Construction site collision risk assessment method based on digital twinborn scene

Abstract

The invention provides a construction site collision risk assessment method based on a digital twinborn scene, and relates to the technical field of building simulation and collision detection. According to the method, after collision risk evaluation factors are screened out by constructing a construction site digital twinborn scene, correlation analysis is carried out by using a weighted Heraonian average operator, and a determined weight is obtained; constructing a membership function of each collision risk evaluation factor, performing fuzzy reasoning, establishing a construction site collision risk evaluation mechanism, and outputting an evaluation result; dynamically generating and updating the working space of the digital twinborn scene according to the evaluation result, reconstructing the digital twinborn scene, and executing voxelization collision detection to obtain an accurate collision detection result; and according to an accurate collision detection result, in combination with the generated dynamic working space, calculating a collision time-space amount to obtain a final risk value. The method is high in collision detection speed and accurate, is applied to a dynamic complex environment of a construction site, and can achieve objective and accurate evaluation of space-time collision risks.

Description

Technical Field

[0001] The present invention relates to the technical field of building simulation and collision detection, and in particular to a construction site collision risk assessment method based on a digital twin scenario and its application. Background Art

[0002] In the construction sites of modern buildings and infrastructure, with the expansion of project scale and the complexity of construction machinery and equipment, how to ensure the reasonable layout and safe operation of various equipment, materials and structures has become a major challenge in project management. Especially in high-density construction environments, when multiple equipment such as large cranes, concrete pump trucks, tower cranes, and robotic arms are operating at the same time, the risk of spatial conflicts and collisions between equipment increases significantly. Traditional construction site collision detection mainly relies on manual monitoring and prejudgment, combined with the use of two-dimensional drawings or three-dimensional models to plan and adjust the location of equipment. However, with the increasing complexity of the construction site, this method shows obvious limitations when facing the real-time dynamic operation of multiple devices. Especially in large-scale, multi-level construction projects, manual work cannot promptly detect and deal with potential spatial conflict risks. Therefore, how to achieve real-time conflict detection and risk warnings between construction site equipment through automation technology to ensure safe and efficient collaborative work of construction equipment is a key issue in construction management.

[0003] There are two main existing construction safety risk assessment methods: one is the weight-based construction safety risk assessment method. This method is based on the traditional indicator system method, which introduces all evaluation indicators into the same scoring system, stipulates the scores of different indicators, selects risk factors with larger weights through statistical data, and introduces weight factors based on universal standard weights to measure the degree of influence of each indicator. The weighted average of all evaluation indicator scores is taken as the risk level classification standard. The second is some improved construction project risk assessment methods. Based on the construction project risk management theory and the characteristics of the project, this method identifies risk influencing factors and groups them according to hierarchical relationships, establishes a construction project risk evaluation index system, and uses the intuitionistic fuzzy set theory to give an improved intuitionistic fuzzy complementary judgment matrix. The intuitionistic fuzzy hierarchical analysis method is used to calculate the attribute weights and give the optimal decision.

[0004] There are many factors to consider in the actual operation of the engineering construction safety risk assessment method and the assessment cycle is long. The existing risk assessment methods, whether the first or the second, are relatively rigid in solving the problem of risk assessment. They are unable to dynamically follow up the assessment over time based on the current construction status, resulting in the assessment results being inconsistent with the actual construction conditions on site.

[0005] In view of this, the applicant filed this application after studying the existing technology. Summary of the invention

[0006] The present invention aims to provide a construction site collision risk assessment method, device, equipment and medium based on a digital twin scenario, so as to realize a systematic analysis of the correlation between risk factors in a dynamic and complex environment such as a construction site, perform more scientific weighting, and dynamically update the size of the working space required for the object according to changes in environmental factors to realize voxelized collision detection of the object, and the collision detection is more accurate.

[0007] In order to solve the above technical problems, the present invention is implemented through the following technical solutions: A construction site collision risk assessment method based on a digital twin scenario, comprising: S1, constructing a digital twin scene of the construction site, screening out collision risk assessment factors, and using a weighted Heronian average operator to perform correlation analysis on the collision risk assessment factors to obtain determined weights; S2, constructing the membership function of each collision risk assessment factor, performing fuzzy reasoning, establishing a construction site collision risk assessment mechanism, and outputting an assessment result; wherein the assessment result is the possibility of occurrence of a risk at the construction site; S3, dynamically generating and updating the workspace of the digital twin scene according to the evaluation results, reconstructing the digital twin scene, and performing voxelized collision detection in the digital twin scene to obtain accurate collision detection results; S4, according to the precise collision detection result, the time and space amount of the collision is calculated in combination with the generated dynamic workspace to obtain the final risk value.

[0008] The present invention also provides a construction site collision risk assessment device based on a digital twin scene, comprising: The digital twin construction safety scenario establishment module is used to build the digital twin scenario of the construction site. After screening out the collision risk assessment factors, the weighted Heronian average operator is used to perform correlation analysis on the collision risk assessment factors to obtain the determined weights. Environmental hazard occurrence probability calculation module, used to construct the membership function of each collision risk assessment factor, perform fuzzy reasoning, establish a construction site collision risk assessment mechanism, and output the assessment result; wherein the assessment result is the probability of the construction site risk; A voxelized precise and fast collision detection module, which is used to dynamically generate and update the workspace of the digital twin scene according to the evaluation results, rebuild the digital twin scene, and perform voxelized collision detection in the digital twin scene to obtain precise collision detection results; The comprehensive risk assessment module is used to calculate the time and space volume of the conflict based on the precise collision detection results and the generated dynamic workspace to obtain the final risk value.

[0009] The present invention also provides a construction site collision risk assessment device based on a digital twin scene, comprising a processor and a memory, wherein a computer program is stored in the memory, and the computer program can be executed by the processor to implement a construction site collision risk assessment method based on a digital twin scene as described above.

[0010] The present invention also provides a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, a construction site collision risk assessment method based on a digital twin scenario as described above is implemented.

[0011] In summary, compared with the prior art, the present invention has the following beneficial effects: (1) The present invention can achieve an objective and accurate assessment of the risk of space-time conflict. The assessment method used is not only based on a single proximity for risk assessment, but also converts the proximity into the opportunity for personnel and machinery to contact danger, quantifies the amount of space-time conflict (i.e., the conflict impact of time and workspace), and pays attention to other environmental factors. The method combines the weighted Heronian average operator with fuzzy reasoning to systematically analyze the correlation between risk factors, perform more scientific weighting, and avoid the impact of extreme environments on the assessment results. In addition, the present invention dynamically updates the size of the workspace required for the object according to the changes in environmental factors for collision detection, so that the more dangerous objects will receive earlier attention and corresponding conflict risk assessments can be performed.

[0012] (2) The present invention combines the voxel tree in the collision detection module, uses the axial bounding box and the spherical voxel model for secondary collision detection, and the collision detection is accurate. In the collision detection process of large-scale complex models such as construction sites, the collision detection calculation time can be kept below 60 milliseconds. At the same time, the time spent on secondary collision detection is monitored to ensure that the total time of the collision detection meets the sensor signal sampling frequency requirements, avoiding a series of problems such as delays and confusions in the overall system due to the calculation time of the collision detection not meeting the signal sampling frequency, resulting in inaccurate collision detection, thereby ensuring the correctness of collision detection at the construction site.

[0013] (3) In the process of generating the voxel tree, the present invention combines the congestion ratio. Compared with the traditional method of dividing the octave voxel tree to the maximum depth or dividing it according to the number of objects, the present invention reduces the computer running memory usage and is more suitable for large-scale and complex construction scenes.

[0014] (4) The present invention uses an circumscribed voxel sphere instead of a voxel cube, which reduces the computer memory usage and eliminates the axial problem in collision detection between cubes.

[0015] (5) The present invention adopts a voxel hollowing method and a method for dynamically calculating voxel ratios to reduce the number of traversals between voxels and speed up the calculation speed during accurate collision detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A schematic diagram of a construction site collision risk assessment method based on a digital twin scenario provided in Example 1.

[0018] Figure 2 A schematic diagram of the framework of the construction site collision risk assessment method based on the digital twin scenario provided in Example 1.

[0019] Figure 3 A schematic diagram of the dynamic expansion and contraction of the experience level in the environmental hazard possibility provided in Example 1.

[0020] Figure 4 A schematic diagram of the recursive process structure of voxel tree generation is provided for the first embodiment.

[0021] Figure 5 This is a schematic diagram of the visualization result generated by the voxel space provided in the first embodiment.

[0022] Figure 6 This is a schematic diagram of the object crowding ratio scenario provided in Example 1.

[0023] Figure 7 This is a schematic diagram of the generated axial bounding box of an object provided in Example 1.

[0024] Figure 8 This is a schematic diagram of the collision detection process between a cube voxel and a sphere voxel provided in the first embodiment.

[0025] Fig. 9 This is a schematic diagram of a humanoid spherical voxel model provided in Example 1.

[0026] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] Embodiment 1 Embodiment 1 of the present invention provides a construction site collision risk assessment method based on a digital twin scenario, which can be implemented by a construction site collision risk assessment device based on a digital twin scenario (hereinafter referred to as a collision risk assessment device), and in particular, executed by one or more processors in the collision risk assessment device.

[0029] In this embodiment, the collision risk assessment device may be an electronic device equipped with a processor, which carries a computer program of the construction site collision risk assessment method based on the digital twin scenario and the computer program can be executed, such as a computer, a smart phone, a smart tablet, a workstation, etc., which is not limited here.

[0030] In this embodiment, Digital Twin refers to the use of digital means to create a virtual model corresponding to a physical entity in a virtual space, and to form a dynamic two-way mapping between the physical world and the digital world through the collection, transmission and analysis of real-time data. Digital Twin technology can not only simulate, monitor, analyze and optimize physical objects, but also make predictions and decision support through virtual models.

[0031] Workspace: refers to the physical or virtual environment used by individuals or teams to complete work tasks safely and efficiently. Existing methods for generating workspaces are mostly based on outsourcing cuboids or cylinders, which cannot truly reflect the essential geometric characteristics of the entity and have an impact on accurate collision detection. In addition, the definition of workspace is usually qualitative, lacking consideration of the current dynamic risk environment around it. For example, the high-altitude falling area near the moving robot arm is also a dangerous area.

[0032] Voxelization: It is a process of representing an object in three-dimensional space as a collection of small cubes to approximate its description.

[0033] Collision Detection: It is a technique used in computer graphics, physics simulation, and game development to detect whether two or more objects are touching or passing through each other. Its main purpose is to determine the physical interaction between objects so that appropriate responses or processing can be performed when a collision occurs.

[0034] Risk Assessment: refers to the identification, analysis and evaluation of potential hazards in specific activities, processes, equipment or substances to determine their possible impact on personnel, property, environment or other interests.

[0035] This embodiment mainly conducts a construction site hazard exposure risk assessment based on the possibility of environmental hazards occurring, the chances of people and machinery approaching the hazards, and the severity of the consequences of the hazards occurring.

[0036] like Figure 1 and 2 As shown, a construction site collision risk assessment method based on a digital twin scenario includes steps S1 to S4.

[0037] S1, construct a digital twin scene of the construction site, screen out the collision risk assessment factors, and use the weighted Heronian average operator to perform correlation analysis on the collision risk assessment factors to obtain determined weights.

[0038] Furthermore, according to the construction technology and safety technical requirements of the physical space in the construction process, the safety operation of the construction process is standardized, the logical flow between the process levels and work steps of the construction process, the safety technical requirements and potential safety risks associated with the work steps are analyzed, and the standardized model and virtual space corresponding to the physical entity of the construction process are established to obtain the attribute data of the physical entity; Deploy IoT sensor devices on physical entities during the construction process to dynamically monitor and obtain index data of the physical entities, including position information and motion information of the monitored physical objects, such as motion information of a robotic arm; Using the indicator data obtained from monitoring as input, the standardized model corresponding to the physical entity of the construction process is simulated in real time, and the rendered model is imported into the virtual reality simulation engine. The physical engine is used to build a visual simulation model to obtain a digital twin construction scene, so as to realize information interaction and data synchronization between the physical entity and the digital twin.

[0039] Real-time data collection, remote communication and real-time updating of multi-source dynamic data are achieved through the communication interface. Real-time interaction and synchronous feedback between the digital twin and the physical entity are achieved through the human-machine interface and database interface.

[0040] Furthermore, when screening collision risk assessment factors, multiple safety managers are required to make scoring evaluations at different time stages. By integrating real-time data, historical data and the scores of each expert, the contribution of various risk factors (such as illumination, speed, congestion, etc.) to the occurrence of accidents is dynamically evaluated, including steps 1 to 6.

[0041] Step 1: Define scoring dimensions for each risk factor (such as lighting, speed, crowding, and experience level). For example, each factor can be divided into multiple levels (such as 1-10 points), where the higher the score, the greater the risk contribution of the factor to the accident.

[0042] Set an initial score for each risk factor based on historical data, expert scoring or industry standards. For example, light intensity: low light 8 points, medium light 5 points, full light 2 points, full score 10 points.

[0043] Speed: 9 points for high speed, 5 points for normal speed, 2 points for low speed, full score is 10 points.

[0044] Crowding: high congestion 9 points, medium congestion 6 points, low congestion 3 points, full score is 10 points.

[0045] Experience level: 2 points for more experience, 8 points for less experience, full score is 10 points.

[0046] In this embodiment, the dangerous collision scenario is a collision between a human and a machine, and four factors, namely, moving speed, light intensity, spatiotemporal congestion, and whether or not workers or machine operators have experience, are selected to evaluate the risk of spatiotemporal conflict at the construction site.

[0047] Step 2: Use a time weighting factor Calculate a score for each collision risk assessment factor. Time weighting factor Expresses the impact of past risks on the current state. It can be defined by an exponential decay function, which means that the risk gradually decays over time and the impact of the past will gradually weaken.

[0048] The time weighting factor Calculated by the following formula: ; in, is the decay rate parameter, and t is the time interval (i.e., the time difference from the last risk score to the current one). Indicates that risk decays rapidly over time, while smaller It means that the impact of past risks lasted for a long time.

[0049] The introduction of a time-weighted factor means that past risk events will gradually affect the current risk score. For example, if a risk factor (such as light intensity) has been in a high risk state for a period of time in the past, even if the current light conditions are good, its impact on the current score will continue to exist, but this impact will gradually weaken over time.

[0050] This mechanism helps to consider historical data in the assessment process, ensuring that certain long-standing risk factors (such as persistent low light or crowding) are captured in a timely manner, rather than relying solely on real-time data.

[0051] Step 3: For each risk factor score, it is necessary to adjust it through the initial score and the time weighting factor to reflect the impact of historical data on it. The specific calculation formula is as follows: ; in, The current initial score for the collision risk assessment factor; The score is weighted by time; This is the historical impact amplitude of the collision risk assessment factor, which is determined by the specific risk factors. For long-term accumulation, a larger value such as 2 can be selected, and a smaller value of 1 can be used for short-term fluctuations.

[0052] Suppose that in the past, the light conditions were poor (high score, indicating high risk), but now the light intensity has improved. When scoring, it is necessary to be able to gradually reflect the decline of historical high risk over time. For example, suppose =3, indicating that the current lighting conditions have improved; the risk level is set to 2 (low lighting conditions in the past led to high risks and had a greater impact), =0.25.

[0053] When t=3: ; (1-0.4724) 0.4724=3.4986; At t = 3, the high risk impact of historical low light conditions slightly increases the current score to 3.4986, still maintaining the effect of current light improvement.

[0054] When t=4: ; (1-0.3679) 0.3679=3.465; At t = 4, the impact of past low light conditions weakened further, and the final weighted score slightly decreased to 3.465. As time increases, the cumulative impact of past high risk gradually decreases.

[0055] Therefore, the current score will be attenuated by the long-term effect of historical low light risk, and this effect will decrease as the interval increases.

[0056] Step 4: Adopt an adaptive learning model. Dynamically adjust the weight and score of each risk factor based on real-time data to ensure that the score can be flexibly updated as the construction environment changes. For example, as the operator's experience improves, the system automatically reduces the risk score of the factor. If the light intensity is monitored to continue to deteriorate over a certain period of time, the score curve will gradually rise over time until the risk value reaches the preset threshold. This process is similar to the weight update in machine learning, which continuously optimizes the score through feedback.

[0057] The score of each collision risk evaluation factor is updated through an adaptive learning model, and the formula of the adaptive learning model is: ; in, represents the score after adaptive adjustment; Indicates the score of the previous stage; Indicates the difference between real-time data and historical data; Represents the learning rate parameter, which is used to control the adjustment speed of the risk assessment factor score. The calculation formula is: ); is the initial learning rate, which is set according to actual needs; Used to measure the fluctuation of historical scores. When the fluctuation is large, the learning rate Reduce, fluctuate hourly learning rate Increase; Indicates the magnitude of the change in the score. When the change is large, the learning rate Reduce, change the hourly learning rate Increase; , They are adjustment coefficients, which are used to control the impact of historical fluctuations and score changes on the learning rate and are set according to actual needs; represents the standard deviation of historical scores; represents the historical score average; Indicates the current stage score.

[0058] Assumptions is 5, is 1, is 5, is 6, is 1, the initial learning rate is 0.01.

[0059] Assumptions =0.5, then the impact of fluctuation on the learning rate is: =1+0.5 =1.1; Rating change range: |6.0-5.0|=1.0; Assumptions =0.3, then the effect of score change on learning rate is: =1-0.3 =0.94; The impact of comprehensive historical fluctuations and rating changes: =0.01 1.1 0.94=0.01034; at last, =5+0.01034 1=5.01034.

[0060] The results are explained as follows: the standard deviation of the historical scores is small, indicating that the environment is relatively stable, and the learning rate is slightly increased; the change in the current score is small (1 unit), so the learning rate is slightly reduced to avoid over-adjustment.

[0061] The learning rate parameters will automatically adjust as environmental data changes. The system adaptively adjusts the score according to the changing trend of lighting at the construction site, thereby achieving a more objective and reasonable evaluation. This method can effectively deal with the uncertainty of lighting conditions at the construction site and improve the accuracy of safety risk assessment. Step 5: Integrate expert scoring and data-driven scoring, and comprehensively consider expert scoring and data-driven scoring to ensure the accuracy of the evaluation model. For example, when real-time data is lacking, expert scoring is used first; after obtaining real-time data, the score is adaptively adjusted, and the two scoring sources are dynamically fused according to the credibility of the data. The weighted average method is used to fuse expert scoring and real-time data scoring. The formula is: ; in, It represents the result of integrating expert scoring and real-time data scoring; It indicates the expert score after frailty treatment. represents adaptive real-time data scoring; and The weights of expert scoring and adaptive real-time data scoring respectively.

[0062] Using a two-layer Bayesian update mechanism, we first perform independent Bayesian updates on the data-driven scoring and expert scoring, and then perform uncertainty quantification analysis based on the two independent update results to generate the final weight distribution. The details are as follows: Set initial weights (Expert scoring weight) and (Data-driven scoring weights). In the early stages of construction, you may rely more on expert scoring, for example, the initial weights are set as =0.7 and =0.3.

[0063] At the same time, a prior distribution is set for each weight. Based on the historical data or experience of the construction site, a Bayesian prior distribution is used to reflect the initial confidence level. For example, if the initial confidence of the expert score is high, it can be set to a Gaussian distribution. , while data-driven scoring can be done using ,in, It means that the expert scores in the initial stage are more credible. Indicates Gaussian distribution, i.e. normal distribution; , They represent the variance of expert ratings and the variance of data-driven ratings, respectively.

[0064] In the two-layer Bayesian update, the first layer performs Bayesian updates on expert scores and data-driven scores respectively, and the second layer integrates the results of the two and dynamically adjusts the weights.

[0065] Bayesian Update Results for Data-Driven Scoring The expression is: ; in, Represents the observation value obtained from the sensor or real-time monitoring data, Indicates how well the data-driven score matches the actual observations. Each new data provided by the sensor triggers an adjustment of the weights. Updates.

[0066] Bayesian update results of expert ratings The expression is: ; in, Indicates the score given by the expert. The update basis is the historical performance of the expert's score, that is, whether the past accident data or on-site feedback can prove the effectiveness of the expert's score.

[0067] After the first layer completes the update of each weight, it enters the second layer and uses uncertainty quantification to deal with the fluctuations and potential uncertainties of data sources. For example, when data quality fluctuates greatly or there is insufficient feedback from expert scoring, uncertainty increases. This uncertainty is further quantified by introducing Bayesian reasoning to generate dynamic risk assessment weights.

[0068] Specifically, the present invention can calculate the variance (i.e., uncertainty) of expert scores and data scores to obtain the results of the second-level Bayesian reasoning. , which is expressed as follows: ; in, It represents the result of comprehensive expert scoring and data-driven scoring. represents the posterior probability of the final weight; It represents the total likelihood function of assessing risk after integrating expert scoring and data-driven scoring. This step introduces uncertainty quantification based on the two-layer update results, so that the final weight can reflect the uncertainty of each scoring source.

[0069] In another preferred embodiment, each weight and Set an uncertainty indicator (such as variance or entropy) to reflect the credibility of the current data for the final weight. For example, if the expert scores are consistent with the historical accident records, the uncertainty of the expert scores is low; while if the sensor quality of the data-driven scoring fluctuates greatly, the uncertainty is high.

[0070] The final weight after each weight adjustment The formula is: ; in, and They represent the variance of expert scores and data scores, that is, uncertainty. The larger the variance, the smaller the weight. In this way, the weight with higher uncertainty will be reduced in the total score.

[0071] Each time new data is obtained, the model automatically adjusts the weights. As construction progresses, the accuracy of data-driven scoring will improve, while expert scoring may be more credible in the early stages. Therefore, adaptive learning will gradually reduce the weight of expert scoring and increase the weight of data scoring.

[0072] The final score can be calculated using the fused weights and score results, using the formula: ; in, Indicates the final score; It indicates the expert score after frailty treatment; Represents adaptive real-time data scoring.

[0073] Step 6: Follow the above steps to calculate the final score and proceed to the next step of relevance consideration.

[0074] Identify and screen out environmental factors that may lead to collisions between people or machinery. Specific factors include but are not limited to light intensity, speed, crowding, and the experience of workers or machinery operators.

[0075] Then, use sensors, computer vision and other monitoring technologies to input data, build virtual models, build secure data channels, and establish a dynamic evaluation framework based on digital twins to promote the flow and communication of information.

[0076] In this embodiment, after confirming the risk factors, a scoring evaluation is performed to obtain the opinions of each expert and data-driven scoring on each risk factor. There is a correlation between the various environmental factors. For example, when the light intensity is extremely low, the movement speed of the workers will be extremely slow. When the crowding is extremely high, the movement speed of the machinery will be extremely slow. In order to avoid the influence of such extreme situations, the influence of extreme values ​​on the weight distribution is reduced to a certain extent to determine a more objective indicator weight.

[0077] According to the scoring results of the collision risk evaluation factors, the weighted Heronian average operator is used to evaluate the scoring of each expert and the real-time data score. The evaluation formula of the weighted Heronian average operator is as follows: ; in, represents the evaluation result of the weighted Heronian average operator; 、 、 、 Respectively indicate j、 k、 、 The importance score of each collision assessment risk factor; , , Respectively represent the j, k, The weights of the collision assessment risk factors; p, q, r are the parameters of the Heronian average operator; represents the number of collision assessment risk factors; λ represents the standardization constant, and the calculation formula of λ is: ; The calculated G values ​​are sorted and the score with the highest G value is selected as the determination weight of the collision assessment risk factor. Weight The weighting method is: .

[0078] S2, constructing the membership function of each collision risk assessment factor, performing fuzzy reasoning, establishing a construction site collision risk assessment mechanism, and outputting an assessment result; wherein the assessment result is the possibility of occurrence of risks at the construction site.

[0079] Furthermore, a risk assessment level is set for each collision assessment risk factor, and a membership function is defined to describe the membership of each collision assessment risk factor at different assessment levels; a fuzzy relationship matrix is ​​obtained according to the membership value of each collision assessment risk factor at the corresponding risk assessment level; and a fuzzy comprehensive operation is performed according to the determined weights and the fuzzy relationship matrix to obtain a comprehensive risk assessment direction.

[0080] Before constructing the membership function of each collision risk assessment factor, we first construct a set of risk assessment levels, namely the evaluation set The risk assessment level can usually be set as the following levels (other levels can also be set according to actual conditions, which are not limited here), such as: Low risk: , Medium risk: ; High risk: , then the evaluation set It is expressed as: , }.

[0081] The risk levels such as "high risk", "medium risk" and "low risk" are converted into numerical form through fuzzy membership function.

[0082] In this embodiment, a membership function is defined for each risk factor to describe the membership of the factor at different evaluation levels. For example, factors such as light intensity, speed, congestion, and experience correspond to different membership functions.

[0083] (1) Insufficient or excessive light intensity will have a negative impact on construction safety, especially during nighttime construction, where the visibility of operators under low light conditions is reduced, which increases the risk of collision. Therefore, the membership function needs to reflect nonlinear risk changes. An asymmetric trapezoidal membership function is used to represent the membership of light intensity. , the formula is: ; in, Indicates light intensity, a represents the lowest critical value of light intensity, which is usually set to a very low light intensity (such as 10 lux); b represents the upper limit of light intensity (such as 100 lux).

[0084] Users can automatically adjust the thresholds a and b of the membership function as needed to highlight risks when there is insufficient light at night.

[0085] (2) Calculation speed membership The faster the speed, the greater the probability and severity of a collision. The expression is: ; in, Indicates speed, It is the critical low speed value of the speed. When the speed is less than or equal to this value, it is considered high risk. For example, it can be set to 2 m / s. It is the critical high speed value of speed. When the speed is higher than this value, it is considered low risk. For example, it is set to 6 m / s.

[0086] (3) Calculate the membership degree of congestion The higher the degree of congestion, the smaller the space between construction workers and equipment, the higher the probability of collision and mutual influence between them, and therefore the greater the risk; on the contrary, when the construction site is not crowded, there is less mutual interference between people and equipment, and the risk is lower.

[0087] A linear ascending membership function is used to represent the impact of crowding on risk. Crowding is represented by c (for example, the crowding index ranges from 0 to 1, 0 represents no crowding at all, and 1 represents extreme crowding).

[0088] ; in, Indicates the critical value of being completely uncrowded and having the lowest risk (e.g., very few people in the venue).

[0089] Indicates the extreme crowding threshold, where the risk is the highest (e.g. the venue is very crowded).

[0090] (4) Experienced operators can usually effectively reduce the probability of accidents, while inexperienced operators are more likely to make operational errors. Therefore, the membership function of experience The inverse relationship between experience level and accident risk should be reflected.

[0091] The linear descent membership function is used for calculation, and the experience is represented by e: ; Indicates the inexperience threshold (e.g., less than 1 year of work experience).

[0092] Indicates the critical point of extensive experience (e.g., more than 5 years of work experience).

[0093] Membership functions can also be set for other risk evaluation factors according to actual conditions, which are not limited here.

[0094] Next, a fuzzy relationship matrix is ​​established to represent the association between each risk factor and the risk level.

[0095] In another preferred embodiment, the membership values ​​of each risk factor under three evaluation levels (other evaluation levels, such as level 4, level 6, etc., are also available, which are not limited here) are organized into a matrix R. Assuming there are n risk factors, the fuzzy relationship matrix R is in the form of: ; Finally, the weight set is combined with the fuzzy relationship matrix R, and the comprehensive risk assessment value is obtained through weighted summation. According to the calculation results, the maximum membership method is used to select the risk level with the largest membership, or the fuzzy distribution method is used to retain the membership of all risk levels, so as to output the possibility of environmental danger.

[0096] The comprehensive risk assessment vector B is obtained by performing fuzzy comprehensive operation on the previously determined weight vector W and the fuzzy relationship matrix R, and the formula is: ; Wherein, W represents the determined weight, represents the determined weight of the nth collision assessment risk factor; R represents the fuzzy relationship matrix; represents the membership degree of the nth collision assessment risk factor; Indicates Level of risk assessment; Indicates the set risk assessment level.

[0097] Under the three evaluation levels, the results of the comprehensive risk assessment vector B are: ; in, is the comprehensive membership of the low risk level, is the comprehensive membership of the medium risk level, It is the comprehensive degree of membership of high risk level.

[0098] According to the comprehensive risk assessment vector B, the comprehensive risk occurrence probability score is obtained by weighted calculation using the centroid method: , The expression is: ; in, Indicates Level of risk assessment; express The comprehensive membership degree corresponding to the risk assessment level, ; It reflects the overall probability of risk occurrence. The larger the value, the greater the possibility of risk occurrence. =1, =2, =3, indicating a risk level from low to high. It is a continuous value that reflects the overall probability of risk occurrence and is between [1,3].

[0099] In particular, based on the risk-driven resource allocation principle, the workspace size of the object is correlated with the risk probability score. The size of the workspace is dynamically updated by association, so that when the collision risk indicator system and the dynamic workspace are generated, people or machines that may be in danger can be detected more quickly, thereby filtering out dangerous objects in advance and leaving sufficient buffer time.

[0100] Analysis of factors that affect the possibility of danger shows that movement speed, illumination, and crowding change dynamically over time, while factors such as the experience of workers or machine operators do not change over time. The minimum value of the workspace is determined by the axial bounding box of the object entity and the dynamic expansion ratio. Figure 3 As shown in the figure, workers or machine operators are divided into two levels of experience, high and low. The vertical axis is the expansion ratio, and the horizontal axis is the possibility of danger in the environment. The linear function for low experience is y=1 / 2x+1 / 2, and the linear function for high experience is y= / 2x+1- / 2.

[0101] Then, computer vision acquisition technology is used to monitor mobile resources, such as the coordinate position, number of targets, and moving speed of workers and machines. Sensors are used to collect data to monitor mobile resources, such as the posture of workers and machines and the light intensity of the environment. All monitoring data is input into the digital twin platform through information communication, and the digital twin scene is built in combination with the three-dimensional model, and evaluation instructions are issued.

[0102] S3, dynamically generate and update the workspace of the digital twin scene according to the evaluation results, rebuild the digital twin scene, and perform voxelized collision detection in the digital twin scene to obtain accurate collision detection results.

[0103] By analyzing the mainstream geometric collision detection strategy based on the workspace bounding box, in this module, the voxelized collision detection method based on the digital twin mode in Example 1 is used to perform accurate and fast collision detection.

[0104] The steps of voxelized collision detection in S3 are: A1, the constructed digital twin scene is voxelized and segmented, and a voxel tree is recursively generated based on the set crowding ratio threshold. The spatial area of ​​each node of the voxel tree is divided axially, and the space of the digital twin scene is recursively divided into smaller subspaces for efficient query, analysis and rendering.

[0105] In the process of generating the voxel tree, the root node of the voxel tree represents the cube area of ​​the entire digital twin scene. The step A1 is specifically as follows: Initialize the root node of the voxel tree to , expressed as: ; ); in, Represents the central coordinates of the digital twin scene, that is, the center point of the root node; is the side length of the digital twin scene, The value of is set to the maximum size of the digital twin scene; Points Values ​​in the X, Y, and Z coordinate directions; like Figure 4 As shown in the figure, each node of the voxel tree is recursively divided into 8 child nodes. According to the set maximum depth of the voxel tree and the set crowding ratio threshold, the voxel tree is dynamically generated to obtain the voxelized twin construction scene, so as to reduce the computer running memory usage. Figure 5 As shown, the resulting voxel space is visualized using green lines.

[0106] The crowding degree is used to represent the degree of occupation of objects in the pre-divided sub-nodes of the current voxel node, and is calculated by calculating the ratio of the occupied volume of the objects in the pre-divided next-level nodes to the volume of the pre-divided next-level nodes; The occupied volume of the object in the pre-divided next-level node is: the volume of the axial bounding box of the object intersecting with the pre-divided next-level node; Among them, the volume of the object's axial bounding box The calculation formula is: ; Then, the volume of the pre-divided next-level node is The calculation formula is: = 8; in, is the volume of the object’s axial bounding box, The width of the axial bounding box AABB, that is, the size on the X axis; The height of the axial bounding box AABB, that is, the size on the Y axis; The depth of the axial bounding box AABB, that is, the size on the Z axis; If the crowding of the objects in the next pre-divided node of the current node exceeds the set crowding ratio threshold, the division continues until the set maximum depth of the voxel tree is reached; nodes at different depths have different crowding ratio thresholds.

[0107] For example, when the depth of the voxel tree is not greater than 3, the crowding ratio threshold can be set to 0.05; otherwise, the crowding ratio threshold can be set to );in, Indicates the depth of the current node.

[0108] The spatial area of ​​each node of the voxel tree is divided axially; for the nodes of the voxel tree , set the node The center point is , the side length is , then the next level child node It is expressed as: ; = +(± / 4, ± / 4, ± / 4); = / 2; Among them, ± determines the different directions of the child nodes. Representation Node The center point of Representation Node The side length of The entire digital twin scene is thus divided into multiple voxel blocks, each of which contains location information, size information and its hierarchical structure; the root node records the basic parameters of the voxel tree, including center coordinates, voxel block size and hierarchical depth.

[0109] By combining the crowding ratio to generate a voxel tree, different from the traditional octave voxel tree that is divided to the maximum depth or divided according to the number of objects, it reduces the computer running memory usage and is more suitable for large-scale and complex construction scenes. It has the following advantages: (1) If Figure 6 As shown in scenes A and B, when the number of objects is small but the crowding ratio is high or the local crowding ratio is high, the traditional octree partitioning method will not continue to partition, but the voxel tree method considering the crowding ratio will perform further voxel tree partitioning, thereby filtering out objects that are truly suspected of collision and speeding up collision detection.

[0110] (2) If Figure 6 As shown in scene C, when there are a large number of objects, but the crowding ratio is low, it means that the volume of the objects is small relative to the scene space. The traditional octree partitioning method will continue to divide multiple layers, greatly increasing the computer memory usage. When the voxel tree method takes crowding into consideration, no further voxel tree partitioning will be performed, but subsequent collision detection will be performed directly, because the objects are small, even if precise collision detection is performed, it will not bring a significant increase in collision detection time.

[0111] (3) If Figure 6 As shown in scenario D, when the number of objects is large but does not reach the number threshold of traditional octree division, the traditional octree division method will not continue to divide due to the high crowding ratio, but the voxel tree method considering crowding will perform further voxel tree division, thereby filtering out objects that are truly suspected of collision and speeding up collision detection.

[0112] A2, determine the current candidate collision object according to whether the received sensor signal satisfies the first preset signal; and generate and trigger the first collision detection instruction for any current candidate collision object in the digital twin scenario, perform rough collision detection on the current candidate collision object and the voxel tree, and obtain a first detection result; wherein the first preset signal is the set time threshold for receiving monitoring data.

[0113] The step A2 is specifically as follows: If the received sensor signal of the physical entity satisfies the first preset signal, the digital twin corresponding to the physical entity is set as the current candidate collision object and added to the candidate collision object group; then a first collision detection instruction related to the current candidate collision object is generated to perform a rough collision detection between the working space and the voxel space of the current candidate collision object; If the received sensor signal of the physical entity does not satisfy the first preset signal, outputting a warning; The first collision detection instruction is to generate an axis-aligned axial bounding box for each component of the current candidate collision object, that is, the working space of the current candidate collision object, and then perform intersection collision detection based on the axial bounding box with the bounding box of each node of the voxel tree, that is, the voxel space; wherein each face of the axial bounding box is parallel to the coordinate axis of the world coordinate system, and the enclosing range of the axial bounding box is defined by the minimum point and the maximum point of the object itself; Suppose the set of all vertices of the current candidate collision object is V={ , ,..., }, each vertex The coordinates in three-dimensional space are ; Then, the minimum point of the axial bounding box is Determined by the minimum value of all vertices of the current candidate collision object in the X, Y, and Z directions: =(min( ),min( ),min( )); Where i=1,2,...,n, n is the number of vertices of the current candidate collision object; Respectively Values ​​in X, Y, and Z directions; The maximum point of the axial bounding box Determined by the maximum value of the X, Y, and Z directions of all vertices of the current candidate collision object: =(max( ),max( ),max( )); The center point C and size S of the axial bounding box are calculated by the minimum point and the maximum point: C=( + ) / 2; S= - ; Among them, C is the center point of the axial bounding box, and S is the size vector of the axial bounding box, that is, the length, width and height.

[0114] The generated object component bounding box, i.e., the workspace, is Figure 7 As shown, an axial bounding box of each component of the object is generated and displayed with a light red transparent bounding box. The dangerous area is also surrounded by an axial bounding box. In this embodiment, the dangerous area is the high-altitude falling area near the edge.

[0115] For each candidate collider in the candidate collider group, each node of the voxel tree is traversed to perform intersection collision detection based on the axial bounding box to filter out objects that are unlikely to collide.

[0116] The intersection collision detection is performed by comparing the relative positions of the minimum and maximum points of the two axial bounding boxes, specifically: Assume that the two axial bounding boxes to be detected are BA and BB, and the intersection condition is: ≥ , ≤ ; ≥ , ≤ ; ≥ , ≤ ; in, , , They are the maximum points in the X, Y, and Z directions of the axial bounding box BA respectively; , , They are the minimum points in the X, Y, and Z directions of the axial bounding box BA respectively; , , They are the maximum points in the X, Y, and Z directions of the axial bounding box BB respectively; , , They are the minimum points in the X, Y and Z directions of the axial bounding box BB respectively; If the above conditions are met in the three directions of X, Y, and Z, BA and BB intersect, otherwise they do not intersect; Traverse each node of the voxel tree until the deepest node of the voxel tree, store the intersection collision information of each candidate collision object and the deepest voxel tree node, and obtain a rough collision detection result, that is, a first detection result; wherein the first detection result includes whether a rough collision occurs and the collision voxel node information of the candidate collision object.

[0117] For example, according to the shape and position of the object, the relevant voxels in the voxel tree are marked as "occupied" or "free". A hierarchical voxel tree structure is formed by recursively merging adjacent "occupied" voxels.

[0118] In the coarse collision detection stage, by comparing whether the root nodes or higher-level nodes of the voxel trees of the two objects intersect, if these nodes do not intersect, it can be determined that the two objects have not collided, thereby avoiding more detailed collision detection and improving the calculation speed.

[0119] For example, the BA object is the deepest node in the voxel tree , If a collision occurs, the collision information of the BA object is stored as , Similarly, the collision information of the BB object is stored as , , BA objects have the same collision information as BB objects. , then the output BA object and BB object are the current quasi-collision group.

[0120] A3, if the first detection result indicates that a rough collision of the current candidate collision object has occurred and has the same collision voxel node, the current candidate collision object is added to the current quasi-collision group; a second collision detection instruction is generated and triggered for the current quasi-collision group, and a precise collision detection is performed to determine whether the collision detection time meets the second preset signal, thereby obtaining a second collision detection result, that is, the output precise collision detection result; wherein the precise collision detection is used to perform traversal collision detection on the voxelized three-dimensional model in the current quasi-collision group; the second preset signal is that the sum of the first detection time required for rough collision detection and the second detection time required for precise collision detection is not greater than the sampling frequency of the received sensor signal.

[0121] The step A3 is specifically as follows: If the first detection result indicates that a rough collision has occurred, and the candidate collision objects have the same collision voxel node, then the candidate collision objects having the same collision voxel node are added to the current quasi-collision object group, and a second collision detection instruction related to the current quasi-collision object group is generated; The second collision detection instruction is to voxelize all objects in the current quasi-collider group, and then traversally perform the intersection collision detection between the current quasi-colliders based on their respective voxels; if the second preset signal condition is met, the result of the precise collision detection is output; if not, a detection timeout warning is output. Perform precise collision detection between objects in the current quasi-collider group, that is, perform precise collision detection between voxels and between voxels and workspace.

[0122] The steps for voxelizing objects in the current quasi-collider group are: Calculate the axial bounding box of the voxel grid of each current quasi-collider in the current quasi-collider group using the formula: = min( ), = max( ); in, and Respectively represent the maximum and minimum values ​​of the axial bounding box; Set the resolution of the voxel grid of the current quasi-collider to N, and divide each dimension of the voxel grid of the current quasi-collider into equal , , For each voxel, the formula is: =( - )x / ; =( - )y / ; =( - )z / ; in, , , Represents the number of voxels in the X, Y, and Z directions respectively; Assign the triangles of each voxel grid to the corresponding voxels, where each triangle will intersect with some voxels, so it is necessary to check whether each voxel contains part of the voxel grid, expressed as: = floor(( - ) / ); in, is the vertex The position in voxel grid space, is the minimum value of the axial bounding box, is the size of the voxel; floor(.) represents the floor function.

[0123] Based on the Separating Axis Theorem (SAT), collision intersection detection is performed by calculating the bounding box of the triangle and the bounding box of the voxel to optimize the intersection detection algorithm between the triangle and the voxel. The formula is: d = max(0, max( , , ) - min( , , )); Where d represents the projection distance deviation between the triangle and the voxel on a given axis. , , Represents the vertex of the voxel, max( , , ) represents the maximum value of the projection range of the voxel on a given axis; , , Represents the three vertices of a triangle, min( , , ) represents the minimum value of the projection extent of the triangle onto a given axis.

[0124] For example, suppose the three vertices of a voxel are , , , the vertices of the triangle are , , , then the formula is: d = max(0, max( , , ) - min( , , )).

[0125] The intersection is determined by calculating the distance deviation when the triangle and the voxel intersect, that is, calculating the minimum overlap of the triangle and the voxel on all possible separation axes based on the separation axis theorem. If d is a positive number, it means there is a gap (i.e., distance) between the two; if it is a negative number or zero, the output is 0, indicating that the two overlap on this axis.

[0126] Then, an circumscribed voxel sphere is used to replace the voxel block for accurate traversal collision detection. When a cubic voxel unit (i.e., a voxel block) is used for representation, multiple parameters need to be stored, such as side length, vertex coordinates, and body center, etc.; when a spherical voxel unit is used to replace the cubic voxel, only the sphere center position and radius need to be stored, which can significantly reduce the computer memory usage and eliminate the axial problem in collision detection between cubes. The voxel block includes an axial bounding box.

[0127] like Figure 8 As shown in (a), when the cube voxel is parallel to the coordinate axis, the collision detection is relatively simple, using the same axial bounding box collision detection as before; when the object moves and turns, such as Figure 8 As shown in (b), the cube voxels are not parallel to the coordinate axes, and collision detection will be performed using the separating axis theorem, requiring a maximum of 15 calculations between the two cubes; Figure 8 As shown in (c), using spherical voxels will not cause axial problems.

[0128] The collision detection method using spherical voxels is as follows: For collision detection between entities, collision detection based on circumscribed voxel spheres is adopted, that is, the relationship between the distance between the two sphere centers and the sum of the two sphere radii is calculated to determine whether they intersect. If they intersect, the output is that the entities collide with each other; when the distance between the two sphere centers is greater than the sum of the two sphere radii, they do not intersect, otherwise, they intersect.

[0129] For the collision detection between the entity and the virtual body, the collision detection between the circumscribed voxel sphere and the axial bounding box is adopted. By calculating the center P of the circumscribed voxel sphere ( , , ) and the relationship between the distance and radius between the nearest point Q within the axial bounding box to determine whether there is a collision. If there is a collision, the output is that the entity and the virtual body collide; among them, the nearest point Q ( , , ) is obtained by the following formula: ; ; ; in, and Respectively represent the minimum and maximum values ​​of the axial bounding box in the X-axis direction; and Respectively represent the minimum and maximum values ​​of the axial bounding box in the Y-axis direction; and Respectively represent the minimum and maximum values ​​of the axial bounding box in the Z-axis direction; The relationship between the distance between the center P and point Q and the radius is calculated to determine whether there is a collision, such as Fig. 9 As shown in the figure, a human spherical voxel model is generated, and the solid model of the human body is replaced by a sphere set. If the distance between the center P and the point Q is greater than the radius, it means that they do not intersect; otherwise, they intersect.

[0130] For collision detection between virtual bodies, collision detection based on axial bounding boxes is used. If there is a collision, the virtual body collision is output.

[0131] It is particularly noted that the so-called entity refers to an object that has physical properties, can occupy space, and can collide and interact with other objects. In this embodiment, it refers to a three-dimensional model corresponding to the physical entity.

[0132] The so-called virtual bodies refer to those objects that do not have physical properties, do not occupy space, or whose collision detection is specially handled, such as trigger areas, high-altitude falling areas near edges, and movement path spaces of robotic arms.

[0133] In this embodiment, the circumscribed sphere used can completely surround the voxel block, so there will be no omissions on the grid surface. The surface details of the object can be retained to the greatest extent, because the circumscribed sphere is larger than the voxel block and can capture all parts of the voxel edge. Since the surface of the sphere is smoother than the voxel block, the surface of the object after voxelization using the circumscribed sphere will be smoother than the result of block voxelization, especially at low resolutions, it can reduce the "staircase effect" (i.e. pixelation).

[0134] In another preferred embodiment, since the generated object voxel grid itself contains a double-layer structure, that is, in addition to the surface of the object, voxels are also generated inside the object. Therefore, before performing accurate collision detection, a voxel hollowing method is used to reduce the number of traversal times of the object voxels to speed up the calculation speed during collision detection, specifically: When generating a number of spherical voxels of a set size for the surface of an object, it is assumed that the vertices of the object's cube voxel Send a ray r( )= + ,in, represents the ray direction vector, Indicates that the control ray moves along The degree of movement in direction; Indicates the starting point coordinates of the ray; Whenever the ray r( ) intersects with the triangle face of the voxel grid of the object itself, it is determined whether the intersection point is on the ray, where ray r( ) intersects with the triangular face under the following conditions: = ( - ) / ; in, represents the plane normal vector; if >0, the ray intersects with the triangle of the voxel grid, and the number of intersections between the ray and the voxel grid is counted; If the number of intersections between the ray and the voxel grid is odd or there is no intersection, the voxel is outside the grid; if there is a ray that intersects the grid with an even number of intersections, the voxel is inside the grid; Delete the voxels inside the mesh and keep only the voxels on the surface of the model.

[0135] In another preferred embodiment, the voxel ratio, i.e., the voxel resolution, is dynamically calculated according to the mutual distance between the objects to be detected and the size of the objects to be detected, so as to speed up the calculation speed during accurate collision detection, specifically: Calculate the centroid distance between objects A and B: ; in, represents the centroid distance between A and B, , denote the centroid of A and B respectively; distance The closer, the smaller the volume of the object to be measured, which means the higher the voxel resolution, then, use the following formula for comprehensive consideration: ; Indicates the dynamic voxel resolution between the objects A and B to be measured; Indicates the maximum voxel resolution. When the object is farther away and the object is large, a lower resolution is used. Indicates the maximum distance threshold that affects the resolution. Beyond this distance, the voxel resolution will no longer change. , represents the geometric mean volume of the objects A and B to be measured, which is used to comprehensively consider the size of the objects; The calibration parameter representing the set object volume determines the effect of the object size on the resolution; The weight factor representing the distance is used to control the effect of distance on voxel resolution; β is the size weight factor, which is used to control the effect of object size on voxel resolution.

[0136] In this embodiment, the output result types include: no collision, virtual body-virtual body collision (ie, collision between workspaces), real body-virtual body collision (ie, collision between a real body and a workspace), and real body-real body collision.

[0137] S4, according to the precise collision detection result, the time and space amount of the collision is calculated in combination with the generated dynamic workspace to obtain the final risk value.

[0138] Furthermore, based on the data results obtained from the calculation of the possibility of environmental hazards and voxelized precise and rapid collision detection, the ratio of the time and space required for the work to the available time and space is used to quantify the degree of workspace conflict to represent the contact opportunities between personnel, machinery and hazards. A comprehensive assessment is conducted, and the results are graded and input into the construction site establishment module of the digital twin for visual display.

[0139] When calculating the time-space quantity of the conflict, the time-space quantity of the conflict is calculated based on the generated dynamic workspace to quantify the chance of human and machine contact danger. The time-space quantity of the conflict refers to the spatial intersection and time intersection of the confirmed exposure, and then the severity is set according to the collision type. The setting is shown in Table 1. The value of the severity can also be set to other values ​​according to the actual situation, which is not limited here. Finally, the final risk level is calculated comprehensively based on the possibility of environmental hazards.

[0140] Table 1. Collision types and set severity

[0141] The calculation formula of the final risk value is: I=H T / ; Wherein, I is the final risk value; H is the set severity, which is set according to the precise collision detection result; T is the exposure duration of the calculated object; is the volume exposed by the object being calculated; is the volume of the working space of the object being calculated.

[0142] The final risk value I is graded and visualized in the digital twin scenario.

[0143] In summary, compared with the prior art, the present invention has the following beneficial effects: (1) The present invention can achieve an objective and accurate assessment of the risk of space-time conflict. The assessment method used is not only based on a single proximity for risk assessment, but also converts the proximity into the opportunity for human and mechanical contact danger, quantifies the amount of space-time conflict (i.e., the conflict impact of the object's time and workspace), and pays attention to other environmental factors. The method combines the weighted Heronian average operator with fuzzy reasoning to systematically analyze the correlation between risk factors, perform more scientific weighting, and avoid the impact of extreme environments on the assessment results. In addition, the present invention dynamically updates the size of the workspace required by the object according to the changes in environmental factors for collision detection, so that the more dangerous objects will receive earlier attention and corresponding conflict risk assessments can be performed.

[0144] (2) The present invention combines the voxel tree in the collision detection module, uses the axial bounding box and the spherical voxel model for secondary collision detection, and the collision detection is accurate. In the collision detection process of large-scale complex models such as construction sites, the collision detection calculation time can be kept below 60 milliseconds. At the same time, the time spent on secondary collision detection is monitored to ensure that the total time of the collision detection meets the sensor signal sampling frequency requirements, avoiding a series of problems such as delays and confusions in the overall system due to the calculation time of the collision detection not meeting the signal sampling frequency, resulting in inaccurate collision detection, thereby ensuring the correctness of collision detection at the construction site.

[0145] (3) In the process of generating the voxel tree, the present invention combines the congestion ratio. Compared with the traditional method of dividing the octave voxel tree to the maximum depth or dividing it according to the number of objects, the present invention reduces the computer running memory usage and is more suitable for large-scale and complex construction scenes.

[0146] (4) The present invention uses an circumscribed voxel sphere instead of a voxel cube, which reduces the computer memory usage and eliminates the axial problem in collision detection between cubes.

[0147] (5) The present invention adopts a voxel hollowing method and a method for dynamically calculating voxel ratios to reduce the number of traversals between voxels and speed up the calculation speed during accurate collision detection.

[0148] Embodiment 2 The second embodiment of the present invention further provides a construction site collision risk assessment device based on a digital twin scene, comprising: The digital twin construction safety scenario establishment module collects the motion data and status parameters of equipment and personnel on the construction site through IoT sensors and generates a digital twin in real time. In the virtual simulation platform, the digital twin is simulated and rendered in real time through the physical engine to ensure that it is synchronized with the physical entity.

[0149] The environmental hazard occurrence possibility calculation module, after screening out the collision evaluation risk factors, uses a system combining the weighted Heronian average operator and fuzzy reasoning to perform correlation analysis and weighting on the selected risk factor indicators, establishes an evaluation mechanism, and constructs a comprehensive evaluation model. The output evaluation result is the possibility of environmental risk occurrence; according to the evaluation results, the workspace is dynamically generated and updated, re-input into the digital twin construction safety scenario establishment module, and provides data support for the voxelized precise and rapid collision detection module.

[0150] The voxelized precise and fast collision detection module divides the entire scene into multiple voxel blocks based on the spatial information of the construction site and the distribution of objects. According to the preliminary geometric form of the object, an axial bounding box is generated for rough collision and precise collision detection. The rough collision detection performs a rapid intersection test between the bounding box and the bounding box of the voxel tree node to screen out potential candidate collision objects. By recording the labels of the voxel nodes where the collision occurs, a preliminary collision detection result is generated. The precise collision detection module further voxelizes the object, performs precise collision detection on each voxel block, and outputs a high-precision collision detection result. In the construction scene, the detection result can be used to analyze in real time whether there are safety hazards, such as potential collisions between equipment and people, and equipment and dangerous areas.

[0151] The comprehensive risk assessment module conducts a comprehensive assessment based on the data results obtained by the environmental hazard occurrence possibility calculation module and the voxelized precise and rapid collision detection module, and uses the ratio of the time and space required for work to the available time and space to quantify the degree of workspace conflict to represent the contact opportunities between personnel, machinery and hazards. The results are graded and input into the digital twin construction site establishment module for visual display.

[0152] The multi-source sensor fusion monitoring module uses high-precision RTK positioning technology combined with lidar, cameras and other sensors to obtain the location information of objects on site. Through fusion algorithms (such as Kalman filtering), the sensor data is denoised to improve the accuracy and stability of the monitoring data.

[0153] The communication interface module supports common wired communication protocols such as CAN bus and RS-232. CAN bus is widely used in industrial automation scenarios and is suitable for real-time status transmission of mechanical equipment such as vehicles and cranes. It has strong anti-interference ability and high stability. RS-232 can be used for simple serial communication, such as data reading of small sensors. It supports modern wireless communication technologies such as LoRa, Zigbee and 5G. Among them, LoRa is suitable for wide-area sensor data transmission in large construction sites due to its long distance and low power consumption; Zigbee is suitable for short-distance device-to-device communication, has low power consumption, and is suitable for scenarios with dense node deployment. 5G provides high bandwidth and low latency, which is suitable for complex operation scenarios with extremely high real-time requirements, such as remote monitoring and automatic control operations.

[0154] Data Storage Module The data storage module is mainly used to process, manage and store sensor data from the construction site, as well as to support the execution of voxelized collision detection algorithms. The server is equipped with multi-core CPUs, high-bandwidth memory and large-capacity storage devices to support large-scale data processing requirements in complex collision detection scenarios.

[0155] The central control server module is the core of the entire digital twin system, responsible for executing complex collision detection algorithms, managing data streams from multiple sensors, and coordinating the work of various modules. To meet the high computing needs under complex scenes on the construction site, the central server is equipped with a high-performance multi-core processor and GPU acceleration unit.

[0156] Embodiment 3 The third embodiment of the present invention also provides a construction site collision risk assessment device based on a digital twin scene, including a processor and a memory, wherein the memory stores a computer program, and the computer program can be executed by the processor to implement a construction site collision risk assessment method based on a digital twin scene as described above.

[0157] Embodiment 4 The fourth embodiment of the present invention also provides a computer-readable storage medium, on which computer-readable storage medium is stored computer-readable instructions. When the computer-readable instructions are executed by a processor of a device where the computer-readable storage medium is located, the voxelized collision detection method based on the digital twin mode as described above is implemented.

[0158] In the several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A construction site collision risk assessment method based on digital twin scenarios, characterized in that: include: S1, constructing a digital twin scene of the construction site, screening out collision risk assessment factors, and using a weighted Heronian average operator to perform correlation analysis on the collision risk assessment factors to obtain determined weights; S2, constructing the membership function of each collision risk assessment factor, performing fuzzy reasoning, establishing a construction site collision risk assessment mechanism, and outputting an assessment result; wherein the assessment result is the possibility of occurrence of a risk at the construction site; S3, dynamically generating and updating the workspace of the digital twin scene according to the evaluation results, reconstructing the digital twin scene, and performing voxelized collision detection in the digital twin scene to obtain accurate collision detection results; S4, according to the precise collision detection result, the time and space amount of the collision is calculated in combination with the generated dynamic workspace to obtain the final risk value.

2. A construction site collision risk assessment method based on a digital twin scenario according to claim 1, characterized in that ,The construction site digital twin scene is constructed through the following steps: Based on the safety accident database, the construction technology and safety technical requirements of the physical space in the construction process, standardize the safety operation of the construction process, analyze the logical flow between the process levels and work steps of the construction process, the safety technical requirements and potential safety risks associated with the work steps, establish a standardized model and virtual space corresponding to the physical entity of the construction process, and obtain the indicator data of the physical entity; Deploy IoT sensor devices on physical entities during the construction process to dynamically monitor and obtain index data of the physical entities, wherein the index data includes location information and motion information of the monitored physical objects; Using the indicator data obtained from monitoring as input, the standardized model corresponding to the physical entity of the construction process is simulated in real time, and the rendered model is imported into the virtual reality simulation engine. The physical engine is used to build a visual simulation model to obtain a digital twin scene of the construction site, so as to realize information interaction and data synchronization between the physical entity and the digital twin.

3. A construction site collision risk assessment method based on a digital twin scenario according to claim 1, characterized in that When screening collision risk assessment factors, the contribution of each risk factor to the occurrence of an accident is dynamically evaluated by integrating real-time data, historical data and the scores of each expert; The scoring formula for each collision risk assessment factor is: ； in, The current initial score for the collision risk assessment factor; The score is weighted by time; The historical impact magnitude of the collision risk assessment factors; is the time weighting factor, which is used to express the impact of the past on the current state; t is the time interval; The time weighting factor Calculated by the following formula: ; in, is the decay rate parameter; The score of each collision risk evaluation factor is updated through an adaptive learning model, and the formula of the adaptive learning model is: ； in, represents the score after adaptive adjustment; Indicates the score of the previous stage; ΔS indicates the difference between real-time data and historical data; Represents the learning rate parameter, which is used to control the adjustment speed of the risk assessment factor score. The calculation formula is: ）； in, is the initial learning rate, which is set according to actual needs; Used to measure the fluctuation of historical scores. When the fluctuation is large, the learning rate Reduce, fluctuation time, learning rate Increase; Used to measure the magnitude of the score change. When the change is large, the learning rate Reduce, change small, learning rate Increase; , They are adjustment coefficients, which are used to control the impact of historical fluctuations and score changes on the learning rate and are set according to actual needs; represents the standard deviation of historical scores; represents the historical score average; Definition of the score for the current stage; The weighted average method is used to integrate expert scores and real-time data scores. The formula is: ； in, It represents the result of integrating expert scoring and real-time data scoring; It indicates the expert score after frailty treatment; represents adaptive real-time data scoring; and The weights of expert scoring and adaptive real-time data scoring respectively.

4. A construction site collision risk assessment method based on a digital twin scenario according to claim 3, characterized in that ,According to the scoring results of the collision risk evaluation factors, the weighted Heronian average operator is used to evaluate the scoring of each expert and real-time data. The evaluation formula of the weighted Heronian average operator is as follows: ； in, represents the evaluation result of the weighted Heronian average operator; 、 、 、 Respectively indicate j, k, 、 The importance score of each collision assessment risk factor; , , Respectively represent the j, k, The weights of the collision assessment risk factors; p, q, r are the parameters of the Heronian average operator; represents the number of collision assessment risk factors; λ represents the standardization constant, and the calculation formula of λ is: ； The calculated G values ​​are sorted and the score with the highest G value is selected as the determination weight of the collision assessment risk factor. The corresponding weight The determination method is: 。 5. A construction site collision risk assessment method based on a digital twin scenario according to claim 4, characterized in that , the S2 is specifically: A risk assessment level is set for each collision assessment risk factor, and a membership function is defined to describe the membership of each collision assessment risk factor at different assessment levels; a fuzzy relationship matrix is ​​obtained according to the membership value of each collision assessment risk factor at the corresponding risk assessment level; a fuzzy comprehensive operation is performed according to the determined weight and the fuzzy relationship matrix to obtain a comprehensive risk assessment vector B; the formula of the comprehensive risk assessment vector B is: ； Wherein, W represents the determined weight, represents the determined weight of the nth collision assessment risk factor; R represents the fuzzy relationship matrix; represents the membership degree of the nth collision assessment risk factor; Indicates Level of risk assessment; Indicates the set risk assessment level; According to the comprehensive risk assessment vector B, the comprehensive risk occurrence probability score is obtained by weighted calculation using the centroid method: , The expression is: ； in, Indicates Level of risk assessment; express The comprehensive membership degree corresponding to the risk assessment level, ; It reflects the overall probability of risk occurrence. The larger the value, the greater the possibility of risk occurrence.

6. A construction site collision risk assessment method based on a digital twin scenario according to claim 5, characterized in that ,The steps of voxelized collision detection in S3 are: A1, the constructed digital twin scene is voxelized and segmented, and a voxel tree is recursively generated based on the set crowding ratio threshold. The spatial area of ​​each node of the voxel tree is divided axially, and the space of the digital twin scene is recursively divided into smaller subspaces for efficient query, analysis and rendering; A2, determining the current candidate collision object according to whether the received sensor signal satisfies the first preset signal; And for any current candidate collision object in the digital twin scene, a first collision detection instruction is generated and triggered, and a rough collision detection is performed on the current candidate collision object and the voxel tree to obtain a first detection result; wherein the first preset signal is a set time threshold for receiving monitoring data; A3, if the first detection result indicates that a rough collision of the current candidate collision object has occurred and has the same collision voxel node, the current candidate collision object is added to the current quasi-collision group; a second collision detection instruction is generated and triggered for the current quasi-collision group, and a precise collision detection is performed to determine whether the collision detection time meets the second preset signal, thereby obtaining a second collision detection result, that is, the output precise collision detection result; wherein the precise collision detection is used to perform traversal collision detection on the voxelized three-dimensional model in the current quasi-collision group; the second preset signal is that the sum of the first detection time required for rough collision detection and the second detection time required for precise collision detection is not greater than the sampling frequency of the received sensor signal.

7. A construction site collision risk assessment method based on a digital twin scenario according to claim 6, characterized in that , the voxelized collision detection is further: In the process of generating the voxel tree, the root node of the voxel tree represents the cube area of ​​the entire digital twin scene. The step A1 is specifically as follows: Initialize the root node of the voxel tree to , expressed as: ； )； in, Represents the central coordinates of the digital twin scene, that is, the center point of the root node; is the side length of the digital twin scene, The value of is set to the maximum size of the digital twin scene; Points Values ​​in the X, Y, and Z coordinate directions; Recursively divide each node of the voxel tree into 8 child nodes, and dynamically complete the generation of the voxel tree according to the set maximum depth of the voxel tree and the set crowding ratio threshold to reduce the occupation of computer running memory; The crowding ratio is used to represent the degree of occupancy of objects in the pre-divided child nodes of the current voxel tree node, and is calculated by calculating the ratio of the occupied volume of the objects in the pre-divided next-level nodes to the volume of the pre-divided next-level nodes; The occupied volume of the object in the pre-divided next-level node is: the volume of the axial bounding box of the object intersecting with the pre-divided next-level node; Among them, the volume calculation formula of the object's axial bounding box is: ； Then, the volume calculation formula of the pre-divided next-level node is: = 8； in, is the volume of the object’s axial bounding box, The width of the axial bounding box AABB, that is, the size on the X axis; The height of the axial bounding box AABB, that is, the size on the Y axis; The depth of the axial bounding box AABB, that is, the size on the Z axis; is the volume of the pre-divided next level node; If the crowding ratio of the object in the next pre-divided node of the current node exceeds the set crowding ratio threshold, the division continues until the set maximum depth of the voxel tree is reached; nodes at different depths have different crowding ratio thresholds; The spatial area of ​​each node of the voxel tree is divided axially; for the nodes of the voxel tree , set the node The center point is , the side length is , then the next level child node It is expressed as: ； = +(± / 4，± / 4，± / 4); = / 2; Among them, ± determines the different directions of the child nodes. Representation Node The center point of Representation Node The side length of The entire digital twin scene is thus divided into multiple voxel blocks, each of which contains location information, size information and its hierarchical structure; the root node records the basic parameters of the voxel tree, which include center coordinates, voxel block size and hierarchical depth; The step A2 is specifically as follows: If the received sensor signal of the physical entity satisfies the first preset signal, the digital twin corresponding to the physical entity is set as the current candidate collision object and added to the candidate collision object group; then a first collision detection instruction related to the current candidate collision object is generated to perform a rough collision detection between the working space and the voxel space of the current candidate collision object; If the received sensor signal of the physical entity does not satisfy the first preset signal, outputting a warning; The first collision detection instruction is to generate an axis-aligned axial bounding box for each component of the current candidate collision object, that is, the working space of the current candidate collision object, and then perform intersection collision detection based on the axial bounding box with the bounding box of each node of the voxel tree, that is, the voxel space; wherein each face of the axial bounding box is parallel to the coordinate axis of the world coordinate system, and the enclosing range of the axial bounding box is defined by the minimum point and the maximum point of the object itself; Suppose the set of all vertices of the current candidate collision object is V={ , ,..., }, each vertex The coordinates in three-dimensional space are ; Then, the minimum point of the axial bounding box is Determined by the minimum value of all vertices of the current candidate collision object in the X, Y, and Z directions: =(min( ),my( ),my( )); Where i=1,2,...,n, n is the number of vertices of the current candidate collision object; Respectively Values ​​in the X, Y, and Z directions; The maximum point of the axial bounding box Determined by the maximum value of the X, Y, and Z directions of all vertices of the current candidate collision object: =(max( ),max( ),max( )); The center point C and size S of the axial bounding box are calculated by the minimum point and the maximum point: C=( + ） / 2； S= - ； Among them, C is the center point of the axial bounding box, and S is the size vector of the axial bounding box, that is, the length, width and height; For each candidate collider in the candidate collider group, traverse each node of the voxel tree and perform intersection collision detection based on the axial bounding box to filter out objects that are unlikely to collide; The intersection collision detection is performed by comparing the relative positions of the minimum and maximum points of the two axial bounding boxes, specifically: Assume that the two axial bounding boxes to be detected are BA and BB, and the intersection condition is: ≥ ， ≤ ； ≥ ， ≤ ； ≥ ， ≤ ； in, , , They are the maximum points in the X, Y, and Z directions of the axial bounding box BA respectively; , , They are the minimum points in the X, Y, and Z directions of the axial bounding box BA respectively; , , They are the maximum points in the X, Y, and Z directions of the axial bounding box BB respectively; , , They are the minimum points in the X, Y and Z directions of the axial bounding box BB respectively; If the above conditions are met in the three directions of X, Y, and Z, BA and BB intersect, otherwise they do not intersect; Traversing each node of the voxel tree until the deepest node of the voxel tree, storing the intersection collision information of each candidate collision object and the deepest voxel tree node, thereby obtaining a rough collision detection result, that is, a first detection result; wherein the first detection result includes whether a rough collision occurs and the collision voxel node information of the candidate collision object; The step A3 is specifically as follows: If the first detection result indicates that a rough collision has occurred, and the candidate collision objects have the same collision voxel node, then the candidate collision objects having the same collision voxel node are added to the current quasi-collision object group, and a second collision detection instruction related to the current quasi-collision object group is generated; The second collision detection instruction is to voxelize all objects in the current quasi-collider group, and then traversally perform intersection collision detection between the current quasi-colliders based on their respective voxels; The steps for voxelizing objects in the current quasi-collider group are: Calculate the axial bounding box of the voxel grid of each current quasi-collider in the current quasi-collider group using the formula: = min( ), = max( ); in, and Respectively represent the maximum and minimum values ​​of the axial bounding box; Set the resolution of the voxel grid of the current quasi-collider to N, and divide each dimension of the voxel grid of the current quasi-collider into equal , , For each voxel, the formula is: =( - )x / ; =( - )y / ； =( - )z / ； in, , , Represents the number of voxels in the X, Y, and Z directions respectively; Assign the triangles of each voxel grid to the corresponding voxels, where each triangle will intersect with some voxels, so it is necessary to check whether each voxel contains part of the voxel grid, expressed as: = floor(( - ) / ); in, is the vertex The position in voxel grid space, is the axial bounding box The minimum value of is the size of the voxel; floor(.) represents the floor function; Based on the Separating Axis Theorem, the intersection collision detection is performed by calculating the bounding box of the triangle and the bounding box of the voxel to optimize the intersection detection algorithm between the triangle and the voxel. The formula is: d = max(0, max( , , ) - min( , , )); Where d represents the projection distance deviation between the triangle and the voxel on a given axis. , , Represents the vertex of the voxel, max( , , ) represents the maximum value of the projection range of the voxel on a given axis; , , Represents the three vertices of a triangle, min( , , ) represents the minimum value of the projection range of the triangle on a given axis; Then, an circumscribed voxel sphere is used instead of a voxel block for accurate traversal collision detection. It is only necessary to store the center position and radius parameters of the circumscribed voxel sphere to reduce the computer memory usage and eliminate the axial problem of collision detection between voxel blocks. The voxel block includes an axial bounding box. The specific collision detection method is as follows: For the collision detection between entities, the collision detection between the circumscribed voxel spheres is adopted, that is, the relationship between the distance between the centers of the two spheres and the sum of the radii of the two spheres is calculated to determine whether they intersect. If they intersect, the collision between the entities is output; For the collision detection between the entity and the virtual body, the collision detection between the circumscribed voxel sphere and the axial bounding box is adopted. By calculating the center P of the circumscribed voxel sphere ( , , ) and the nearest point Q within the axial bounding box and the radius to determine whether there is a collision. If there is a collision, the output is that the entity and the virtual body collide; among them, the nearest point Q ( , , ) is obtained by the following formula: ； ； ； in, and Respectively represent the minimum and maximum values ​​of the axial bounding box in the X-axis direction; and Respectively represent the minimum and maximum values ​​of the axial bounding box in the Y-axis direction; and Respectively represent the minimum and maximum values ​​of the axial bounding box in the Z-axis direction; For collision detection between virtual bodies, intersection collision detection based on axial bounding boxes is used. If there is a collision, the virtual body collision is output.

8. A construction site collision risk assessment method based on a digital twin scenario according to claim 7, characterized in that , also includes: Since the generated object voxel grid itself contains a double-layer structure, that is, in addition to the surface of the object, voxels are also generated inside the object. Therefore, before performing accurate collision detection, the voxel hollowing method is used to reduce the number of traversal times of the object voxels to speed up the calculation speed during collision detection. Specifically: When generating a number of spherical voxels of a set size for the surface of an object, it is assumed that the vertices of the object's cube voxel Send a ray r( )= + ,in, represents the ray direction vector, Indicates that the control ray moves along The degree of movement in direction; Indicates the starting point coordinates of the ray; Whenever the ray r( ) intersects with the triangle face of the voxel grid of the object itself, it is determined whether the intersection point is on the ray, where ray r( ) intersects with the triangular face under the following conditions: ) / ； in, represents the plane normal vector; if >0, the ray intersects with the triangle of the voxel grid, and the number of intersections between the ray and the voxel grid is counted; If the number of intersections between the ray and the voxel grid is odd or there is no intersection, the voxel is outside the grid; if the number of intersections between the ray and the grid is even, the voxel is inside the grid; Delete the voxels inside the mesh and keep only the voxels on the surface of the model.

9. A construction site collision risk assessment method based on a digital twin scenario according to claim 7, characterized in that , also includes: dynamically calculating the voxel ratio, i.e., the voxel resolution, according to the mutual distance between the objects to be detected and the size of the objects to be detected, so as to speed up the calculation speed of accurate collision detection, specifically: Calculate the centroid distance between objects A and B: ； in, represents the centroid distance between A and B, , denote the centroid of A and B respectively; distance The closer, the smaller the volume of the object to be measured, which means the higher the voxel resolution, then, use the following formula for comprehensive consideration: ； Indicates the dynamic voxel resolution between the objects A and B to be measured; Indicates the maximum voxel resolution. When the object is farther away and the object is large, a lower resolution is used. Indicates the maximum distance threshold that affects the resolution. Beyond this distance, the voxel resolution will no longer change. , represents the geometric mean volume of the objects A and B to be measured, which is used to comprehensively consider the size of the objects; The calibration parameter representing the set object volume determines the effect of the object size on the resolution; The weight factor representing the distance is used to control the effect of distance on voxel resolution; β is the size weight factor, which is used to control the effect of object size on voxel resolution.

10. A construction site collision risk assessment method based on a digital twin scenario according to any one of claims 1 to 9, characterized in that , the calculation formula of the final risk value is: I=H T / ; Wherein, I is the final risk value; H is the set severity, which is set according to the precise collision detection result; T is the exposure duration of the calculated object; is the volume exposed by the object being calculated; is the volume of the working space of the object being calculated, It is the comprehensive risk occurrence probability score.

Images