improved Transformer model is combined with a spatiotemporal self-attention mechanism, which can effectively improve the accuracy of wire breakage location identification and reduce the number density error; based on torque-strain joint calculation, dynamic updates of eccentricity are implemented to meet the real-time control requirements of hoisting operations; a three-dimensional risk tensor integrating structural defects, load offset, and environmental interference is constructed to achieve multi-dimensional risk assessment. Compared with the existing single overload warning, this can effectively reduce the false alarm rate and realize full-process risk management including pre-event prevention, in-event monitoring, and post-event traceability. Through the above steps, end-to-end modeling from multimodal data to risk decision-making is achieved, which can provide core technical support for intelligent safety monitoring of hoisting systems.

[0025] The risk prediction and assessment dynamic control module analyzes the three-dimensional risk tensor using a pre-trained chaotic prediction model to obtain the deviation between the predicted trajectory and the baseline trajectory. It then analyzes the deviation and dynamically triggers the electromagnetic locking mechanism of the fall arrestor based on the analysis results, simultaneously pushing the data to the hoisting monitoring platform. Specific steps include: Mapping the three-dimensional risk tensor R to the typical initial region (−20, 20) of the Lorenz system involves the following expressions: x0 = 40 × S − 20; y0 = 40 × C − 20; z0 = 40 × W − 20; where x0, y0, and z0 are the initial state variables of the Lorenz system, corresponding to the nonlinear mapping of S, C, and W in the three-dimensional risk tensor; the value 40 is the scaling factor, and the value -20 is the translation amount, ensuring that the initial values ​​cover the core region (−20, 20) of the Lorenz attractor; the Lorenz system is a dynamical system consisting of three nonlinear ordinary differential equations, designed to study the convectional behavior of fluids under heating. The evolution of the risk trajectory is described using the standard Lorenz equation, involving the following expression: in, These are Prandtl number, Rayleigh number, and geometric parameter, with values ​​of 10, 28, and 8 / 3, respectively. All of these are state variables, specifically the intensity of convective motion, the horizontal temperature gradient, and the vertical temperature gradient, such as the velocity difference between updrafts and downdrafts, the temperature difference between the left and right sides of the fluid layer, and the temperature difference between the upper and lower surfaces of the fluid layer. It should be explained that the Prandtl number, the diffusion rate of the control system, was experimentally calibrated to 10 to match the nonlinear characteristics of the lifting risk; Rayleigh number controls the convection intensity of the system, and 28 is the classical chaos threshold of the Lorenz attractor; Geometric parameters control the attractor shape, and are fixed at 8 / 3 to ensure trajectory stability; When pre-training the chaos prediction model, the acquired multidimensional fusion dataset is processed to simulate 1000 risk-failure samples, including normal state, slight risk, and severe risk; the Lorenz trajectory generated for each sample is used as model training data; 5 seconds corresponds to 50 steps. The model structure uses a hybrid LSTM-ResNet network, with the initial state (x0, y0, z0) as input and the predicted trajectory for 50 steps as output. Mixed time steps =1, 2, ..., 50; LSTM layer: captures time series dependencies, 2 layers, hidden dimension 64; ResNet layer: extracts nonlinear features of the trajectory, 3 residual blocks, 3×3 convolutional kernels; Output layer: Fully connected layer, output dimension 3×50, corresponding to 3 variables and 50 steps of prediction; Fine-tuning mechanism: During online deployment, the model parameters are fine-tuned every hour using the latest risk-free samples with a deviation of <0.3, 5e−5, and a batch size of 16, to ensure that the prediction accuracy dynamically adapts as the equipment ages; It should be noted that training and optimizing the chaotic prediction model is a conventional technique. The specific implementation steps will not be described here, only the key parameters will be explained. Based on historical risk-free data, specifically samples with a deviation <0.3, the time average of the Lorenz trajectory is calculated using the following expression: ;in, M represents the baseline trajectory; M represents the number of historical risk-free samples; m represents the index of historical risk-free samples. For the m-th group of risk-free samples at the mixed time step The state variables of the Lorenz system correspond to the standard Lorenz equations. ; Obtain the risk trajectory for the next 5 seconds output by the chaotic model and convert it into a risk tensor dimension: in, The predicted trajectories at the mixed time steps are respectively Defect severity, load eccentricity coefficient, and environmental interference weight; These are the predicted values ​​of the Lorenz system state variables output by the chaotic prediction model; The weighted Euclidean distance is used to measure the deviation D between the predicted trajectory and the baseline trajectory. The relevant expression is: ; in, These are the defect risk weight, load risk weight, and environmental risk weight, respectively, with default values ​​of 0.5, 0.3, and 0.2. To predict the deviation between the defect severity risk component and the baseline defect severity risk component; To predict the deviation between the load eccentricity risk component and the reference load eccentricity risk component; To predict the deviation between the environmental disturbance risk component and the baseline environmental disturbance risk component; The calculated deviation is analyzed, and the electromagnetic locking mechanism is dynamically triggered based on the analysis results. If the deviation is greater than Y1 and the duration is greater than or equal to Y2, the locking signal of the fall arrestor will be triggered; Y1 and Y2 are both real numbers greater than 0, with default values ​​of 0.6 and 0.5 respectively. In other cases, normal operation will continue; The superconducting magnet is locked according to the triggered locking signal; by utilizing the zero resistance characteristics and Meissner effect of superconducting materials in low temperature environment, a strong magnetic field is quickly generated, and the physical locking of the load is achieved by magnetic force, such as adsorbing and fixing the track and clamping the steel wire rope. Its braking force can reach more than 10 times that of traditional mechanical braking, and the response time is <10ms. As the execution terminal of the safety system, the superconducting magnet receives the braking command from the fall arrestor and instantly switches from the non-braking state (low magnetic field) to the braking state (strong magnetic field), forcibly stopping the hoisting movement through electromagnetic force to avoid a fall accident.

[0026] In this embodiment of the invention, the three-dimensional risk tensor is converted into the initial conditions of the Lorenz system through state mapping, and the trajectory in the next 5 seconds is predicted through a pre-trained model, realizing end-to-end modeling of dynamic risk and nonlinear evolution. By implementing weighted deviation calculation between the baseline trajectory and the predicted trajectory, the degree of risk anomaly can be quantified, solving the problem of missed detection of sudden nonlinear risks by the traditional threshold method. By calculating the deviation and analyzing whether it exceeds the standard, the superconducting magnetic locking mechanism is dynamically triggered to quickly realize physical braking, realizing early warning of nonlinear risks. Compared with traditional vibration monitoring, it can effectively improve the warning time window, avoid accidents caused by insufficient braking, and improve adaptive robustness.

[0027] In the several embodiments provided by this invention, it should be understood that the disclosed methods can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0028] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0029] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0030] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A large equipment hoisting risk intelligent perception system, characterized in that, include: The hoisting dynamic feature data acquisition and processing module is deployed through a multi-modal sensor network to acquire and process data, obtaining the strain distribution sequence of the wire rope, the relative displacement sequence of the hook and the load, and the torque sequence output by the torque limiter in real time. It also collects environmental interference parameters simultaneously and preprocesses the acquired sequence data to generate a multi-dimensional fusion dataset. The multidimensional risk data processing and fusion module, based on the generated multidimensional fusion dataset, extracts features from the strain distribution sequence of the wire rope body using an improved Transformer model, identifies the location and number density of internal broken wire defects, and simultaneously calculates the dynamic load eccentricity by fusing the relative displacement sequence of the hook and the load with the torque sequence, and constructs a three-dimensional risk tensor by combining wind speed and turbulence intensity. The risk prediction and assessment dynamic control module analyzes the three-dimensional risk tensor through a pre-trained chaotic prediction model to obtain the deviation between the predicted trajectory and the baseline trajectory. It then analyzes the deviation and dynamically triggers the electromagnetic locking mechanism of the fall arrestor based on the analysis results, which are simultaneously pushed to the hoisting monitoring platform.

2. The intelligent risk perception system for hoisting large equipment according to claim 1, characterized in that, When collecting the strain distribution sequence of a steel wire rope body using a fiber optic grating sensor array, the strain data of the rope body under tensile and bending combined loads during the hoisting process are collected, and the strain distribution sequence is output. The connection point between the hook and the load is continuously tracked by millimeter-wave radar, and a relative displacement polar coordinate sequence is output. By transforming the polar coordinates into three-dimensional displacements in the Cartesian coordinate system, the obtained three-dimensional displacements are combined to obtain a relative displacement sequence. When acquiring the torque sequence output in real time by the torque limiter using a high-precision torque sensor, the voltage signal output by the high-precision torque sensor is also acquired. The torque-voltage linear relationship is fitted and the actual torque is output. The actual torques are sorted and combined according to the acquisition time sequence to obtain the torque sequence. When collecting environmental interference parameters, an ultrasonic anemometer is deployed next to the radar installation location to collect wind speed and turbulence intensity. Air humidity is collected using a temperature and humidity sensor; Wind speed, turbulence intensity, and air humidity constitute environmental disturbance parameters.

3. The intelligent risk perception system for hoisting large equipment according to claim 2, characterized in that, The acquired sequence data is preprocessed, including temperature drift compensation for strain distribution sequence, clutter suppression and smoothing for relative displacement sequence, and correction for displacement caused by wind speed interference.

4. The intelligent risk perception system for hoisting large equipment according to claim 3, characterized in that, When generating the multidimensional fusion dataset, the timestamps of each sensor are linearly interpolated and corrected to obtain the corrected timestamps; the axial position x of the fiber optic grating sensor is converted into global coordinates; the Cartesian coordinates of the millimeter-wave radar are directly incorporated into the global coordinate system to realize the spatial correlation between strain distribution and displacement. By integrating the preprocessed strain distribution sequence, corrected displacement sequence, torque sequence, and environmental disturbance parameters, a structured multidimensional fusion dataset is generated.

5. The intelligent risk perception system for hoisting large equipment according to claim 4, characterized in that, When extracting features from the strain distribution sequence of a wire rope, the strain feature vector is output from the last layer of the improved Transformer encoder; the number of broken wires K is calculated using the following expression: It is a rounding function; For quantity density.

6. The intelligent risk perception system for hoisting large equipment according to claim 5, characterized in that, The load eccentricity is calculated based on the output torque of the torque limiter and the total lifting weight. Convert the load eccentricity into a dimensionless load eccentricity coefficient C.

7. The intelligent risk perception system for hoisting large equipment according to claim 6, characterized in that, When constructing the three-dimensional risk tensor, the defect severity S is obtained by weighting the number density of broken wires and their location. The environmental disturbance weight W is obtained based on the calculation of wind speed and turbulence intensity. The obtained defect severity S, load eccentricity coefficient C, and environmental disturbance weight W are used as three dimensions to construct a three-dimensional risk tensor: ;in, This is an element-wise product operation.

8. The intelligent risk perception system for hoisting large equipment according to claim 7, characterized in that, The three-dimensional risk tensor R is mapped to the typical initial interval of the Lorenz system; the evolution of the risk trajectory is described using the standard Lorenz equation.

9. The intelligent risk perception system for hoisting large equipment according to claim 8, characterized in that, Obtain the future risk trajectory output by the chaotic model and convert it into a risk tensor dimension: ； in, The predicted trajectories at the mixed time steps are respectively Defect severity, load eccentricity coefficient, and environmental interference weight; These are the predicted values ​​of the Lorenz system state variables output by the chaotic prediction model.

10. The intelligent risk perception system for hoisting large equipment according to claim 9, characterized in that, The deviation between the predicted trajectory and the reference trajectory is measured by weighted Euclidean distance. The calculated deviation is analyzed, and the electromagnetic locking mechanism is dynamically triggered based on the analysis results. If the deviation is greater than Y1 and the duration is greater than or equal to Y2, the locking signal of the fall arrestor is triggered; Y1 and Y2 are both real numbers greater than 0.

Images