A digital-based method for safe construction management

Abstract

This application discloses a digital-based safety construction management method in the field of construction management, including: collecting circular saw operation data at the construction site; preprocessing the collected circular saw operation data; determining the safe operation status of the circular saw using a Support Vector Machine (SVM) algorithm based on the preprocessed circular saw operation data; conducting a risk assessment based on a risk assessment method combining evidence theory and fuzzy set theory to obtain a safety risk level based on the safe operation status and safety risk level; and managing the circular saw's safety through a multi-level early warning control strategy based on reinforcement learning, addressing the low accuracy of risk assessment in existing circular saw construction technologies. This application improves the accuracy of risk assessment by combining evidence theory and fuzzy set theory, and by utilizing a reinforcement learning early warning control strategy.

Description

Technical Field

[0001] This application relates to the field of construction management, and in particular to a digital-based method for safe construction management. Background Technology

[0002] With the rapid development of the construction industry and the continuous expansion of project scale, construction safety issues have become increasingly prominent, posing a bottleneck to the industry's sustainable and healthy development. Statistics show that construction safety accidents account for more than 20% of all production safety accidents nationwide, with a large number of these accidents occurring during the operation of construction machinery such as circular saws. Circular saws, with their high efficiency and convenience, are widely used in construction, but they also present numerous safety hazards and difficulties in risk control. Once an accident occurs, it often results in serious casualties and economic losses.

[0003] Traditional circular saw construction safety management relies primarily on manual experience and qualitative analysis, which suffers from drawbacks such as strong subjectivity, high randomness, and significant lag. On the one hand, the complex construction site environment and numerous interfering factors make it difficult for operators to accurately identify and assess the safety risks of circular saws. Often, countermeasures are only taken after an accident occurs, resulting in a passive and delayed state of safety management. On the other hand, the safety status of circular saws is affected by various factors, such as saw blade speed, saw blade height, and the position of the protective cover. These factors are difficult to monitor and judge in real time and comprehensively using only manual methods. Risk assessment and early warning often become mere formalities, lacking scientific rigor and effectiveness.

[0004] To improve the safety management of circular saw construction, some scholars and enterprises have conducted fruitful research and practice. The "Handbook of Safety Protection Technology for Metal Cutting Machine Tools" published by the Machinery Industry Press systematically summarizes the safety protection design methods and operating procedures for machine tools such as circular saws; the "Construction Safety Risk Classification and Control and Hidden Danger Investigation and Treatment System" of the China Academy of Building Research has established safety inspection standards and management processes for construction machinery, including circular saws; a construction company in Shenzhen has developed an "Intelligent Safety Monitoring System for Circular Saws," which achieves real-time monitoring and visual management of equipment status through sensor data collection and big data analysis. These research results have improved the current safety management of circular saws to some extent, but problems such as low accuracy in risk assessment and lack of intelligent early warning control still exist, making it difficult to meet the growing safety management needs. Summary of the Invention

[0005] To address the issue of low risk assessment accuracy in circular saw construction in existing technologies, this application provides a digital-based safety construction management method. This method improves risk assessment accuracy by combining evidence theory and fuzzy set theory, and by utilizing reinforcement learning early warning and control strategies.

[0006] The purpose of this application is achieved through the following technical solution.

[0007] This application provides a digital-based safety construction management method, comprising: collecting circular saw operation data at the construction site; the operation data including saw blade speed, saw blade height, and protective cover position; preprocessing the collected circular saw operation data; determining the safe operation status of the circular saw using a Support Vector Machine (SVM) algorithm based on the preprocessed circular saw operation data; the input of the SVM algorithm being the circular saw operation data, and the output being the safe operation status of the circular saw; performing a risk assessment based on a risk assessment method combining evidence theory and fuzzy set theory to obtain a safety risk level based on the safe operation status and the safety risk level; and managing the circular saw for safety using a multi-level early warning control strategy based on reinforcement learning based on the safe operation status and the safety risk level.

[0008] A circular saw mainly consists of a motor, saw blade, and safety guard. Its working principle is that a high-speed rotating circular saw blade cuts the workpiece. Circular saws are widely used in wood processing, construction, and other fields, but they also pose certain safety risks. Saw blade speed refers to the rotational speed of the circular saw blade, usually expressed in revolutions per minute (rpm). Speed ​​is one of the key parameters affecting cutting efficiency and safety. Excessive speed may lead to overheating, deformation, or breakage of the saw blade, while insufficient speed will reduce cutting efficiency. Saw blade height refers to the height of the saw blade protruding from the worktable. A suitable saw blade height ensures cutting depth, improving cutting efficiency and quality. However, an excessively high saw blade height increases the exposed area of ​​the saw blade, increasing the risk of accidental contact. The safety guard is a crucial safety device for circular saws, covering the saw blade when not in use to prevent accidental contact. The position of the safety guard refers to the extent to which the guard covers the saw blade; a suitable position maximizes the isolation of the saw blade and reduces the risk of accidents. This application employs a risk assessment method that combines evidence theory and fuzzy set theory. Evidence theory is used to represent and combine the importance of different risk indicators with uncertainty, while fuzzy set theory is used to describe the risk level of each indicator with membership degree. The results of evidence combination and fuzzy assessment are then fused to obtain a comprehensive risk level. This method can simultaneously handle random uncertainty and fuzzy uncertainty, thereby improving the accuracy and reliability of the assessment.

[0009] Furthermore, based on the preprocessed circular saw operation data, the safe operating status of the circular saw is determined using the Support Vector Machine (SVM) algorithm. This includes: selecting the Radial Basis Function (RBF) as the kernel function of the SVM algorithm to construct an SVM classification model; optimizing the hyperparameters of the SVM classification model using a grid search method, wherein the hyperparameters include the penalty coefficient C and the parameter γ of the RBF; constructing a feature vector x based on the preprocessed circular saw operation data; and calculating the distance between the feature vector x and the classification hyperplane using the decision function of the SVM classification model to obtain the safe operating status of the circular saw.

[0010] Furthermore, the decision function expression for the SVM classification model is: Where x represents the feature vector formed by the circular saw's operating data. x = [ x 1 , x 2 , x 3 ] ,in, For saw blade rotation speed, The height of the saw blade. The shield position is shown; w represents the normal vector of the SVM classification hyperplane, and represents the direction and slope of the hyperplane, obtained through SVM model training; The RBF kernel function represents the nonlinear transformation performed on the eigenvector x, mapping x to a high-dimensional feature space. , where γ is a parameter of the RBF kernel function, controlling the width of the Gaussian function. λ represents the training samples; b represents the intercept of the SVM classification hyperplane, which is the distance between the hyperplane and the origin, obtained through SVM model training; λ represents the safety state correction factor, used to adjust the original SVM classification results to better reflect the actual safety state of the circular saw. The value of λ can be set based on experience or optimized through data analysis; g(x) represents the safety state correction function, which adjusts the original SVM classification results according to the characteristics of the circular saw's operating data.

[0011] Furthermore, ,in, The weighting coefficients for the three characteristic parameters represent their degree of influence on the safety status and can be determined based on expert experience or data analysis. The safety state correction subfunction for the three characteristic parameters: , , ,in, These are the safety thresholds for saw blade speed, saw blade height, and protective cover position, respectively. These represent the upper and lower limits of the safety range for the three characteristic parameters.

[0012] By introducing a safety state correction factor λ and a correction function Based on the actual characteristics of the circular saw's operating data, the original SVM classification results can be appropriately adjusted to make the decision function more closely reflect the actual safety status of the circular saw. When a feature parameter exceeds the safety range, its corresponding correction function will produce a large value, thereby reducing the value of the entire decision function, indicating that the feature parameter has a negative impact on the safety status. When all feature parameters are within the safety range, the value of the correction function is close to 0, and the value of the decision function is mainly determined by the original SVM classification results. In this way, the accuracy and reliability of the circular saw's safety status judgment can be further improved based on SVM classification.

[0013] Furthermore, the safety risk level is obtained, including: determining the initial risk level based on the safe operating status of the circular saw; establishing a risk assessment index system comprising three indicators: accident probability, severity of accident consequences, and saw blade exposure; setting the assessment level and fuzzy membership function for each indicator; and, based on historical data, setting the importance weights for the three indicators—accident probability, severity of accident consequences, and saw blade exposure—and using the probability assignment function (BPA) in evidence theory to quantify the importance weights, thus obtaining the evidence body for the importance of each indicator. According to the evidence Calculate the comprehensive body of evidence using Dempster's combination rule. Based on the circular saw's operating data, fuzzy evaluation vectors for each indicator are obtained through membership function matching; where the evaluation level of the i-th indicator is... The membership value obtained from the matching is Then the fuzzy evaluation vector of the corresponding indicator Using the fuzzy weighted average operator, the comprehensive evidence body and fuzzy evaluation vector By fusion, the fuzzy membership distribution of the comprehensive risk assessment level is obtained. : U ( x ) = ∑ i = 1 3 [ m ( A i ) × R i ( x ) ] Where x is the risk level variable, This is a comprehensive body of evidence regarding the importance of indicator i. Let i be the fuzzy evaluation vector of indicator i; using the centroid defuzzification method in fuzzy mathematics, the fuzzy membership distribution U(x) is transformed into a quantified safety risk level value. Based on the initial risk level and the safety risk level value By adopting the Yager fusion rule in the DS evidence theory, the final safety risk level is obtained by comprehensively considering the safety status and risk assessment of the circular saw.

[0014] In evidence theory, BPA is a basic probability assignment function used to describe the uncertainty of evidence. It divides the uncertainty of evidence into several disjoint subsets and assigns a basic probability value between [0, 1] to each subset, representing the degree to which the evidence supports that subset. The sum of the probability assignments for all subsets equals 1. Accident probability refers to the likelihood of a safety accident occurring under given conditions. It is a key indicator in risk assessment, reflecting the random uncertainty of accident occurrence. In this scheme, accident probability is used as an important dimension for assessing the safety risk of the circular saw.

[0015] The severity of accident consequences refers to the extent of the negative consequences, such as personal injury, equipment damage, and environmental impact, that would result from a safety accident. It is another important indicator for measuring risk, reflecting the potential hazards of an accident. Saw blade exposure reflects the size of the area of ​​the saw blade exposed during the operation of a circular saw, directly affecting the risk level of accidental contact with the saw blade.

[0016] Dempster's combination rule is an evidence fusion method in evidence theory used to synthesize information from multiple different sources of evidence. It calculates the degree of consistency among different pieces of evidence to obtain a comprehensive piece of evidence that reflects the proposition commonly supported by each piece of evidence. The fuzzy weighted average operator is a commonly used method for fuzzy information integration. It assigns weights to each fuzzy evaluation result and then performs a weighted average to obtain a comprehensive evaluation result. The weights are usually determined based on the importance of the evaluation indicators, reflecting the degree of influence of different indicators on the overall evaluation. The centroid method is a commonly used fuzzy mathematics defuzzification method used to transform a fuzzy set into a deterministic numerical value. It calculates the centroid abscissa of the membership function of the fuzzy set as the representative value of the fuzzy set. The centroid value comprehensively considers the membership degree of each element and can reasonably reflect the overall position of the fuzzy set.

[0017] The Yager fusion rule is an important extension of Dempster's evidence theory, used to address the problem of contra-intuitive results in cases of severe evidence conflict. It avoids overly negative results by introducing an unknown set to which the probabilities of conflicting evidence are assigned.

[0018] Furthermore, a risk assessment index system is established, comprising three indicators: accident probability, accident consequence severity, and saw blade exposure. An assessment level and fuzzy membership function are set for each indicator, including: the assessment level for accident probability is {very low, low, medium, high, very high}, with a triangular membership function; the assessment level for accident consequence severity is {very low, low, medium, high, very high}, with a trapezoidal membership function; and the assessment level for saw blade exposure is {very weak, weak, medium, strong, very strong}, with a Gaussian membership function.

[0019] Furthermore, the Dempster combination rule is used to calculate the comprehensive body of evidence. This includes: calculating the degree of conflict K between the evidence pieces: K = ∑ B 1 ∩ B 2 ∩ B 3 =∅ [ m 1 ( B 1 ) × m 2 ( B 2 ) × m 3 ( B 3 ) ] Among them, B1, B2, and B3 are subsets of the three indicator importance assessment frameworks, and That is, K is the sum of confidence levels of incompatibility among the evidence bodies; calculate the comprehensive evidence body of the non-zero subset A. : m ( A ) = 1 1 − K × ∑ B 1 ∩ B 2 ∩ B 3 = A [ m 1 ( B 1 ) × m 2 ( B 2 ) × m 3 ( B 3 ) ] Where A is a non-zero subset of the indicator importance assessment framework. These are subsets of the three indicator importance assessment frameworks, and ;Right now It is the sum of the confidence products of each piece of evidence with respect to subset A after normalization.

[0020] The evaluation framework for indicator importance is defined as Θ = {A1, A2, ..., An}, where A1, A2, ..., An represent n levels of indicator importance, such as {very low, low, medium, high, very high}. Subsets of this framework encompass all possible combinations of indicator importance levels, for example: {A1}, {A2}, ..., {An}: single importance level; {A1, A2}, {A1, A3}, ...: pairwise combinations; {A1, A2, A3}, {A1, A2, A4}, ...: triadic combinations; ...; {A1, A2, ..., An-1}, {A2, A3, ..., An}: combinations of n-1 levels; Θ: combinations of all levels. The evidence bodies m1, m2, and m3 for indicator importance are essentially probability assignment functions defined on these subsets. They represent the degree of support each piece of evidence (such as expert opinion) provides for various combinations of importance levels.

[0021] According to evidence theory, a subset of evidence is considered a zero subset if the support of a given piece of evidence for that subset is zero (i.e., its probability is assigned a value of 0). Conversely, a subset is considered a non-zero subset if its probability is greater than zero. For the indicator importance assessment framework Θ, its non-zero subsets refer to those combinations of indicator importance levels supported by at least one piece of evidence. For example, if the non-zero subsets corresponding to the three evidence bodies are: m1: {A1}, {A2, A3}, {A4, A5}; m2: {A1, A2}, {A3}, {A4}; m3: {A2}, {A3, A4}, {A5}; then the non-zero subsets of the entire evaluation framework are the union of these subsets: {A1}, {A2}, {A3}, {A4}, {A5}, {A1, A2}, {A2, A3}, {A3, A4}, {A4, A5}; when calculating the synthesized evidence body m(A), A is to traverse all these non-zero subsets. The calculation of K is also performed on these non-zero subsets. The purpose of this approach is to reduce the computational load, because the product of m1, m2, and m3 corresponding to the zero subsets must be 0, which does not affect the result.

[0022] Furthermore, using the centroid defuzzification method in fuzzy mathematics, the fuzzy membership distribution U(x) is transformed into a quantified safety risk level value x*, including: calculating the centroid x* of the fuzzy membership distribution U(x). Where x is the risk level variable, Let x be the membership distribution function for the comprehensive risk assessment, where [a, b] is the universe of discourse interval of x, and a and b are the lower and upper bounds of the universe of discourse, respectively; x* represents the centroid of the membership distribution, reflecting the overall position of the comprehensive risk level in the universe of discourse.

[0023] In this context, the universe of discourse refers to the range of values ​​that elements in a fuzzy set can take. It defines the boundaries and range of the fuzzy set. In this scheme, the universe of discourse interval [a, b] refers to the range of values ​​for the safety risk level.

[0024] Furthermore, based on the safe operating status and safety risk level, a multi-level early warning control strategy based on reinforcement learning is used to manage the safety of the circular saw. This includes: constructing a Markov Decision Process (MDP) model for the safety early warning control of the circular saw; and using the Q-learning algorithm to solve the MDP model to obtain the optimal early warning control strategy. According to the optimal early warning and control strategy This generates corresponding early warning and control commands.

[0025] Furthermore, a Markov decision process (MDP) model for the safety early warning control of the circular saw is constructed, including:

[0026] Based on the safe operating status and safety risk level of the circular saw, the safety status of the circular saw is divided into three states: safe, warning, and dangerous, forming a state space. ;in, This indicates a safe status, corresponding to an SVM classification result of "safe" and a risk level of "low risk". This indicates a warning status, with the corresponding SVM classification result being safe and the risk level being medium risk. This indicates a dangerous state, corresponding to an SVM classification result of "dangerous," with a risk level of high risk or extremely high risk.

[0027] Based on the safety status of the circular saw, corresponding early warning control actions are set to create a range of motion. ;in, This indicates that the current situation will be maintained and no early warning or control measures will be taken. This indicates the issuance of warning signals, including audible and visual alarms, and human-computer interaction prompts; This indicates forced speed reduction, limiting the maximum rotational speed of the circular saw via the controller; This indicates emergency braking, where the controller cuts off the power to the circular saw and applies braking force. Based on historical operating data, the frequency of the circular saw transitioning to other states after taking different warning control actions under different safety conditions is statistically analyzed, resulting in a state transition frequency matrix. According to the state transition frequency matrix Calculate state transition probability : Set the reward function for early warning control. , represents the immediate reward obtained by the circular saw after taking action a in state s; ,in, Indicates a security risk reward. This indicates a production efficiency bonus. Indicates an energy loss reward; The weighting coefficients for the three reward components satisfy the following conditions: ; ,in, This represents the risk cost under state s, and different cost values ​​can be set according to the risk level. This represents the probability of an accident occurring after taking action a in state s, which can be estimated based on historical data and expert experience. The calculation formula is: ,in, This represents the time cost saved after taking action a in state s. Different cost values ​​can be set according to the degree of impact of different actions on production efficiency. This represents the probability of resuming normal production after taking action a in state s, which can be estimated based on the state transition probability. The calculation formula is: ,in, This represents the energy cost consumed after taking action a in state s. Different cost values ​​can be set according to the energy consumption characteristics of different actions.

[0028] The energy loss reward refers to the additional energy cost, such as electrical energy, incurred due to forced speed reduction or emergency braking of equipment after taking warning control action a in state s. It is a negative reward, meaning the greater the energy loss, the lower the reward value. In the reward function R(s, a), the energy loss reward is included as a component, along with the safety risk reward and production efficiency reward, to comprehensively evaluate the effectiveness of the warning control action. The purpose of introducing the energy loss reward is to balance safety and economy, avoiding excessive sacrifice of energy efficiency in pursuit of absolute safety.

[0029] The probability of restoring normal production refers to the likelihood that a circular saw will return to normal production status after taking a warning control action a in state s. It is a conditional probability, representing the chance of the system recovering from an abnormal state to a normal state given the current state and the control action. The probability of restoring normal production reflects the effectiveness of the warning control and is a key factor in the production efficiency reward Refficiency(s, a). If a control action can restore the system to normal with a high probability, it means that the production efficiency reward for that action is high. Specifically, it involves calculating the frequency of successfully restoring to a safe state after taking action a in state s, and then dividing by the total frequency of taking action a in state s.

[0030] Energy cost refers to the economic cost of electricity and other energy consumed by a circular saw after taking a warning control action a in state s. It is related to factors such as the energy consumption characteristics of the equipment, energy prices, and the duration of the control action. When calculating the energy loss reward Renergy(s,a), the energy cost Cenergy(s,a) is a fundamental variable. It is typically estimated by multiplying the rated power of the equipment, the duration of the control action, and the unit energy price. The introduction of energy cost allows the reward function to more comprehensively evaluate the effectiveness of warning control, balancing the impact of safety, production efficiency, and energy loss. This helps the Markov decision process model learn an optimal warning control strategy.

[0031] Compared to existing technologies, the advantages of this application are:

[0032] This application employs the Support Vector Machine (SVM) algorithm to determine the safe operating status of a circular saw. By selecting the Radial Basis Function (RBF) kernel function and optimizing hyperparameters using a grid search method, a safety status correction factor and a correction function are introduced, thereby improving the accuracy of safety status judgment from both data and algorithmic perspectives. The SVM algorithm can effectively handle high-dimensional, small-sample data; the RBF kernel function enables nonlinear classification; grid search and cross-validation optimize model parameters; and the safety status correction mechanism adaptively adjusts the judgment results based on the actual operating conditions of the circular saw.

[0033] This application innovatively proposes a risk assessment method based on a combination of evidence theory and fuzzy set theory. It constructs a risk assessment index system comprising three indicators: accident probability, accident consequence severity, and saw blade exposure. Fuzzy membership functions are used to quantify each indicator, and probability assignment functions from evidence theory are used to model the importance of the indicators. Dempster's combination rule and a fuzzy weighted average operator are used to fuse evidence information and assessment results. Finally, a centroid defuzzification method is used to obtain the quantified risk level. This method comprehensively considers uncertainties in the risk assessment process, improving assessment accuracy.

[0034] Based on the safe operating status and risk assessment level of a circular saw, this application constructs a Markov Decision Process (MDP) model. By defining the state space, action space, state transition probabilities, and reward function, the dynamic stochastic characteristics of the early warning control process are characterized. On this basis, a Q-learning reinforcement learning algorithm is used to solve the MDP model. Through continuous exploration and learning, the optimal early warning control strategy is obtained, and corresponding control commands are generated. This method can autonomously select the optimal early warning control action based on the real-time safety status and risk level of the circular saw. Through multi-level early warning and coordinated control, timely intervention measures can be taken before an accident occurs, thereby effectively avoiding or mitigating the consequences of accidents and improving the timeliness and effectiveness of safety early warning control. Attached Figure Description

[0035] This application will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0036] Figure 1 This is an exemplary flowchart of a digital-based safe construction management method according to some embodiments of this application;

[0037] Figure 2 This is an exemplary flowchart illustrating the safe operating state of a circular saw according to some embodiments of this application;

[0038] Figure 3 This is an exemplary flowchart illustrating the generation of a final security risk level according to some embodiments of this application;

[0039] Figure 4 This is an exemplary flowchart illustrating the setting of early warning control instructions according to some embodiments of this application. Detailed Implementation

[0040] The methods and systems provided in the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0041] like Figure 1As shown, the operation data of the circular saw at the construction site is collected. The operation data includes saw blade speed, saw blade height, and protective cover position. The collected circular saw operation data is preprocessed. Based on the preprocessed circular saw operation data, the safe operation status of the circular saw is determined using a Support Vector Machine (SVM) algorithm. The input of the SVM algorithm is the circular saw operation data, and the output is the safe operation status of the circular saw. Based on the safe operation status, a risk assessment method combining evidence theory and fuzzy set theory is used to conduct a risk assessment and obtain a safety risk level. Based on the safe operation status and the safety risk level, a multi-level early warning control strategy based on reinforcement learning is used to manage the safety of the circular saw.

[0042] Collect operational data from the circular saw at the construction site. This data includes saw blade speed, saw blade height, and the position of the protective cover. Specifically, select appropriate sensors based on the data type and accuracy requirements, such as Hall effect sensors for measuring saw blade speed, laser rangefinders for measuring saw blade height, and proximity switches for detecting the position of the protective cover. Install the sensors at their respective locations on the circular saw, ensuring that their measurement range, installation angle, and fixing method meet the data acquisition needs. Design the sensor data acquisition circuit and module, including sensor signal conditioning circuits, analog-to-digital conversion circuits, and data storage circuits. Based on the sensor's output characteristics and the data acquisition time interval, appropriately set parameters such as sampling frequency, quantization accuracy, and data buffer size. Select appropriate communication protocols and modules based on the site environment and data transmission distance. For example, Zigbee and Bluetooth can be used for short distances, while LoRa and NB-IoT low-power wide-area network protocols can be used for long distances. The communication module is responsible for transmitting the collected sensor data to an edge gateway or cloud server.

[0043] The collected circular saw operation data undergoes preprocessing. Specifically, based on the communication protocol and data format, the received raw data packets are parsed to extract effective data such as saw blade speed, saw blade height, and protective cover position. The data is converted into a unified physical quantity unit and numerical range to facilitate subsequent data analysis and modeling. Since the sampling frequencies and timestamps of different sensors may be inconsistent, time synchronization and alignment of multi-source heterogeneous data are necessary. Using timestamps or sampling sequences as a benchmark, data from different sensors are interpolated, fitted, or resampled to ensure time axis consistency, facilitating data fusion and correlation analysis. Because the physical quantities measured by different sensors have different dimensions and numerical ranges, data normalization or standardization is required to eliminate dimensional influences and facilitate data comparison and model training. Common methods include maximum-minimum normalization and zero-mean unit variance standardization, ensuring that data with different characteristics are distributed on the same scale.

[0044] like Figure 2As shown, based on the preprocessed circular saw operation data, the safe operating status of the circular saw is determined using the Support Vector Machine (SVM) algorithm. The input of the SVM algorithm is the circular saw operation data, and the output is the safe operating status of the circular saw. The Radial Basis Function (RBF) is selected as the kernel function of the SVM algorithm to construct the SVM classification model. The form of the RBF kernel function is: , where x and Given two feature vectors, γ is a parameter controlling the width of the Gaussian function. The RBF kernel function maps the original feature space to an infinite-dimensional high-dimensional feature space, making originally linearly inseparable data linearly separable in the high-dimensional space, thereby improving classification performance. A training sample set is constructed based on the preprocessed circular saw running data. ,in Let be the feature vector of the i-th sample. Let be the corresponding safety status label, -1 representing a dangerous state and +1 representing a safe state, and N be the total number of training samples. An SVM classification model is trained using the training sample set D to obtain the normal vector w and intercept b of the classification hyperplane, as well as the parameters γ of the RBF kernel function.

[0045] A grid search method is used to optimize the hyperparameters of the SVM classification model. The hyperparameters include the penalty coefficient C and the parameter γ of the radial basis function (RBF). The ranges of C and γ are divided into several discrete intervals to form a hyperparameter grid, such as C ∈ {0.1, 1, 10, 100} and γ ∈ {0.01, 0.1, 1, 10}. For each combination (C, γ) in the hyperparameter grid, the corresponding SVM classification model is trained, and its classification performance is evaluated on the validation set, using methods such as cross-validation to calculate evaluation metrics such as average classification accuracy and F1 score. The hyperparameter combination with the optimal classification performance is selected. , as the optimal hyperparameter of the SVM classification model, is used for subsequent security status determination.

[0046] Based on the preprocessed circular saw operation data, a feature vector x is constructed; three key feature parameters—saw blade speed, saw blade height, and protective cover position—are extracted from the preprocessed operation data, denoted as [x, y, x], [x, y ... The three feature parameters are combined into a feature vector. x = [ x 1 , x 2 , x 3 ] The feature vector is used as the input to the SVM classification model. In order to eliminate the influence of different features in terms of their dimensions, the feature vector can be normalized, such as by using maximum and minimum value normalization, which scales the value range of each feature to the interval [0, 1].

[0047] The safe operating status of the circular saw is obtained by calculating the distance between the feature vector x and the classification hyperplane using the decision function of the SVM classification model. The expression of the decision function of the SVM classification model is: Where x represents the feature vector formed by the circular saw's operating data. x = [ x 1 , x 2 , x 3 ] ,in For saw blade rotation speed, The height of the saw blade. The shield position is shown; w represents the normal vector of the SVM classification hyperplane, and represents the direction and slope of the hyperplane, obtained through SVM model training; The RBF kernel function represents the nonlinear transformation performed on the eigenvector x, mapping x to a high-dimensional feature space. , where γ is a parameter of the RBF kernel function, controlling the width of the Gaussian function. λ represents the training samples; b represents the intercept of the SVM classification hyperplane, which is the distance between the hyperplane and the origin, obtained through SVM model training; λ represents the safety state correction factor, used to adjust the original SVM classification results to better match the actual safety state of the circular saw. The value of λ can be set based on experience or optimized through data analysis. This represents the safety state correction function, which adjusts the original SVM classification results based on the characteristics of the circular saw's operating data. ,in, The weighting coefficients for the three characteristic parameters represent their degree of influence on the safety status and can be determined based on expert experience or data analysis. The safety state correction subfunction for the three characteristic parameters: , , ,in, These are the safety thresholds for saw blade speed, saw blade height, and protective cover position, respectively. These represent the upper and lower limits of the safety range for the three characteristic parameters.

[0048] like Figure 3 As shown, based on the safe operating status, a risk assessment method combining evidence theory and fuzzy set theory is used to conduct a risk assessment and obtain a safety risk level. Based on the safe operating status of the circular saw, an initial risk level is determined. According to the output of the SVM classification model, the safe operating status of the circular saw is divided into three levels: safe, warning, and dangerous, corresponding to low risk, medium risk, and high risk, respectively. The initial risk level reflects the current safety status of the circular saw and serves as the basis for subsequent risk assessments. By setting an initial risk level, the real-time safety status of the equipment can be considered during the risk assessment process, improving the relevance and timeliness of the assessment results.

[0049] Establish a risk assessment index system that includes three indicators: accident probability, severity of accident consequences, and saw blade exposure. Set the assessment level and fuzzy membership function for each indicator. Specifically, the accident probability index has the following assessment levels: {very small, small, medium, large, very large}. The membership function adopts a triangular membership function. Based on historical accident data and expert experience, set the quantitative scoring range for the accident probability index, such as [0, 10]. The scoring interval is divided into five sub-intervals, corresponding to five evaluation levels: very small [0, 2], small (2, 4], medium (4, 6], large (6, 8], and very large (8, 10). A triangular membership function is defined to describe the membership degree of the score value to each evaluation level. The parameters of the triangular function include the peak point and the left and right boundary points, which correspond to score values ​​with membership degrees of 1 and 0, respectively. For example, the triangular function for the "very small" level can be defined as: peak point: 0, membership degree 1; right boundary point: 2, membership degree 0; function expression: μ(x) = 1 - x / 2, x ∈ [0, 2]. Similarly, triangular membership functions can be defined for other levels. The triangular function has a simple shape and high computational efficiency, making it suitable for describing fuzzy sets with clear boundaries. By adjusting the peak point and boundary points, the shape of the membership function can be flexibly controlled to adapt to different evaluation needs.

[0050] Accident consequence severity index: Assessment level: {Very low, Low, Medium, High, Very high}, Membership function: A trapezoidal membership function is used. Based on the qualitative description and quantitative analysis of the accident consequences, a quantitative scoring interval for the severity index is set, such as [0, 10]. The scoring interval is divided into five sub-intervals, corresponding to the five assessment levels: Very low [0, 2], Low [2, 4], Medium [4, 6], High [6, 8], Very high [8, 10]. A trapezoidal membership function is defined to describe the membership degree of the score value to each assessment level. The parameters of the trapezoidal function include the upper and lower boundary points and the left and right boundary points, which correspond to score values ​​with membership degrees of 1 and 0, respectively. For example, the trapezoidal membership function for the "medium" level can be defined as follows: Lower base left boundary point: 4, membership degree 0; Upper base left boundary point: 5, membership degree 1; Upper base right boundary point: 6, membership degree 1; Lower base right boundary point: 7, membership degree 0; Function expression: μ(x)=0, x<4 or x>7; μ(x)=(x-4), x∈[4,5]; μ(x)=1, x∈[5,6]; μ(x)=(7-x), x∈[6,7]; Similarly, trapezoidal membership functions for other levels can be defined. The trapezoidal function can represent the core region and transition region of a fuzzy set, and is suitable for describing the fuzziness of continuous variables. By adjusting the upper and lower bases and boundary points, the size of the core region and the width of the transition region of the membership function can be controlled, improving the flexibility of the description.

[0051] Saw blade exposure index: Assessment levels: {Very weak, Weak, Medium, Strong, Very strong}, Membership function: A Gaussian membership function is used. Based on the saw blade exposure questionnaire and behavioral observation, a quantitative scoring range for the risk awareness index is set, such as [0, 10]. The scoring range is divided into five sub-ranges, corresponding to the five assessment levels: Very weak [0, 2], Weak (2, 4], Medium (4, 6], Strong (6, 8], Very strong (8, 10). A Gaussian membership function is defined to describe the membership degree of the score value to each assessment level. The parameters of the Gaussian function include the mean and variance, which control the center position and width of the function, respectively. For example, the Gaussian function for the "Strong" level can be defined as: Mean: 7, corresponding to the maximum membership degree of 1; Variance: 0.8, controlling the width of the function; Function expression: Similarly, other levels of Gaussian membership functions can be defined. The Gaussian function is a bell-shaped curve, with the highest membership at the center point and gradually decreasing towards the sides, making it suitable for describing the ambiguity of subjective human judgment. By adjusting the mean and variance, the concentration and coverage of the membership function can be controlled, improving its ability to characterize risk awareness.

[0052] Using historical accident data of circular saws, this study analyzes the impact of three indicators on safety risks: Historical accident data for circular saws is collected, including information such as the time, location, cause, consequences, and personnel involved. Statistical analysis is performed on the accident data to calculate the frequency and contribution of each indicator in accidents. The indicators are: Accident Probability Indicator: The total number of accidents is counted, and the percentage of accidents at different probability levels (e.g., very low, low, medium, high, very high) is calculated. Accident Consequence Severity Indicator: The percentage of accidents at different severity levels (e.g., very low, low, medium, high, very high) is counted, and the impact of accident consequences on personnel injuries, equipment damage, and production delays is analyzed. Saw Blade Exposure Indicator: The proportion of human factors in accidents is counted, and the frequency and consequences of behaviors such as violations of operating procedures, ignoring warnings, and improper use of protective devices due to insufficient saw blade exposure are analyzed. Based on the frequency and contribution analysis results, the impact of the three indicators on safety risks is ranked, and their importance weights are obtained. Weights can be expressed as percentages or fractions, such as {Accident Probability: 0.5, Accident Consequence Severity: 0.3, Saw Blade Exposure: 0.2}. The greater the weight, the greater the impact of the indicator on safety risks, and the higher the priority it should be given in risk assessment and early warning control.

[0053] The probability assignment function (BPA) from evidence theory is used to quantify the indicator weights into evidence bodies. Evidence theory, also known as Dempster-Shafer theory, is a mathematical theory for dealing with uncertain information. It defines a basic probability as a sign (BPA) that assigns confidence levels to a power set of events, representing the degree of confidence in different combinations of events. The BPA function satisfies the following properties: The confidence level of the empty set is 0. Let A be all subsets of the power set, and let the sum of their confidence scores be 1. The confidence level of any subset A is between 0 and 1. Based on the indicator weights, a BPA function is constructed, assigning the weights to the single-element subsets of the indicators. Let the weights of the three indicators be 0.5, 0.3, and 0.2 respectively. The corresponding BPA function can be expressed as: m1({accident probability}) = 0.5, m2({accident consequence severity}) = 0.3, m3({saw blade exposure}) = 0.2. A single-element subset represents the independent confidence level of that indicator, that is, the degree to which the indicator is considered to influence the safety risk is the corresponding weight value. For other subsets of the indicators, if there is no additional supporting evidence, the confidence level is 0. For example: m({accident probability, accident consequence severity}) = 0, m({accident probability, saw blade exposure}) = 0, m({accident consequence severity, saw blade exposure}) = 0, m({accident probability, accident consequence severity, saw blade exposure}) = 0. This application employs the BPA function from evidence theory to transform indicator weights into evidence bodies, representing the degree of confidence in different combinations of indicators. Evidence theory can flexibly model uncertain information, quantifying the impact of indicator weights on security risks through the allocation and combination of confidence levels. By assigning confidence levels to the power set of indicators through the BPA function, it not only represents the confidence level of a single indicator but also the confidence level of a combination of indicators. This approach can more comprehensively characterize the uncertainty of indicator weights, demonstrating the advantages of evidence theory in handling incomplete and imprecise information.

[0054] Calculate the conflict degree K among the evidence pieces, where K represents the sum of the confidence levels of incompatibility among the evidence pieces. A higher conflict degree indicates a greater contradiction between the evidence pieces, and a lower overall confidence level. Calculate the overall evidence m(A) for the non-zero subset A, where m(A) is the sum of the normalized products of the confidence levels of each evidence piece for subset A. m(A) integrates the support levels of each evidence piece and represents the overall confidence level for subset A. Let m1({probability of the accident}) = 0.5, m2({severity of the accident consequences}) = 0.3, and m3({saw blade exposure}) = 0.2. Then, the comprehensive evidence body m(A) can be calculated as: K = m1({probability of the accident})m2({severity of the accident consequences})m3({saw blade exposure}) = 0.03; m({probability of the accident, severity of the accident consequences}) = [m1({probability of the accident})m2({severity of the accident consequences}) + m1({probability of the accident})m3({saw blade exposure}) + m 2({Severity of Accident Consequences})m3({Saw Blade Exposure})] / (1-K)=0.35; m({Probability of Accident, Saw Blade Exposure})=[m1({Probability of Accidents})m3({Saw Blade Exposure})] / (1-K)=0.10; m({Severity of Accident Consequences, Saw Blade Exposure})=[m2({Severity of Accident Consequences})m3({Saw Blade Exposure})] / (1-K)=0.06; Dempster's combination rule can integrate information from multiple sources of evidence to obtain a unified understanding. By calculating the degree of conflict and the combination confidence, the consistency of the evidence can be assessed, and the importance of different combinations of indicators can be quantified, making the risk assessment more comprehensive and reliable.

[0055] Based on the circular saw's operating data, fuzzy evaluation vectors for each indicator are obtained through membership function matching: Based on the circular saw's real-time operating parameters (such as current and vibration), fuzzy membership functions are used to calculate the membership degree of each parameter value under different evaluation levels. For the i-th indicator, let its evaluation level be... The corresponding membership value is Then the fuzzy evaluation vector of this indicator is For example, assuming the assessment level of the accident probability index is {very small, small, medium, large, very large}, and the corresponding membership values ​​are {0.2, 0.5, 0.3, 0.0, 0.0}, then its fuzzy assessment vector is R1 = (0.2, 0.5, 0.3, 0.0, 0.0). Membership function matching transforms the operating parameters into a fuzzy assessment vector, quantifying the probability of parameter values ​​at different risk levels, and providing a fuzzy information representation for comprehensive risk assessment.

[0056] Using the fuzzy weighted average operator, the integrated evidence body m(A) and the fuzzy evaluation vector are combined. The fuzzy membership distribution U(x) of the comprehensive risk assessment level is obtained by fusion: the fuzzy weighted average operator linearly combines the evidence body and the fuzzy assessment vector to obtain the membership distribution of the risk level. Let the comprehensive evidence body for the importance of the i-th indicator be... The fuzzy evaluation vector is Then the fuzzy membership distribution of the comprehensive risk assessment level is: U ( x ) = ∑ i = 1 3 [ m ( A i ) × R i ( x ) ] Membership distribution By integrating the importance of indicators and fuzzy assessment information, this method reflects the probability of different risk levels, providing a basis for the final risk level decision. The fuzzy weighted average integrates evidence theory and fuzzy set theory, using evidence to assign indicator weights and membership functions for fuzzy assessment. This ensures that risk assessment takes into account both subjective and objective factors, resulting in more reasonable and reliable assessment results.

[0057] Using the centroid defuzzification method in fuzzy mathematics, the fuzzy membership distribution is... Converted into quantified safety risk level values The centroid method transforms fuzzy quantities into definite quantities by calculating the centroid's abscissa of the membership distribution, thus obtaining a quantified risk level value. The formula for calculating the centroid's abscissa x is: Where a and b are the upper and lower bounds of the risk level domain. x reflects the overall position of the comprehensive risk level in the domain; the larger x is, the higher the risk level. The smaller the value, the lower the risk level. The centroid defuzzification method is simple and intuitive. It directly reflects the overall characteristics of the membership distribution through the centroid position, quantifying fuzzy risks into definite level values, which facilitates risk comparison and ranking.

[0058] Initial risk level and quantified risk level value Two sources of evidence are considered: the initial risk level originates from the equipment's safety status information, such as sensor data and fault records, reflecting the equipment's current operating status and potential risks. The risk level value is then quantified. Derived from comprehensive risk assessment models, such as the evidence theory-fuzzy assessment model, which comprehensively considers factors such as the probability of an accident, the severity of its consequences, and personnel's risk awareness, a quantitative score for the risk level is given. Two sources of evidence provide a basis for judging the safety risk level of the circular saw from the perspectives of equipment status and comprehensive assessment, respectively.

[0059] Evidence fusion using Yager's rule: Yager's rule is a conflict-of-evidence fusion method in DS evidence theory, applicable to situations where there is high conflict between pieces of evidence. Let the evidence corresponding to the initial risk level be... Quantify risk level value The corresponding evidence is The fusion process of Yager's rule is as follows: Calculate the conflict coefficient k between evidence bodies: ,in For a non-empty subset C, calculate the confidence score after fusion. : Where A∩B=C, for the entire universe of discourse Calculate the confidence level after fusion. : The combined confidence level m(C) represents the final level of support for subset C. This represents uncertainty about the entire domain. Through evidence fusion using Yager's rule, the initial risk level and the quantified risk level value are combined. The judgment results are combined to obtain the final risk level that takes into account both the equipment status and the overall assessment.

[0060] The final risk level is determined by interpreting the fusion results: based on the confidence distribution after fusion, the final risk level is determined. The risk level can be selected from the subset with the highest confidence, or the risk level can be divided according to a preset threshold based on the confidence level. For example, if the subset with the highest confidence after fusion is "high risk," then the current safety risk level of the circular saw is determined to be high risk, requiring emergency measures. The final risk level comprehensively considers equipment status information and comprehensive risk assessment results, providing a more comprehensive and objective reflection of the safety risk level of the circular saw, and offering a reliable basis for subsequent early warning decisions.

[0061] This application integrates equipment status information and comprehensive risk assessment results by fusing initial risk levels and quantified risk level values, resulting in a final risk judgment that considers both factors and improves the comprehensiveness and accuracy of risk assessment. The Yager rule, through the redistribution of confidence levels, allocates the confidence levels of conflicting evidence across the entire domain, effectively reducing the impact of conflicting evidence on the fusion result, making the risk level judgment more robust and reliable. The fused confidence level distribution not only provides the final risk level judgment but also reflects the degree to which the evidence supports the result and the uncertainty regarding the entire domain, providing decision-makers with richer information.

[0062] like Figure 4 As shown, based on the safe operating status and safety risk level, a multi-level early warning control strategy based on reinforcement learning is used to manage the safety of the circular saw. A Markov decision process model (MDP) for the safety early warning control of the circular saw is constructed. According to the safe operating status and safety risk level of the circular saw, the safety status of the circular saw is divided into three states: safe, warning, and dangerous, forming a state space. The SVM classification model is used to classify the real-time operating data of the circular saw and determine its current safe operating status, classifying it into two categories: safe and dangerous. An evidence theory-fuzzy assessment model is then used to evaluate the safety risk of the circular saw, obtaining risk levels categorized into low risk, medium risk, high risk, and extremely high risk. The SVM classification results and risk assessment levels are combined to define three safety states: This indicates a safe state, corresponding to an SVM classification result of "safe" and a risk level of "low risk". This indicates a warning status, with the corresponding SVM classification result being safe and the risk level being medium risk. This indicates a dangerous state, corresponding to an SVM classification result of "dangerous," with a risk level of high or extremely high risk. The state classification comprehensively considers the equipment's real-time operating status and potential risk level, providing a more comprehensive and accurate reflection of the equipment's safety condition and offering a basis for early warning and control decisions.

[0063] Based on the safety status of the circular saw, corresponding early warning control actions are set to create a range of motion. Four warning control actions are defined for different safety states: This indicates that the current state will be maintained, and no early warning or control measures will be taken. Applicable to a safe state. The equipment is operating normally, the risk level is low, and no intervention is required. This indicates the issuance of a warning signal, including audible and visual alarms, and human-machine interaction prompts. It applies to warning states. The equipment is operating normally, but the risk level is high, and operators need to be reminded to pay attention. This indicates forced speed reduction, limiting the maximum rotational speed of the circular saw via the controller. Suitable for hazardous situations. The equipment is malfunctioning and poses a high risk level, requiring restrictions on equipment performance to reduce the risk of accidents. This indicates emergency braking; the controller cuts off power to the circular saw and applies braking force. Suitable for hazardous situations. The equipment is operating abnormally, posing an extremely high risk level, requiring immediate shutdown to prevent an accident. The early warning and control system employs intervention measures of varying intensities based on the severity of the safety situation, achieving multi-level early warning and control.

[0064] Collect historical operating data of the circular saw, including equipment status, control actions, and state transitions. For each state-action pair (s, a), count the frequency of transitions to different states s' after taking action a, and obtain a frequency matrix. The frequency matrix reflects the empirical distribution of transitions to different states after taking a specific action in a specific state, demonstrating the statistical regularity of state transitions.

[0065] Based on the state transition frequency matrix Calculate state transition probability For each state-action pair (s, a), the frequency matrix will be... Normalize each row to obtain the state transition probability. State transition probability This represents the probability of transitioning to state s' after taking action a in state s. It transforms the frequency distribution into a probability distribution, reflecting the uncertainty of state transitions. The state transition probability is one of the core parameters of the MDP model, determining the dynamic changes in state when the agent interacts with the environment.

[0066] Set the reward function for early warning control , represents the immediate reward obtained by the circular saw after taking action a in state s; ,

[0067] in, Indicates a security risk reward. This indicates a production efficiency bonus. Indicates an energy loss reward; The weighting coefficients for the three reward components satisfy the following conditions: ; ,in, This represents the risk cost under state s, and different cost values ​​can be set according to the risk level. This represents the probability of an accident occurring after taking action a in state s, which can be estimated based on historical data and expert experience.

[0068] The calculation formula is: ,in, This represents the time cost saved after taking action a in state s. Different cost values ​​can be set according to the degree of impact of different actions on production efficiency. This represents the probability of resuming normal production after taking action a in state s, which can be estimated based on the state transition probability. The calculation formula is: ,in, This represents the energy cost consumed after taking action a in state s. Different cost values ​​can be set according to the energy consumption characteristics of different actions.

[0069] The MDP model formalizes the circular saw safety early warning control problem into a sequential decision problem by defining elements such as state space, action space, state transition probabilities, and reward function. The state space comprehensively considers the equipment's operating status and risk level, the action space covers multi-level early warning control measures, and the reward function takes into account multiple objectives such as safety, production, and energy consumption, thus comprehensively characterizing the internal mechanism and decision-making process of circular saw safety early warning control.

[0070] The Q-learning algorithm is used to solve the Markov Decision Process (MDP) model to obtain the optimal early warning control strategy. Q-learning is a model-free, heterogeneous policy temporal difference learning algorithm that iteratively updates the state-action value function Q(s, a) to gradually approach the optimal policy. Initialize Q(s, a) to arbitrary values ​​(e.g., all zeros), and set the learning rate α and discount factor γ. Repeat the following steps until Q(s, a) converges or reaches the maximum number of iterations: Select action a according to an ε-greedy policy, i.e., select the action with the largest Q value with a probability of 1-ε, and explore randomly with a probability of ε. Execute action a, observe the next state s' and the immediate reward r. Update Q(s, a) according to the Q-learning update formula: Q ( s , a ) ← Q ( s , a ) + α [ r + γ max α ' Q ( s ', α ') − Q ( s , α ) ] Update the current state s to Repeat the above steps. After the learning process is complete, for each state s, select the action a with the largest Q value as the optimal policy π(s): The Q-learning algorithm continuously interacts with the environment, accumulating experience samples of state transitions and reward feedback, dynamically adjusting the state-action value function, and ultimately learning the optimal early warning control strategy. This algorithm can learn autonomously in unknown environments, exhibiting strong adaptability and robustness, and can cope with random fluctuations in the safety state of the circular saw, making optimal decisions.

[0071] Based on the optimal early warning and control strategy Generate corresponding early warning and control commands; based on the optimal strategy Select the appropriate early warning control action. The safety state 's' and risk level of the circular saw are calculated in real time using a safety state discrimination model (such as an SVM classification model) and a risk assessment model (such as an evidence theory-fuzzy assessment model). The real-time state 's' is then input into the optimal early warning control strategy 'π' to obtain the corresponding optimal control action 'a'. The optimal policy π can be implemented through table lookup or function mapping, directly outputting the optimal action a based on the state s. This avoids recalculating the Q-value and action selection each time, improving the real-time performance of decision-making. This is then converted into specific early warning and control commands, such as warning signals, speed reduction commands, and braking commands: Based on the type of the optimal action 'a', a corresponding early warning and control command is generated. For example, if 'a' is a warning signal, an audible and visual alarm command is generated; if... To force a speed reduction, a speed limit command and a target rotational speed value are generated. The warning control command must conform to the communication protocol and command format of the circular saw control system, such as Modbus or CANopen. The command content includes parameters such as command type, target value, and execution time. The warning control command is encapsulated into standardized data packets, such as JSON or XML formats, for easy transmission and parsing between different systems.

Claims

1. A digital-based method for safe construction management, characterized in that, include: Collect circular saw operation data at the construction site; the operation data includes saw blade speed, saw blade height, and protective cover position; Preprocess the collected circular saw operation data; Based on the preprocessed circular saw operation data, the safe operation status of the circular saw is determined by the Support Vector Machine (SVM) algorithm; the input of the SVM algorithm is the circular saw operation data, and the output is the safe operation status of the circular saw. The decision function expression for the SVM classification model is: , where x represents the feature vector formed by the circular saw's running data; w represents the normal vector of the SVM classification hyperplane; λ represents the nonlinear transformation performed by the RBF kernel function on the feature vector x; b represents the intercept of the SVM classification hyperplane; λ represents the safety state correction factor. Represents the safety state correction function; Safety state correction function The expression is: ,in, These are the weighting coefficients for the three feature parameters; The safety state correction subfunction for the three characteristic parameters: ; ; ;in, These are the safety thresholds for saw blade speed, saw blade height, and protective cover position, respectively. These are the upper and lower limits of the safety range for the three characteristic parameters; among them, The saw blade rotation speed, The height of the saw blade. Location of the protective shield; Based on the safe operation status, a risk assessment method combining evidence theory and fuzzy set theory is used to conduct a risk assessment and obtain the safety risk level. Based on the safe operating status and safety risk level, a multi-level early warning control strategy based on reinforcement learning is used to manage the safety of the circular saw. A Markov decision process (MDP) model for the safety early warning control of the circular saw is constructed, including: dividing the safety status of the circular saw into three states—safe, warning, and dangerous—based on its safe operating status and safety risk level, forming a state space. ;in, Indicates a safe state; Indicates a warning status; Indicate a dangerous state; based on the safety status of the circular saw, set corresponding early warning control actions to create a range of motion. ;in, This indicates that the current situation will be maintained and no early warning or control measures will be taken. This indicates that a warning signal has been issued; This indicates a forced speed reduction; Indicates emergency braking.

2. The digital-based safety construction management method according to claim 1, characterized in that: The safe operating status of the circular saw is determined using the Support Vector Machine (SVM) algorithm, including: The radial basis function (RBF) is selected as the kernel function of the SVM algorithm to construct an SVM classification model. The hyperparameters of the SVM classification model are optimized using a grid search method. The hyperparameters include the penalty coefficient C and the parameter γ of the radial basis function (RBF). Based on the preprocessed circular saw operation data, construct the feature vector x; The safe operating status of the circular saw is obtained by calculating the distance between the feature vector x and the classification hyperplane using the decision function of the SVM classification model.

3. The digital-based safety construction management method according to claim 2, characterized in that: Risk assessment is conducted using a risk assessment method that combines evidence theory and fuzzy set theory, including: Determine the initial risk level based on the safe operating status of the circular saw; Establish a risk assessment index system that includes three indicators: accident probability, severity of accident consequences, and saw blade exposure; and set the assessment level and fuzzy membership function for each indicator. Based on historical data, three importance weights were assigned to the indicators of accident probability, accident consequence severity, and saw blade exposure. The probability assignment function (BPA) from evidence theory was then used to quantify these importance weights, resulting in the evidentiary body for the importance of each indicator. ; According to the evidence Calculate the comprehensive body of evidence using Dempster's combination rule. ; Based on the circular saw's operational data, fuzzy evaluation vectors for each indicator are obtained through membership function matching; where the evaluation level of the i-th indicator is... The membership value obtained from the matching is Then the fuzzy evaluation vector of the corresponding indicator Where n represents the number of evaluation levels; Using the fuzzy weighted average operator, the comprehensive evidence body and fuzzy evaluation vector By fusion, the fuzzy membership distribution of the comprehensive risk assessment level is obtained. : in, The evidence body for the importance of indicator i, wherein, This represents the identification framework for the i-th evaluation indicator; Let x be the fuzzy evaluation vector for indicator i, where x represents the risk level; Using the centroid defuzzification method in fuzzy mathematics, the fuzzy membership distribution is... Converted into quantified safety risk level values ; Based on the initial risk level and the safety risk level value By adopting the Yager fusion rule in the DS evidence theory, the final safety risk level is obtained by comprehensively considering the safety status and risk assessment of the circular saw.

4. The digital-based safety construction management method according to claim 3, characterized in that: Set the evaluation level and fuzzy membership function for each indicator, including: The assessment levels for the probability of an accident include very small, small, medium, large, and very large, and the membership function adopts a triangular membership function. The severity of the accident consequences is assessed in levels of very low, low, medium, high, and very high, and the membership function uses a trapezoidal membership function. The assessment levels for saw blade exposure include very weak, weak, medium, strong, and very strong, and the membership function uses a Gaussian membership function.

5. The digital-based safety construction management method according to claim 3, characterized in that: Calculating the composite body of evidence using Dempster's combination rule ,include: Calculate the degree of conflict K between the evidence pieces: in, These are subsets of the three indicator importance assessment frameworks; Calculate the comprehensive evidence body of the non-zero subset A : Where A is a non-zero subset of the indicator importance assessment framework.

6. The digital-based safety construction management method according to claim 3, characterized in that: Fuzzy membership distribution Converted into quantified safety risk level values ,include: Calculate the fuzzy membership distribution The x-coordinate of the centroid : in, Let be the membership distribution function for comprehensive risk assessment, and [a, b] be the universe of discourse interval of x, where a and b are the lower and upper bounds of the universe of discourse, respectively.

7. The digital-based safety construction management method according to claim 6, characterized in that: Safety management of circular saws is achieved through a multi-level early warning and control strategy based on reinforcement learning, including: Construct a Markov decision process (MDP) model for safety early warning control of circular saws; The Q-learning algorithm is used to solve the Markov Decision Process (MDP) model to obtain the optimal early warning control strategy. ; Based on the optimal early warning and control strategy This generates corresponding early warning and control commands.

8. The digital-based safety construction management method according to claim 7, characterized in that: A Markov decision process (MDP) model for safety early warning control of circular saws is constructed, including: Based on historical operational data, the frequency of transitions from different safe states to other states after the circular saw takes different early warning control actions was statistically analyzed, resulting in a state transition frequency matrix. ;in, Indicates the state after the transition; Based on the state transition frequency matrix Calculate state transition probability : Set the reward function for early warning control , represents the immediate reward obtained by the circular saw after taking action a in state s; in, Indicates a security risk reward. This indicates a production efficiency bonus. Indicates an energy loss reward; These are the weighting coefficients for the three reward components; in, This represents the risk cost in state s; This represents the probability of an accident occurring after taking action a in state s; The calculation formula is: in, This represents the time cost saved by taking action a in state s; This represents the probability that normal production will resume after taking action a in state s; The calculation formula is: in, This represents the energy cost consumed after taking action a in state s.

Images