Project quality intelligent evaluation system based on AI big data

Abstract

The invention discloses a project quality intelligent evaluation system based on AI big data, and relates to the technical field of project quality evaluation. The system comprises a data acquisition module, a data processing module, an AI analysis module, a judgment module, an application module and a feedback optimization module. The data acquisition module acquires project quality data through the multi-source sensor unit and the image acquisition unit; the data processing module performs data preprocessing and feature extraction on the project quality data; the AI analysis module generates a comprehensive quality score by adopting a multi-modal fusion neural network based on the data processed by the data processing module; the judgment module performs multi-level threshold comparison on the comprehensive quality score to obtain a project quality evaluation result; the application module displays a comprehensive quality score of the project and a project quality evaluation result; and the feedback optimization module dynamically updates the multi-level threshold value and the feature weight according to manual correction. According to the invention, automatic and multi-dimensional evaluation of the engineering quality is realized, and the accuracy and efficiency of evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering quality assessment, and specifically to an intelligent engineering quality assessment system based on AI big data. Background Art

[0002] In the field of engineering construction, engineering quality assessment is the core link to ensure structural safety. Traditional engineering quality assessment relies on manual inspections and single-point monitoring, and has the defects of one-sided data collection, subjective feature analysis, insufficient fusion of multi-source data, and poor real-time performance. Traditional engineering quality assessment can only obtain sensor data from a limited number of points, lacks global visual monitoring of structural surface defects, and image feature extraction relies on manual interpretation, without utilizing texture analysis technology. At the same time, sensor time series data and image data are processed independently, multimodal fusion is not achieved through deep learning models, and there is a lack of dynamic feedback mechanism. The present invention aims to solve the shortcomings of existing assessment methods in comprehensiveness, intelligence and adaptability through AI big data technology, and to improve the automation and accuracy of engineering quality assessment. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent engineering quality assessment system based on AI big data to solve the problems raised in the above background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0005] An intelligent engineering quality assessment system based on AI big data, including a data acquisition module, a data processing module, an AI analysis module, a judgment module and an application module;

[0006] The data acquisition module is used to acquire engineering quality data of the project;

[0007] The data processing module is used to perform data preprocessing on the engineering quality data, extract features from the preprocessed engineering quality data, obtain engineering quality feature data, and store the data in a database;

[0008] The AI ​​analysis module is used to input the project quality feature data and the pre-processed project quality data into the pre-trained AI model to generate a comprehensive quality score for the project;

[0009] The judgment module is used to construct a multi-level threshold for comprehensive quality score, compare the multi-level threshold for comprehensive quality score with the comprehensive quality score, and obtain the engineering quality assessment result of the project;

[0010] The application module is used to display the comprehensive quality score of the project and the project quality assessment results.

[0011] Preferably, the data acquisition module includes:

[0012] A multi-source sensor unit, used to collect multi-source sensor time series data of the project, wherein the multi-source sensor time series data includes structural stress, structural settlement and structural vibration frequency data;

[0013] An image acquisition unit, used to obtain images of the structural surface of the project through a drone or a camera;

[0014] The engineering quality data includes multi-source sensor time series data and structure surface images.

[0015] Preferably, in the data processing module, the process of preprocessing the engineering quality data is as follows:

[0016] The multi-source sensor time series data in the engineering quality data is processed for outliers using the 3σ principle, and outliers exceeding ±3 times the standard deviation of the mean are eliminated. The multi-source sensor time series data after outlier processing is then filled with missing values ​​using linear interpolation. The multi-source sensor time series data after missing value filling is then standardized to obtain data preprocessed multi-source sensor time series data. The linear interpolation method is a mathematical method that estimates intermediate values ​​by constructing a linear function using known data points, and the standardization process is Z-Score standardization.

[0017] The structural surface image in the engineering quality data is denoised by non-local mean denoising, and the denoised structural surface image is subjected to illumination equalization by contrast-limited adaptive histogram equalization to obtain a data-preprocessed structural surface image. The non-local mean denoising is a denoising algorithm based on non-local similarity of images, and the contrast-limited adaptive histogram equalization is an improved image contrast enhancement technology.

[0018] Based on the data preprocessed multi-source sensor time series data and the structure surface image, the data preprocessed engineering quality data is obtained.

[0019] Preferably, in the data processing module, the process of extracting features from the engineering quality data that has undergone data preprocessing to obtain engineering quality feature data is as follows:

[0020] The engineering quality feature data includes engineering image features and sensor timing features, and the engineering image features include length l, average width w, fractal dimension d f , contrast c and energy e, the sensor timing characteristics include mean μ i ,variance Frequency domain characteristics η (i) ;

[0021] For the structure surface image I∈R after data preprocessing H×W×3Generate a binary mask image I through the U-shaped network mask , the binary mask image I mask Through the thinning algorithm, a crack skeleton with a single pixel width is generated. skeleton , and count the crack skeletons with a single pixel width I skeleton The total number of non-zero pixels N piexl , for the total number of non-zero pixels N piexl The length l is calculated by a physical length formula. The U-shaped network is a convolutional neural network designed specifically for image segmentation tasks. The thinning algorithm is a morphological operation that reduces the foreground object in the binary image to a center line with a single pixel width.

[0022] The physical length formula is:

[0023] l=N piexl ×r;

[0024] Where r is the pixel resolution, l is the length, and N piexl is the total number of non-zero pixels;

[0025] For a crack skeleton with a single pixel width I skeleton Through the skeleton coordinate indexing method, the width sequence is obtained And the width sequence The average width w is calculated by the average width calculation formula. The skeleton coordinate indexing method is a method for obtaining a crack width sequence in image processing;

[0026] The average width calculation formula is:

[0027]

[0028] Where w is the average width, r is the pixel resolution, and N is the crack skeleton with a single pixel width. skeleton The total number of pixels on d i is the crack skeleton representing the width of a single pixel I skeleton The unilateral distance from the i-th point to the crack edge;

[0029] The binary mask image I is converted into mask Divide into square grids with a side length of s, and count the number of grids N(S) with a side length of s covering the crack. Gradually increase the side length s of the square grid, and count the number of grids N(S) with different side lengths of s. Based on the square grids with different side lengths s and the number of grids N(S), calculate the fractal dimension d using the fractal dimension formula. f ;

[0030] The fractal dimension formula is:

[0031]

[0032] Among them, d f is the fractal dimension, N(S) is the number of grid cells N(S) of a square grid with side length s, s is the side length;

[0033] For the structure surface image I∈R after data preprocessing H×W×3 The gray-level co-occurrence probability matrix P(i, j) is obtained by the GLCM algorithm. The gray-level co-occurrence probability matrix P(i, j) is calculated by the contrast characteristic formula and the energy characteristic formula respectively to obtain the contrast c and energy e. Based on this, [l, w, d f ,c,e]∈R 5 Engineering image features The GLCM algorithm is a classic statistical method for analyzing image texture features;

[0034] The contrast characteristic formula is:

[0035] c=∑ i，j |ij| 2 P(i,j);

[0036] Where c is the contrast, P(i,j) is the gray-level co-occurrence probability matrix, i and j are the row index and column index in the gray-level co-occurrence probability matrix respectively;

[0037] The energy characteristic formula is:

[0038] e=∑ i，j P(i,j) 2 ;

[0039] Where e is the energy, P(i,j) is the gray-level co-occurrence probability matrix, i and j are the row index and column index in the gray-level co-occurrence probability matrix respectively;

[0040] The multi-source sensor time series data X that has been preprocessed raw ∈R T×m , respectively calculated by the mean characteristic formula and variance characteristic formula, and the mean μ is obtained i and variance Simultaneously analyze the multi-source sensor time series data X raw ∈R T×m The frequency domain characteristics are calculated by the frequency domain characteristic formula. η (i) , based on which we get That is, sensor timing characteristics

[0041] The mean characteristic formula is:

[0042]

[0043] Among them, μ i is the mean value of the i-th sensor in the time window, T is the time step, m is the number of sensors, x ti is the measurement value of the i-th sensor at the t-th moment;

[0044] The variance characteristic formula is:

[0045]

[0046] in, is the variance of the i-sensor signal, T is the time step, m is the number of sensors, x ti is the measurement value of the i-th sensor at the t-th moment, μ i is the mean value of the i-th sensor in the time window;

[0047] The frequency domain characteristic formula is:

[0048]

[0049] in, η (i) is the frequency domain feature of the i sensor signal, FTT(x 1:T ,i) is the fast Fourier transform of the time series data of the i-th sensor.

[0050] Preferably, the pre-trained AI model is a multimodal fusion neural network, including a CNN branch, an LSTM branch and a feature fusion layer.

[0051] Preferably, in the AI ​​analysis module, the process of inputting the engineering quality feature data and the engineering quality data that has undergone data preprocessing into a pre-trained AI model to generate a comprehensive quality score of the engineering is as follows:

[0052] The CNN branch extracts deep features from the pre-processed structure surface image through the pre-trained ResNet50 model with the fully connected layer removed, and obtains the deep semantic features f cnn , and combine the deep semantic features and engineering image features to obtain image fusion features The ResNet50 model is a classic residual neural network in deep learning;

[0053] The LSTM branch extracts deep features from the pre-processed multi-source sensor time series data through the long short-term memory network LSTM to obtain the time series implicit feature h T And the time series implicit features and sensor time series features are spliced ​​together to obtain the time series fusion features The long short-term memory network LSTM is a classic variant of the recurrent neural network RNN;

[0054] Input the image fusion feature f img and the time series fusion feature f sensor into the feature fusion layer, which calculates the image fusion feature f img and the time series fusion feature f sensor through the feature-level fusion formula and outputs the comprehensive quality score S∈[0,100] of the project;

[0055] The feature-level fusion formula is as follows:

[0056] S = 100×σ(W s ·(W i f img +W t f sensor +b));

[0057] Where, W i is the image feature weight, W t is the time series feature weight, f img is the image fusion feature, f sensor is the time series fusion feature, σ() is the Sigmoid activation function, S is the comprehensive quality score of the project, b is the bias term, and W s is the weight matrix; <00​​​​​​​​​​​​​​​​​​​​​The comprehensive quality score S of the project is displayed in the form of a dashboard, and the project quality assessment results of the project are corresponded by color labels. If the project quality assessment result is excellent, the color label is green; if the project quality assessment result is medium, the color label is blue; if the project quality assessment result is poor, the color label is yellow; if the project quality assessment result is critical, the color label is red.

[0066] Preferably, an intelligent engineering quality assessment system based on AI big data includes, in addition to a data acquisition module, a data processing module, an AI analysis module, a judgment module, and an application module, a feedback optimization module for dynamically updating the multi-level threshold of the comprehensive quality score based on the manually corrected engineering quality assessment results, and automatically adjusting the image feature weight W of the feature-level fusion formula of the feature fusion layer according to the type of project. i and time series feature weight W t ;

[0067] The types of projects mentioned include bridges, high-rise buildings, tunnels, dams, transmission towers and nuclear power plants. i is 0.7, the W of high-rise buildings i is 0.4, the tunnel's W i is 0.6, the W of the dam i is 0.5, the W of the transmission tower i is 0.3, the W of the nuclear power plant i is 0.2, the W of the bridge t is 0.3, the W of high-rise buildings t is 0.6, the tunnel's W t is 0.4, the W of the dam t is 0.5, the W of the transmission tower t is 0.7, the W of the nuclear power plant t is 0.8.

[0068] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0069] 1. This invention realizes the deep fusion evaluation of multimodal data. By constructing a multimodal fusion neural network through ResNet50 and LSTM, the structural surface image features and the time series sensor data features are cascaded and integrated, breaking through the limitations of independent analysis of multi-source data in traditional methods, and significantly improving the accuracy and comprehensiveness of engineering quality assessment.

[0070] 2. The present invention implements a dynamic adaptive optimization mechanism. The feedback optimization module of the present invention can dynamically update the scoring threshold according to the manual correction results, and automatically adjust the image and time series feature weights for different project types, thereby solving the problem of insufficient adaptability of traditional evaluation systems and realizing intelligent calibration of model parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0072] Figure 1 It is a schematic diagram of the system function module flow of the present invention. DETAILED DESCRIPTION

[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0074] Examples, such as Figure 1 The system is an intelligent engineering quality assessment system based on AI big data, comprising a data acquisition module, a data processing module, an AI analysis module, a judgment module, an application module and a feedback optimization module, which work together to complete an intelligent assessment of the process quality of the project.

[0075] The data acquisition module is used to acquire engineering quality data of the project;

[0076] The data processing module is used to perform data preprocessing on the engineering quality data, extract features from the preprocessed engineering quality data, obtain engineering quality feature data, and store the data in a database;

[0077] The AI ​​analysis module is used to input the project quality feature data and the pre-processed project quality data into the pre-trained AI model to generate a comprehensive quality score for the project;

[0078] The judgment module is used to construct a multi-level threshold for comprehensive quality score, compare the multi-level threshold for comprehensive quality score with the comprehensive quality score, and obtain the engineering quality assessment result of the project;

[0079] The application module is used to display the comprehensive quality score of the project and the project quality assessment results;

[0080] The feedback optimization module is used to dynamically update the multi-level threshold of the comprehensive quality score based on the manually corrected project quality assessment results, and automatically adjust the image feature weight W of the feature-level fusion formula of the feature fusion layer according to the type of project. iand time series feature weight W t .

[0081] Furthermore, the working principle of the present invention is described below by way of examples:

[0082] Taking a cross-river bridge in a certain city as an example, a quality assessment was conducted on the crack damage and vibration response of the bridge structure. The system achieved real-time monitoring of the bridge's health status by deploying stress sensors, settlement monitors, and vibration accelerometers at key nodes of the bridge, namely multi-source sensor units, and drones that regularly take aerial images of the bridge deck and pier surfaces, namely image acquisition units.

[0083] The stress data, settlement data, and vibration frequency data of the bridge structure are collected through a multi-source sensor unit with a sampling frequency of 10 Hz. Time series data with a duration of 24 hours is generated daily. The T of the time series data is 86400 and the m is 3. The drone is used to vertically photograph the bridge deck at a height of 50 meters to obtain RGB images with a resolution of 2048×1536, that is, H×W RGB images, focusing on detecting the distribution of cracks on the bridge deck.

[0084] For sensor time series data X raw ∈R 86400×3 The 3σ principle is used to eliminate abnormal points in the stress data that exceed ±3 times the standard deviation of the mean, such as the instantaneous stress peak caused by vehicle overloading; the missing points in the settlement data caused by equipment failure are supplemented by linear interpolation. For example, if the settlement data of a certain period is missing for 20 minutes, it is linearly estimated by the mean of the previous and next moments; the vibration frequency data is Z-Score standardized to make the mean 0 and the standard deviation 1. 2048×1536×3 , that is, the structural surface image, uses the non-local mean denoising algorithm to eliminate the noise in the drone image, such as pixel anomalies caused by weather interference. The denoised RGB image is subjected to contrast-limited adaptive histogram equalization to enhance the contrast of the crack area and solve the influence of shadows on crack identification. The crack area of ​​the RGB image after data preprocessing is segmented by a U-shaped network to generate a binary mask image I mask , the binary mask image I mask The crack skeleton with a single pixel width is generated by Zhang-Suen thinning algorithm. skeleton , and count the crack skeletons with a single pixel width I skeleton The total number of non-zero pixels N piexl is 500, the total number of non-zero pixels N piexl The length l is calculated by the physical length formula to be 50 mm; for the crack skeleton with a single pixel width I skeleton Through the skeleton coordinate indexing method, the width sequence is obtained And the width sequence The average width w is calculated by the average width calculation formula to be 0.4 mm; the binary mask image I is converted into mask Divide into square grids with a side length of s, and count the number of grids N(S) with a side length of s covering the crack. Gradually increase the side length s of the square grid. When s = 2, 4, 6, 8, 32, count the number of grids N(S) of square grids with different side lengths s as [200, 120, 65, 35, 18]. Based on the square grids with different side lengths s and the number of grids N(S), the fractal dimension d is calculated using the fractal dimension formula. f is 1.72, based on which the engineering image features are obtained The grayscale co-occurrence probability matrix P(i, j) is obtained by the GLCM algorithm for the structure surface image after data preprocessing. The grayscale co-occurrence probability matrix P(i, j) is calculated by the contrast characteristic formula and the energy characteristic formula respectively, and the contrast c is 8.5 and the energy e is 0.15. The sensor time series data after data preprocessing is calculated by the mean characteristic formula, variance characteristic formula and frequency domain characteristic formula respectively, and the mean μ1 of the stress data is 15.2MPa and the variance is 0. is 0.8, is 5.5Hz, η (1) is 0.12; the mean μ2 of the settlement data is 0.3 mm, and the variance is 0.05, is 0.1Hz, η (2) is 0.0; the mean μ3 of the stress data is 0.02g, and the variance is 0.001, is 8.2Hz, η (3) is 0.0, based on which the sensor timing characteristics are obtained That is, sensor timing characteristics

[0085] The CNN branch extracts deep features from the pre-processed structure surface image through the pre-trained ResNet50 model with the fully connected layer removed, and obtains the deep semantic features f cnn , and concatenate the deep semantic features and engineering image features to obtain the image fusion feature f img ∈R 5+2048 The LSTM branch extracts deep features from the sensor time series data that has been preprocessed through the long short-term memory network LSTM to obtain the time series implicit feature h T , and concatenate the temporal implicit features and sensor temporal features to obtain the temporal fusion feature f sensor ∈R 12+128, the image feature weights and temporal feature weights are automatically set according to the bridge type, and the comprehensive quality score S of the output project is calculated by the feature-level fusion formula to be 88.

[0086] The overall quality score S of the project is 88, which corresponds to the "excellent" grade. The dashboard displays the score S as 88 in real time, and the color label is green.

[0087] If manual review reveals that cracks in a certain area have been missed, the system will automatically correct the assessment results and dynamically adjust the threshold of the crack detection model. At the same time, it will increase the image feature weight from 0.7 to 0.8, thereby increasing the proportion of image data in the assessment and improving the accuracy of subsequent detection of similar diseases.

[0088] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An intelligent engineering quality assessment system based on AI big data, characterized by: include: Data acquisition module, used to obtain engineering quality data of the project; The data processing module is used to perform data preprocessing on the engineering quality data, extract features from the preprocessed engineering quality data, obtain engineering quality feature data, and store it in the database; The AI ​​analysis module is used to input project quality feature data and pre-processed project quality data into a pre-trained AI model to generate a comprehensive quality score for the project; The judgment module is used to construct a multi-level threshold for the comprehensive quality score, compare the multi-level threshold for the comprehensive quality score with the comprehensive quality score, and obtain the engineering quality assessment result of the project; Application module, used to display the comprehensive quality score of the project and the project quality assessment results.

2. The intelligent engineering quality assessment system based on AI big data according to claim 1 is characterized in that: The data acquisition module includes: A multi-source sensor unit, used to collect multi-source sensor time series data of the project, wherein the multi-source sensor time series data includes structural stress, structural settlement and structural vibration frequency data; An image acquisition unit, used to obtain structural surface images of the project through a drone or camera; The engineering quality data includes multi-source sensor time series data and structure surface images.

3. The intelligent engineering quality assessment system based on AI big data according to claim 2 is characterized in that: In the data processing module, the process of preprocessing the engineering quality data is as follows: The multi-source sensor time series data in the engineering quality data is processed for outliers using the 3σ principle, and outliers exceeding ±3 times the standard deviation of the mean are eliminated. The multi-source sensor time series data processed for outliers are then filled with missing values ​​using linear interpolation. The multi-source sensor time series data after missing value filling is then standardized to obtain the multi-source sensor time series data after data preprocessing. The structural surface image in the engineering quality data is denoised by non-local mean denoising, and the denoised structural surface image is subjected to illumination equalization by contrast-limited adaptive histogram equalization to obtain a data-preprocessed structural surface image; Based on the pre-processed multi-source sensor time series data and the structure surface image, the pre-processed engineering quality data is obtained; The non-local mean denoising is a denoising algorithm based on non-local similarity of images; The limited contrast adaptive histogram equalization is an improved image contrast enhancement technology.

4. The intelligent engineering quality assessment system based on AI big data according to claim 3 is characterized in that: In the data processing module, the process of extracting features from the engineering quality data that has undergone data preprocessing to obtain engineering quality feature data is as follows: The engineering quality feature data includes engineering image features and sensor timing features, and the engineering image features include length l, average width w, fractal dimension d f , contrast c and energy e, the sensor timing characteristics include mean μ i ,variance Frequency domain characteristics η (i) ; The structural surface image I∈R after data preprocessing H×W×3 Generate a binary mask image I through the U-shaped network mask , the binary mask image I mask Through the thinning algorithm, a crack skeleton with a single pixel width is generated. skeleton , and count the crack skeletons with a single pixel width I skeleton The total number of non-zero pixels N piexl , for the total number of non-zero pixels N piexl Calculate the length l using the physical length formula; For a crack skeleton with a single pixel width I skeleton Through the skeleton coordinate indexing method, the width sequence is obtained And the width sequence The average width w is obtained by calculating the average width calculation formula; The binary mask image I is converted into mask Divide into square grids with a side length of s, and count the number of grids N(S) with a side length of s covering the crack. Gradually increase the side length s of the square grid, and count the number of grids N(S) with different side lengths of s. Based on the square grids with different side lengths s and the number of grids N(S), calculate the fractal dimension d using the fractal dimension formula. f ; The structural surface image I∈R after data preprocessing H×W×3 Through the GLCM algorithm, the gray-level co-occurrence probability matrix P(i, j) is obtained. The gray-level co-occurrence probability matrix P(i, j) is calculated by the contrast characteristic formula and the energy characteristic formula respectively to obtain the contrast c and energy e. Based on this, [l, w, d f ,c,e]∈R 5 Engineering image features The multi-source sensor time series data X that has been preprocessed raw ∈R T×m Calculate the mean characteristic formula and variance characteristic formula respectively to get the mean μ i and variance Simultaneously analyze the multi-source sensor time series data X raw ∈R T×m The frequency domain characteristics are calculated by the frequency domain characteristic formula. η (i) , based on which we get That is, sensor timing characteristics The U-type network is a convolutional neural network designed specifically for image segmentation tasks; The thinning algorithm is a morphological operation that reduces foreground objects in a binary image to a centerline of single pixel width; The skeleton coordinate indexing method is a method for obtaining crack width sequences in image processing; The GLCM algorithm is a classic statistical method for analyzing image texture features.

5. The intelligent engineering quality assessment system based on AI big data according to claim 1 is characterized in that: The pre-trained AI model is a multimodal fusion neural network, including a CNN branch, an LSTM branch and a feature fusion layer.

6. The intelligent engineering quality assessment system based on AI big data according to claim 5 is characterized in that: In the AI ​​analysis module, the project quality feature data and the pre-processed project quality data are input into the pre-trained AI model to generate a comprehensive quality score for the project: The CNN branch extracts deep features from the pre-processed structure surface image through the pre-trained ResNet50 model with the fully connected layer removed, and obtains the deep semantic features f cnn , and concatenate the deep semantic features and engineering image features to obtain the image fusion feature f img ; The LSTM branch extracts deep features from the pre-processed multi-source sensor time series data through the long short-term memory network LSTM to obtain the time series implicit feature h T , and concatenate the temporal implicit features and sensor temporal features to obtain the temporal fusion feature f sensor ; The image fusion feature f img and temporal fusion feature f sensor Input feature fusion layer, feature fusion layer fuses image features f img and temporal fusion feature f sensor The comprehensive quality score S∈[0,100] of the output project is calculated through the feature-level fusion formula; The ResNet50 model is a classic residual neural network in deep learning; The long short-term memory network LSTM is a classic variant of the recurrent neural network RNN.

7. The intelligent engineering quality assessment system based on AI big data according to claim 6 is characterized in that: In the judgment module, the process of obtaining the engineering quality evaluation result of the project: Compare the multi-level threshold of the comprehensive quality score with the comprehensive quality score S of the project. If 85 < S ≤ 100, the engineering quality evaluation result of the project is excellent; If 60 < S ≤ 85, the engineering quality evaluation result of the project is medium; If 40 < S ≤ 60, the engineering quality evaluation result of the project is poor; If 0 < S ≤ 40, the engineering quality evaluation result of the project is critical.

8. The intelligent engineering quality assessment system based on AI big data according to claim 7 is characterized in that: In the application module, the process of displaying the comprehensive quality score and the engineering quality evaluation result of the project: Display the comprehensive quality score S of the project in the form of a dashboard, and correspond the engineering quality evaluation result of the project through color labels. If the engineering quality evaluation result is excellent, the color label is green; if the engineering quality evaluation result is medium, the color label is blue; if the engineering quality evaluation result is poor, the color label is yellow; if the engineering quality evaluation result is critical, the color label is red.

9. The intelligent engineering quality assessment system based on AI big data according to claim 1 is characterized in that: It also includes a feedback optimization module for dynamically updating the multi-level threshold of the comprehensive quality score based on the manually corrected project quality assessment results, and automatically adjusting the image feature weight W of the feature-level fusion formula of the feature fusion layer according to the type of project. i and time series feature weight W t ; The types of projects mentioned include bridges, high-rise buildings, tunnels, dams, transmission towers and nuclear power plants. i is 0.7, the W of high-rise buildings i is 0.4, the tunnel's W i is 0.6, the W of the dam i is 0.5, the W of the transmission tower i is 0.3, the W of the nuclear power plant i is 0.2, the W of the bridge t is 0.3, the W of high-rise buildings t is 0.6, the tunnel's W t is 0.4, the W of the dam t is 0.5, the W of the transmission tower t is 0.7, the W of the nuclear power plant t is 0.8.

Images