Intelligent safety inspection system for construction site

Abstract

The invention discloses an intelligent safety inspection system for a construction site, and relates to the technical field of construction inspection, multiple data sources of the construction site are obtained through a data acquisition module, and a data fusion and processing module dynamically optimizes heterogeneous data through time synchronization, format standardization and noise filtering in combination with a self-adaptive weighted fusion algorithm, so that the construction site safety is improved. The intelligent analysis and early warning module evaluates the data difference degree based on machine learning, adopts clustering analysis to mine an abnormal mode for high-difference data, carries out safety risk identification on low-difference data, and dynamically adjusts an early warning strategy. The terminal interaction module pushes the inspection result and the early warning information to the mobile terminal APP, the monitoring center display screen and the voice broadcast device, so that a manager can remotely check the construction condition and quickly make a decision, the intelligent level of construction site safety inspection can be remarkably improved, efficient and accurate potential safety hazard recognition and dynamic early warning are achieved, and the safety inspection efficiency is improved. And the safety accident risk is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of construction inspection, and in particular to an intelligent safety inspection system for a construction site. Background Art

[0002] With the rapid development of the construction industry, safety management of construction sites has become particularly important. Traditional safety inspections mainly rely on manual inspections and regular inspections, but this approach has many problems, such as low inspection frequency, strong subjectivity of personnel, and untimely discovery of hidden dangers. In recent years, the rapid development of intelligent technology has provided new ideas for safety inspections at construction sites. By integrating technologies such as the Internet of Things, artificial intelligence, and big data analysis, real-time monitoring and intelligent management of construction sites can be achieved, thereby improving the efficiency and accuracy of safety inspections.

[0003] The existing technology has the following shortcomings: In the intelligent safety inspection system, there may be multiple sources of safety risks at the construction site, including illegal operation by personnel, falling objects from high altitude, mechanical equipment failure, etc. These risks need to be monitored through multiple data sources such as video surveillance, sensors, RFID, etc. However, there are large differences in the information format, sampling frequency, transmission delay and other characteristics of different data sources. How to efficiently integrate and analyze these heterogeneous data in a high-noise and dynamically changing construction environment to ensure the real-time and accuracy of the inspection system is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent safety inspection system for a construction site to solve the deficiencies in the background technology.

[0005] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: an intelligent safety inspection system for a construction site, comprising a data acquisition module, a data fusion and processing module, an intelligent analysis and early warning module and a terminal interaction module; A data acquisition module is used to obtain multi-source data at the construction site, including video surveillance data, sensor data, RFID identification information and environmental data, wherein the sensor data includes temperature, humidity, dust concentration and vibration intensity data; The data fusion and processing module is used to synchronize the time, standardize the data format and filter the noise of the collected multi-source data, and dynamically optimize and fuse different data sources based on the adaptive weighted fusion algorithm; The intelligent analysis and early warning module is used to evaluate the difference degree of the fused multi-source data based on the machine learning model. For multi-source data with high difference degree, it further mines abnormal patterns through cluster analysis; for multi-source data with low difference degree, it identifies security risks and dynamically adjusts the early warning strategies of different levels based on the risk level and the change of the difference degree of multi-source data to achieve real-time security early warning; The terminal interaction module is used to push the safety inspection results and warning information to the terminals of management personnel, including the mobile APP, the monitoring center display screen and the voice broadcast device. The management personnel can remotely view the construction site conditions and make safety decisions through the terminal interaction module.

[0006] Preferably, in the intelligent analysis and warning module, after analyzing the vibration frequency change rate through fast Fourier transform, a vibration frequency change rate fluctuation index is generated. The acquisition method of the vibration frequency change rate fluctuation index is as follows: Use a high-precision acceleration sensor to collect the vibration signal of the construction structure in the time series. Perform FFT transformation on the collected time-domain vibration signal A(t) and convert it to the frequency domain to obtain the frequency spectrum A(f): ; The amplitude spectrum calculated by FFT: ; where, is the k-th frequency component, k = 0, 1, 2,..., N / 2, are respectively the real part and the imaginary part of; Find the frequency component with the largest amplitude in the FFT result, that is, the main vibration frequency : ; Compare the change of the main vibration frequency in adjacent time windows and calculate the vibration frequency change rate , and the expression is: ; where: is the main vibration frequency of the i-th time window, is the main vibration frequency of the i-1-th time window, and Δt is the time window interval; Calculate the vibration frequency change rate fluctuation index VFVFI, and the expression is: ; where: M is the total number of time windows in the calculation period, is the average value of the vibration frequency change rates in all windows.

[0007] Preferably, after analyzing the change of the movement pattern stability of construction workers through the LSTM model, a movement stability anomaly index is generated. The acquisition method of the movement stability anomaly index is as follows: Use an inertial sensor to sample the movement state of construction workers. The sensor data at each moment is expressed as: ; where: is the three-axis acceleration, is the three-axis angular velocity, is the three-axis magnetometer data; The LSTM network training includes: input layer: the input data is in the shape; LSTM layer: extract the time series features and output the hidden state ; fully connected layer: map to the stability score ; Loss function: Calculate the error between the LSTM predicted value and the true stability score using the mean squared error to optimize the model; Use the LSTM model to predict the stability score within each time window H After that, calculate the change rate of motion pattern stability , and the expression is: ; where: is the stability score of the current window, is the stability score of the previous window, and Δt is the window interval; Calculate the motion stability anomaly index MSAI, and the calculation formula is: ; where: Q is the total number of time windows, is the change rate of stability in the i-th window, is the mean value of the change rates of stability within all windows.

[0008] Preferably, convert the vibration frequency change rate fluctuation index and the motion stability anomaly index into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes predicting the difference degree value label of the fused multi-source data as the prediction target, and takes minimizing the sum of the prediction errors of the difference degree value labels of all fused multi-source data as the training target to train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the difference degree value of the fused multi-source data according to the model output result, where the machine learning model is a polynomial regression model.

[0009] Preferably, compare the obtained difference degree value of the fused multi-source data with the difference degree reference threshold preset according to historical data. If the difference degree value of the fused multi-source data is greater than or equal to the preset difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is high, and mark it as multi-source data with a high difference degree; if the difference degree value of the fused multi-source data is less than the preset difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is low, and mark it as multi-source data with a low difference degree.

[0010] Preferably, for the multi-source data with a high difference degree, further mine the abnormal patterns through cluster analysis. Specifically: The input data for cluster analysis comes from the multi-source data samples with a high difference degree, and each sample contains multiple feature values: ; where: VFVFI is the vibration frequency change rate fluctuation index, and MSAI is the motion stability anomaly index; is the main vibration frequency, is the vibration frequency change rate, is the standard deviation of the vibration change rate, is the stability score predicted by LSTM, is the change rate of motion stability; all samples constitute a high-dimensional feature dataset: , J is the total number of features; select the value of K, initialize K cluster centers, and iterate and optimize: ; among them, is the center of the k-th category, is the set of samples belonging to this category; calculate the Euclidean distance from each data point to the cluster center , and classify it into the nearest cluster: ; represents the j-th feature value of the i-th sample; represents the value of the k-th cluster center on the j-th feature dimension; iterate and update until convergence; the high-difference data at the construction site is clustered into categories such as abnormal equipment vibration, unstable construction structure, and abnormal personnel behavior.

[0011] Preferably, based on the risk level and the change of the difference degree of multi-source data, dynamically adjust early warning strategies at different levels to achieve real-time safety early warning, specifically: Take the risk level and the difference degree value of multi-source data as the input items of fuzzy logic, and take early warning strategies at different levels as the output items of fuzzy logic, and perform dynamic regulation through fuzzy logic; Use the fuzzy membership function to fuzzify the input, use the fuzzy rule base for reasoning, and use the centroid method for defuzzification to calculate the final early warning level value , ; among them: is the membership degree value corresponding to different rules, is the numerical mapping of different early warning strategies; According to the fuzzy inference result, dynamically execute the corresponding early warning measures.

[0012] In the above technical solution, the technical effects and advantages provided by the present invention: 1. Through the data acquisition module, the present invention systematically integrates various data sources such as video monitoring, sensors, and RFID to ensure comprehensive perception of the construction site; the data fusion and processing module uses the adaptive weighted fusion algorithm to solve the differences in format, sampling frequency, and transmission delay of different data sources, and improves data consistency and reliability. The intelligent analysis and early warning module is based on machine learning and polynomial regression models. By calculating the vibration frequency change rate fluctuation index (VFVFI) and the motion stability anomaly index (MSAI), it accurately evaluates the data difference degree, and further mines patterns such as abnormal equipment vibration, unstable construction structure, and abnormal personnel behavior through cluster analysis, so as to achieve efficient and accurate risk identification and hierarchical early warning.

[0013] 2. The present invention realizes dynamic warning adjustment under different risk levels through fuzzy logic reasoning, avoiding the limitations of the traditional fixed-threshold warning method, enabling the system to intelligently adjust the warning level according to the actual on-site situation, and improving the warning accuracy and response speed. The terminal interaction module supports mobile APP, monitoring center display screen and voice broadcast. Managers can remotely view the inspection results and warning information in real time and make safety decisions, improving management efficiency. The present invention can significantly improve the intelligence, precision and real-time performance of construction safety monitoring, reduce the risk of construction accidents, improve the overall safety management level of the construction site, and contribute to the construction of an intelligent construction site. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0015] Figure 1 It is the system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0017] Embodiment, please refer to Figure 1 As shown, the intelligent safety inspection system for a construction site in this embodiment includes a data acquisition module, a data fusion and processing module, an intelligent analysis and warning module, and a terminal interaction module; The data acquisition module is used to obtain multi-source data of the construction site, including video monitoring data, sensor data, RFID identification information and environmental data. Among them, the sensor data includes temperature, humidity, dust concentration and vibration intensity data; The data fusion and processing module is used to perform time synchronization, data format standardization and noise filtering on the acquired multi-source data, and perform dynamic optimization fusion on different data sources based on the adaptive weighted fusion algorithm; The intelligent analysis and early warning module is used to evaluate the degree of difference of the fused multi-source data based on a machine learning model. For multi-source data with a high degree of difference, abnormal patterns are further mined through cluster analysis; for multi-source data with a low degree of difference, security risks are identified, and different levels of early warning strategies are dynamically adjusted based on the risk level and the change of the degree of difference of the multi-source data to achieve real-time security early warning. The terminal interaction module is used to push the safety inspection results and early warning information to the management terminal, including the mobile APP, the monitoring center display screen and the voice broadcast device. The management personnel can remotely view the construction site conditions and make safety decisions through the terminal interaction module.

[0018] The data acquisition module is responsible for obtaining various data from the construction site to comprehensively perceive the construction environment and the status of personnel and equipment. Among them, video surveillance data is collected in real time through high-definition cameras deployed in key areas. These cameras can not only provide high-definition video streams, but also support night vision and zoom functions to ensure that details of the construction site can be accurately captured under various lighting conditions. Video data can be used to detect dangerous scenarios such as personnel's illegal operations, abnormal equipment operation status, and high-altitude operations.

[0019] The sensor data includes environmental variables such as temperature, humidity, dust concentration, and vibration intensity. Temperature and humidity sensors are mainly installed at different heights and positions in the construction area to monitor environmental changes in real time and prevent extreme weather from affecting construction safety. The dust concentration sensor is used to detect the content of dust particles in the air, especially in high-dust operation areas such as welding, cutting, and grinding, to timely identify the situation of excessive air quality and reduce the risk of workers inhaling harmful substances. The vibration intensity sensor is usually installed on building structures, scaffolding, and mechanical equipment to monitor the vibration situation during the construction process and prevent structural instability or equipment failure caused by excessive vibration.

[0020] The RFID identification information is used to track the personnel and equipment on the construction site. Construction personnel wear RFID tags. When they enter or leave the designated area, the RFID reader will automatically record the personnel's entry and exit situation, and combined with the attendance management system, ensure that construction personnel work in the correct construction area within the specified time. In addition, RFID technology can also be used for intelligent identification and positioning of key equipment, such as large mechanical equipment like cranes and excavators, to record the equipment's operation status, maintenance time, and user information through RFID tags, improving the intelligent level of equipment management.

[0021] Environmental data mainly includes information such as wind speed and noise. Wind speed sensors are usually installed in high-altitude working areas, such as tower cranes and scaffolding, to monitor the wind intensity in real time. When the wind speed exceeds the safety standard, the system will automatically issue a warning to remind construction workers to suspend high-altitude operations. Noise sensors are used to monitor the noise level at the construction site, especially when constructing near residential areas. By controlling noise pollution, the impact on the surrounding environment can be reduced.

[0022] Through the multi-source data acquisition ability of the data acquisition module, the system can perceive the overall state of the construction site in real time, providing high-quality information input for subsequent data fusion and intelligent analysis, and ensuring the comprehensiveness and accuracy of safety inspections.

[0023] The data fusion and processing module is used to deeply process the multi-source data collected at the construction site to ensure the accuracy, timeliness, and consistency of the data, providing high-quality data support for subsequent intelligent analysis. This module mainly includes time synchronization, data format standardization, noise filtering, and data optimization fusion based on the adaptive weighted fusion algorithm.

[0024] First of all, time synchronization is a key step to ensure the effective fusion of multi-source data. The sampling frequencies and timestamps of various sensors, video monitoring devices, and RFID readers at the construction site may be different. Direct use may lead to data asynchronization and affect the accuracy of the analysis results. To solve this problem, the system adopts a global time synchronization mechanism. Through the Network Time Protocol (NTP) or GPS timing technology, it ensures that the timestamps of all data sources are consistent with the system time. In addition, for sensors with clock drift, the system will perform time calibration regularly and use interpolation algorithms to align the data during the data processing stage to ensure that the data is analyzed under the same time reference.

[0025] Secondly, data format standardization is used to solve the problem of inconsistent formats of different data sources. The data types at the construction site are diverse. For example, video data is usually in a streaming media format, sensor data may be discrete numerical values, and RFID information is an identifier plus a time record. The system converts all data into a standardized format through a unified data interface, such as using JSON or database storage format, so that different types of data can be processed and stored on the same platform. In addition, for video data, the system will extract key frames and perform feature encoding to convert the video information into structured data for correlation analysis with sensor data.

[0026] Noise filtering is an important part of data processing. Especially in a complex environment such as a construction site, data may be affected by electromagnetic interference, environmental changes, or equipment errors. The system adopts a variety of data cleaning and denoising algorithms. For example, the Kalman filtering algorithm is used for sensor data to smooth out sudden changes in data and improve data stability; for video image data, a convolutional neural network (CNN) is used for denoising to filter out interference information caused by light changes or dust occlusion; for RFID signals, average filtering and outlier detection are used to eliminate misread situations caused by signal reflection or occlusion.

[0027] Finally, the system dynamically optimizes and fuses different data sources based on an adaptive weighted fusion algorithm to improve the reliability and integrity of the data. This algorithm dynamically adjusts the weights according to the credibility, historical error rate, and environmental adaptability of different data sources. For example, in sufficient light, the weight of video surveillance data is higher, while at night or in bad weather, the weights of RFID and sensor data will increase accordingly. In addition, the system will also combine historical data and continuously optimize the weight allocation through a machine learning model to make the results of data fusion more accurate and stable.

[0028] Through the data fusion and processing module, the system can effectively solve problems such as out-of-sync, format mismatch, and noise interference caused by multi-source heterogeneous data at the construction site, provide high-quality input data for intelligent analysis and early warning, and thus improve the accuracy and real-time performance of safety inspections.

[0029] The intelligent analysis and early warning module is used to evaluate the degree of difference of the fused multi-source data based on a machine learning model, including: The vibration frequency change rate is one of the important parameters for the structural safety of the construction site, mainly used to monitor the stability of buildings, scaffolding, or large equipment. Different from traditional vibration intensity detection, the vibration frequency change rate focuses on the dynamic changes of vibration modes, such as vibration mode drift caused by uneven loads, mechanical shocks, or structural looseness. This parameter can be obtained through high-precision acceleration sensors or microelectromechanical system (MEMS) sensors and analyzed using signal processing techniques such as the fast Fourier transform (FFT).

[0030] When the vibration frequency change rate remains stable or fluctuates within a preset safe range, it usually indicates that the construction structure or equipment is operating normally.

[0031] When the vibration frequency change rate increases abnormally (for example, the change exceeds the set threshold within a short period of time), it may mean early signs of scaffolding loosening, foundation settlement, or even equipment failure.

[0032] When the vibration frequency change rate decreases but the amplitude increases, it may indicate that the structure has undergone non-linear deformation, such as the expansion of concrete cracks or fatigue damage of steel structures.

[0033] After analyzing the change rate of vibration frequency through fast Fourier transform, a vibration frequency change rate fluctuation index is generated. The method for obtaining the vibration frequency change rate fluctuation index is as follows: Use a high-precision acceleration sensor to collect the vibration signal a(t) of the construction structure in a time series, and preprocess the data (denoising, normalization). Assume the sampling frequency is , and the sampling duration is T. Then the total number of data points is: ; The sampling signal data set is expressed as: ; Perform FFT transformation on the collected time-domain vibration signal A(t) to convert it to the frequency domain and obtain the frequency spectrum A(f): ; The amplitude spectrum calculated by FFT: ; where is the kth frequency component, k = 0, 1, 2,..., N / 2, are respectively the real and imaginary parts of; Find the frequency component with the largest amplitude in the FFT result, that is, the main vibration frequency : ; The main vibration frequency represents the main vibration mode of the construction structure.

[0034] Compare the change of the main vibration frequency in adjacent time windows, and calculate the change rate of vibration frequency , and the expression is: ; where: is the main vibration frequency of the ith time window, is the main vibration frequency of the (i - 1)th time window, and Δt is the time window interval (e.g., 5 seconds); Calculate the vibration frequency change rate fluctuation index VFVFI, and the expression is: ; where: M is the total number of time windows within the calculation period, is the mean value of the vibration frequency change rate in all windows.

[0035] The stability of the movement pattern of the operator is a key indicator to measure whether the construction personnel are in a safe state. Different from traditional attendance or positioning information, this parameter focuses on the coherence and rationality of actions such as walking, standing, and climbing of personnel. It mainly analyzes the movement characteristics of personnel through inertial sensors (IMU) or depth cameras, and uses machine learning models (such as LSTM or HMM) for pattern recognition.

[0036] Under normal circumstances, the movement pattern of construction personnel has a certain regularity. For example, walking is relatively stable in the same work area, and climbing actions comply with the operation specifications.

[0037] When the stability of the motion pattern decreases (e.g., sudden discontinuous movements, abnormal stagnation, or fall signals occur), it may indicate that the person has slipped, lost balance, or even been affected by sudden dangerous events, such as falling from a height or being hit by an object.

[0038] If there are long-term instability phenomena in the motion pattern, it may be caused by personnel fatigue, violation of regulations (such as climbing without wearing a safety belt), or complex construction environments (such as slippery floors and increased obstacles). The system needs to further analyze its risk level.

[0039] After analyzing the change in the stability of the construction worker's motion pattern through the LSTM model, a motion stability anomaly index is generated. The method for obtaining the motion stability anomaly index is as follows: Use an inertial sensor (IMU, including accelerometer, gyroscope, magnetometer) to sample the motion state of the construction worker. The sensor data at each moment is represented as: ; where: is the three-axis acceleration (m / s²), is the three-axis angular velocity (° / s), is the three-axis magnetometer data (μT).

[0040] The LSTM network training includes: Input layer: The input data has a shape of (batch_size, N, 9), where 9 is the dimension of the sensor data; LSTM layer: Extract temporal features and output the hidden state ; Fully connected layer: Map to the stability score (between 0 and 1, representing motion stability). Loss function: Use the mean squared error (MSE) to calculate the error between the LSTM predicted value and the true stability score to optimize the model.

[0041] Use the LSTM model to predict the stability score within each time window H After that, calculate the change rate of the motion pattern stability , and the expression is: ; where: is the stability score of the current window, is the stability score of the previous window, and Δt is the window interval (unit: second); Calculate the motion stability anomaly index MSAI. The motion stability anomaly index (MSAI) is used to measure the degree of change in the person's motion pattern and identify abnormal behaviors. The calculation formula is: ; where: Q is the total number of time windows, is the change rate of stability in the i-th window, is the mean value of the change rates of stability within all windows.

[0042] Convert the vibration frequency change rate fluctuation index and the motion stability anomaly index into a comprehensive feature vector, use the comprehensive feature vector as the input of the machine learning model, and use the machine learning model to predict the difference degree value label of the fused multi-source data for each group of comprehensive feature vectors as the prediction target. Use minimizing the sum of the prediction errors for the difference degree value labels of all fused multi-source data as the training target to train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the difference degree value of the fused multi-source data according to the model output result, where the machine learning model is a polynomial regression model.

[0043] The method for obtaining the difference degree value of the fused multi-source data is as follows: Obtain the corresponding function expression from the comprehensive feature vector training data of the trained machine learning model: ; In the formula, is the output function of the model, is the vibration frequency change rate fluctuation index, MSAI is the motion stability anomaly index, is the difference degree value of the fused multi-source data.

[0044] Compare the obtained difference degree value of the fused multi-source data with the pre-set difference degree reference threshold according to historical data. If the difference degree value of the fused multi-source data is greater than or equal to the pre-set difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is high, and mark it as multi-source data with a high difference degree; if the difference degree value of the fused multi-source data is less than the pre-set difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is low, and mark it as multi-source data with a low difference degree.

[0045] For the multi-source data with a high difference degree, further mine abnormal patterns through cluster analysis. Specifically: The input data for cluster analysis comes from the multi-source data samples with a high difference degree, and each sample contains multiple feature values. For example: ; Where: VFVFI is the vibration frequency change rate fluctuation index, MSAI is the motion stability anomaly index; is the main vibration frequency (Hz), is the vibration frequency change rate, is the standard deviation of the vibration change rate, is the stability score predicted by LSTM, is the motion stability change rate; All samples constitute a high-dimensional feature data set: , where J is the total number of features; when looking for abnormal patterns in high-variance data, common clustering algorithms include: K-Means Clustering: suitable for finding different types of abnormal patterns, such as equipment anomalies and personnel anomalies. DBSCAN (Density-Based Spatial Clustering of Applications with Noise): suitable for mining rare but critical abnormal situations, such as extreme vibrations or sudden fall events. Gaussian Mixture Model (GMM): suitable for handling cases where data may overlap, such as personnel behavior anomalies that may be related to environmental factors. Divide the high-variance data into multiple categories through K-Means clustering analysis to identify the main abnormal patterns.

[0046] Select the value of K (the Elbow Method or Silhouette Score can be used).

[0047] Initialize K cluster centers and iterate for optimization: ; where is the center of the k-th category, is the set of samples belonging to this category.

[0048] Calculate the Euclidean distance from each data point to the cluster center , and classify it into the nearest cluster: ; represents the j-th feature value of the i-th sample; represents the value of the k-th cluster center on the j-th feature dimension; iterate and update until convergence.

[0049] The high-variance data at the construction site may be clustered into categories such as equipment vibration anomalies, construction structure instability, and personnel behavior anomalies.

[0050] Each category can be further analyzed through the feature mean. For example: Category A (equipment abnormal vibration): high VFVFI, high , low ; Category B (personnel behavior anomaly): low VFVFI, high MSAI, high .

[0051] Based on the results of the clustering analysis, an intelligent early warning strategy can be formulated: For high-risk categories (such as the abnormal patterns found by K-Means): If a certain cluster exists for a long time, it indicates that there may be systematic risks at the construction site, such as equipment aging or personnel violation of regulations. Take measures such as equipment maintenance and construction specification inspection.

[0052] Identify security risks for multi-source data with a low degree of difference, and dynamically adjust early warning strategies at different levels based on the risk level and the change of the difference degree of multi-source data to achieve real-time security early warning, including: In the case of a low degree of difference, the sensor data and personnel movement data at the construction site are generally stable, but there may still be subtle risk signals. Therefore, it is necessary to conduct a fine-grained analysis of the following features: The vibration frequency variation rate VFVFI with low fluctuations but continuous increase may indicate long-term wear of equipment or structures, rather than sudden failures. The main vibration frequency offset trend may mean that the operating state of construction equipment is gradually deteriorating and maintenance is required.

[0053] Minor anomalies in the movement pattern (e.g., a slight increase in MSAI but not reaching the high-difference threshold) may indicate that construction workers are in a fatigued state and a safety accident may occur in the future.

[0054] Repeated violations (such as not wearing a safety helmet, short-term illegal operations) may not trigger a high-risk alarm at a single moment, but may increase the probability of accidents after accumulation.

[0055] Slight exceedances of air quality, temperature, humidity, and dust concentration may not immediately cause a safety accident, but if they persist, they may have a long-term impact on the health of workers. A high noise level but not exceeding the danger threshold may imply the gradual aging of construction equipment or the existence of hidden dangers that are not easily detected. Through the analysis of these subtle features, the system can identify potential risk signals and classify the risk levels.

[0056] According to the security risk assessment results of low-difference data, the risks can be classified into low risks, medium risks, and early warning risks.

[0057] Low risk (safe state): The vibration frequency variation rate (VFVFI) is stable and there is no obvious trend of increase. The movement pattern anomaly index (MSAI) has small fluctuations and no abnormal trend. The environmental parameters are within the safe range and there are no obvious fluctuations. The behavior patterns of construction workers are normal and there are no violation records. Disposal measures: The system continuously monitors, no intervention is required, and only routine records are made.

[0058] Medium risk (state to be concerned about): The vibration frequency variation rate (VFVFI) has a slow upward trend but still does not exceed the safety threshold. For example, 0.1 < VFVFI < 0.3 indicates that construction equipment or structures may be in an early aging state.

[0059] The Motion Pattern Abnormality Index (MSAI) shows slight fluctuations but does not reach the level of falls or extreme behaviors, e.g., 0.1 < MSAI < 0.25; this may indicate an unstable work rhythm for construction workers. Construction workers have minor violations (such as not wearing safety helmets for a short period), but these do not occur continuously. Environmental factors are close to but do not exceed the safety threshold, e.g., the dust concentration increases but does not reach the dangerous level. Disposal measures: Record potential risks and regularly review the data trends. If it is found that the trend continues to deteriorate, escalate to the warning-level risk. Construction workers and equipment managers receive reminders to pay attention to safe operation and equipment maintenance.

[0060] Warning-level risks (requiring measures) include: The rate of change of vibration frequency (VFVFI) exceeds the medium-risk threshold and continues to rise: 0.3 ≤ VFVFI < 0.5. This may indicate that the equipment or structure has entered the stage of precursor to failure.

[0061] The Motion Pattern Abnormality Index (MSAI) continues to rise but does not reach the level of sudden falls: 0.25 ≤ MSAI < 0.4; this may indicate that construction workers are in a state of excessive fatigue and safety accidents may occur in the future. Construction workers have repetitive violations, such as not wearing safety equipment multiple times, or show brief but frequent unstable behaviors (such as abnormal gait). Environmental parameters are close to dangerous values, such as noise levels and dust concentrations, which have approached the upper limit of safety standards. Disposal measures: Trigger an automatic warning and notify the safety administrator. Require construction workers to take a short break to reduce the probability of fatigue-related accidents. Conduct early maintenance on equipment to prevent small problems from evolving into serious failures. If environmental factors continue to approach dangerous values, adjust the construction arrangement to reduce the exposure risk.

[0062] Use the risk level and the degree of difference value of multi-source data as input items for fuzzy logic, and use different levels of warning strategies as output items of fuzzy logic for dynamic regulation; The fuzzy logic system needs to fuzzify the input items (risk level and the degree of difference value CS), that is, define their membership functions for reasoning.

[0063] Input item 1: Risk level , and the risk level is defined as: Low (low risk): ≤ 0.3; Medium (medium risk): 0.3 < ≤ 0.7; High (high risk): > 0.7.

[0064] Represented by a triangular membership function: ; ; ; The difference degree value CS of multi-source data, Low (low difference): CS ≤ 0.3; Medium (medium difference): 0.3 < CS ≤ 0.7; High (high difference): CS > 0.7; The definition of its membership function is similar to the expression of

[0065] Output warning strategy: Mild Warning (mild warning): Triggered when the risk is low and the difference is low. Moderate Warning (moderate warning): Triggered when the risk is medium or the data has medium differences. Severe Warning (severe warning): Triggered when the risk is high or the difference is high, and immediate action is required.

[0066] When the risk level is Low: The data difference degree is Low: The construction status is stable and the risk is small. Therefore, only a mild warning (Mild Warning) is required, and the system can operate normally without additional intervention.

[0067] The data difference degree is Medium: Although the data fluctuates slightly, it is still within the acceptable range. A mild warning (Mild Warning) can still be used to remind the management to pay attention to construction safety.

[0068] The data difference degree is High: Although the risk level is low, the data difference degree is large, which may imply potential problems. Therefore, a moderate warning (Moderate Warning) is required, and it is recommended to inspect the construction site to ensure no safety hazards.

[0069] When the risk level is Medium: The data difference degree is Low: The overall construction status is relatively stable, but due to the increased risk level, vigilance should still be maintained. Therefore, a mild warning (Mild Warning) is used, and it is recommended to regularly check the construction status.

[0070] The data difference degree is Medium: Both the risk level and the data difference degree are at a medium level, indicating that potential safety hazards may be accumulating. Therefore, a moderate warning (Moderate Warning) should be adopted, and it is recommended that on-site management focus on it and conduct equipment inspections and safety checks.

[0071] The data difference degree is High: In the case of a medium risk level, if the data difference degree is large, it may mean that a safety accident is about to occur. Therefore, a severe warning (Severe Warning) must be taken, requiring the management to intervene immediately, investigate potential problems, and formulate safety prevention measures.

[0072] When the risk level is High: Low data difference: Although the data change is small, due to the high risk level itself, a Moderate Warning still needs to be issued to remind the management to remain highly vigilant and conduct inspections in the construction area.

[0073] Medium data difference: At a high risk level, a medium data difference indicates that the construction safety situation may be deteriorating. Therefore, a Severe Warning needs to be issued to strengthen construction supervision, and it may be necessary to temporarily adjust the construction plan or conduct equipment maintenance.

[0074] High data difference: When the risk level is already very high and the data difference also reaches a high level, it means that the probability of an accident is extremely high. Therefore, a Severe Warning must be immediately executed, and it may be necessary to suspend construction, evacuate personnel, or take emergency measures to avoid serious safety accidents.

[0075] The Mamdani fuzzy inference method is used for dynamic adjustment, calculating the weights of different rules to obtain the final warning level.

[0076] The Centroid Method is used to calculate the numerical value of the final warning level: ; where: is the membership value corresponding to different rules, is the numerical mapping of different warning strategies (e.g., Mild = 1, Moderate = 2, Severe = 3).

[0077] According to the fuzzy inference results, the system dynamically executes corresponding warning measures: Mild Warning: Only send a reminder message to alert the management. Moderate Warning: It is recommended to conduct equipment inspections and personnel safety checks. Severe Warning: Immediately notify the management, suspend construction, and activate the safety emergency plan.

[0078] The terminal interaction module is the core output interface of the intelligent safety inspection system at the construction site. It is responsible for promptly pushing the safety inspection results and early warning information to the terminals of management personnel to achieve remote monitoring and efficient decision-making. This module integrates the mobile APP, the display screen in the monitoring center, and voice broadcast devices, ensuring that management personnel can obtain the safety status of the construction site at any time and place and take corresponding safety management measures. Through multi-terminal information synchronization, management personnel can real-time master the safety inspection results, early warning levels, risk assessment data, and historical records of the construction area, thereby improving the scientific nature and accuracy of construction safety management.

[0079] In terms of the mobile APP, the terminal interaction module supports remote access to the real-time monitoring data of the construction site, including functions such as video surveillance images, safety inspection reports, equipment operation status, and personnel behavior analysis. Management personnel can receive real-time pushed early warning information through the APP, view the detailed data of various safety incidents, and remotely confirm or assign handling tasks. The system also supports voice or text notification functions, enabling management personnel to obtain key safety information immediately even when not in the monitoring center, so as to quickly respond to emergencies.

[0080] In terms of the display screen in the monitoring center and voice broadcast devices, the terminal interaction module visually presents the safety status of the construction site on a large display screen, such as safety status distribution maps, risk trend analysis, and alarm event lists, enabling the on-duty personnel to intuitively view the safety situation of the entire construction area. At the same time, the system supports voice broadcast. When high-risk events (such as high-altitude falls, equipment failures, or fire hazards) are detected, it automatically triggers a voice alarm to remind the on-site personnel and management personnel to quickly take measures. Through these interaction methods, the terminal interaction module can significantly improve the safety supervision efficiency at the construction site and reduce the probability of safety accidents.

[0081] In this embodiment, it includes four modules: data collection, data fusion and processing, intelligent analysis and early warning, and terminal interaction to achieve intelligent safety inspection at the construction site. The data collection module is responsible for obtaining multi-source data at the construction site, including video surveillance, sensors (temperature, humidity, dust concentration, vibration intensity), RFID identification information, and environmental data. The data fusion and processing module performs time synchronization, format standardization, and noise filtering on the collected data, and uses an adaptive weighted fusion algorithm to optimize multi-source data fusion. The intelligent analysis and early warning module uses machine learning to evaluate the degree of data difference. For high-difference data, it mines abnormal patterns through clustering analysis; for low-difference data, it conducts safety risk identification and dynamically adjusts the early warning strategy in combination with the risk level and data change situation to achieve real-time safety early warning. The terminal interaction module pushes the inspection results and early warning information to the mobile APP, the display screen in the monitoring center, and voice broadcast devices. Management personnel can remotely view the construction status and make safety decisions, improving the safety supervision ability at the construction site.

[0082] The above formulas are all dimensionless and only take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0083] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wire (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains a set of one or more available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0084] As described above, the above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, and all of them should be covered by the protection scope of the present application.

Claims

1. An intelligent safety inspection system for construction sites, characterized in that: It includes a data acquisition module, a data fusion and processing module, an intelligent analysis and warning module, and a terminal interaction module; The data acquisition module is used to obtain multi-source data at the construction site, including video surveillance data, sensor data, RFID identification information, and environmental data. Among them, the sensor data includes temperature, humidity, dust concentration, and vibration intensity data; The data fusion and processing module is used to perform time synchronization, data format standardization, and noise filtering on the acquired multi-source data, and dynamically optimize the fusion of different data sources based on the adaptive weighted fusion algorithm; The intelligent analysis and warning module is used to evaluate the degree of difference of the fused multi-source data based on a machine learning model. For multi-source data with a high degree of difference, abnormal patterns are further mined through cluster analysis; for multi-source data with a low degree of difference, safety risks are identified, and different levels of warning strategies are dynamically adjusted based on the risk level and the change of the degree of difference of the multi-source data to achieve real-time safety warning; The terminal interaction module is used to push the safety inspection results and warning information to the management terminal, including the mobile APP, the monitoring center display screen, and the voice broadcast device. The management personnel can remotely view the construction site conditions and make safety decisions through the terminal interaction module.

2. The intelligent safety inspection system for construction sites according to claim 1, characterized in that: In the intelligent analysis and warning module, after analyzing the vibration frequency change rate through the fast Fourier transform, a vibration frequency change rate fluctuation index is generated. The acquisition method of the vibration frequency change rate fluctuation index is: Collect the vibration signals of the construction structure over time using a high-precision acceleration sensor, perform an FFT transformation on the collected vibration signals A(t) in the time domain, convert it to the frequency domain, and obtain the spectrum A(f): ; Amplitude spectrum calculated by FFT: ; where is the k-th frequency component, k = 0, 1, 2, ..., N / 2, are respectively the real and imaginary parts of; find the frequency component with the largest amplitude in the FFT result, i.e., the main vibration frequency : ; compare the changes in the main vibration frequency within adjacent time windows and calculate the vibration frequency change rate , and the expression is: ; where: is the main vibration frequency of the i-th time window, is the main vibration frequency of the (i - 1)-th time window, and Δt is the time window interval; calculate the vibration frequency change rate fluctuation index VFVFI, and the expression is: ; where: M is the total number of time windows within the calculation period, is the mean value of the vibration frequency change rates within all windows.

3. The intelligent safety inspection system for construction sites according to claim 2, wherein: After analyzing the change of the movement mode stability of the construction personnel through the LSTM model, a movement stability abnormal index is generated. The acquisition method of the movement stability abnormal index is: The motion state of construction workers is sampled using inertial sensors, and the sensor data at each moment is represented as: where: is the three-axis acceleration, is the three-axis angular velocity, is the three-axis magnetometer data; The training of the LSTM network includes: Input layer: The input data is in the shape; LSTM layer: Extract temporal features and output the hidden state ; Fully connected layer: Map to the stability score ; Loss function: Calculate the error between the LSTM predicted value and the true stability score using the mean squared error to optimize the model; Use the LSTM model to predict the stability score within each time window H After that, calculate the change rate of motion pattern stability , and the expression is: ; where: is the stability score of the current window, is the stability score of the previous window, and Δt is the window interval; calculate the motion stability anomaly index MSAI, and the calculation formula is: ; where: Q is the total number of time windows, is the change rate of stability of the i-th window, is the mean value of the change rates of stability within all windows.

4. The intelligent safety inspection system for construction sites according to claim 3, wherein: Convert the vibration frequency change rate fluctuation index and the movement stability abnormal index into a comprehensive feature vector, use the comprehensive feature vector as the input of the machine learning model, and use the machine learning model to predict the difference degree value label of the fused multi-source data for each group of comprehensive feature vectors as the prediction target, and minimize the sum of the prediction errors of the difference degree value labels of all fused multi-source data as the training target, and train the machine learning model until the sum of the prediction errors reaches convergence and stop the model training. Determine the difference degree value of the fused multi-source data according to the model output result, where the machine learning model is a polynomial regression model.

5. The intelligent safety inspection system for construction sites according to claim 4, characterized in that: Compare the obtained difference degree value of the fused multi-source data with the pre-set difference degree reference threshold according to historical data. If the difference degree value of the fused multi-source data is greater than or equal to the pre-set difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is high, and it is marked as multi-source data with a high degree of difference; if the difference degree value of the fused multi-source data is less than the pre-set difference degree reference threshold, it indicates that the difference degree of the fused multi-source data is low, and it is marked as multi-source data with a low degree of difference.

6. The intelligent safety inspection system for construction sites according to claim 5, characterized in that: For multi-source data with a high degree of difference, abnormal patterns are further mined through cluster analysis. Specifically, the input data for cluster analysis comes from multi-source data samples with a high degree of difference, and each sample contains multiple feature values: ; where: VFVFI is the vibration frequency change rate fluctuation index, and MSAI is the motion stability abnormality index; is the main vibration frequency, is the vibration frequency change rate, is the standard deviation of the vibration change rate, is the stability score predicted by LSTM, is the motion stability change rate; all samples constitute a high-dimensional feature data set: , J is the total number of features; select the value of K, initialize K cluster centers, and iterate and optimize: ; where, is the center of the k-th category, is the set of samples belonging to this category; calculate the Euclidean distance from each data point to the cluster center , and classify it into the nearest cluster: ; represents the j-th feature value of the i-th sample; represents the value of the k-th cluster center on the j-th feature dimension; iterate and update until convergence; the high-difference data at the construction site is clustered into categories of equipment vibration abnormality, construction structure instability, and personnel behavior abnormality.

7. An intelligent safety inspection system for construction sites according to claim 6, characterized in that: Dynamically adjust different levels of warning strategies based on the risk level and the change of the degree of difference of the multi-source data to achieve real-time safety warning. Specifically: Take the risk level and the difference degree value of multi-source data as the input items of fuzzy logic, and take the warning strategies at different levels as the output items of fuzzy logic, and perform dynamic regulation through fuzzy logic; Fuzzify the input using a fuzzy membership function, perform reasoning using a fuzzy rule base, and defuzzify using the centroid method to calculate the final early warning level value , ; where: is the membership degree value corresponding to different rules, is the numerical mapping of different early warning strategies; Dynamically execute corresponding warning measures according to the fuzzy inference results.

Images