Novel crane stress automatic monitoring method

Abstract

The invention provides a novel crane stress automatic monitoring method which is characterized in that a data acquisition module and a stress monitoring module are included, the method identifies the current working state of a crane according to a monitoring system signal, and when it is judged that the current working state of the crane is in a high-risk operation stage, the stress monitoring module is started; starting stress monitoring aiming at a key stress part of the crane; the invention belongs to an intelligent stress monitoring method based on signals of a crane monitoring system, and the method can automatically start stress monitoring when needed and automatically stop monitoring when not needed.

Description

Technical Field

[0001] The invention relates to the field of safety monitoring of lifting machinery, in particular to a novel automatic monitoring method for crane stress. Background Art

[0002] Traditional crane stress monitoring systems usually monitor stress continuously, which not only consumes a lot of energy but also may cause difficulties in data processing due to excessive data volume.

[0003] In the daily operation of a crane, the need for stress monitoring is relatively rare.

[0004] The present invention aims to provide an intelligent monitoring method to reduce unnecessary stress monitoring, save energy and improve data processing efficiency. Summary of the invention

[0005] The present invention proposes a novel automatic crane stress monitoring method, which belongs to an intelligent stress monitoring method based on crane monitoring system signals. The method can automatically start stress monitoring when needed and automatically stop monitoring when not needed.

[0006] The present invention adopts the following technical solutions.

[0007] A novel automatic crane stress monitoring method includes a data acquisition module and a stress monitoring module. The method identifies the current working state of a portal crane according to a monitoring system signal. When it is determined that the current working state of the crane is in a high-risk operation stage, stress monitoring of key stress-bearing parts of the portal crane is started.

[0008] When deployed, the data acquisition module is integrated into the original crane monitoring system and uses the existing sensor data of the crane to identify the working status of the crane; the sensors include load sensors and position sensors;

[0009] The method analyzes the sensor data through a data analysis and processing module to determine whether the crane is in a high-risk operation stage, thereby determining whether stress monitoring needs to be turned on;

[0010] The sensor of the stress monitoring module includes a strain gauge and a data collector, which are used to monitor the stress changes of the key stress-bearing parts of the crane in real time. The strain gauge is fixed on the crane with a matching tooling;

[0011] When the data analysis and processing module determines that the current crane is in a high-risk operation stage and needs to be monitored, the stress monitoring module is activated through a start signal to start collecting stress data.

[0012] Use male and female cables with anti-loosening function to connect the strain gauge to the collector of the data acquisition module. The strain gauge is fixed to the surface of the crane by welding. Pass the signal line through the plug hole on the protective shell, and weld the signal line and the cable of the strain gauge in the corresponding order.

[0013] The installation method of the sensor is as follows: put the protective shell on the sensor, fix it on the crane surface through multiple fixing feet of the protective shell, and finally seal the surrounding of the protective shell and the fixing feet with silicone glue;

[0014] The automatic monitoring method can select a periodic monitoring mode or an online monitoring mode according to the needs of users and equipment. When the periodic monitoring mode is selected, the sensor is installed on the crane in a quick plug-in manner to shorten the installation time and ensure the installation quality.

[0015] The data acquisition module forms the data acquisition layer of the automatic monitoring method, and its various sensors include an S-type tension sensor installed at the hinge point at the root of the crane boom, an inclination sensor installed at the connection between the crane revolving platform and the column, an ultrasonic anemometer installed at the top of the crane boom, and a welded strain gauge installed at the mid-span of the crane main beam and the weld of the door leg.

[0016] The sensors in the data acquisition layer form a sensor group. The collected data of the sensor group is processed by the signal conditioning module and then input into the embedded controller (STM32H743, supporting multi-protocol communication). The data of the tension sensor and the tilt sensor are transmitted to the embedded controller via the CAN bus. The data of the wind speed sensor is transmitted to the embedded controller via the RS485 link.

[0017] Sampling frequency: tension / tilt angle (10Hz), wind speed (1Hz), strain (50Hz);

[0018] The data analysis and processing module forms the edge computing layer and control layer of the automatic monitoring method. The edge computing layer includes an industrial computer (Intel NUC, running Ubuntu+ROS) for performing data preprocessing, feature extraction, and state classification;

[0019] The control layer includes PLC (Siemens S7-1200), which is used to control the start and stop of the stress monitoring module and the alarm output.

[0020] The data analysis and processing module processes and analyzes the data of the data acquisition module in real time to predict and evaluate the stress state of the crane, and uses data analysis technology to improve the accuracy of the prediction, including the following steps:

[0021] Step S1, collecting data from sensors in the crane monitoring system: lifting weight Q1, lifting moment F1, rotation angle A1, real-time wind speed V1;

[0022] Step S2, when the data collected by the sensor has the following characteristics (which can be changed according to user needs), the crane is judged to be in a high-risk operating state: Q1≥0.9*rated lifting weight ∪F1≥0.9*rated lifting moment weight ∪A1≥90°∪real-time wind speed V1≥10M / S, start data analysis, and when it is determined that the crane is in a high-risk state, a start signal is given to the stress monitoring module.

[0023] In step S2, the data analysis process uses a random forest classifier algorithm, including a data preprocessing algorithm, a feature extraction and selection algorithm; the specific goals of the data preprocessing algorithm are: to clean, standardize and extract features from the collected raw data to ensure the accuracy of subsequent analysis;

[0024] The steps of the data preprocessing algorithm include;

[0025] Step A1, data cleaning: remove invalid data, fill missing values, and remove noise;

[0026] Step A2, data standardization: In order to unify the data scales of different sensors, a standardization method (such as Z-score standardization) is used to normalize all sensor data; the formula is: Among them, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data;

[0027] During data preprocessing, the standardized Q1, A1, V1, σ are input, and the feature matrix X∈R is output. NX4 N = number of samples in the time window;

[0028] The specific goals of the feature extraction and selection algorithm are: extract meaningful features from the cleaned data for subsequent status assessment and anomaly detection;

[0029] The feature extraction and selection algorithm includes the following steps;

[0030] Step B1, basic statistical characteristics: Calculate statistical characteristics such as mean, standard deviation, maximum value, minimum value, kurtosis, skewness, etc. for sensor data.

[0031] For example

[0032]

[0033] Where Zi is each data point, N is the number of data points, and μ is the mean;

[0034] Step B2: Use machine learning models to perform multi-dimensional evaluation of the working status; predict whether the crane is in a normal state or a high-risk state by fusing multiple sensor data (load, wind speed, position, etc.); use the random forest boosting decision tree ensemble learning method to build a multi-classifier to evaluate the crane status at each moment, including the above four data features Q1 F1A1 V1;

[0035] Step B3: Use random forest for classification, specifically:

[0036] Training model: Use labeled historical data (such as normal, overload, excessive wind speed, excessive rotation angle, etc.) to train the random forest model;

[0037] When Q1≥0.9*rated lifting weight, it is marked as overload; when Q1<0.9*rated lifting weight, it is marked as normal;

[0038] When F1≥0.9*rated lifting moment, it is marked as overload; when F1<0.9*rated lifting moment, it is marked as normal;

[0039] When A1≥90°, it is marked as risk data; when A1<90°, it is marked as normal;

[0040] When V1≥10m / s, it is marked as high wind speed; when V1<10m / s, it is marked as normal;

[0041] Model output: For each moment of data, a status label is output, i.e., “normal”, “overload” or “high wind speed”;

[0042] When the data analysis and processing module performs risk assessment, it combines the outputs of all features and models to determine whether the current crane is in a high-risk state; when any one is marked as abnormal, the crane is judged to be in a high-risk state; if none of them is marked as abnormal, the crane is judged to be in a normal state.

[0043] The crane stress automatic monitoring method also includes a control module. The data analysis and processing module transmits the analysis results and risk assessment results to the control module to obtain corresponding control instructions. The control module automatically controls the opening and closing of the stress monitoring module according to the results of the data processing and analysis module and the data in the intelligent identification module. It realizes seamless integration with the crane monitoring system to ensure rapid response when needed.

[0044] Normal status: If the monitored data is within a safe range, the system will continue to monitor and wait for the next data update.

[0045] High-risk status: If an abnormal or high-risk status is detected, the system triggers the alarm mechanism, starts the stress monitoring module, and provides alarm information through the user interface.

[0046] When the control module starts the stress monitoring module after the threshold is triggered, the stress monitoring module is started with a delay of 2 seconds to avoid instantaneous interference.

[0047] When there is no high-risk signal for 5 consecutive minutes and σ<80MPa, the control module turns off the stress monitoring module.

[0048] The advantages of the present invention are:

[0049] 1. Energy-saving and efficient: Stress monitoring is enabled only when needed, reducing energy consumption and data processing burden.

[0050] 2. Intelligent safety: Intelligent monitoring ensures that the crane is fully monitored during high-risk operations to improve safety.

[0051] 3. Flexibility: Operators can manually adjust the monitoring strategy according to actual conditions to improve the adaptability of the system.

[0052] In the present invention, the algorithm used uses a grading strategy of "threshold judgment priority + model prediction supplement", which can take into account both real-time performance and accuracy:

[0053] In the present invention, the risk (such as overload) is directly intercepted by threshold judgment, which can shorten the response time;

[0054] In the present invention, the random forest model is used to solve complex working conditions (such as multi-parameter coupling risks), which can reduce the false negative rate;

[0055] In the present invention, dynamic start-stop control balances safety and energy consumption, and can ensure long-term reliable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0057] Attached Figure 1 It is a schematic diagram of the supporting tooling for installing the strain gauge;

[0058] Attached Figure 2 Schematic diagram of the installation position of the sensor in the present invention (in the figure, ▽ represents the measuring point formed by the sensor). DETAILED DESCRIPTION

[0059] As shown in the figure, a new type of automatic crane stress monitoring method includes a data acquisition module and a stress monitoring module. The method identifies the current working status of the portal crane according to the monitoring system signal. When it is determined that the current working status of the crane is in a high-risk operation stage, stress monitoring of key stress-bearing parts of the portal crane is started.

[0060] When deployed, the data acquisition module is integrated into the original crane monitoring system and uses the existing sensor data of the crane to identify the working status of the crane; the sensors include load sensors and position sensors;

[0061] The method analyzes the sensor data through a data analysis and processing module to determine whether the crane is in a high-risk operation stage, thereby determining whether stress monitoring needs to be turned on;

[0062] The sensor of the stress monitoring module includes a strain gauge and a data collector, which are used to monitor the stress changes of the key stress-bearing parts of the crane in real time. The strain gauge is fixed on the crane with a matching tooling;

[0063] When the data analysis and processing module determines that the current crane is in a high-risk operation stage and needs to be monitored, the stress monitoring module is activated through a start signal to start collecting stress data.

[0064] Use male and female cables with anti-loosening function to connect the strain gauge to the collector of the data acquisition module. The strain gauge is fixed to the surface of the crane by welding. Pass the signal line through the plug hole on the protective shell, and weld the signal line and the cable of the strain gauge in the corresponding order.

[0065] The installation method of the sensor is as follows: put the protective shell on the sensor, fix it on the crane surface through multiple fixing feet of the protective shell, and finally seal the surrounding of the protective shell and the fixing feet with silicone glue;

[0066] The automatic monitoring method can select a periodic monitoring mode or an online monitoring mode according to the needs of users and equipment. When the periodic monitoring mode is selected, the sensor is installed on the crane in a quick plug-in manner to shorten the installation time and ensure the installation quality.

[0067] The data acquisition module forms the data acquisition layer of the automatic monitoring method, and its various sensors include an S-type tension sensor installed at the hinge point at the root of the crane boom, an inclination sensor installed at the connection between the crane revolving platform and the column, an ultrasonic anemometer installed at the top of the crane boom, and a welded strain gauge installed at the mid-span of the crane main beam and the weld of the door leg.

[0068] The sensors in the data acquisition layer form a sensor group. The collected data of the sensor group is processed by the signal conditioning module and then input into the embedded controller (STM32H743, supporting multi-protocol communication). The data of the tension sensor and the tilt sensor are transmitted to the embedded controller via the CAN bus. The data of the wind speed sensor is transmitted to the embedded controller via the RS485 link.

[0069] Sampling frequency: tension / tilt angle (10Hz), wind speed (1Hz), strain (50Hz);

[0070] The data analysis and processing module forms the edge computing layer and control layer of the automatic monitoring method. The edge computing layer includes an industrial computer (Intel NUC, running Ubuntu+ROS) for performing data preprocessing, feature extraction, and state classification;

[0071] The control layer includes PLC (Siemens S7-1200), which is used to control the start and stop of the stress monitoring module and the alarm output.

[0072] The data analysis and processing module processes and analyzes the data of the data acquisition module in real time to predict and evaluate the stress state of the crane, and uses data analysis technology to improve the accuracy of the prediction, including the following steps:

[0073] Step S1, collecting data from sensors in the crane monitoring system: lifting weight Q1, lifting moment F1, rotation angle A1, real-time wind speed V1;

[0074] Step S2, when the data collected by the sensor has the following characteristics (which can be changed according to user needs), the crane is judged to be in a high-risk operating state: Q1≥0.9*rated lifting weight ∪F1≥0.9*rated lifting moment weight ∪A1≥90°∪real-time wind speed V1≥10M / S, start data analysis, and when it is determined that the crane is in a high-risk state, a start signal is given to the stress monitoring module.

[0075] In step S2, the data analysis process uses a random forest classifier algorithm, including a data preprocessing algorithm, a feature extraction and selection algorithm;

[0076] The specific goals of the data preprocessing algorithm are: to clean, standardize, and extract features from the collected raw data to ensure the accuracy of subsequent analysis;

[0077] The steps of the data preprocessing algorithm include;

[0078] Step A1, data cleaning: remove invalid data, fill missing values, and remove noise;

[0079] Step A2, data standardization: In order to unify the data scales of different sensors, a standardization method (such as Z-score standardization) is used to normalize all sensor data; the formula is: Among them, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data;

[0080] During data preprocessing, the standardized Q1, A1, V1, σ are input, and the feature matrix X∈R is output. NX4 N = number of samples in the time window;

[0081] The specific goals of the feature extraction and selection algorithm are: extract meaningful features from the cleaned data for subsequent status assessment and anomaly detection;

[0082] The feature extraction and selection algorithm includes the following steps;

[0083] Step B1, basic statistical characteristics: Calculate statistical characteristics such as mean, standard deviation, maximum value, minimum value, kurtosis, skewness, etc. for sensor data.

[0084] For example

[0085]

[0086] Where Zi is each data point, N is the number of data points, and μ is the mean;

[0087] Step B2: Use machine learning models to perform multi-dimensional evaluation of the working status; predict whether the crane is in a normal state or a high-risk state by fusing multiple sensor data (load, wind speed, position, etc.); use the random forest boosting decision tree ensemble learning method to build a multi-classifier to evaluate the crane status at each moment, including the above four data features Q1 F1A1 V1;

[0088] Step B3: Use random forest for classification, specifically:

[0089] Training model: Use labeled historical data (such as normal, overload, excessive wind speed, excessive rotation angle, etc.) to train the random forest model;

[0090] When Q1≥0.9*rated lifting weight, it is marked as overload; when Q1<0.9*rated lifting weight, it is marked as normal;

[0091] When F1≥0.9*rated lifting moment, it is marked as overload; when F1<0.9*rated lifting moment, it is marked as normal;

[0092] When A1≥90°, it is marked as risk data; when A1<90°, it is marked as normal;

[0093] When V1≥10m / s, it is marked as high wind speed; when V1<10m / s, it is marked as normal;

[0094] Model output: For each moment of data, a status label is output, i.e., “normal”, “overload” or “high wind speed”;

[0095] When the data analysis and processing module performs risk assessment, it combines the outputs of all features and models to determine whether the current crane is in a high-risk state; when any one is marked as abnormal, the crane is judged to be in a high-risk state; if none of them is marked as abnormal, the crane is judged to be in a normal state.

[0096] The crane stress automatic monitoring method also includes a control module. The data analysis and processing module transmits the analysis results and risk assessment results to the control module to obtain corresponding control instructions. The control module automatically controls the opening and closing of the stress monitoring module according to the results of the data processing and analysis module and the data in the intelligent identification module. It realizes seamless integration with the crane monitoring system to ensure rapid response when needed.

[0097] Normal status: If the monitored data is within a safe range, the system will continue to monitor and wait for the next data update.

[0098] High-risk status: If an abnormal or high-risk status is detected, the system triggers the alarm mechanism, starts the stress monitoring module, and provides alarm information through the user interface.

[0099] When the control module starts the stress monitoring module after the threshold is triggered, the stress monitoring module is started with a delay of 2 seconds to avoid instantaneous interference.

[0100] When there is no high-risk signal for 5 consecutive minutes and σ<80MPa, the control module turns off the stress monitoring module.

[0101] Embodiment 1,

[0102] Based on the above technical solution, this example proposes an intelligent stress monitoring technical solution for portal cranes, which uses multi-sensor data fusion and dynamic start-stop control; specifically:

[0103] 1. Portal crane specifications and sensor configuration

[0104] 1.1 Crane specifications

[0105] Model: MQ4035 portal crane

[0106] Rated parameters:

[0107] Maximum lifting capacity (Q_max): 40 tons

[0108] Maximum amplitude: 35 meters

[0109] Allowable rotation angle (A_max): ±180° (actual operation limit is ±90°)

[0110] Allowable working wind speed (V_max): ≤10m / s (instantaneous wind speed ≤20m / s requires emergency shutdown)

[0111] Metal structure material: Q345B steel (yield strength 345MPa, fatigue limit σ_lim=120MPa)

[0112] 1.2 Sensor installation location and parameters

[0113]

[0114] 2. System architecture and data flow

[0115] 2.1 Hardware Architecture

[0116] Data acquisition layer: sensor group (tension, inclination, wind speed, strain gauge) → signal conditioning module → embedded controller (STM32H743, supporting multi-protocol communication);

[0117] Edge computing layer: Industrial computer (Intel NUC, running Ubuntu+ROS) → performs data preprocessing, feature extraction, and status classification;

[0118] Control layer: PLC (Siemens S7-1200) → Control the start and stop of the stress monitoring module and the alarm output.

[0119] 2.2 Data Flow Design

[0120] 1. Real-time data collection:

[0121] The sensor data is transmitted to the embedded controller via CAN bus (tension, inclination) and RS485 (wind speed);

[0122] Sampling frequency: tension / tilt angle (10Hz), wind speed (1Hz), strain (50Hz).

[0123] 2. Data preprocessing:

[0124] Data standardization: In order to unify the data scales of different sensors, a standardization method (such as Z-score standardization) can be used to normalize all sensor data.

[0125] formula

[0126]

[0127] Among them, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data.

[0128] 3. Feature extraction

[0129] Time domain feature mean Q avg Variance Q var Peak Q peak

[0130] Frequency domain characteristics: FFT analysis of main frequency components (0-5Hz).

[0131] 3. Working status recognition algorithm

[0132] 3.1 State Definition and Threshold

[0133]

[0134] 3.2 Algorithm Process (Random Forest Classifier)

[0135] Step 1: Data Preprocessing

[0136] Input: Q1, A1, V1, σ (normalized);

[0137] Output: feature matrix X∈R NX4 (N = number of samples in the time window).

[0138] Step 2: Model training data set: historical data (labels: normal / overload / high wind / combined risk), divided into training set and test set in a ratio of 7:3;

[0139] Hyperparameters: n_estimators = 200, max_depth = 10, min_samples_split = 5; Feature importance ranking: Q1 (40%), A1 (30%), V1 (20%), σ (10%).

[0140] Step 3: Real-time classification

[0141] Output probability: Alarm is triggered when P(y=emergency)>0.7;

[0142] Decision logic:

[0143] if(Q1≥36)or(F1≥36×L):

[0144] label="Overload"

[0145] elif(A1≥90)and(V1≥10):

[0146] label="Compound Risk"

[0147] else:

[0148] label=random_forest.predict(X)

[0149] 3.3 Algorithm Verification (Confusion Matrix)

[0150]

[0151] 4. Dynamic start-stop control logic

[0152] 4.1 Control Flow Diagram

[0153] graph TD A[real-time acquisition of Q1, A1, V1]-->B{does it meet any threshold?}

[0154] B-->|Yes|C[Activate stress monitoring module]

[0155] B-->|No|D[Stay dormant]

[0156] C-->E[Collect σ data and analyze]

[0157] E-->F{σ≥120MPa?}

[0158] F-->|Yes|G[Trigger emergency stop]

[0159] F-->|No|H[Continue monitoring until the status is restored]

[0160] H-->I[Turn off stress monitoring module]

[0161] 4.2 Control parameter start delay: after the threshold is triggered, the stress monitoring is started with a delay of 2s (to avoid instantaneous interference); shutdown condition: no high-risk signal for 5 consecutive minutes and σ<80MPa;

[0162] Alarm strategy:

[0163] Level 1 alarm (LED flashing): overload / strong wind;

[0164] Level 2 alarm (sound and light + SMS notification): combined risk or σ≥120MPa.

[0165] The solution described in this example is for the MQ4035 portal crane. It achieves high-precision state recognition and energy-saving monitoring through multi-sensor data fusion, random forest classification and dynamic control logic. The design takes into account environmental adaptability and ease of operation, and can be extended to other large crane models.

[0166] Embodiment 2,

[0167] This example is based on Example 1. A gantry crane performs container lifting operations at a port, and sensors collect lifting weight (Q1), rotation angle (A1), wind speed (V1) and stress (σ) of key parts in real time.

[0168] The overload detection in this example is:

[0169] If the current lifting capacity Q1 ≥ 36 tons (90% of the rated 40 tons), or the calculated lifting torque F1 (Q1 × operating range) exceeds the safety threshold, the system will immediately determine it as "overload" and trigger a level 1 alarm (flashing yellow light).

[0170] Example: When a crane lifts 38 tons of cargo, Q1 = 38 tons. The system skips model prediction and directly marks "overload".

[0171] Composite risk detection:

[0172] If the slewing angle A1≥90° (such as the crane swings horizontally to the extreme position) and the wind speed V1≥10m / s (strong wind environment) are met at the same time, the system determines it as a "compound risk" and triggers a secondary alarm (red light + buzzer). In this example, the scenarios where the machine learning model intervenes are:

[0173] When the sensor data does not trigger the threshold but there is potential risk (such as Q1 = 34 tons, A1 = 85°, V1 = 9m / s), the risk needs to be further assessed through the random forest model.

[0174] Algorithm execution steps:

[0175] Data Standardization:

[0176] Convert Q1, A1, V1, σ real-time data into Z-score values. For example:

[0177] Q1 = 34 tons → Z = (34-20) / 8 = 1.75 (assuming historical mean μ = 20 tons, standard deviation σ = 8 tons);

[0178] A1=85°→Z=(85-45) / 30≈1.33 (assuming historical mean μ=45°, standard deviation σ=30°).

[0179] Feature input and model prediction:

[0180] The standardized data is input into the pre-trained random forest model, and the model outputs the probability of each state: P(normal) = 10%, P(overload) = 15%, P(strong wind) = 5%, and P(compound risk) = 70%.

[0181] If the probability of "compound risk" is >70% (such as 70% in this example), it is judged as a high-risk state.

[0182] In this example, the scenarios for dynamic start-stop control and stress monitoring are:

[0183] After the system determines that it is a high risk, the stress monitoring module needs to be activated to track the stress changes of the metal structure in real time. Algorithm execution steps:

[0184] Delayed Start:

[0185] To prevent instantaneous interference (such as short gusts of wind), the system waits 2 seconds before re-verifying the data. If the risk persists, the stress monitoring module is activated.

[0186] Stress data collection and analysis:

[0187] The welding strain gauge data is collected at a frequency of 50 Hz, and the stress σ and its rate of change Δσ / Δt are calculated in real time.

[0188] Example:

[0189] σ increases from 100MPa to 125MPa (metamaterial fatigue limit 120MPa), Δσ / Δt=5MPa / s (rapid rising trend).

[0190] Graded Response:

[0191] Level 1 response (σ<120MPa): Continuously monitor and record data, and issue an alarm to prompt the operator to reduce the load or adjust the posture;

[0192] Secondary response (σ≥120MPa): Immediately cut off the power to the crane, trigger an emergency stop, and send a text message to notify maintenance personnel.

[0193] In this example, the scenarios of state recovery and module hibernation are:

[0194] When the risk is eliminated (such as wind speed drops, cargo is unloaded), the system needs to turn off stress monitoring to save energy.

[0195] Algorithm execution steps:

[0196] Recovery condition detection:

[0197] If Q1<36 tons, A1<90°, V1<10m / s and σ<80MPa are met for 5 consecutive minutes, it is judged as a "safe state".

[0198] Module shutdown:

[0199] Turn off the stress monitoring module, only keep the basic sensors running, and restore the system to low power consumption mode.

[0200] Example: After unloading, Q1=25 tons, wind speed V1=8m / s, σ drops back to 70MPa, and the system automatically goes into sleep mode after 5 minutes.

[0201] In this example, the algorithm can take into account both real-time performance and accuracy through the hierarchical strategy of "threshold judgment priority + model prediction supplement":

[0202] In this case, the threshold judgment directly intercepts clear risks (such as overload), which can shorten the response time;

[0203] In this case, the random forest model solves complex working conditions (such as multi-parameter coupling risks) and can reduce the false negative rate;

[0204] In this example, dynamic start-stop control balances safety and energy consumption, ensuring long-term reliable operation of the system.

[0205] The data preprocessing process in this example is as follows:

[0206] 1.1 Data cleaning and filtering Input data: raw data from the sensor group (Q1, A1, V1, σ), with sampling frequencies of 10 Hz (Q1 / A1), 1 Hz (V1), and 50 Hz (a), respectively.

[0207] 1. Sliding window filtering

[0208] Purpose: To suppress high-frequency noise (such as mechanical vibration, electromagnetic interference).

[0209] Method: A Hamming window moving average filter was used with a window length of 10 seconds (covering 100 Q1 / A1 samples).

[0210] formula:

[0211] Wherein, N=100, w(i) is the Hamming window coefficient, satisfying ∑w(i)=1.

[0212] 2. Outlier processing

[0213] Threshold rejection: If the instantaneous value of Q1 exceeds 110% of the rated value (44 tons), it is marked as abnormal and filled with linear interpolation;

[0214] Wild value of wind speed: If V1>30m / s (upper limit of sensor range), it is regarded as invalid data and replaced by the average value of the previous 10 seconds.

[0215] 1.2 Purpose of data standardization: to eliminate dimensional differences and improve model convergence speed.

[0216] 1. Z-score standardization:

[0217] Calculate the mean (μ) and standard deviation (a) of each parameter:

[0218]

[0219] Example: Q1's μ = 20 tons, σ = 8 tons, then Q1 = 36 tons is standardized to Z = (36-20) / 8 = 2.0.

[0220] 2. Detailed explanation of the model training process

[0221] 2.1 Dataset Construction

[0222] 1. Data source: historical operation data (including normal, overload, high wind and other labels), totaling 100,000 samples.

[0223] 2. Feature Engineering:

[0224] Time domain characteristics: mean (Q avg ), variance (Q var ), peak value (Q PEAK =max(Q1));.

[0225] Frequency domain features: extract the energy proportion of the 0-5Hz frequency band through FFT (reflecting the mechanical vibration mode); composite features: such as "torque-amplitude ratio" (Detection torque exceeding limit).

[0226] 2.2 Random Forest Model Training

[0227] 1. Hyperparameter settings:

[0228] n_estimators = 200: number of decision trees in the forest;

[0229] max_depth=10: maximum depth of a single tree to prevent overfitting;

[0230] min_samples_split=5: minimum number of samples required for node splitting.

[0231] 2. Training steps:

[0232] Data division: divided into training set (70,000 records) and test set (30,000 records) in a ratio of 7:3;

[0233] Cross-validation: 5-fold cross-validation to optimize hyperparameters;

[0234] Feature importance evaluation: Based on the Gini index ranking, the top four features (Q1, A1, V1, σ) are selected.

[0235] 3. Model evaluation:

[0236] Confusion matrix (see previous table): overall accuracy ≥ 95%;

[0237] ROC curve: AUC value ≥ 0.98, showing high classification discrimination.

[0238] 3.1 The algorithm flow is expressed in pseudo code as follows:

[0239]

[0240]

Claims

1. A novel crane stress automatic monitoring method, characterized in that: The method comprises a data acquisition module and a stress monitoring module. The method identifies the current working state of the crane according to the monitoring system signal. When it is determined that the current working state of the crane is in a high-risk operation stage, stress monitoring of key stress-bearing parts of the crane is started.

2. A novel crane stress automatic monitoring method according to claim 1, characterized in that: The crane is a portal crane. When deployed, the data acquisition module is integrated into the original crane monitoring system, and the working status of the crane is identified by using the existing sensor data of the crane; the sensors include load sensors and position sensors; The method analyzes the sensor data through a data analysis and processing module to determine whether the crane is in a high-risk operation stage, thereby determining whether stress monitoring needs to be turned on; The sensor of the stress monitoring module includes a strain gauge and a data collector, which are used to monitor the stress changes of the key stress-bearing parts of the crane in real time. The strain gauge is fixed on the crane with a matching tooling; When the data analysis and processing module determines that the current crane is in a high-risk operation stage and needs to be monitored, the stress monitoring module is activated through a start signal to start collecting stress data.

3. A novel crane stress automatic monitoring method according to claim 2, characterized in that: The installation method of the strain gauge is as follows: the strain gauge is connected to the collector of the data acquisition module by using a male and female cable with anti-loosening function, the strain gauge is fixed to the surface of the crane by welding, the signal line is passed through the plug hole on the protective shell, and the signal line and the cable of the strain gauge are welded in the corresponding order; The installation method of the sensor is as follows: put the protective shell on the sensor, fix it on the crane surface through multiple fixing feet of the protective shell, and finally seal the surrounding of the protective shell and the fixing feet with silicone glue; The automatic monitoring method can select a periodic monitoring mode or an online monitoring mode according to the needs of users and equipment. When the periodic monitoring mode is selected, the sensor is installed on the crane in a quick plug-in manner to shorten the installation time and ensure the installation quality.

4. A novel crane stress automatic monitoring method according to claim 2, characterized in that: The data acquisition module forms the data acquisition layer of the automatic monitoring method, and its various sensors include an S-type tension sensor installed at the hinge point at the root of the crane boom, an inclination sensor installed at the connection between the crane revolving platform and the column, an ultrasonic anemometer installed at the top of the crane boom, and a welded strain gauge installed at the mid-span of the crane main beam and the weld of the door leg.

5. A novel crane stress automatic monitoring method according to claim 4, characterized in that: The sensors in the data acquisition layer form a sensor group. The collected data of the sensor group is processed by the signal conditioning module and then input into the embedded controller. The data of the tension sensor and the tilt sensor are transmitted to the embedded controller via the CAN bus. The data of the wind speed sensor is transmitted to the embedded controller via the RS485 link. The data analysis and processing modules form the edge computing layer and control layer of the automatic monitoring method. The edge computing layer includes industrial computers, which are used to perform data preprocessing, feature extraction, and status classification; The control layer includes PLC, which is used to control the start and stop of the stress monitoring module and the alarm output.

6. A novel crane stress automatic monitoring method according to claim 4, characterized in that: The data analysis and processing module processes and analyzes the data of the data acquisition module in real time to predict and evaluate the stress state of the crane, and uses data analysis technology to improve the accuracy of the prediction, including the following steps: Step S1, collecting data from sensors in the crane monitoring system: lifting weight Q1, lifting moment F1, rotation angle A1, real-time wind speed V1; Step S2, when the data collected by the sensor has the following characteristics, the crane is judged to be in a high-risk operating state: Q1≥0.9*rated lifting weight ∪F1≥0.9*rated lifting moment weight ∪A1≥90°∪real-time wind speed V1≥10M / S, start data analysis, and when it is determined that the crane is in a high-risk state, a start signal is given to the stress monitoring module.

7. A novel crane stress automatic monitoring method according to claim 6, characterized in that: In step S2, the data analysis process uses a random forest classifier algorithm, including a data preprocessing algorithm, a feature extraction and selection algorithm; The specific goals of the data preprocessing algorithm are: to clean, standardize, and extract features from the collected raw data to ensure the accuracy of subsequent analysis; The steps of the data preprocessing algorithm include; Step A1, data cleaning: remove invalid data, fill missing values, and remove noise; Step A2, data standardization: In order to unify the data scales of different sensors, a standardization method is used to normalize all sensor data; the formula is: Among them, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data; During data preprocessing, the standardized Q1, A1, V1, σ are input, and the feature matrix X∈R is output. NX4 N = number of samples in the time window; The specific goals of the feature extraction and selection algorithm are: extract meaningful features from the cleaned data for subsequent status assessment and anomaly detection; The feature extraction and selection algorithm includes the following steps; Step B1, basic statistical characteristics: calculate statistical characteristics such as mean, standard deviation, maximum value, minimum value, kurtosis, skewness, etc. of sensor data; Step B2: Use machine learning models to perform multi-dimensional evaluation of the working status; use multiple sensor data fusion to predict whether the crane is in a normal state or a high-risk state; use the random forest boosting decision tree ensemble learning method to build a multi-classifier to evaluate the crane status at each moment, including the above four data features Q1 F1A1 V1; Step B3: Use random forest for classification, specifically: Training model: Use labeled historical data to train a random forest model; When Q1≥0.9*rated lifting weight, it is marked as overload; when Q1<0.9*rated lifting weight, it is marked as normal; When F1≥0.9*rated lifting moment, it is marked as overload; when F1<0.9*rated lifting moment, it is marked as normal; When A1≥90°, it is marked as risk data; when A1<90°, it is marked as normal; When V1≥10m / s, it is marked as high wind speed; when V1<10m / s, it is marked as normal; Model output: For each moment of data, a status label is output, i.e. "normal", "overload" or "high wind speed"; When the data analysis and processing module performs risk assessment, it combines the outputs of all features and models to determine whether the current crane is in a high-risk state; when any one is marked as abnormal, the crane is judged to be in a high-risk state; if none of them is marked as abnormal, the crane is judged to be in a normal state.

8. A novel crane stress automatic monitoring method according to claim 7, characterized in that: The crane stress automatic monitoring method also includes a control module. The data analysis and processing module transmits the analysis results and risk assessment results to the control module to obtain corresponding control instructions. The control module automatically controls the opening and closing of the stress monitoring module according to the results of the data processing and analysis module and the data in the intelligent identification module. It realizes seamless integration with the crane monitoring system to ensure rapid response when needed. Normal status: If the monitored data is within a safe range, the system will continue to monitor and wait for the next data update. High-risk status: If an abnormal or high-risk status is detected, the system triggers the alarm mechanism, starts the stress monitoring module, and provides alarm information through the user interface.

9. A novel crane stress automatic monitoring method according to claim 8, characterized in that: When the control module starts the stress monitoring module after the threshold is triggered, the stress monitoring module is started with a delay of 2 seconds to avoid instantaneous interference.

10. A novel crane stress automatic monitoring method according to claim 8, characterized in that: When there is no high-risk signal for 5 consecutive minutes and σ<80MPa, the control module turns off the stress monitoring module.

Images