An engineering quality inspection system and method based on multi-source sensor fusion
By using multi-source sensor fusion technology, the risk value and correction coefficient of engineering nodes are calculated in real time, which solves the problems of lag and fixed parameters in traditional detection methods, realizes the forward prediction and dynamic optimization of engineering quality, and improves the controllability and accuracy of the construction process.
Patent Information
- Application Number
- CN202611367118.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-04
- Publication Date
- 2026-11-17
AI Technical Summary
Traditional engineering quality inspection relies on manual on-site acceptance, which is time-consuming and makes it difficult to capture quality deviations during construction. It also makes it impossible to make forward-looking predictions, and the benchmark parameters of the inspection model cannot be optimized autonomously, resulting in rework and increased costs.
The engineering quality inspection method based on multi-source sensor fusion collects engineering data through multi-source sensor devices, calculates the correlation coefficient and real-time risk value of engineering nodes, dynamically calibrates the predicted risk of subsequent nodes, establishes risk correction coefficients, and realizes risk prediction and control throughout the construction process.
It significantly improves the reliability and accuracy of engineering quality inspection, identifies high-risk processes in advance, reduces the probability of rework, ensures that the construction period and cost are controllable, and the inspection model is continuously optimized as engineering data accumulates.
Smart Images

Figure CN123117673A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engineering quality testing technology, specifically to an engineering quality testing system and method based on multi-source sensor fusion. Background Technology
[0002] In the field of engineering construction, full-process quality inspection is a core link in ensuring the acceptance of completed projects and controlling construction costs and schedules. Traditional engineering quality inspection relies heavily on manual on-site inspections after each work process node is completed. Data collection and quality judgment are significantly delayed, making it difficult to capture quality deviations during construction in a timely manner. Often, problems are only discovered after they have accumulated and become apparent, easily leading to rework, schedule delays, and increased construction costs. Most existing inspection solutions can only determine whether the quality of completed nodes is up to standard, and cannot proactively predict the acceptance status of subsequent unconstructed nodes and the overall project acceptance risks based on the actual quality level of the completed nodes. This makes it difficult to provide on-site management personnel with guidance for early intervention, resulting in insufficient proactive prevention capabilities for quality control. In addition, the benchmark parameters and judgment thresholds of most inspection models are fixed and cannot be autonomously optimized and iterated with the accumulation of engineering data. The accuracy of inspection and prediction cannot be continuously improved with the enrichment of engineering experience. Summary of the Invention
[0003] The purpose of this invention is to provide an engineering quality inspection system and method based on multi-source sensor fusion to solve the problems mentioned in the background art.
[0004] To address the aforementioned technical problems, this invention provides the following technical solution: an engineering quality inspection method based on multi-source sensor fusion, comprising the following steps: S1. Divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. S2. Based on the sample of engineering projects that failed historical acceptance, count the number of samples that failed acceptance at each engineering node, and calculate the correlation coefficient between the node and the project acceptance results. S3. Determine the basic risk coefficient by combining the correlation coefficient of the preceding unqualified nodes, calculate the real-time sensing risk coefficient based on the real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. S4. Based on the comprehensive quality risk value, determine and output the quality status of the current project node; S5. Calculate the risk correction coefficient based on the inspection results of the completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. S6. After the project is completed, the data of this project will be stored in the historical database, and the node correlation coefficient, risk benchmark and judgment threshold will be updated. Based on historical acceptance big data, the correlation degree of node risks is quantified. The maximum correlation coefficient of the preceding unqualified node is used as the basic risk base. Real-time sensor data is integrated through progressive risk superposition rules. The predicted risks of subsequent nodes are dynamically calibrated and predicted in advance through risk correction coefficients, thus controlling the acceptance risks of the entire construction process.
[0005] Furthermore, in step S1, after authorization, for any type of engineering project, engineering nodes are established for the engineering project. For any engineering node, engineering project data is collected based on multi-sensor devices, and the engineering project data is preprocessed and normalized to obtain the engineering node features of any engineering node. Here, the number of engineering nodes is X, and the number of engineering node features of any engineering node is Y. Historical engineering project records are reviewed to obtain the project acceptance records of the engineering projects. The project acceptance records include the project acceptance status of the first acceptance after the completion of the engineering project. The project acceptance status is either project acceptance passed or project acceptance failed. The project acceptance status is determined by the actual project acceptance situation and the project acceptance indicators. If the actual project acceptance situation meets the project acceptance indicators, the project acceptance is passed; otherwise, the project acceptance is failed.
[0006] Further, in step S2, the M project acceptance records that failed project acceptance are retrieved. For the M projects corresponding to the M project acceptance records, the node acceptance records of the project nodes are obtained. The node acceptance records include the node acceptance status of the first acceptance after the completion of the project node. The node acceptance status is either node acceptance passed or node acceptance failed. The node acceptance status is determined by the node acceptance status and node acceptance indicators. If the node acceptance status meets the node acceptance indicators, the node acceptance passes; otherwise, the node acceptance fails. Among the M projects, the number of projects that failed the x-th project node acceptance is Cx, and thus the number of projects that failed the X-th project node acceptance is {C1,C2,…,Cx,…,CX}, where x=1,2,…,X. Then, the correlation coefficient Kx of the x-th project node to the project acceptance records is obtained, where Kx is the ratio of Cx to M. Substituting x=1,2,…,X one by one, the correlation coefficients {K1,K2,…,Kx,…,KX} of the X-th project nodes to the project acceptance records are obtained.
[0007] Furthermore, in step S3, when the x-th project node is reached, the project node acceptance records of the previous x-1 project nodes are retrieved. For project nodes that failed acceptance in the x-1 project node acceptance records, the largest correlation coefficient corresponding to the project node is used as the basic risk coefficient D1. If x=1, then D1=0. Project data is collected in real time to obtain real-time project node features, which are {E1, E2, ..., Ey, ..., EY}, where Ey is the feature of the y-th real-time project node. Based on historical samples of similar project nodes that have passed acceptance, the risk coefficient is predicted to be... Obtain the normal baseline interval corresponding to each engineering node feature. The upper limit of the normal baseline interval for the y-th engineering node feature is Uy, and the lower limit is Ly. If Ey falls within the normal baseline interval, set the relative deviation Py of the y-th engineering node feature to 0; if Ey is greater than the upper limit of the normal baseline interval, set Py to (Ey-Uy) / (Uy-Ly); otherwise, set Py to (Ly-Ey) / (Uy-Ly). Use the relative deviation Py of the y-th engineering node feature as the feature risk value Sy of the y-th engineering node feature, and then obtain the real-time sensing risk coefficient D2 of the x-th engineering node. ; Where Wy is the relevant weight of the y-th engineering node feature to be calculated, and the number of samples Ny in which the y-th engineering node feature exceeds the normal baseline range among the Cx samples that fail the acceptance of the x-th engineering node is counted. The original importance coefficient Fy of the y-th engineering node feature is calculated, Fy=Ny / Cx. Substituting each value into y=1,2,…,Y, the original importance coefficients of the Y engineering node features are obtained, and then the relevant weight of the y-th engineering node feature is obtained. The relevant weight of the y-th engineering node feature is the ratio of Fy to the sum of the original importance coefficients of the Y engineering node features.
[0008] Furthermore, in step S4, the comprehensive quality risk value of the current x-th engineering node is compared with the preset node-level risk judgment threshold; if the comprehensive quality risk value is greater than or equal to the node-level risk judgment threshold, the current engineering node is judged to be unqualified; if the comprehensive quality risk value is less than the node-level risk judgment threshold, the current engineering node is judged to be qualified, and the quality judgment result of the current engineering node is output.
[0009] Furthermore, in step S5, if the current completed project node number x=1, then subsequent node prediction is not performed, and only the quality judgment result of the current first project node is output; otherwise, based on the detection result of the current x-th project node, the node acceptance status of the remaining Xx project nodes is predicted and the comprehensive project acceptance risk is calculated. The range of project nodes to be predicted is determined to be from the (x+1)th to the Xth project node, a total of Xx project nodes to be predicted. The number of the g-th project node to be predicted is denoted as g, where g=x+1,x+2,…,X. The correlation coefficient Kg of each project node to be predicted is obtained, and the correlation coefficient of the first x completed project nodes is calculated. The sum of the numbers J is used as the risk correction coefficient U, which is the ratio of the comprehensive quality risk value R of the current x-th project node to the sum of the correlation coefficients J of the previous x nodes. R = D1 + (1-D1)*D2 is used to limit the risk correction coefficient to fall within a preset range. For each historical project, the risk correction coefficient corresponding to the 2nd to Xth project nodes is calculated sequentially. All calculated risk correction coefficients are summarized to form a historical correction coefficient sample set. The historical correction coefficient sample set is sorted in ascending order of value, and the value corresponding to the 5th percentile is taken as the minimum value of the preset range, and the 95th percentile is taken as the minimum value. The corresponding value is taken as the maximum value of the preset interval. If the risk correction coefficient is less than the minimum value of the preset interval, U is set to the minimum value of the preset interval; if the risk correction coefficient is greater than the maximum value of the preset interval, U is set to the maximum value of the preset interval; otherwise, U remains unchanged, and the predicted risk value Vg of the g-th node to be predicted is obtained, Vg = U * Kg. The predicted risk value Vg of each node to be predicted is compared with the preset node acceptance risk threshold. If Vg is greater than or equal to the node acceptance risk threshold, the node to be predicted is predicted to fail acceptance; otherwise, the node to be predicted is predicted to pass acceptance. The predicted risk values of all nodes to be predicted are accumulated to obtain the remaining value. The cumulative predicted risk D3 of each node is used to obtain the total predicted risk value Rto for the entire project lifecycle, where Rto = R + D3. For each historical project that has passed acceptance, the final actual total risk value after the completion of all nodes of the project is calculated. The sample set is sorted in ascending order of values, and the value corresponding to the 95th percentile is taken as the project-level acceptance risk threshold. If Rto is greater than or equal to the project-level acceptance risk threshold, the project is predicted to fail final acceptance; otherwise, the project is predicted to pass final acceptance. For all nodes that are predicted to fail acceptance, key quality control prompts are output to remind project managers to focus on monitoring and maintenance.
[0010] Furthermore, in step S6, after the completion of each project, the project acceptance record of this project, the node acceptance record of each project node, and the corresponding collected project node feature data are stored in the historical project record database; the correlation coefficient of each project node, the normal benchmark interval and related weight of each node feature, the preset interval of the risk correction coefficient, and the project-level acceptance risk threshold are recalculated.
[0011] An engineering quality inspection system based on multi-source sensor fusion, the system comprising: a node division and data acquisition module, a risk correlation coefficient calculation module, a multi-source risk fusion calculation module, a node quality judgment output module, an acceptance risk prediction and early warning module, and a model parameter iterative update module; The node division and data acquisition module is used to divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. The risk correlation coefficient calculation module is used to count the number of samples that failed the acceptance of each project node based on the sample of historical projects that failed the acceptance, and to calculate the correlation coefficient between the node and the project acceptance result. The multi-source risk fusion calculation module is used to determine the basic risk coefficient by combining the correlation coefficient of the preceding unqualified nodes, calculate the real-time sensing risk coefficient based on the real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. The node quality assessment output module is used to determine and output the quality status of the current engineering node based on the comprehensive quality risk value. The acceptance risk prediction and early warning module is used to calculate the risk correction coefficient based on the inspection results of completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. The model parameter iterative update module is used to store the project data into the historical database after the project is completed, and update the node correlation coefficient, risk benchmark and judgment threshold.
[0012] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: On the one hand, based on historical engineering acceptance data, the correlation between each construction node and the final acceptance result of the project is quantified, and the value with the highest risk correlation among the previous unqualified nodes is selected as the basic risk base. Progressive risk fusion is carried out in combination with real-time multi-source sensor data. At the same time, based on the abnormal frequency of each sensor feature in the historical unqualified samples, the risk weight of each feature is objectively determined, avoiding the subjective bias of traditional schemes that rely on manual experience to set weights. This makes the risk assessment results more in line with the actual quality status of the project and significantly improves the reliability and accuracy of node quality detection. On the one hand, based on the actual inspection results of completed construction nodes, the historical baseline risk of subsequent unconstructed nodes is dynamically calibrated through a risk correction coefficient. This accurately predicts the acceptance status and cumulative risk of remaining nodes, resulting in an overall acceptance risk assessment for the entire project lifecycle. High-risk procedures are identified in advance, and targeted control measures are provided. Compared to traditional solutions that only assess the quality of completed nodes, this method moves the quality control checkpoint forward, shifting from post-construction acceptance to pre-construction prevention. This facilitates early intervention and rectification by management personnel, effectively reducing the probability of rework and ensuring controllable construction period and costs.
[0013] On the other hand, after the project is completed, the acceptance records, node quality data, and sensor feature data of this project are automatically stored in the historical database. Simultaneously, core parameters such as node risk correlation, feature benchmark intervals, and risk judgment thresholds are re-statistically updated, allowing the detection model to continuously self-optimize as project data accumulates. Compared to traditional detection schemes with fixed parameters, this method can continuously improve detection and prediction accuracy as engineering practice experience grows, adapting to the quality control needs of different types of engineering projects, and exhibiting superior adaptability and stability for long-term use. Attached Figure Description
[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of an engineering quality inspection system based on multi-source sensor fusion according to the present invention; Figure 2 This is a flowchart of an engineering quality inspection method based on multi-source sensor fusion according to the present invention. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0016] Please see Figure 1 and Figure 2 This invention provides a technical solution: an engineering quality inspection method based on multi-source sensor fusion, comprising the following steps: S1. Divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. S2. Based on the sample of engineering projects that failed historical acceptance, count the number of samples that failed acceptance at each engineering node, and calculate the correlation coefficient between the node and the project acceptance results. S3. Determine the basic risk coefficient by combining the correlation coefficient of the preceding unqualified nodes, calculate the real-time sensing risk coefficient based on the real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. S4. Based on the comprehensive quality risk value, determine and output the quality status of the current project node; S5. Calculate the risk correction coefficient based on the inspection results of the completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. S6. After the project is completed, the data of this project will be stored in the historical database, and the node correlation coefficient, risk benchmark and judgment threshold will be updated. Based on historical acceptance big data, the correlation degree of node risks is quantified. The maximum correlation coefficient of the preceding unqualified node is used as the basic risk base. Real-time sensor data is integrated through progressive risk superposition rules. The predicted risks of subsequent nodes are dynamically calibrated and predicted in advance through risk correction coefficients, thus controlling the acceptance risks of the entire construction process.
[0017] In step S1, after authorization, for any type of engineering project, engineering nodes are established for the engineering project. For any engineering node, engineering project data is collected based on multi-sensor devices. The engineering project data is preprocessed and normalized to obtain the engineering node features of any engineering node. The number of engineering nodes is X, and the number of engineering node features of any engineering node is Y. Historical engineering project records are reviewed to obtain the project acceptance records of the engineering projects. The project acceptance records include the project acceptance status of the first acceptance after the completion of the engineering project. The project acceptance status is either project acceptance passed or project acceptance failed. The project acceptance status is determined by the actual project acceptance situation and the project acceptance indicators. If the actual project acceptance situation meets the project acceptance indicators, the project acceptance is passed; otherwise, the project acceptance is failed.
[0018] In step S2, M project acceptance records that failed project acceptance are retrieved. For the M projects corresponding to the M project acceptance records, node acceptance records of the project nodes are obtained. The node acceptance records include the node acceptance status of the first acceptance after the completion of the project node. The node acceptance status is either node acceptance passed or node acceptance failed. The node acceptance status is determined by the node acceptance status and node acceptance indicators. If the node acceptance status meets the node acceptance indicators, the node acceptance passes; otherwise, the node acceptance fails. Among the M projects, the number of projects that failed the x-th project node acceptance is Cx. Then, the number of projects that failed the X-th project node acceptance is {C1,C2,…,Cx,…,CX}, where x=1,2,…,X. Then, the correlation coefficient Kx of the x-th project node to the project acceptance records is obtained. Kx is the ratio of Cx to M. Substituting x=1,2,…,X one by one, the correlation coefficients {K1,K2,…,Kx,…,KX} of the X-th project nodes to the project acceptance records are obtained.
[0019] In step S3, when the x-th project node is reached, the project node acceptance records of the previous x-1 project nodes are retrieved. For project nodes that failed acceptance in the x-1 project node acceptance records, the largest correlation coefficient corresponding to the project node is used as the basic risk coefficient D1. If x=1, then D1=0. Project data is collected in real time to obtain real-time project node features, which are {E1, E2, ..., Ey, ..., EY}, where Ey is the feature of the y-th real-time project node. Based on historical samples of similar project nodes that have passed acceptance, each... For each engineering node feature, the normal baseline interval is defined. The upper limit of the normal baseline interval for the y-th engineering node feature is Uy, and the lower limit is Ly. If Ey falls within the normal baseline interval, the relative deviation Py of the y-th engineering node feature is set to 0. If Ey is greater than the upper limit of the normal baseline interval, Py is set to (Ey-Uy) / (Uy-Ly). Otherwise, Py is set to (Ly-Ey) / (Uy-Ly). The relative deviation Py of the y-th engineering node feature is taken as the feature risk value Sy of the y-th engineering node feature, thus obtaining the real-time sensing risk coefficient D2 of the x-th engineering node. ; Where Wy is the relevant weight of the y-th engineering node feature to be calculated, and the number of samples Ny in which the y-th engineering node feature exceeds the normal baseline range among the Cx samples that fail the acceptance of the x-th engineering node is counted. The original importance coefficient Fy of the y-th engineering node feature is calculated, Fy=Ny / Cx. Substituting each value into y=1,2,…,Y, the original importance coefficients of the Y engineering node features are obtained, and then the relevant weight of the y-th engineering node feature is obtained. The relevant weight of the y-th engineering node feature is the ratio of Fy to the sum of the original importance coefficients of the Y engineering node features.
[0020] In step S4, the comprehensive quality risk value of the current x-th engineering node is compared with the preset node-level risk judgment threshold. If the comprehensive quality risk value is greater than or equal to the node-level risk judgment threshold, the current engineering node is determined to be unqualified. If the comprehensive quality risk value is less than the node-level risk judgment threshold, the current engineering node is determined to be qualified, and the quality judgment result of the current engineering node is output.
[0021] In step S5, if the current completed project node number x=1, then subsequent node prediction is not performed, and only the quality judgment result of the current first project node is output; otherwise, based on the detection result of the current x-th project node, the node acceptance status of the remaining Xx project nodes is predicted and the comprehensive project acceptance risk is calculated. The range of project nodes to be predicted is determined to be from the (x+1)th to the Xth project node, a total of Xx project nodes to be predicted. The number of the g-th project node to be predicted is denoted as g, where g=x+1,x+2,…,X. The correlation coefficient Kg of each project node to be predicted is obtained, and the sum of the correlation coefficients J of the first x completed project nodes is calculated. The ratio of the comprehensive quality risk value R of the current x-th project node to the sum of the correlation coefficients J of the previous x nodes is used as the risk correction coefficient U, where R = D1 + (1-D1)*D2. This sets the boundary of the risk correction coefficient to ensure it falls within a preset range. For each historical project, the risk correction coefficient corresponding to the 2nd to Xth project nodes is calculated sequentially. All calculated risk correction coefficients are summarized to form a historical correction coefficient sample set. This sample set is then sorted in ascending order of value. The value corresponding to the 5th percentile is taken as the minimum value of the preset range, and the value corresponding to the 95th percentile is taken as the minimum value. The value is set as the maximum value of the preset interval. If the risk correction coefficient is less than the minimum value of the preset interval, U is set to the minimum value of the preset interval; if the risk correction coefficient is greater than the maximum value of the preset interval, U is set to the maximum value of the preset interval; otherwise, U remains unchanged. This yields the predicted risk value Vg of the g-th node to be predicted, where Vg = U * Kg. The predicted risk value Vg of each node to be predicted is compared with the preset node acceptance risk threshold. If Vg is greater than or equal to the node acceptance risk threshold, the node to be predicted is predicted to fail acceptance; otherwise, the node to be predicted is predicted to pass acceptance. The predicted risk values of all nodes to be predicted are summed to obtain the remaining nodes. The cumulative predicted risk D3 is used to obtain the total predicted risk value Rto for the entire project lifecycle, Rto = R + D3. For each historical project that has passed acceptance, the final actual total risk value after the completion of all nodes of the project is calculated. The sample set is sorted in ascending order of values, and the value corresponding to the 95th percentile is taken as the project-level acceptance risk threshold. If Rto is greater than or equal to the project-level acceptance risk threshold, the project is predicted to fail the final acceptance; otherwise, the project is predicted to pass the final acceptance. For all nodes that are predicted to fail the acceptance, key quality control prompts are output to remind project managers to focus on monitoring and maintenance.
[0022] In step S6, after the completion of each project, the project acceptance record of this project, the node acceptance record of each project node, and the corresponding collected project node feature data are stored in the historical project record database; the correlation coefficient of each project node, the normal benchmark interval and related weight of each node feature, the preset interval of the risk correction coefficient, and the project-level acceptance risk threshold are recalculated.
[0023] An engineering quality inspection system based on multi-source sensor fusion, the system comprising: a node division and data acquisition module, a risk correlation coefficient calculation module, a multi-source risk fusion calculation module, a node quality judgment output module, an acceptance risk prediction and early warning module, and a model parameter iterative update module; The node division and data acquisition module is used to divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. The risk correlation coefficient calculation module is used to count the number of samples that failed the acceptance of each engineering node based on a sample of engineering projects that failed the acceptance in the past, and to calculate the correlation coefficient between the node and the project acceptance result. The multi-source risk fusion calculation module is used to determine the basic risk coefficient by combining the correlation coefficient of the preceding non-conforming nodes, calculate the real-time sensing risk coefficient based on real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. The node quality assessment output module is used to determine and output the quality status of the current engineering node based on the comprehensive quality risk value; The acceptance risk prediction and early warning module is used to calculate the risk correction coefficient based on the inspection results of completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. The model parameter iteration and update module is used to store the project data into the historical database after the project is completed, and update the node correlation coefficient, risk benchmark and judgment threshold.
[0024] Example 1: Taking a prefabricated residential building project as the application object, the project is divided into five engineering nodes: foundation construction, first-floor component assembly, second-floor component assembly, third-floor component assembly, and overall completion acceptance. Each engineering node is equipped with four types of multi-element sensors for stress, displacement, temperature, and verticality, corresponding to four engineering node characteristics. Historical project records of the same type are retrieved, including one hundred records of projects failing acceptance, used to statistically analyze node risk correlation parameters and judgment criteria. After system permission verification, the above five engineering nodes are established for this project. Construction data for each node is collected in real time through the four types of sensors. The raw data is denoised, missing value filled, and normalized to obtain standardized engineering node characteristics. Simultaneously, historical project acceptance records are retrieved. Each record includes the final acceptance status of the project and the acceptance status of each node. The acceptance status is divided into pass and fail, judged according to national and industry-specific engineering acceptance indicators. One hundred historical project records of failed acceptance are retrieved, and the number of projects failing acceptance at each engineering node is counted. Statistics show that 30 projects failed the basic construction node acceptance, 40 failed the first-floor assembly node, 35 failed the second-floor assembly node, 25 failed the third-floor assembly node, and 50 failed the overall completion acceptance node. The correlation coefficient between each node and the project acceptance results was calculated, i.e., the ratio of the number of unqualified items at each node to the total number of unqualified items. The correlation coefficients for the five nodes were 0.3, 0.4, 0.35, 0.25, and 0.5, respectively. Real-time quality inspection was conducted when the project reached the second engineering node (first-floor assembly node). First, the foundation risk coefficient was determined: since there was only one previous node (foundation construction), and in this example, the foundation construction node passed acceptance with no previous unqualified nodes, the foundation risk coefficient was set to 0.
[0025] Four sensor feature data points of the first-layer assembly node are collected in real time. Based on the normal baseline range pre-obtained from historical acceptance samples, the relative deviation of each feature is calculated. The calculated relative deviations of the four features are 0, 0.2, 0, and 0.15, respectively, and the feature risk values are consistent with the relative deviation values.
[0026] In the historical samples of failed first-level assembly node acceptance, the number of samples with each feature exceeding the baseline range was counted, and the original importance coefficients were calculated. The original importance coefficients for the four features were 0.2, 0.6, 0.3, and 0.5, respectively. After summing the original importance coefficients and normalizing them, the relevant weights of the four features were obtained as 0.125, 0.375, 0.1875, and 0.3125, respectively. The risk value of each feature was weighted and summed with its corresponding weight to obtain the real-time sensing risk coefficient, which was calculated as: 0.125×0 + 0.375×0.2 + 0.1875×0 + 0.3125×0.15 = 0.1219.
[0027] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
Claims
1. An engineering quality inspection method based on multi-source sensor fusion, characterized in that: The method includes the following steps: S1. Divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. S2. Based on the sample of engineering projects that failed historical acceptance, count the number of samples that failed acceptance at each engineering node, and calculate the correlation coefficient between the node and the project acceptance results. S3. Determine the basic risk coefficient by combining the correlation coefficient of the preceding unqualified nodes, calculate the real-time sensing risk coefficient based on the real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. S4. Based on the comprehensive quality risk value, determine and output the quality status of the current project node; S5. Calculate the risk correction coefficient based on the inspection results of the completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. S6. After the project is completed, the data of this project will be stored in the historical database, and the node correlation coefficient, risk benchmark and judgment threshold will be updated.
2. The engineering quality inspection method based on multi-source sensor fusion according to claim 1, characterized in that: Based on historical acceptance big data, the correlation degree of node risks is quantified. The maximum correlation coefficient of the preceding unqualified node is used as the basic risk base. Real-time sensor data is integrated through progressive risk superposition rules. The predicted risks of subsequent nodes are dynamically calibrated and predicted in advance through risk correction coefficients, thus controlling the acceptance risks of the entire construction process.
3. The engineering quality inspection method based on multi-source sensor fusion according to claim 2, characterized in that: In step S1, after authorization, for any type of engineering project, engineering nodes are established for the engineering project. For any engineering node, engineering project data is collected based on multi-sensor devices. The engineering project data is preprocessed and normalized to obtain the engineering node features of any engineering node. The number of engineering nodes is X, and the number of engineering node features of any engineering node is Y. Historical engineering project records are reviewed to obtain the project acceptance records of the engineering projects. The project acceptance records include the project acceptance status of the first acceptance after the completion of the engineering project. The project acceptance status is either project acceptance passed or project acceptance failed. The project acceptance status is determined by the actual project acceptance situation and the project acceptance indicators. If the actual project acceptance situation meets the project acceptance indicators, the project acceptance is passed; otherwise, the project acceptance is failed.
4. The engineering quality inspection method based on multi-source sensor fusion according to claim 3, characterized in that: In step S2, M project acceptance records that failed project acceptance are retrieved. For the M projects corresponding to the M project acceptance records, node acceptance records of the project nodes are obtained. The node acceptance records include the node acceptance status of the first acceptance after the completion of the project node. The node acceptance status is either node acceptance passed or node acceptance failed. The node acceptance status is determined by the node acceptance status and node acceptance indicators. If the node acceptance status meets the node acceptance indicators, the node acceptance passes; otherwise, the node acceptance fails. Among the M projects, the number of projects that failed the x-th project node acceptance is Cx. Then, the number of projects that failed the X-th project node acceptance is {C1,C2,…,Cx,…,CX}, where x=1,2,…,X. Then, the correlation coefficient Kx of the x-th project node to the project acceptance records is obtained. Kx is the ratio of Cx to M. Substituting x=1,2,…,X one by one, the correlation coefficients {K1,K2,…,Kx,…,KX} of the X-th project nodes to the project acceptance records are obtained.
5. The engineering quality inspection method based on multi-source sensor fusion according to claim 4, characterized in that: In step S3, when the x-th project node is reached, the project node acceptance records of the previous x-1 project nodes are retrieved. For project nodes that failed acceptance in the x-1 project node acceptance records, the largest correlation coefficient corresponding to the project node is used as the basic risk coefficient D1. If x=1, then D1=0. Project data is collected in real time to obtain real-time project node features, which are {E1, E2, ..., Ey, ..., EY}, where Ey is the feature of the y-th real-time project node. Based on historical samples of similar project nodes that have passed acceptance, each... For each engineering node feature, the normal baseline interval is defined. The upper limit of the normal baseline interval for the y-th engineering node feature is Uy, and the lower limit is Ly. If Ey falls within the normal baseline interval, the relative deviation Py of the y-th engineering node feature is set to 0. If Ey is greater than the upper limit of the normal baseline interval, Py is set to (Ey-Uy) / (Uy-Ly). Otherwise, Py is set to (Ly-Ey) / (Uy-Ly). The relative deviation Py of the y-th engineering node feature is taken as the feature risk value Sy of the y-th engineering node feature, thus obtaining the real-time sensing risk coefficient D2 of the x-th engineering node. ; Where Wy is the relevant weight of the y-th engineering node feature to be calculated, and the number of samples Ny in which the y-th engineering node feature exceeds the normal baseline range among the Cx samples that fail the acceptance of the x-th engineering node is counted. The original importance coefficient Fy of the y-th engineering node feature is calculated, Fy=Ny / Cx. Substituting each value into y=1,2,…,Y, the original importance coefficients of the Y engineering node features are obtained, and then the relevant weight of the y-th engineering node feature is obtained. The relevant weight of the y-th engineering node feature is the ratio of Fy to the sum of the original importance coefficients of the Y engineering node features.
6. The engineering quality inspection method based on multi-source sensor fusion according to claim 5, characterized in that: In step S4, the comprehensive quality risk value of the current x-th engineering node is compared with the preset node-level risk judgment threshold; If the overall quality risk value is greater than or equal to the node-level risk assessment threshold, then the quality of the current project node is determined to be unqualified. If the overall quality risk value is less than the node-level risk assessment threshold, the current project node is deemed to be of acceptable quality, and the quality assessment result of the current project node is output.
7. The engineering quality inspection method based on multi-source sensor fusion according to claim 6, characterized in that: In step S5, if the current completed project node number x=1, then the prediction of subsequent nodes is not performed, and only the quality judgment result of the current first project node is output; otherwise, based on the detection result of the current xth project node, the node acceptance status of the remaining Xx project nodes is predicted and the comprehensive project acceptance risk is calculated. The range of project nodes to be predicted is determined to be from the (x+1)th to the Xth project node, a total of Xx project nodes to be predicted. The number of the gth project node to be predicted is denoted as g, where g=x+1,x+2,…,X. The correlation coefficient Kg of each project node to be predicted is obtained. The sum of the correlation coefficients J of the previous x completed project nodes is calculated. The ratio of the comprehensive quality risk value R of the current xth project node to the sum of the correlation coefficients J of the previous x nodes is used as the risk correction coefficient U. R=D1+(1-D1)*D2. The risk correction coefficient is boundary-limited so that the risk correction coefficient falls within the preset interval. If the risk correction coefficient is less than the minimum value of the preset interval, U is the minimum value of the preset interval. If the risk correction coefficient is greater than the maximum value of the preset interval, then U is the maximum value of the preset interval; Otherwise, keep U unchanged, and obtain the predicted risk value Vg of the g-th node to be predicted, Vg=U*Kg. Compare the predicted risk value Vg of each node to be predicted with the preset node acceptance risk threshold. If Vg is greater than or equal to the node acceptance risk threshold, the node to be predicted is predicted to fail acceptance; otherwise, the node to be predicted is predicted to pass acceptance. Accumulate the predicted risk values of all nodes to be predicted to obtain the cumulative predicted risk D3 of the remaining nodes, and then obtain the total predicted risk value Rto for the entire project cycle, Rto=R+D3. Take the value corresponding to the 95th percentile as the project-level acceptance risk threshold. If Rto is greater than or equal to the project-level acceptance risk threshold, the project is predicted to fail final acceptance; otherwise, the project is predicted to pass final acceptance. For all nodes to be predicted to fail acceptance, output key quality control prompts to remind project managers to focus on monitoring and maintenance.
8. The engineering quality inspection method based on multi-source sensor fusion according to claim 7, characterized in that: In step S6, after the completion of each project, the project acceptance record of this project, the node acceptance record of each project node, and the corresponding collected project node feature data are stored in the historical project record database. The correlation coefficients of each engineering node, the normal baseline range and relevant weights of each node's characteristics, the preset range of risk correction coefficients, and the project-level acceptance risk thresholds were recalculated.
9. An engineering quality inspection system based on multi-source sensor fusion, wherein the system is applied to the engineering quality inspection method based on multi-source sensor fusion as described in any one of claims 1-8, characterized in that: The system includes: a node division and data acquisition module, a risk correlation coefficient calculation module, a multi-source risk fusion calculation module, a node quality judgment output module, an acceptance risk prediction and early warning module, and a model parameter iterative update module; The node division and data acquisition module is used to divide the target project into several engineering nodes, collect the engineering data of the nodes through multi-source sensing devices, obtain the engineering node characteristics of the nodes, and establish project acceptance records and node acceptance records. The risk correlation coefficient calculation module is used to count the number of samples that failed the acceptance of each project node based on the sample of historical projects that failed the acceptance, and to calculate the correlation coefficient between the node and the project acceptance result. The multi-source risk fusion calculation module is used to determine the basic risk coefficient by combining the correlation coefficient of the preceding unqualified nodes, calculate the real-time sensing risk coefficient based on the real-time sensing characteristics, and fuse them to obtain the comprehensive quality risk value of the current node. The node quality assessment output module is used to determine and output the quality status of the current engineering node based on the comprehensive quality risk value. The acceptance risk prediction and early warning module is used to calculate the risk correction coefficient based on the inspection results of completed nodes, predict the acceptance status and cumulative risk of the remaining nodes, obtain the total risk of the entire project cycle, and output the project acceptance prediction results and control prompts. The model parameter iterative update module is used to store the project data into the historical database after the project is completed, and update the node correlation coefficient, risk benchmark and judgment threshold.
