A fault prediction method and system for a smart tower machine

Abstract

The application discloses a kind of fault prediction method and system for intelligent tower crane, it is related to tower crane technical field, including the following steps: S100, the multi-source data in the process of tower crane jacking is collected in real time, multi-source data include external environment data and equipment inherent state data;S200, based on multi-source data, jacking risk is dynamically predicted by prediction model, and risk is classified into multiple grades according to preset threshold value by outputting risk score;S300, according to risk classification result, jacking operation parameter is dynamically adjusted, including jacking rate and balance parameter;S400, based on the jacking parameter after adjustment and real-time multi-source data, implement hierarchical risk response action;S500, by feedback verification module, compare predicted risk with actual monitoring data, optimize prediction model and adjustment parameter, form closed-loop control.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tower cranes, in particular to a fault prediction method and system for an intelligent tower crane. BACKGROUND

[0002] As indispensable heavy lifting equipment in construction operations, tower cranes have increasingly expanded application ranges and have become essential core equipment on construction sites due to the large-scale emergence of high-rise and super high-rise buildings. However, due to the complex structure of tower cranes, the poor working environment, and the large dynamic load variation, the jacking process is one of the most risky operations in all operation links of the tower crane. The traditional tower crane fault diagnosis highly depends on the personal experience and regular inspection of maintenance personnel. In actual application, maintenance personnel judge the equipment state by observing the sound, vibration and other surface characteristics of the tower crane during operation. This method is highly subjective and low in accuracy, and cannot realize quantitative analysis.

[0003] A tower crane intelligent jacking monitoring control method and system based on a ROS platform are disclosed in Chinese Patent No. 2022105159150. The method includes: S1, parallel monitoring before jacking starts: sleeve monitoring, hanging shoe in place monitoring, balancing monitoring, and pin shaft monitoring; S2, increasing the following abnormal parallel monitoring during the jacking process: oil cylinder monitoring, overtop monitoring, personnel and rotation action monitoring; and S3, increasing the following monitoring after jacking in place: standard section connection monitoring. The above-mentioned invention monitors the positions and working states of each component prone to failure in the intelligent jacking process of the tower crane and the tower crane personnel, thereby improving the safety in the intelligent jacking process of the tower crane.

[0004] In the tower crane jacking scenario, the existing technology relies on static monitoring models or offline historical data for fault prediction, but significant defects are exposed in actual application. First, the data sources are fused by multiple sensors such as recognition cameras and laser range finders, but in the high-dust and strong-vibration construction site environment, sensors are easily disturbed, such as camera misidentification of guide wheel position in rain and fog, or current sensor output distortion under electromagnetic noise, which amplifies the error in the fusion process, making the warehouse flow simulation or state prediction deviation large. In addition, the existing system modules are isolated, such as the disconnection of monitoring, early warning and control links, and lack of closed-loop interaction. When the sensor suddenly fails due to extreme working conditions, the system has insufficient redundant data, relies on manual response of operators to introduce delay, and is prone to false alarms or missed alarms, increasing the risk of accidents.

[0005] Meanwhile, for example, the ROS platform-based monitoring system described above can collect data through sensors and trigger control instructions, but its model training process is complex and requires a large amount of historical data for offline batch processing, with a long update cycle and unable to adapt to minute-level dynamic changes in the jacking process in real time. When encountering sudden adverse environments such as sudden increase in instantaneous wind speed or intensified mechanical vibration, the system needs to quickly adjust the jacking parameters to prevent the sleeve from falling or the oil cylinder from failing, but the prediction model in the prior art responds slowly and often adjusts through post-calibration or manual intervention, resulting in insufficient real-time performance.

[0006] Based on the above difficulties, how to effectively solve the fusion error of multi-source data, real-time response delay and false alarm in the dynamic jacking environment has become a core technical barrier that needs to be broken through in building a high-reliability tower crane intelligent jacking system.

[0007] Therefore, it is necessary to invent a fault prediction method and system for intelligent tower cranes to solve the above problems. SUMMARY

[0008] The purpose of the present application is to provide a fault prediction method and system for intelligent tower cranes to solve the problems raised in the background art.

[0009] To achieve the above purpose, the present application provides the following technical solution: a fault prediction method for intelligent tower cranes, comprising the following steps:

[0010] S100, real-time collection of multi-source data in the jacking process of the tower crane, the multi-source data including external environmental data and internal state data of the equipment;

[0011] S200, based on the multi-source data, dynamically predicting the jacking risk through a prediction model, outputting a risk score and classifying the risk into multiple levels according to a preset threshold;

[0012] S300, dynamically adjusting jacking operation parameters including jacking speed and balance parameters according to the risk classification result;

[0013] S400, implementing graded risk response actions based on the adjusted jacking parameters and real-time multi-source data;

[0014] S500, comparing the predicted risk with the actual monitoring data through a feedback verification module, optimizing the prediction model and the adjustment parameters, and forming a closed-loop control.

[0015] Preferably, the specific steps of dynamically predicting the jacking risk in step S200 include:

[0016] S210, pre-processing the multi-source data, including data filtering and normalization;

[0017] S220, processing the preprocessed data using an LSTM neural network model to output a risk score;

[0018] S230, classifying the risk into a first type of unworkable, a second type of controllable risk, or a third type of no risk according to a comparison of the risk score with a preset threshold.

[0019] Preferably, the specific steps of dynamically adjusting the jacking parameters in step S300 include:

[0020] S310, calculating a parameter adjustment amount according to the risk classification result, wherein the adjustment of the jacking rate is based on a standard rate, an adjustment coefficient, and a risk score;

[0021] S320, dynamically calculating the balancing parameters by a PID controller based on real-time wind direction and tower body inclination data;

[0022] S330, real-time input of the adjusted parameters into the risk response module.

[0023] Preferably, in S400, when the risk is classified as the first type of unworkable, an emergency braking action is triggered; when the risk type is the second type of controllable risk, the jacking parameters are automatically adjusted and the risk change is detected, and a fault alarm signal is issued; when the risk type is the third type of no risk, the work is performed according to the preset parameters;

[0024] The response action results performed based on different risk types are fed back to the feedback verification module in real time.

[0025] Preferably, the specific steps of optimizing the prediction model and the adjustment parameters in step S500 include:

[0026] S510, calculating a deviation value of the predicted risk score and the actual monitoring data;

[0027] S520, updating the LSTM model weight using an incremental learning algorithm when the deviation value exceeds a preset threshold;

[0028] S530, optimizing the adjustment coefficient in the parameter adjustment algorithm based on historical adjustment effects.

[0029] Preferably, the parameters of the PID controller in step S320 are dynamically optimized by a reinforcement learning algorithm, and a reward function is designed based on the risk reduction degree.

[0030] Preferably, the emergency braking action includes stopping jacking and retreating to the initial position, and the braking data is used to calibrate the prediction model.

[0031] The jacking work includes the following steps:

[0032] N100, balance tower crane superstructure: by operating the jib to the direction of jacking, adjusting the load cart position to make the tower crane superstructure center of gravity fall on the jacking cylinder, keeping the fuselage balance, and removing the connecting bolts of the tower body and the lower support;

[0033] N200, jacking sleeve and introducing standard section: operating the hydraulic system to make the jacking crossbeam hang on the tower body steps, the cylinder jacks up the part above the sleeve, and uses the introduction device on the sleeve to introduce the standard section to the tower body directly above;

[0034] N300, connection and fastening: the cylinder retracts to make the new standard section align with the original tower body, and connects with high-strength bolts or pins;

[0035] For each step, the system monitors multi-source data in real time, dynamically assesses the risk level, and executes specific processing methods according to the risk level: when the assessment is a mild risk, the system issues a warning signal and continues the current step operation, while fine-tuning the parameters to reduce the risk;

[0036] When the assessment is a moderate risk, the system automatically adjusts the jacking parameters and monitors the risk changes in real time, if the risk is not reduced after adjustment, the processing method is upgraded;

[0037] When the assessment is a high risk, the system immediately triggers an emergency braking action to stop the current step and return to a safe position;

[0038] The results of the processing method are fed back to the feedback verification module in real time for optimizing the prediction model and parameter adjustment.

[0039] Preferably, the external environment data in the multi-source data includes wind speed, wind direction, temperature and humidity, and the internal state data of the equipment includes cylinder pressure, vibration amplitude and motor current.

[0040] The application also provides a fault prediction system for an intelligent tower crane, which is used to realize the tower crane jacking intelligent fault prediction method described above, so that the tower crane completes the jacking operation, and comprises a multi-source perception unit, an intelligent prediction unit, a parameter adjustment unit, a risk response unit and a closed-loop verification unit.

[0041] Preferably, the multi-source perception unit comprises a wind speed sensor and a cylinder pressure sensor integrated module, which is integrated through multi-sensor fusion and data common board design, and outputs a bound data stream to a unified space-time, and the vibration compensation module is activated when the wind speed and pressure data deviation is greater than a threshold value, the intelligent prediction unit comprises a risk response module for receiving a jacking feature vector and outputting an adjustment instruction interlocked with a jacking risk coefficient, and a risk response decision chain module for receiving an adjustment effectiveness index and outputting a control signal interlocked with a response action amplitude.

[0042] The technical effects and advantages of the application are as follows:

[0043] The application forms efficient collaborative optimization capability through real-time data sharing and instruction interlocking of the intelligent prediction unit, the parameter adjustment unit, the risk response unit and the closed-loop verification unit, optimizes the traditional emergency braking into a gradual response chain of controlled deceleration-dynamic balance-safe return based on real-time data, effectively avoids the secondary risk caused by emergency stop, realizes smooth and accurate control of risk response, thereby ensuring that the system can realize smooth processing of risk mutation through multi-module collaboration even if it encounters sudden severe environment after the jacking operation starts, and introducing the incremental learning algorithm and the closed-loop verification unit completely changes the model updating mode of traditional offline batch processing depending on a large amount of historical data. When the prediction deviation exceeds the safety threshold due to sudden severe environment, the system can immediately trigger online calibration of the model parameters, ensuring that the prediction model can continuously adapt to the dynamic changes of the jacking operation. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 It is a schematic diagram of the overall structure of the application.

[0045] Figure 2 It is a system architecture design diagram of the application. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0047] First embodiment

[0048] The prior art depends on static monitoring models or offline historical data for fault prediction, but in actual application, it has significant defects. First, the data sources are often fused through multiple sensors such as recognition cameras and laser range finders, but in high-dust and strong-vibration construction environments, the sensors are easily disturbed, such as the camera misrecognizing the position of the guide wheel in rain and fog, or the current sensor outputting distortion under electromagnetic noise, the fusion process amplifies the error, and the simulation of the warehouse inlet flow or the state prediction deviation is large. In addition, the functions of the existing system modules are isolated, such as the disconnection between monitoring, early warning and control, and lack of closed-loop interaction: when the sensor suddenly fails due to extreme working conditions, the system lacks redundant data, relies on manual response of the operator, introduces delay, and easily causes false alarm or missed alarm, increasing the risk of accidents.

[0049] Therefore, to solve the above problems, such as Figure 1 Figure 2 ​The application discloses a fault prediction system for a smart tower crane.

[0050] In the embodiment, the multi-source sensing unit comprises a wind speed sensor and a cylinder pressure sensor integrated module, which is fused with a data common substrate design through multiple sensors, and outputs a bound data stream unified in time and space; the vibration compensation module is activated when the wind speed and pressure data deviation is greater than a threshold value; the intelligent prediction unit comprises a risk response module for receiving a jacking feature vector and outputting a regulation instruction interlocked with a jacking risk coefficient; and a risk response decision chain module for receiving a regulation efficiency index and outputting a control signal interlocked with a response action amplitude.

[0051] It should be noted that the precise clock protocol is adopted to control the timestamp deviation of the multi-source data within ±1ms, so as to ensure the simultaneity of the data; the data with spatial attributes such as the tower body inclination is mapped to the tower crane base coordinate system through coordinate transformation; and the data without spatial attributes such as the current is associated with the spatial data through a label, for example, after the wind speed sensor data and the inclination sensor data are aligned through the timestamp, the wind load moment is calculated in combination with the inclination angle, so that effective fusion is realized.

[0052] In the embodiment, the multi-source sensing unit further comprises inclination sensors, vibration sensors and motor current detection modules and various data acquisition modules, which jointly construct a three-dimensional monitoring network and can collect key parameters such as the height, the moment, the wind speed and the inclination in the jacking process of the tower crane at a millisecond-level refresh rate; the unit realizes hardware-level integration through the data common substrate design, so as to ensure that all sensor data is bound to the data stream unified in time and space when being output, and provide a consistent data basis for subsequent analysis; when the system detects that the wind speed and the cylinder pressure data deviation exceeds a preset threshold value, the vibration compensation module is automatically activated to calibrate and compensate the data, so as to cope with the interference in the complex construction site environment.

[0053] In the embodiment, the intelligent prediction unit is provided with a risk response module and a parameter optimization algorithm, the input end of the unit acquires parameters such as the jacking speed, the tower body inclination and the cylinder pressure in real time through a multi-source data fusion interface; the unit adopts an LSTM neural network architecture to compare the real-time monitoring data with historical fault modes, dynamically calculates the risk coefficient of the jacking operation based on the regulation stability data fed back by the closed-loop verification unit, and outputs three decisions including risk grading, early warning publishing and regulation instruction generation, so as to form a complete jacking control chain.

[0054] The risk response module is responsible for receiving the lifting feature vector generated by the multi-source perception unit and outputting adjustment instructions associated with the lifting risk coefficient interlock; the risk response decision chain module receives the adjustment efficiency index from the closed-loop verification unit and outputs control signals associated with the response action amplitude interlock.

[0055] In this embodiment, the parameter adjustment unit integrates an adaptive control algorithm, which can receive real-time adjustment instructions from the intelligent prediction unit. The unit synchronously processes actual lifting execution data and output results of the prediction model and extracts performance indicators including lifting rate deviation and balance maintenance error through a PID controller. The adjustment algorithm further integrates the device health state parameters provided by the digital twin module, and finally outputs the precise adjustment amount of the lifting parameters (such as lifting rate and trim value).

[0056] In this embodiment, the risk response unit adopts a hierarchical triggering mechanism. Its instruction synthesizer receives the risk level judgment result of the prediction unit and generates specific response action instructions based on the output of the parameter adjustment unit. The unit can perform multi-level responses from sound and light warning, automatic adjustment of lifting parameters to emergency braking. At the same time, the unit records the execution effect of all response actions and packages these data as efficiency index to feedback to the closed-loop verification unit.

[0057] In this embodiment, the closed-loop verification unit adopts a dynamic data assimilation algorithm. Its model updater receives real-time monitoring data from the perception unit and assimilates physical models and observation data using Kalman filtering algorithm. The unit generates performance evaluation reports including prediction accuracy, response delay, and fault omission rate by comparing predicted risks and actual monitoring results. Once it finds that the prediction deviation exceeds the safety threshold, the unit will immediately trigger model parameter calibration or start emergency warning protocol and feedback the verification results back to the intelligent prediction unit and parameter adjustment unit for optimizing subsequent decisions.

[0058] In use, the closed-loop verification unit, intelligent prediction unit, and parameter adjustment unit form an efficiency evaluation information exchange channel. In this process, the closed-loop verification unit processes the adjustment efficiency data which is transmitted to the intelligent decision center. The decision center receives the parameter adjustment effect data returned by the parameter adjustment unit and finally transmits the optimized instructions to the lifting execution mechanism to complete the real-time optimization of the strategy.

[0059] In the embodiment, the system further comprises an auxiliary monitoring module composed of a high-precision displacement sensor and an intelligent visual recognition camera. When the system detects that the deviation between different sensor data is greater than a set threshold value (for example, the oil cylinder displacement and the lifting height logic are inconsistent) through the multi-source perception unit, the auxiliary module is activated to provide redundant verification. The risk response module of the intelligent prediction unit is used to receive the lifting process feature vector and output the adjustment instruction interlocked with the lifting risk coefficient. The risk response decision chain module is used to receive the adjustment efficiency index and output the control signal interlocked with the adjustment action amplitude.

[0060] It should be noted that the intelligent prediction unit realizes multi-module collaborative iterative optimization by relying on the closed-loop verification mechanism and the quality monitoring module of the digital twin system. Through the verification interface, three key performance indicators are obtained in real time: the first is the running stability index calculated based on the lifting trajectory tracking deviation; the second is the model reliability measured by the risk prediction accuracy; and the third is the system response timeliness index, including data acquisition delay and decision calculation time. According to the monitored data, the system will automatically start a three-level dynamic adjustment strategy. When the adjustment effect exceeds the preset threshold, the cross-module joint optimization process will be started. If the system response timeliness breaks through the upper limit, algorithm simplification operation will be performed, and the lifting working condition prediction model will be combined to allocate and schedule the computing resources in advance.

[0061] In actual use, the system completes the initialization of the multi-source perception unit, the intelligent prediction unit, the parameter adjustment unit, the risk response unit and the closed-loop verification unit by synchronously accessing the environmental data collected by the wind speed sensor, the internal state data of the equipment collected by the oil cylinder pressure sensor, the tower posture data monitored by the inclination sensor and the historical lifting records.

[0062] The multi-source perception unit processes multi-source heterogeneous data fusion based on the space-time alignment mechanism, outputs the lifting feature vector to the prediction unit, and receives the weight adjustment parameters from the closed-loop verification unit to dynamically optimize the data acquisition strategy.

[0063] The intelligent prediction unit calculates the lifting risk coefficient by means of the LSTM neural network, and optimizes the generation of prediction decisions by fusing the equipment state parameters provided by the digital twin module.

[0064] The closed-loop verification unit uses an effect evaluation algorithm to generate performance indicators based on actual lifting data, and feeds back the verification results to the intelligent prediction unit. The parameter adjustment unit adaptively adjusts according to the prediction results and real-time state throughout the process. Once it is found that the prediction deviation exceeds the standard, it will trigger multi-module collaborative correction or start the emergency warning protocol. Finally, a tower crane lifting operation scheme with controllable risk, optimal efficiency and high reliability is output.

[0065] Second embodiment

[0066] Based on the system framework constructed above, the specific steps of the fault prediction system for the intelligent tower crane are elaborated in detail, thereby effectively solving the fusion error of multi-source data, real-time response delay and false alarm in the dynamic jacking environment.

[0067] The system comprises the following steps:

[0068] S100, real-time collection of multi-source data in the jacking process of the tower crane, the multi-source data comprising external environment data and internal state data of the equipment.

[0069] In the embodiment, the external environment data in the multi-source data comprises wind speed, wind direction, temperature and humidity, and the internal state data of the equipment comprises cylinder pressure, vibration amplitude and motor current.

[0070] S200, dynamic prediction of jacking risk based on the multi-source data through a prediction model, output of a risk score and classification of the risk into multiple levels according to a preset threshold.

[0071] In the embodiment, the specific steps of dynamically predicting the jacking risk in the step S200 comprise:

[0072] S210, pre-processing of the multi-source data, comprising data filtering and normalization.

[0073] It should be noted that when the data is pre-processed, the multi-source data is dimensionless. The minimum-maximum normalization method is adopted to convert each physical quantity to the interval [0, 1], and the conversion method is that the normalized value is equal to the original measured value minus the minimum value of the physical quantity, and the obtained difference is divided by the difference between the maximum and minimum values of the physical quantity.

[0074] wherein the original measured value is the actual measured value of the physical quantity, which can be the real-time sampling value of wind speed (unit: meter / second), cylinder pressure (unit: megapascal) or current (unit: ampere) in the tower crane jacking system.

[0075] The minimum value of the physical quantity refers to the lower limit of the physical quantity in the historical data set or the preset range, for example, the minimum value of the wind speed is set to zero meter per second, which corresponds to the no wind state, and the minimum value of the cylinder pressure can be set to zero megapascal, which corresponds to the system idle state.

[0076] The maximum value of the physical quantity refers to the upper limit of the physical quantity in the historical data set or the preset range, for example, the maximum value of the wind speed is set to twenty meters per second, which is based on the safety standard, and the maximum value of the cylinder pressure is set to sixty megapascal, which is based on the rated value of the equipment.

[0077] The normalized value is the result of the above calculation, which is a dimensionless number without unit, ranging from zero to one, indicating the relative position of the original measured value in its possible value range, used to eliminate the influence of different physical quantities due to different units, facilitating comparison and analysis between them.

[0078] S220, processing the pre-processed data using the LSTM neural network model to output a risk score.

[0079] S230, according to the comparison of the risk score and the preset threshold, classifying the risk into the first type of unworkable, the second type of controllable risk or the third type of no risk.

[0080] S300, dynamically adjusting the jacking operation parameters including jacking rate and balance parameters according to the risk classification result.

[0081] In the embodiment, the specific steps of dynamically adjusting the jacking parameters in step S300 include:

[0082] S310, calculating the parameter adjustment amount according to the risk classification result, wherein the adjustment of the jacking rate is based on the standard rate, the adjustment coefficient and the risk score, and the adjustment mode is that the adjusted actual jacking speed is the maximum value of zero and the final calculated speed value, and the final calculated speed value is the smaller one of the standard speed and the expected target speed calculated after risk discount, and the expected target speed is calculated by reducing the ideal speed by a certain proportion according to the real-time evaluated risk level, so as to calculate a new target speed.

[0083] It should be noted that the adjusted actual jacking speed refers to the final jacking speed execution value after intelligent calculation of the system.

[0084] The standard jacking speed refers to the reference jacking speed preset under ideal or standard working conditions.

[0085] The adjustment coefficient is a risk influence weight coefficient, which reflects the sensitivity of the control system to the risk level.

[0086] The risk score is a numerical value output by the LSTM neural network prediction model for quantifying the jacking operation risk, with a value range from zero to one, and the larger the value, the higher the risk.

[0087] It should be noted that the adjustment coefficient will be dynamically optimized: for example, a preset value is calculated by a reinforcement learning algorithm such as Q-learning, to ensure that even in extreme cases, the adjusted actual jacking speed will not be lower than half of the standard jacking speed, thereby maintaining a basic safety operation threshold.

[0088] S320, dynamically calculating the balance parameter through the PID controller based on the real-time wind direction and tower body inclination data.

[0089] It should be noted that the parameters of the PID controller in step S320 are dynamically optimized through a reinforcement learning algorithm, and the reward function is designed based on the risk reduction degree.

[0090] S330, real-time input of the adjusted parameter into the risk response module.

[0091] S400, implementing a hierarchical risk response action based on the adjusted jacking parameter and real-time multi-source data.

[0092] In this embodiment, in S400, when the risk is classified as the first type of non-operable, an emergency braking action is triggered; when the risk type is the second type of controllable risk, the jacking parameter is automatically adjusted and the risk change is detected, and a fault alarm signal is issued; when the risk type is the third type of no risk, the operation is performed according to the preset parameter.

[0093] The response action results executed based on different risk types are fed back to the feedback verification module in real time.

[0094] S500, through the feedback verification module, comparing the predicted risk with the actual monitoring data, optimizing the prediction model and the adjustment parameter, forming a closed loop control.

[0095] In this embodiment, the specific steps of optimizing the prediction model and the adjustment parameter in step S500 include:

[0096] S510, calculating the deviation value of the predicted risk score and the actual monitoring data.

[0097] S520, when the deviation value exceeds the preset threshold, updating the LSTM model weight using the incremental learning algorithm.

[0098] It should be noted that the specific architecture of the LSTM neural network adopts a three-layer hidden layer structure, and the number of neurons is 128, 64 and 32 respectively; the input layer includes 20 nodes corresponding to the following feature dimensions: environmental data: wind speed, wind direction, temperature, humidity; mechanical state: cylinder pressure, vibration amplitude X / Y / Z axis, inclination angle; electrical parameters: motor current, voltage, frequency converter temperature; time series features: historical 10 time step jacking rate change trend; the output layer uses a Sigmoid activation function, and outputs a 1-dimensional risk score R∈[0,1].

[0099] The training data set and the training data of the process are derived from 10,000 groups of historical jacking operation records, the data preprocessing includes outlier elimination and sequence alignment, and time series cross validation is adopted, the training set / validation set ratio is 8:2, the optimizer uses Adam, the initial learning rate is 0.001, the batch size is 32, and the training stop condition is that the validation set loss does not decrease for 5 consecutive epochs.

[0100] The data set contains 10,000 groups of historical jacking operation records, and the data annotation rules are as follows:

[0101]

[0102] For ease of understanding, a specific application scenario example is provided: in the case of sudden change of wind speed, the system data flow presents a complete closed-loop response: when the wind speed sensor detects that the wind speed suddenly increases from 5 m / s to 15 m / s, the multi-source perception unit synchronously collects abnormal data that the oil cylinder pressure fluctuates from 28 MPa to 35 MPa (exceeding 10%) and the tower body inclination increases from 0.5° to 1.8°. After time and space alignment, these real-time parameters generate a feature vector input into the LSTM neural network; then, the neural network calculates the risk score through three hidden layers and triggers the second type of controllable risk warning; the parameter adjustment unit dynamically adjusts the jacking speed according to the actual jacking speed execution value after intelligent adjustment of the system, and the reinforcement learning algorithm optimizes the PID parameters according to the reward function; finally, the closed-loop verification unit observes that the actual inclination decreases to 1.2° and the risk score decreases to the preset range through Kalman filtering, confirms that the adjustment is effective, and feeds back the efficiency index to the digital twin module to update the equipment health status parameters, completing the whole-link collaborative control from perception, decision, execution to verification.

[0103] S530, based on the historical adjustment effect, optimizing the adjustment coefficient in the parameter adjustment algorithm.

[0104] In the present embodiment, the emergency braking action includes stopping jacking and retreating to the initial position, and the braking data is used for calibrating the prediction model;

[0105] The jacking operation includes the following steps:

[0106] N100, balance the tower crane superstructure: rotate the hoisting arm to the jacking direction by operation, adjust the position of the load trolley to make the center of gravity of the tower crane superstructure fall on the jacking cylinder, keep the body balanced, and remove the connecting bolts between the tower body and the lower support.

[0107] N200, jacking the sleeve frame and introducing the standard section: operating the hydraulic system to make the jacking beam hang on the tower body steps, the cylinder jacks up the part above the sleeve frame, and uses the introduction device on the sleeve frame to introduce the standard section to the tower body.

[0108] N300, connection and fastening: the retracting of the oil cylinder makes the new standard section align with the original tower body, and connects them with high-strength bolts or pins.

[0109] It should be noted that for the implementation steps of the above jacking operation, when the wind speed suddenly increases from 10 m / s to 20 m / s, the risk score predicted by the LSTM neural network may jump from the second type of risk to the first type of risk in an instant. If the system mechanically executes the preset first type of risk emergency braking instruction, the risk may be aggravated due to the following reasons: dynamic load impact, emergency braking may stop jacking instantly, resulting in a sharp increase in hydraulic system pressure, which may damage the oil cylinder or connecting parts; tower inertia instability, in high-speed jacking, the upper part of the tower may swing due to inertia, increasing the risk of overturning; operation interruption risk, braking at N200 or N300 stage may make the standard section in a semi-connected state, introducing new safety hazards.

[0110] Therefore, when the jacking operation starts, the system needs to optimize the emergency braking into a gradual response chain of controlled deceleration-dynamic balance-safe return through the cooperation of multiple modules.

[0111] In the above case, through the real-time interaction of the intelligent prediction unit, the parameter adjustment unit, the risk response unit and the closed-loop verification unit, the smooth processing of risk mutation is realized, and the specific adjustment process is as follows:

[0112] When the multi-source perception unit detects a sudden change in wind speed (such as an increase of more than 5 m / s per second), the priority calculation of the intelligent prediction unit is triggered immediately: the LSTM model recalculates the risk score within a unit of time and marks the "risk mutation flag", and at the same time the prediction unit shares the mutation signal to the parameter adjustment unit and the risk response unit, triggering the cooperative response mode.

[0113] When a risk mutation is detected, an attenuation factor is introduced, which is learned based on historical mutation data, so that the jacking speed is stepped down smoothly.

[0114] The adjusted actual jacking speed step: the value of the adjusted actual jacking speed is the maximum of zero and the final calculated speed value, and the final calculated speed value is the smaller one of the standard speed and the expected target speed calculated after risk discount. The expected target speed is reduced in proportion to the ideal speed according to the real-time evaluation of the risk level, and an attenuation factor is introduced to calculate a new target speed. As time goes by, the value of the attenuation factor decreases exponentially, and the discounting effect on the speed will also become weaker and weaker, so that the jacking speed can be smoothly and gradually restored.

[0115] The risk response unit performs a step-by-step action according to the adjustment effect:

[0116] The first step: when the current state is in high risk based on the risk score, the system will issue an audible and visual alarm, and at the same time, it will execute a slow descent mode, reducing the jacking rate to a pre-set safe lower limit value, and immediately activate the balancing mechanism to compensate for the effect of wind load.

[0117] The second step: when the risk score continues to increase, and this state continues for 3 seconds, the system will start a controlled retreat program, controlling the oil cylinder to retract to the last recorded safe position at a lower speed, thereby avoiding the impact caused by emergency stop.

[0118] The third step: when the risk score continues to increase, or the device inclination exceeds five degrees, the system will eventually trigger an emergency brake, but at the same time, it will activate the hydraulic buffer device to reduce the impact of dynamic load on the device.

[0119] In this example, for each step, the system monitors multi-source data in real time, dynamically assesses the risk level, and executes specific processing methods according to the risk level: when the assessment is a mild risk, the system issues a warning signal and continues the current step operation, while fine-tuning the parameters to reduce the risk.

[0120] When the assessment is a moderate risk, the system automatically adjusts the jacking parameters and monitors the risk changes in real time. If the risk does not decrease after adjustment, the processing method is upgraded.

[0121] When the assessment is a high risk, the system immediately triggers an emergency brake action, stops the current step and retreats to a safe position.

[0122] The results of the processing method are fed back to the feedback verification module in real time for optimizing the prediction model and parameter adjustment.

[0123] In use, when the tower crane jacking data is collected in real time by the multi-source perception unit, the data collected by the wind speed sensor is synchronized in real time to the risk calculation module of the intelligent prediction unit, and is shared through the data bus to the wind load compensation module of the parameter adjustment unit. When a sudden change in wind speed is detected, the vibration compensation module will be immediately activated, and the calibrated data will be sent to the prediction unit and the verification unit at the same time, forming a three-way data collaboration of perception-prediction-verification, so as to ensure that in the case of sudden change of wind speed, the prediction unit receives reliable data that has been verified multiple times, rather than isolated raw data.

[0124] Secondly, the intelligent prediction unit and the parameter adjustment unit realize decision closed loop through dynamic interaction, when the risk score output by the LSTM neural network exceeds the threshold value, not only the early warning instruction is triggered, but also the risk feature vector is transmitted to the adaptive algorithm module of the parameter adjustment unit in real time, the adjustment unit adjusts the PID parameters dynamically based on the received risk data through the reinforcement learning algorithm, and meanwhile, the adjustment scheme is fed back to the prediction unit for effect pre-evaluation, the cyclic interaction of prediction-adjustment-pre-evaluation enables the system to complete multiple rounds of optimization iteration in a short time, and the accuracy of decision is significantly improved.

[0125] In the control execution phase, the parameter adjustment unit and the risk response unit establish an instruction interlocking mechanism, when the jacking rate risk score after adjustment is still higher than the threshold value, the system will immediately start cross-module cooperation: the parameter adjustment unit shares the real-time adjustment data to the risk response unit, and the latter adjusts the response strategy based on the data; at the same time, the execution effect data of the risk response unit is fed back to the prediction unit in real time through the verification unit for updating the risk model.

[0126] Finally, the closed-loop verification unit as the core hub of the system realizes the deep nesting of the whole process, this unit not only receives the running data of each module, but also verifies in multiple dimensions through the digital twin model: compares the output of the prediction unit with the actual monitoring data, associates the calculation results of the parameter adjustment unit with the execution effect, and associates the actions of the risk response unit with the final risk change, and the verification result is fed back to each module in real time - guiding the perception unit to optimize the data acquisition strategy, assisting the prediction unit to adjust the model parameters, and helping the adjustment unit to improve the control algorithm, the whole-link closed-loop verification makes the system form a benign cycle of self-optimization.

[0127] It is particularly emphasized that the present application realizes the whole-process cooperative optimization from data acquisition to risk control by constructing the intelligent fault prediction system deeply nested among modules, forms an organic whole through data sharing, decision interaction and verification closed loop, effectively improves the reliability of the system, and effectively solves the problems of fusion error of multi-source data, real-time response delay and false alarm in the dynamic jacking environment.

[0128] Finally, it should be pointed out that: the above-mentioned is only the preferred embodiment of the present application, and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced equivalently, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A fault prediction method for a smart tower, characterized in that, The method comprises the following steps: S100, collecting multi-source data in the jacking process of the tower crane in real time, wherein the multi-source data comprises external environment data and internal state data of the equipment; S200, dynamically predicting the jacking risk based on the multi-source data through a prediction model, outputting a risk score, and classifying the risk into multiple levels according to a preset threshold; S300, dynamically adjusting jacking operation parameters, including jacking speed and balance parameters, according to the risk classification result; S400, implementing a hierarchical risk response action based on the adjusted jacking parameters and real-time multi-source data; when the risk is classified as the first type of unworkable, triggering an emergency braking action; when the risk type is the second type of controllable risk, automatically adjusting the jacking parameters and detecting the risk change, and simultaneously issuing a fault alarm signal; when the risk type is the third type of no risk, operating according to preset parameters; The response action results executed based on different risk types are fed back to a feedback verification module in real time; the emergency braking action comprises stopping jacking and retreating to an initial position, and braking data is used for calibrating the prediction model; wherein the jacking operation comprises the following steps: N100, balancing the tower crane superstructure: rotating the jib to the jacking direction by operation, adjusting the position of the load trolley to make the center of gravity of the tower crane superstructure fall on the jacking cylinder, keeping the body balanced, and removing the connecting bolts between the tower body and the lower support; N200, jacking the sleeve frame and introducing a standard section: operating the hydraulic system to make the jacking crossbeam hang on the tower body steps, the cylinder jacks up the part above the sleeve frame, and the introduction device on the sleeve frame is used to introduce the standard section to the tower body; N300, connection and fastening: the new standard section is aligned with the original tower body by cylinder retraction, and is connected by high-strength bolts or pins; For each step, the system monitors multi-source data in real time, dynamically evaluates the risk level, and executes specific processing methods according to the risk level: when the evaluation is a mild risk, the system issues a warning signal and continues the current step operation, while fine-tuning the parameters to reduce the risk; when the evaluation is a moderate risk, the system automatically adjusts the jacking parameters and monitors the risk change in real time, and if the risk does not decrease after adjustment, the processing method is upgraded; when the evaluation is a high risk, the system immediately triggers an emergency braking action to stop the current step and retreat to a safe position; The results of the processing method are fed back to the feedback verification module in real time for optimizing the prediction model and parameter adjustment; S500, comparing the predicted risk and the actual monitoring data through the feedback verification module to optimize the prediction model and the adjustment parameters, forming a closed-loop control.

2. The fault prediction method of the intelligent tower crane according to claim 1, characterized in that, The specific steps of dynamically predicting the jacking risk in step S200 comprise: S210, preprocessing the multi-source data, including data filtering and normalization; S220, using an LSTM neural network model to process the preprocessed data to output a risk score; S230, classifying the risk into the first type of unworkable, the second type of controllable risk or the third type of no risk according to the comparison of the risk score and the preset threshold.

3. The method of claim 1, wherein, The specific steps of dynamically adjusting the jacking parameters in step S300 comprise: S310, calculating the parameter adjustment amount according to the risk classification result, wherein the adjustment of the jacking speed is based on the comprehensive calculation of the standard speed, the adjustment coefficient and the risk score; S320, dynamically calculating the balance parameter through the PID controller based on the real-time wind direction and tower body inclination data; S330, real-time inputting the adjusted parameter into the risk response module.

4. The fault prediction method of the intelligent tower crane according to claim 1, characterized in that, The specific steps of optimizing the prediction model and adjusting the parameter in the step S500 include: S510, calculating the deviation value of the predicted risk score and the actual monitoring data; S520, updating the LSTM model weight using the incremental learning algorithm when the deviation value exceeds the preset threshold value; S530, optimizing the adjustment coefficient in the parameter adjustment algorithm based on the historical adjustment effect.

5. The fault prediction method of the intelligent tower crane according to claim 3, characterized in that, The parameters of the PID controller in the step S320 are dynamically optimized through the reinforcement learning algorithm, and the reward function is designed based on the risk reduction degree.

6. The fault prediction method of the intelligent tower crane according to claim 1, characterized in that, The external environment data in the multi-source data includes wind speed, wind direction, temperature, and humidity, and the internal state data of the equipment includes cylinder pressure, vibration amplitude, and motor current.

7. A fault prediction system for a smart tower crane, the system being used to implement the fault prediction method of any one of claims 1-6, so that the tower crane completes the jacking operation, characterized in that, It includes: A multi-source perception unit, an intelligent prediction unit, a parameter adjustment unit, a risk response unit, and a closed-loop verification unit.

8. The fault prediction system of the intelligent tower crane according to claim 7, characterized in that, The multi-source perception unit includes a wind speed sensor and a cylinder pressure sensor integrated module, which is designed through multi-sensor fusion and data common substrate, and outputs binding to the unified data stream in space and time. The vibration compensation module is activated when the deviation between wind speed and pressure data is greater than the threshold value. The intelligent prediction unit includes a risk response module for receiving a jacking feature vector and outputting an adjustment instruction interlocked with a jacking risk coefficient, and a risk response decision chain module for receiving an adjustment efficiency index and outputting a control signal interlocked with a response action amplitude.

Images