A method and system for intelligent control of cranes

Abstract

This invention relates to a method and system for intelligent control of cranes, belonging to the field of crane control technology. It collects crane operating status data through sensors, analyzes and models the data, and simultaneously calculates dynamic response weights and static feature weights. The dynamic response weights quickly capture key factors affecting the crane's status and trigger rapid responses; the static feature weights maintain basic data classification and reduce redundant calculations. Combining dynamic and static risks, a risk index is calculated in real time, and the risk level is adaptively adjusted. Furthermore, the risk assessment model is expanded through multi-source data acquisition, operational difficulty assessment, and scene-sensitive feature extraction. Adaptive Kalman filtering eliminates sensor noise, and weight allocation is dynamically adjusted. This invention significantly improves data accuracy and assessment flexibility, achieving intelligent control of the crane's operating status, enhancing operational safety and efficiency, and optimizing maintenance costs and resource allocation.

Description

Technical Field

[0001] This invention relates to a method and system for intelligent control of cranes, belonging to the field of crane control technology. Background Technology

[0002] With the continuous development of existing industries, cranes have become a widely used type of mechanical equipment in industrial and mining enterprises, such as mining, metallurgy, coal smelting, cement, and power industries. Specifically, cranes are mechanical equipment used for lifting, transporting, loading, unloading, and installing materials.

[0003] However, in existing technologies, crane control mainly relies on sensor data acquisition and static model analysis, which cannot dynamically adapt to the time-varying nature of sensor noise, leading to data distortion and affecting the accuracy of subsequent analysis. Sensor weights in traditional models are mostly preset fixed values ​​and cannot be dynamically adjusted according to real-time operating conditions. For example, when there are sudden load changes or a sharp increase in ambient wind speed, static weights are difficult to capture key risk indicators, resulting in delayed risk assessment. Traditional maintenance strategies lack a refined grading mechanism, often leading to increased costs due to over-maintenance or safety hazards due to insufficient maintenance. Summary of the Invention

[0004] This invention provides a method and system for intelligent control of cranes to solve the problems of insufficient sensor noise processing, rigid weight allocation, and imbalance between maintenance costs and safety in the prior art.

[0005] This invention provides a method and system for intelligent control of cranes, comprising:

[0006] 8) Collect crane operating status data through sensors and analyze the collected data;

[0007] 9) Build a working state model based on the processed collected data and calculate the weights corresponding to the sensor data;

[0008] 10) The sensor weights are analyzed, including dynamic response weights and static feature weights. The dynamic response weights comprehensively analyze the factors affecting the working state of the tower crane and trigger a rapid response. The static feature weights analyze the motion parameters, maintain the basic data classification, and reduce redundant calculations.

[0009] 11) Combine dynamic and static risks to calculate the real-time risk index, classify risk levels, and adaptively adjust according to the degree of deviation of real-time risk from the safety threshold. Control the adjustment rate, balance the weight of sudden risks and long-term trends, and prevent a single indicator from dominating decision-making.

[0010] 12) Conduct multi-source data collection, assess the difficulty of operation, calculate the entropy value of action combination, expand the static weighted Bayesian network, and integrate the assessment results into the risk assessment model;

[0011] 13) Extract scene-sensitive features, expand the static weighted Bayesian network, compare the predicted lifetime with the actual damage data, and dynamically update the model parameters;

[0012] 14) Send the analyzed working status data of the crane to the operator for data interaction.

[0013] Preferably, in step 1), the acquired data is subjected to adaptive Kalman filtering to eliminate sensor noise, and time synchronization is performed through multi-source data spatiotemporal registration, followed by filtering iteration.

[0014] Preferably, in step 2), the state equation and observation equation are constructed, the crane coordinate system is established, the installation error is eliminated by multi-sensor data fusion, the absolute coordinates of the origin of the sensor coordinate system are obtained, and the working state model is built based on the processed collected data.

[0015] Preferably, in step 3), initial weight values ​​are assigned to each indicator that affects the safety of crane operation based on historical accident data. Dynamic response weights are calculated and updated in real time by collecting data of each monitoring indicator. Static feature benchmark weights are assigned to different sensor data. When the crane undergoes major maintenance, component replacement, or significant changes in the working environment, the weights are reassessed and adjusted. A relationship model is constructed based on the dynamic response weights and static feature weights.

[0016] Preferably, in step 4), a Bayesian network causal relationship graph is established based on static weights. When major maintenance or environmental changes occur, the Bayesian network parameters are updated, and the maintenance cost is minimized under the constraints of the particle swarm optimization algorithm, thereby balancing cost and security.

[0017] Preferably, in step 5), the complexity score is calculated by a quantitative calculation model, and the hierarchical clustering method is used to divide the work into 3 levels to establish work shift groups; the groups are integrated into the dynamic weight calculation, and the grouping nodes of work complexity and work shifts are added to the Bayesian network to optimize the risk assessment model, so as to avoid misidentifying environmental interference as equipment failure and causing accidental shutdown.

[0018] Preferably, in step 6), the real-time risk assessment model introduces a scenario-specific risk index, expands the static weighted Bayesian network, adds a construction stage and environmental complexity root node, establishes a physical-data fusion model and a vibration fatigue damage model to enhance the life prediction model, simulates under extreme working conditions, compares the predicted life with the actual damage data, and dynamically updates the model parameters.

[0019] Preferably, in step 7), when the crane's working data is greater than the first threshold, the current tower crane's working mode is considered normal; when the crane's working data is less than the first threshold, the current crane's working mode is considered abnormal, and the working status control and maintenance data are sent to the backend server.

[0020] Preferably, the extracted features are standardized, a time window is set to divide the continuous data stream into time series segments, and the segments are input into a bidirectional LSTM neural network. The forward layer captures historical risk accumulation, the backward layer captures future abnormal trends, the features at each time step are dynamically weighted, and a real-time risk index is output.

[0021] Preferably, the real-time monitoring of load weight and volume, the inclinometer recording the path curvature radius and operation sequence, the environmental sensor collecting wind speed and temperature, the continuous variables being Z-score standardized, and the discrete variables being 0-1 normalized.

[0022] This application also provides a system for intelligent control of a crane, including sensors and inclinometers for recording. The system collects working status data through the sensors, analyzes the collected data, builds a working status model, calculates the dynamic response weights and static feature weights of the corresponding data, calculates a real-time risk index by combining dynamic and static risks, classifies risk levels, and the sensors and inclinometers collect multi-source data, expands the static weighted Bayesian network, extracts scene-sensitive features, dynamically updates model parameters, and sends the working status data to the operator for data interaction.

[0023] The beneficial effects of this invention are:

[0024] This invention provides a method and system for intelligent crane control. Through adaptive Kalman filtering, noise parameters are dynamically adjusted to adapt to data changes, effectively eliminating sensor noise and significantly improving data accuracy. The introduction of a fading factor method or square root adaptive filtering further enhances the filtering effect, ensuring more reliable sensor data after denoising. Employing timestamp-based linear or cubic spline interpolation, combined with a timestamp error compensation model and least squares method, asynchronous sensor data can be efficiently aligned to a unified time series, significantly reducing synchronization errors. A crane coordinate system is established, and a coordinate transformation matrix is ​​used to achieve spatial alignment of multi-sensor data, simplifying the data processing flow and improving the accuracy and efficiency of data fusion. Dynamic response weights are calculated and updated in real time based on factors affecting the crane's operating state, ensuring the timeliness and accuracy of weight allocation. Static feature weights are allocated based on key motion parameter data and re-evaluated and adjusted under specific conditions, avoiding the limitations of fixed traditional weight allocation and improving the accuracy and flexibility of evaluation. The combination of dynamic and static weights balances real-time response with basic logic. This system reduces redundant computation and improves overall efficiency. The application of a bidirectional LSTM neural network can simultaneously process forward and reverse time series information, capture historical risk accumulation and future abnormal trends, dynamically assign weights to features at each time step, and output a real-time risk index, improving the flexibility and accuracy of assessment. By dynamically calculating risk levels and dynamically adjusting coefficients, it achieves refined risk level classification. According to different risk levels, corresponding maintenance measures are taken, which not only ensures the safe operation of the crane but also reduces maintenance costs. Based on the comparison results of crane working data with preset thresholds, it can promptly determine whether the crane's working mode is normal, providing operators with effective fault warnings. Data is exchanged with operators and the back-end server, realizing real-time data sharing and remote monitoring, facilitating timely maintenance measures and fault repair, and improving the safety and reliability of the crane. Based on the particle swarm optimization algorithm, it minimizes maintenance costs under the constraint of risk levels below a certain threshold, realizing refined resource allocation. The introduction of a scenario-specific risk index can more accurately assess the risk level in the current scenario, providing strong support for decision-making. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method and system for intelligent control of a crane according to the present invention. Detailed Implementation

[0026] Example 1

[0027] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0028] This invention provides a method and system for intelligent control of a crane, comprising:

[0029] The crane's operating status data is collected through sensors, and the collected data is then analyzed.

[0030] Specifically, sensors collect data on the crane's load changes, motion parameters, environmental parameters, and structural stress. Adaptive Kalman filtering is applied to the collected data to eliminate sensor noise. Time synchronization is achieved through multi-source data spatiotemporal registration. Specifically, based on the data collected by each sensor, its measurement noise covariance matrix and process noise covariance matrix are estimated online. The noise parameters are dynamically adjusted to adapt to data changes using the fading factor method or square root adaptive filtering. State equations and observation equations are then constructed, as detailed below:

[0031]

[0032] in

[0033] : System state vector (lifting weight, motion parameters);

[0034] : Observation vector;

[0035] Process noise;

[0036] : Measurement noise;

[0037] The filtering iteration is performed, and the denoised sensor data is output by executing prediction and update steps. The measurement noise covariance matrix and process noise covariance matrix are dynamically optimized by residual analysis.

[0038] Timestamp-based linear interpolation or cubic spline interpolation aligns asynchronous sensor data to a unified time series. For asynchronous sampling sensors, a timestamp error compensation model is introduced, and the synchronization accuracy is optimized using the least squares method.

[0039] Establish a crane coordinate system, convert the data from each sensor into physical quantities in a unified coordinate system, and use a coordinate transformation matrix. To achieve spatial alignment of multi-sensor data, external measuring equipment can be used as the direct translation vector when the crane is stationary. By utilizing multi-sensor data fusion, optimization is achieved through the least squares method. Parameters are used to eliminate installation errors and obtain the absolute coordinates of the sensor coordinate system origin; a working state model is built based on the processed acquired data, specifically:

[0040]

[0041] in:

[0042] Overall work status assessment value;

[0043] Number of sensors;

[0044] : No. Data collected and processed by a single sensor;

[0045] : No. The weights corresponding to each sensor's data;

[0046] The weights corresponding to the sensor data include dynamic response weights and static feature weights. The dynamic response weights comprehensively analyze the factors affecting the working status of the tower crane and trigger rapid responses, including load changes, environmental parameters, and structural stress. For each factor, corresponding monitoring indicators are defined. Load changes include lifting torque and lifting weight. Environmental parameters use ambient wind speed as the direct monitoring indicator. Structural stress includes the stress values ​​of the boom and standard tower sections. Based on historical accident data, initial weight values ​​are assigned to each indicator that affects the safety of crane operation. By collecting data of each monitoring indicator in real time, the dynamic response weights are calculated and updated in real time. The weights are updated after each collection to ensure the timeliness of the weights.

[0047] Static feature weights are used to analyze motion parameters, maintain basic data classification, and reduce redundant calculations. Motion parameters include three working modes of the crane: extension, rotation, and support. Key data types are determined for each mode. During extension, the extension displacement and extension speed of the components are key data. During rotation, the rotation angle and rotation speed are key data. During support, the pressure and verticality of the support components are key data. Based on the key data, baseline weights are assigned to different sensor data. When the crane undergoes major maintenance, component replacement, or significant changes in the working environment, the weights are reassessed and adjusted.

[0048] A relational model is constructed based on dynamic response weights and static feature weights. Dynamic response weights ( The static feature weights are adjusted based on real-time changes in the working status. Based on a fixed working mode preset. Introducing dynamic weighting coefficients ( and static weight coefficients Both satisfy ,and The dynamic response weights and static feature weights are combined through the following relationship model:

[0049]

[0050] Dynamic weights quickly capture anomalies, while static weights maintain basic logic, ensuring the depth of key data while accelerating the overall state determination.

[0051] The parsed working status data of the crane is sent to the operator for data interaction. Specifically, when the crane's working data is greater than the first threshold, the current tower crane's working mode is considered normal, and the crane's working status in the current environment is within the normal range. When the crane's working data is less than the first threshold, the current crane's working mode is considered abnormal. The operator then disables the crane's operation and performs fault repair, and sends the working status control and maintenance data to the backend server.

[0052] In practical use, sensors are used to collect working status data of the crane, such as load changes (e.g., lifting torque, lifting weight), motion parameters (extension displacement, rotation angle, etc.), environmental parameters (ambient wind speed), and structural stress (stress values ​​of the boom and standard tower sections). Sensor noise is eliminated through adaptive Kalman filtering, adapting to real-time data changes. The measurement noise covariance matrix and process noise covariance matrix are estimated online. The noise parameters are dynamically adjusted using the fading factor method or square root adaptive filtering. In the filtering iteration, the noise parameters are dynamically optimized through prediction and update steps to improve the noise reduction accuracy. Based on timestamp-based linear or cubic spline interpolation, combined with a timestamp error compensation model and least squares method, asynchronous sensor data is aligned to a unified time series. This ensures a unified time series, reduces synchronization errors, and avoids time differences during data fusion caused by different sensor sampling rates or asynchronous timestamps, which affect real-time performance and decision accuracy. A crane coordinate system is established, and a coordinate transformation matrix is ​​used to convert the data from each sensor into physical quantities in a unified coordinate system. Parameters are optimized through external measurement equipment and multi-sensor data fusion to obtain the absolute coordinates of the origin of the sensor coordinate system, determining the dynamic response weights and static feature weights corresponding to the sensor data. The dynamic response weights are calculated and updated in real time based on factors affecting the crane's operating state, while the static feature weights are allocated based on key motion parameter data and reassessed and adjusted under specific conditions. This avoids the inaccurate assessments in dynamic environments caused by traditional fixed weight allocations that cannot be adjusted according to real-time operating conditions. The combination of dynamic and static weights balances real-time response and basic logic, reduces redundant calculations, and improves efficiency. Both are combined through a relational model to build a working state model based on the processed collected data, and the parsed working state data is sent to the operator. When the crane's operating data is greater than the first threshold, the operating mode is considered normal; when it is less than the first threshold, the operating mode is considered abnormal, the operator stops the crane from operating and performs fault repair, and at the same time sends the operating status control and maintenance data to the backend server.

[0053] Compared with existing technologies, adaptive Kalman filtering can dynamically adjust noise parameters to adapt to data changes, effectively eliminating sensor noise and improving data accuracy. Introducing a fading factor method or square root adaptive filtering further enhances the filtering effect, ensuring more reliable denoised sensor data. Using timestamp-based linear or cubic spline interpolation, combined with a timestamp error compensation model and least squares method, asynchronous sensor data can be efficiently aligned to a unified time series, significantly reducing synchronization errors. Establishing a crane coordinate system and using a coordinate transformation matrix to achieve spatial alignment of multi-sensor data not only simplifies the data processing flow but also improves the accuracy and efficiency of data fusion. Dynamic response weights are determined based on factors affecting crane operation. The state factors are calculated and updated in real time, ensuring the timeliness and accuracy of weight allocation. Static feature weights are allocated based on key motion parameter data and re-evaluated and adjusted under specific circumstances, avoiding the limitations of fixed weight allocation in traditional methods and improving the accuracy and flexibility of evaluation. The combination of dynamic and static weights balances real-time response and basic logic, reduces redundant calculations, and improves overall efficiency. Based on the comparison results between crane working data and preset thresholds, the crane's working mode is promptly determined to be normal, providing operators with effective fault warnings. By exchanging data with operators and the back-end server, real-time data sharing and remote monitoring are achieved, facilitating timely maintenance measures and fault repair, and improving the safety and reliability of the crane.

[0054] Example 2

[0055] In the above embodiments, the dynamic response weights are calculated and updated in real time according to the factors affecting the working state of the crane, ensuring the timeliness and accuracy of the weight allocation. The static feature weights are allocated based on the key data of motion parameters and are re-evaluated and adjusted under specific circumstances, avoiding the limitations of the fixed weight allocation in traditional methods and improving the accuracy and flexibility of the evaluation. The embodiments of this application are optimized based on the above embodiments.

[0056] In this embodiment, the dynamic response weight extracts the load mutation rate, wind speed gradient, and stress growth rate. The extracted features are standardized by Z-score. A time window is set to divide the continuous data stream into time series segments and input them into a bidirectional LSTM neural network. The bidirectional LSTM neural network processes both forward and backward time series information. The forward layer captures historical risk accumulation, and the backward layer captures future abnormal trends. The features of each time step are dynamically weighted, and a real-time risk index is output.

[0057] Static weight offline modeling is performed based on static weights. Historical maintenance records, fault data, and component replacement logs are collected to establish a Bayesian network causal relationship graph. The root node represents the component aging stage and working mode, the intermediate nodes represent sensor data, and the leaf nodes represent fault types. The maximum likelihood estimation method is used to learn the conditional probabilities between nodes from historical data. When major maintenance or environmental changes occur, the Bayesian network parameters are updated.

[0058] Real-time risk index calculation combining dynamic and static risks:

[0059]

[0060] in:

[0061] Real-time risk index calculation;

[0062] Dynamic risks;

[0063] Static risks

[0064] Dynamically calculated using the Sigmoid function Dynamically adjust the coefficients:

[0065]

[0066] in:

[0067] k=0.1;

[0068] Safety threshold;

[0069] =1- ;

[0070] Risk level classification, Extend maintenance cycle by 30%; 30% ≤ For samples with a mortality rate below 70%, maintain the original cycle and increase the frequency of online monitoring. ≥70% requires immediate shutdown for maintenance; based on particle swarm optimization algorithm, in To minimize maintenance costs under the constraint of <70%, after each maintenance, the actual fault data is compared with the predicted results, and the parameters of the LSTM and Bayesian networks are updated for data feedback loop. When the dynamic weights detect an anomaly but the static weights show that the remaining lifespan of the component is >80%, an "observation period" is triggered. Decisions are made after verification through multi-source data to avoid frequent maintenance when the dynamic weights detect anomalies, which would increase costs. Historical data from the static weights can help determine whether immediate maintenance is needed or can be postponed, thus balancing cost and safety.

[0071] Dynamic data acquisition during operation: Real-time monitoring of crane operating status parameters, including dynamic features such as load mutation rate, wind speed gradient, and stress growth rate. Static data collection: Offline organization of historical maintenance records, fault data, component replacement logs, and other static information. Dynamic features are Z-score standardized to eliminate dimensional differences. Continuously acquired dynamic data is segmented into fixed-length time series segments to adapt to the temporal characteristics of equipment operation, improving sensitivity to transient anomalies. Processing is achieved through a bidirectional LSTM neural network: the forward layer analyzes historical data to capture risk accumulation trends (e.g., continuous stress growth), and the backward layer predicts future abnormal signals (e.g., the probability of wind speed mutations). Weights are assigned to features at each time step, outputting a real-time dynamic risk index. Maximum likelihood estimation is used to learn the conditional probabilities between nodes from historical data. When major maintenance or environmental changes occur, the Bayesian network is retrained, outputting a static risk index, dynamically calculated using the Sigmoid function, and adaptively adjusted based on the degree of real-time risk deviation from the safety threshold. The adjustment rate is controlled to balance the weights of sudden risks and long-term trends, preventing a single indicator from dominating decision-making. Under the constraint of <70%, the particle swarm optimization algorithm is used to generate a maintenance plan to minimize the total cost. After each maintenance, the actual fault data is compared with the prediction results, and the parameters of the bidirectional LSTM and Bayesian network are updated. If the dynamic weights detect an anomaly, but the static weights show that the remaining life of the component is >80%, the "observation period" is triggered to verify the authenticity of the risk by integrating multi-sensor data, avoiding blind maintenance and realizing fine-grained resource allocation.

[0072] Compared to existing technologies, this method uses dynamic response weights, which are calculated and updated in real time based on factors affecting the crane's operating status. This ensures the timeliness and accuracy of weight allocation. Static feature weights are reassessed and adjusted under specific circumstances, avoiding the limitations of fixed weight allocation in traditional methods. The application of a bidirectional LSTM neural network can simultaneously process forward and reverse time series information, capturing historical risk accumulation and future abnormal trends. It dynamically assigns weights to features at each time step and outputs a real-time risk index, improving the flexibility and accuracy of the assessment. This is further enhanced by using a Sigmoid function. The function dynamically calculates the risk level and dynamically adjusts the coefficients, achieving refined risk level classification. Based on different risk levels, corresponding maintenance measures are taken, such as extending the maintenance cycle, maintaining the original cycle while increasing the online monitoring frequency, or immediately shutting down for maintenance. This ensures the safe operation of the crane while reducing maintenance costs. When the dynamic weight detects an anomaly but the static weight shows that the component's remaining lifespan is >80%, an "observation period" is triggered. Decisions are made after verification using multi-source data, avoiding frequent maintenance that increases costs. The introduction of static weights helps determine whether immediate maintenance is needed or can be postponed, thus achieving a balance between cost and safety. Based on the particle swarm optimization algorithm, maintenance costs are minimized under the constraint of a risk level below 70%, achieving refined resource allocation. After each maintenance, the actual fault data is compared with the predicted results, and the parameters of the bidirectional LSTM and Bayesian network are updated, improving the accuracy of prediction and maintenance.

[0073] Example 3

[0074] The above embodiments take corresponding maintenance measures according to different risk levels and make decisions after verification through multi-source data, thus avoiding frequent maintenance and increased costs. The embodiments of this application are based on the above embodiments and are optimized to a certain extent.

[0075] In this embodiment, multi-source data acquisition is performed to obtain the obstacle density of the hoisting path, real-time monitoring of load weight and volume, inclinometer recording of path curvature radius and operation sequence, environmental sensors to collect wind speed and temperature, Z-score standardization is performed on continuous variables such as weight, volume, and curvature radius, and 0-1 normalization is performed on discrete variables (such as the number of obstacles).

[0076] To assess the difficulty of the operation, the entropy value of the action combination is calculated:

[0077]

[0078] in:

[0079] : for the first The probability of occurrence of a certain action sequence;

[0080] The complexity score is calculated using a quantitative computational model.

[0081] ;

[0082] in:

[0083] : weight coefficient;

[0084] Volume index;

[0085] Obstacle density;

[0086] Curvature points;

[0087] Environmental factors;

[0088] Hierarchical clustering was used to divide the data into three levels: low complexity (score < 0.4), medium complexity (score < 0.7, score ≥ 0.7), and high complexity (score ≥ 0.7).

[0089] Spatiotemporal features are extracted to establish shift grouping; these groups are integrated into dynamic weight calculation, with a real-time weight update formula:

[0090]

[0091] in:

[0092] : This is the grouped sensitivity coefficient matrix;

[0093] : 0.1-0.3 (dynamically adjusted);

[0094] The group sensitivity coefficient includes high-complexity operations and night shift operations. High-complexity operations include stress sensor weights, while night shift operations include visibility sensor weights.

[0095] Extend the static weighted Bayesian network by adding grouping nodes for job complexity and shift, where job complexity includes low, medium, and high, and shifts include day, afternoon, and night. Calculate the joint probability of grouping and fault type:

[0096]

[0097] The optimized risk assessment model and the real-time risk index are as follows:

[0098]

[0099] in;

[0100] Complexity correction factor;

[0101] : Shift adjustment factor;

[0102] Grouping correction factor for high-complexity tasks: Dynamic risk R × 1.2;

[0103] Grouping correction factor for night shift work: static risk S × 1.1;

[0104] During use, obstacle density is obtained, external environmental parameters are collected, the inclinometer records the path curvature radius, and the load weight and volume are monitored in real time. The weight, volume, and curvature radius are standardized using Z-score, and the number of obstacles is normalized using 0-1 to achieve data preprocessing.

[0105] Analyze historical action sequences and calculate the probability of occurrence for each action sequence. The system calculates the entropy value of action combinations; the higher the entropy value, the more complex the operation. It then calculates complexity scores and classifies complexity levels using hierarchical clustering to achieve a three-level complexity classification. A scenario-adaptive strategy is implemented, and spatiotemporal feature extraction is combined with shift assignments and complexity levels to group work shifts. This avoids the model omitting factors such as reduced visibility and personnel fatigue during night shifts, which could lead to environmental interference being mistaken for equipment malfunctions and causing unintended shutdowns. The system establishes work shift groups, captures day-night environmental differences, updates dynamic weights, expands Bayesian networks and static risk modeling, adds work complexity and work shift nodes, and calculates joint probabilities. When new work data is added, the conditional probability table is recalculated to optimize fault prediction.

[0106] Compared with existing technologies, this method comprehensively assesses the crane's operating status by collecting multi-source data, including information such as obstacle density along the lifting path, load weight and volume, path curvature radius, operation sequence, wind speed, and temperature. This provides a more complete picture of the crane's operation and improves the comprehensiveness and accuracy of the assessment. By calculating the entropy value and complexity score of the action combination, the method evaluates the operational difficulty and integrates the assessment results into the risk assessment model. This approach more accurately reflects the impact of operational difficulty on the crane's operating status and further improves the accuracy of the assessment.

[0107] Example 4

[0108] The above embodiments perform macro-grouping of dynamic weight adjustment, and macro-group the complexity of the work plan and the number of work shifts to optimize the safety balance and ultimately reduce maintenance costs. The embodiments in this application are based on the above embodiments with certain optimizations.

[0109] This embodiment extracts scene-sensitive features, including features from water conservancy projects, traffic engineering projects, and cross-scene features. Features from water conservancy projects include water level change rate, water flow impact force, and scour depth. Features from traffic engineering projects include traffic flow in adjacent lanes, vibration interference, and dynamic loads. Features from cross-scene features include multi-device collaboration. The specific complexity is as follows:

[0110]

[0111] in

[0112] : Calculate the distance between devices;

[0113] S: Synchronization rate of actions between devices;

[0114] Z-score standardization is used for continuous variables (such as water level and traffic flow); 0-1 normalization is used for discrete variables (such as scour depth level).

[0115] Extending dynamic weights:

[0116]

[0117] in:

[0118] Scene For sensors Sensitivity coefficient;

[0119] Risk response coefficient (0.2-0.4, dynamically adjusted);

[0120] The real-time risk assessment model introduces a scenario-specific risk index, specifically:

[0121]

[0122] in:

[0123] Basic risk index (output of the original model);

[0124] : Scenario risk amplification factor;

[0125] The static weighted Bayesian network is extended by adding a construction phase and an environmental complexity root node, and the conditional probability table is updated, specifically as follows:

[0126]

[0127] A physical-data fusion model and a vibration fatigue damage model were established to enhance the life prediction model. The specific physical-data fusion model is as follows:

[0128]

[0129] in:

[0130] Theoretical life based on fracture mechanics;

[0131] LSTM-based prediction of remaining lifetime;

[0132] The vibration fatigue damage model is as follows:

[0133]

[0134] in:

[0135] : Actual number of vibrations;

[0136] Fatigue life vibration count;

[0137] Extreme working conditions are simulated in the simulation model, the predicted life is compared with the actual fatigue damage, and the model parameters are updated accordingly.

[0138] During use, features of water conservancy projects, transportation projects, and cross-scenario features are collected. Z-score standardization is used for continuous variables (such as water level and traffic flow), and 0-1 normalization is used for discrete variables (such as scour depth level). This enables multi-scenario feature data collection and preprocessing. Dynamic weight expansion and scenario-specific risk assessment, as well as static weight Bayesian network expansion and fault prediction optimization, physical-data fusion life prediction and vibration fatigue modeling are performed. Simulations are conducted under extreme working conditions, and the predicted life and actual damage data are compared to dynamically update the model parameters.

[0139] Compared with existing technologies, this method extracts features from hydraulic engineering (such as water level change rate, water flow impact force, and scour depth), traffic engineering (such as traffic flow in adjacent lanes, vibration interference, and dynamic loads), and cross-scenario features (such as the complexity of multi-equipment collaboration). This allows for a more accurate reflection of the operating environment and requirements under different scenarios, aiding in subsequent risk assessment and weight allocation, thereby improving operational safety and efficiency. By introducing the sensitivity coefficient and risk response coefficient of the scene to sensors, dynamic weights are extended, enabling more flexible responses to risk changes under different scenarios. The introduction of a real-time risk assessment model, combined with a scenario-specific risk index, can more accurately assess the risk level under the current scenario, providing strong support for decision-making. By extending the static weighted Bayesian network, adding root nodes such as construction stage and environmental complexity, and updating the conditional probability table, the model becomes more complete and can more accurately reflect the actual situation. The introduction of a physical-data fusion model and a vibration fatigue damage model further enhances the accuracy of the life prediction model. This fusion model combines the theoretical life of fracture mechanics with the remaining life prediction of LSTM, enabling a more comprehensive consideration of the impact of various factors on equipment life.

[0140] Example 5

[0141] This embodiment provides a system for intelligent control of a crane, including sensors and an inclinometer. The sensors collect data on load changes, motion parameters, environmental parameters, and structural stress of the crane. The collected data is analyzed to build a working state model, calculating dynamic response weights and static feature weights for the corresponding data. The dynamic response weights comprehensively identify factors affecting the working state of the tower crane, triggering rapid responses, including load changes, environmental parameters, and structural stress. For each factor, corresponding monitoring indicators are defined. The static feature weights analyze motion parameters, maintaining basic data classification and reducing redundant calculations. Motion parameters include three working modes: crane extension, rotation, and support. Key data types are determined for each mode. A real-time risk index is calculated by combining dynamic and static risks. The system calculates and classifies risk levels. The sensors and inclinometers collect multi-source data, acquiring obstacle density along the hoisting path, monitoring load weight and volume in real time, recording the path curvature radius and operational sequence, and using environmental sensors to collect wind speed and temperature. An extended static weighted Bayesian network is used to extract scene-sensitive features, dynamically update model parameters, and send operational status data to the operator for data interaction. When the crane's operational data exceeds a first threshold, the current tower crane's operating mode is considered normal, and the crane's operating status in the current environment is within the normal range. When the crane's operational data falls below the first threshold, the current crane's operating mode is considered abnormal. The operator then disables the crane's operation and performs fault repair, sending operational status control and maintenance data to the backend server.

[0142] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for intelligent control of a crane, characterized in that, Comprise: 1) Collect the working state data of the crane through the sensor, and analyze the collected data; 2) According to the processed collected data, a working state model is built, as follows: Wherein: : working state comprehensive evaluation value; : sensor quantity; : data collected and processed by the first sensor; : data collected and processed by the first sensor corresponding weight; 3) Calculate the weight corresponding to the sensor data, analyze the sensor weight, including dynamic response weight and static feature weight, wherein the dynamic response weight comprehensively analyzes the factors affecting the working state of the tower crane, triggers a rapid response, and the static feature weight analyzes the motion parameters to maintain the classification of basic data and reduce redundant calculations; the dynamic response weight ( ) changes according to the real-time working state, and the static feature weight ( ) is preset based on the fixed working mode; introduce dynamic weight coefficient ( and static weight coefficient , the dynamic response weight and the static feature weight are combined through the following relationship model: 4) Real-time risk index calculation is performed in combination with dynamic risk and static risk, risk level division is performed, self-adaptive adjustment is performed according to the degree of deviation of the real-time risk from the safety threshold, control adjustment rate is performed, the weight of sudden risk and long-term trend is balanced, and single index dominant decision is prevented; Real-time risk index calculation is performed in combination with dynamic risk and static risk: Wherein: : Real-time risk index calculation; : Dynamic risk; : Static risk; dynamically calculated using Sigmoid function , dynamic adjustment of coefficients: wherein: k = 0.1; : safety threshold; = 1- ; risk level classification, <30%: extend maintenance cycle; 30%≤ <70% maintain original cycle, increase online monitoring frequency; ≥70% immediate shutdown for maintenance; 5) Multi-source data collection is performed, operation difficulty evaluation is performed, action combination entropy value is calculated, static weight Bayesian network is expanded, and the evaluation result is integrated into the risk evaluation model; Action combination entropy value is calculated: wherein: : is the probability of occurrence of the sequence of actions of the first kind; the complexity score is calculated by means of a quantification model. ; wherein: : weight coefficient; : volume coefficient; : obstacle density; : curvature point number; : environmental factor; using hierarchical clustering method to divide 3 levels, wherein low complexity: score <0.4, medium complexity: 0.4≤score<0.7, high complexity: score≥0.7; by calculating the action combination entropy value and complexity score, the operation difficulty is evaluated; the work shift grouping is established; the grouping is integrated into dynamic weight calculation, the work complexity and the work shift grouping nodes are added in the static Bayesian network, the risk assessment model is optimized, and the environmental disturbance is avoided from being identified as equipment failure to cause false shutdown; Optimize the risk evaluation model, and the real-time risk index is specifically: wherein; : complexity correction coefficient; : shift correction coefficient; 6) Scene sensitive features are extracted, the scene sensitive features include water conservancy engineering feature extraction, traffic engineering feature extraction, and cross scene feature extraction, wherein the water conservancy engineering feature extraction includes water level change rate, water flow impact force and scouring depth; The traffic engineering feature extraction includes adjacent lane traffic flow, vibration interference and dynamic load; The cross scene feature extraction includes multi-device cooperation complexity, and the specific physical-data fusion model is: wherein, : inter-computing device distance; S: inter-device motion synchronization rate; extension of dynamic weights; introduction of real-time risk assessment models into scenario-specific risk indices; extension of static weight Bayesian networks; The vibration fatigue damage model is specifically: wherein: : Theoretical life based on fracture mechanics; : Residual life prediction based on LSTM; 7) The analyzed working state data of the crane is sent to the operator for data interaction, when the crane working data is greater than the first threshold value, it is determined that the current tower crane working mode is normal; When the crane working data is less than the first threshold value, it is determined that the current crane working mode is not normal, and the working state control and maintenance data are sent to the background server. Wherein: : actual number of vibrations; : fatigue life vibration number; simulate extreme working conditions in the simulation model, compare the predicted life with the actual fatigue damage, and update the model parameters; In step 1), the collected data is adaptively Kalman filtered to eliminate sensor noise, time synchronization is performed through multi-source data space registration, and filtering iteration is performed.

2. A method for intelligent control of a crane according to claim 1, characterized in that: In step 2), the state equation and the observation equation are constructed, the crane coordinate system is established, the installation error is eliminated by using multi-sensor data fusion, the absolute coordinates of the sensor coordinate system origin are obtained, and the working state model is built according to the processed collected data.

3. A method for intelligent control of a crane as claimed in claim 1, characterized in that: In step 3), according to the historical accident data, the influence degree of each index affecting the safety of the crane is assigned an initial weight value, the dynamic response weight is calculated and updated in real time by real-time collection of each monitoring index data, the static characteristic reference weight is allocated to different sensor data, when the crane is subjected to major maintenance, component replacement or major changes in working environment, the weight is reevaluated and adjusted, and the relationship model is constructed according to the dynamic response weight and the static characteristic weight.

4. A method for intelligent control of a crane as claimed in claim 1, characterized in that: In step 4), a Bayesian network causal relationship diagram is established according to the static weight, when major maintenance or environmental changes occur, the Bayesian network parameter update is triggered, and the maintenance cost is minimized based on the particle swarm optimization algorithm constraint, so as to balance the cost and safety.

5. A method for intelligent control of a crane as claimed in claim 1, characterized in that: ​ 6. A method for intelligent control of a crane according to claim 5, characterized in that: The extracted features are standardized, a time window is set to segment the continuous data stream into time series segments, and the segments are input into a bidirectional LSTM neural network. The forward layer captures the historical risk accumulation, and the backward layer captures the future abnormal trend. The features at each time step are dynamically assigned weights, and a real-time risk index is output.

7. A method for intelligent control of a crane as claimed in claim 1, characterized in that: Real-time monitoring of load weight and volume, inclinometer recording of path curvature radius and operation action sequence, environmental sensor collection of wind speed and temperature, Z-score standardization of continuous variables, and 0-1 normalization of discrete variables.

8. A system for intelligent control of a crane, applied to the method according to any one of claims 1 to 7, characterized in that, Including sensors, inclinometer recording, collecting work status data through sensors, analyzing the collected data, and building a work status model; Calculating the dynamic response weight and static feature weight of the corresponding data; Real-time risk index calculation combining dynamic risk and static risk, risk level division; the sensor, inclinometer collects multi-source data, expands the static weight Bayesian network, and extracts scene sensitive features; Dynamic updating of model parameters, and sending work status data to operators for data interaction.

Images