Artificial intelligence-based construction engineering material detection method and system

Abstract

The present application relates to the technical field of building material detection, in particular to a building engineering material detection method and system based on artificial intelligence, in the present application, the crack edge is determined by boundary pixel extraction and gray difference, and the crack area is screened by the connection relationship, so that the morphological information of the crack is converted from scattered pixels to a structured boundary set, the integrity of crack identification and the accuracy of boundary positioning are significantly improved, the concrete load cycle number and the brick humidity sequence are called by taking the geometric feature parameter as the core, the upper and lower limit mapping table is established and the life interval parameter is output, the quantitative correlation between crack change and time effect is realized, the continuity and reliability of life prediction are improved, the detection result and batch data are directly corresponding, a closed-loop data chain from structure characteristics to batch judgment is formed, the quantitative correlation between crack change and time effect is realized, and the consistency of material management and the operability of traceability are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building material detection, and in particular to a building engineering material detection method and system based on artificial intelligence. BACKGROUND

[0002] Building material detection is specifically the field of building engineering material performance and quality detection, covering the testing and analysis of the physical properties, mechanical strength, stability and durability of commonly used materials such as concrete, steel, bricks, sand and composite building materials in building engineering, to ensure that the building safety and engineering design standards are met, and the development trend of the field has gradually shifted from traditional experimental detection to automated detection, intelligent discrimination and computer-aided analysis.

[0003] The building engineering material detection method based on artificial intelligence refers to using an artificial intelligence model to process data and recognize patterns of multiple performance indicators of building engineering materials, thereby replacing traditional experience-based judgment to achieve rapid, accurate and standardized detection of material quality, with the purpose of improving the efficiency and consistency of building engineering material detection, reducing operational errors, and achieving high-frequency batch detection in large-scale engineering projects. The method can shorten the detection period, reduce the detection cost, improve the traceability of detection data, and ensure the overall safety and reliability of the building structure.

[0004] The prior art has many shortcomings in actual operation. The detection process is mainly based on single test and scattered index recording, and the crack morphology and operating conditions lack a unified carrier, resulting in separation of image information, structure information and time sequence information, reliance on manual splicing for cross-stage comparison, deviation of recording caliber and judgment scale, life evaluation mainly based on section estimation and static parameter comparison, lack of explicit labeling and mapping relationship of upper and lower limits of the interval, frequent boundary sample attribution swing, low recheck consistency, lack of systematic expression of weight basis and threshold comparison, difficulty in supporting large-scale batch comparison of the same caliber, batch management mainly based on scattered account records, lack of index mapping and set merging of numbers and grades, broken traceability path, leading to increased time consumption and misjudgment correction cost of recheck, unstable life evaluation and grade, and inability to form consistent caliber conclusions for materials across sections and construction periods. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to provide a building engineering material detection method and system based on artificial intelligence.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: a building engineering material detection method based on artificial intelligence, comprising the following steps:

[0007] S1: Obtain building material surface image data through a sensor, read the gray values of concrete and steel bars, extract and determine edges through boundary pixels and gray differences, screen crack regions by comparing connectivity, record brick and gravel boundary point sets, and generate crack profile data;

[0008] S2: Based on the crack profile data, extract concrete and brick skeletons, record path endpoint positions and intersection point positions, extract steel path peak values and strike angles, count and integrate gravel intersection point sets, and establish crack geometric feature parameters;

[0009] S3: Based on the crack geometric feature parameters, call concrete load cycle numbers and brick humidity sequences, compare time sequences and crack length and width to establish a corresponding relationship, combine steel and gravel sequence matrices, record upper and lower limit boundaries and establish a mapping table, output life interval parameters, and obtain life distribution interval results;

[0010] S4: Based on the life distribution interval results, sort brick stability and gravel strength, divide them into layers according to numerical size and integrate them into a set, record the corresponding weight values of the set, and accumulate the weights of each layer to obtain an integral value. Then, compare the integral value with the interval critical value to output a comprehensive grade determination value;

[0011] S5: Based on the comprehensive grade determination value, classify and sort concrete and steel batch numbers, judge the eligibility of the numbers after comparing the grade boundary points, establish a brick and gravel batch index correspondence table, output a mapping set and complete grade classification, and construct a material grade classification result.

[0012] As a further scheme of the present application, the crack profile data specifically comprises a data set composed of concrete crack boundaries, steel crack boundaries, brick crack boundaries, and gravel crack boundaries. The crack geometric feature parameters include concrete skeleton endpoints, brick skeleton intersection points, steel path direction angles, and gravel intersection point numbers. The life distribution interval results specifically comprise a distribution set composed of concrete life intervals, brick life intervals, steel life intervals, and gravel life intervals. The comprehensive grade determination value includes brick grade values, gravel grade values, concrete grade values, and steel grade values. The material grade classification result specifically comprises a classification set composed of concrete batch classification, steel batch classification, brick batch classification, and gravel batch classification.

[0013] As a further scheme of the present application, the specific steps for generating the crack profile data are as follows:

[0014] Obtain building material surface image data through a sensor, read the gray values of concrete and steel bars and record coordinates, calculate the gray difference values of adjacent pixels to determine edges, check the continuity of boundary pixels and aggregate, and generate a boundary connected point set;

[0015] Based on the boundary connection point set, the region size and the shape proportion are judged, noise sets are removed, the boundary sequence is tracked to form a polyline contour and close the gap, adjacent contours are merged, and the brick and stone are labeled to generate crack contour data.

[0016] As a further scheme of the present application, the specific steps for establishing the crack geometric feature parameters are:

[0017] Based on the crack contour data, the concrete boundary is scanned and isolated line segments are removed, the connected main path is retained, the convolutional neural network is used for pixel skeleton tracking in the brick area and line extraction, the endpoint coordinates and intersection point coordinates are recorded, and the skeleton node data is generated;

[0018] Based on the skeleton node data, the steel path pixel gray scale is detected and the peak value position is identified, the corresponding coordinates are recorded, the straight line connection between the endpoints is calculated to obtain the path direction angle and establish the direction sequence, and the path direction data is generated;

[0019] Based on the path direction data, the number of sand and stone intersection points is counted and numbered uniformly with the endpoint set, integrated into an integrated structure, the direction sequence and endpoint number are jointly stored to form a multi-parameter set, and the crack geometric feature parameters are established.

[0020] As a further scheme of the present application, the specific process of the convolutional neural network is that first, the obtained crack contour data is input into the input layer of the convolutional neural network, the input tensor dimension is set as the image length, width and channel number, in the feature extraction stage, the edge features are extracted through multi-layer convolution operation, the image size is kept unchanged by zero padding, the gradient distribution is stabilized after each layer of convolution by connecting the batch normalization layer, the non-linear features are preserved by connecting the rectified linear unit activation function, the feature map is compressed and the crack main area is highlighted, in the deep feature aggregation stage, high-order structure information is extracted by using multi-layer convolution superposition, and the corresponding relationship between the shallow boundary features and the deep abstract features is preserved by using the jump connection method, in the decoding stage, the convolution features are deconvolved layer by layer to restore to the original size, the skeleton pixel probability map is output, the probability map is binarized according to the set threshold to form the skeleton path, and finally the lines in the skeleton path are detected and extracted, the pixel coordinates of the path endpoints and intersection points are recorded, and the skeleton node data is generated.

[0021] As a further scheme of the present application, the specific steps for obtaining the life distribution interval result are:

[0022] Based on the crack geometric feature parameters, the concrete load cycle number and brick humidity sequence are read and time index rows and columns are generated, the crack length and width values are compared and written into the corresponding row and column positions, the matrix is numbered and stored in segments, and the time corresponding matrix is generated.

[0023] Based on the time corresponding matrix, the reinforcement sequence and the gravel sequence are inserted column by column to generate a multi-dimensional matrix, the upper and lower limit boundaries are extracted and written into a mapping table, the boundary interval is calibrated and the parameters are summarized, the life interval parameters are output, and the life distribution interval result is obtained.

[0024] As a further scheme of the present application, the specific step of outputting the comprehensive grade determination value is:

[0025] Based on the life distribution interval result, the long short-term memory network is used to process time series data, the long-term dependence feature is extracted, the brick stability value is arranged in ascending order and the index position is recorded, the gravel strength value is aligned and then sorted synchronously, the same interval values are grouped into a set and a set number is generated, and the sorted set data is obtained.

[0026] Based on the sorted set data, the sum of each set value is calculated and the weight value is recorded, the weight value is accumulated item by item according to the set number sequence and an accumulated sequence is generated, the correspondence between the accumulated sequence and the set number is stored, and the accumulated score data is generated.

[0027] Based on the accumulated score data, the score value and the interval critical value are compared item by item and the difference position is recorded, the comparison result is classified and a grade number index table is established, the output grade determination sequence of the grade number is confirmed, and the comprehensive grade determination value is obtained.

[0028] As a further scheme of the present application, the long short-term memory network execution process is that the brick stability sequence and the gravel strength sequence arranged in time in the life distribution interval result are combined with the time index to form an input sequence, which is sent into a memory unit composed of an input gate, a forget gate and an output gate at each time step, the gating weight is calculated according to the current input and the previous time step hidden state at the time step, the memory unit content at the previous time step is selectively retained and cleared by the forget gate, the current time step information is written into the memory unit by the input gate to form an updated memory, and the current time step hidden state is generated based on the updated memory by the output gate, after the full sequence traversal is completed, the intermediate hidden state and the terminal hidden state are spliced in time sequence to generate time, and the time feature sequence used for constructing the sorted set is obtained by inputting into a linear mapping layer, the brick stability value is arranged in ascending order according to the time feature sequence and the original index position is recorded, the gravel strength value is aligned according to the same time index and sorted synchronously, the values in the same interval boundary are grouped into a set and a set number is generated, and the sorted set data is output.

[0029] As a further scheme of the present application, the specific step of constructing the material grade classification result is:

[0030] Based on the comprehensive grade judgment value, the concrete batch number and the steel bar batch number are extracted item by item, and the number sequence is established, the number and the grade demarcation point range are compared one by one, and the qualified and unqualified states are recorded and written into the corresponding number position, the sorting sequence is generated in the number order and state value, and the number classification data is generated;

[0031] Based on the number classification data, the brick batch number and the sandstone batch number are extracted and an index table is established, the mapping set is generated after comparing the index table and the number state, the unified mapping is formed by merging the set in turn, and the grade classification sequence is output, and the material grade classification result is constructed.

[0032] The building engineering material detection system based on artificial intelligence is used for executing the building engineering material detection method based on artificial intelligence, and the system comprises:

[0033] The crack recognition module: based on the surface image data collected by the sensor, the gray values of the concrete and the steel bar are read, the boundary area is determined through the pixel gray difference, the crack area is screened in combination with the connectivity, the brick and sandstone boundary points are recorded, and the crack contour data is generated;

[0034] The skeleton modeling module: based on the crack contour data, the skeleton lines are extracted by scanning the concrete and brick area, the path end points and intersection points are marked, the steel bar path peak value is extracted and the strike angle is calculated, the number of sandstone intersection points is counted, and the crack geometric feature parameters are established;

[0035] The life evaluation module: based on the crack geometric feature parameters, the concrete load cycle number and the brick humidity sequence are called, the crack length and width values are written into the matrix, the mapping table is established by combining the steel bar and sandstone sequence, the life interval parameters are output, and the life distribution interval result is obtained;

[0036] The grade judgment module: based on the life distribution interval result, the brick stability and the sandstone strength are sorted and layered, the set weight is recorded and accumulated to form the integral value, the integral value and the interval critical value are compared, and the comprehensive grade judgment value is output;

[0037] The batch classification module: based on the comprehensive grade judgment value, the concrete and steel bar batch numbers are classified and sorted, the qualified state is judged by comparing the grade demarcation point, the brick and sandstone batch index corresponding table is established, the mapping set is output, and the material grade classification result is constructed.

[0038] Compared with the prior art, the advantages and positive effects of the present application are that:

[0039] 1. In the present application, the crack edge is determined by boundary pixel extraction and gray difference, the crack area is screened by connected relationship, and the crack contour data is formed by recording the brick and sand boundary point set, so that the morphological information of the crack is converted from scattered pixels to structured boundary set, which significantly improves the integrity of crack recognition and the accuracy of boundary positioning;

[0040] 2. In the present application, the time index matrix is constructed by calling the concrete load cycle number and the brick humidity sequence with the geometric feature parameter as the core, the upper and lower limit mapping table is established and the life interval parameters are output, the quantitative correlation between crack change and time effect is realized, and the continuity and reliability of life prediction are improved;

[0041] 3. In the present application, the direct corresponding relationship between the detection result and the batch data is established to form a closed-loop data chain from structural characteristics to batch judgment, realize the quantitative correlation between crack change and time effect, improve the continuity and reliability of life prediction, and enhance the consistency of material management and the operability of traceability. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a schematic diagram of the working process of the present application;

[0043] Figure 2 is a system flowchart of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0045] Example 1

[0046] Please refer to Figure 1 The present application provides a technical scheme: a building engineering material detection method based on artificial intelligence, comprising the following steps:

[0047] S1: obtaining building material surface image data through a sensor, reading the gray value of concrete and steel, extracting and determining the edge through boundary pixel extraction and gray difference, screening the crack area by comparing the connected relationship, recording the brick and sand boundary point set, and generating crack contour data;

[0048] S2: based on the crack contour data, extracting the concrete and brick skeleton, recording the path endpoint position and intersection point position, extracting the steel path peak value and strike angle, counting the number of sandstone intersection points and integrating the set, and establishing the crack geometric feature parameter;

[0049] S3: Based on the crack geometry parameter, the concrete load cycle number and the brick humidity sequence are called to compare the time sequence and the crack length width to establish a corresponding relationship, the steel and sandstone sequence matrix are combined, the upper and lower limit boundaries are recorded and a mapping table is established, the life interval parameter is output, and the life distribution interval result is obtained;

[0050] S4: Based on the life distribution interval result, the brick stability and the sandstone strength are sorted and layered into a set according to the numerical size, the corresponding weight value of the set is recorded, the weight of each layer is added item by item to obtain the integral value, and then the integral value is compared with the interval critical value to output the comprehensive grade judgment value;

[0051] S5: Based on the comprehensive grade judgment value, the concrete and steel batch numbers are classified and sorted, the grade boundary points are compared to determine the eligibility of the numbers, a brick and sandstone batch index correspondence table is established, a mapping set is output and the grade classification is completed, and a material grade classification result is constructed.

[0052] The crack profile data is a data set composed of concrete crack boundaries, steel crack boundaries, brick crack boundaries and sandstone crack boundaries, the crack geometry parameter includes concrete skeleton endpoints, brick skeleton intersection points, steel path direction angles and sandstone intersection point numbers, the life distribution interval result is a distribution set composed of concrete life interval, brick life interval, steel life interval and sandstone life interval, the comprehensive grade judgment value includes brick grade value, sandstone grade value, concrete grade value and steel grade value, and the material grade classification result is a classification set composed of concrete batch classification, steel batch classification, brick batch classification and sandstone batch classification.

[0053] The specific steps for generating crack profile data are as follows:

[0054] The building material surface image data is obtained by the sensor, the concrete and steel gray values are read and the coordinates are recorded, the edge is determined by calculating the gray difference value of adjacent pixels, the boundary pixel continuity is checked and aggregated, and the boundary connected point set is generated;

[0055] Based on the boundary connected point set, the region size and shape ratio are judged and the noise set is removed, the boundary sequence is tracked to form a polyline profile and close the gap, the adjacent profiles are merged and the brick and sandstone are labeled, and the crack profile data is generated;

[0056] Based on the building material surface image data obtained by the sensor, the low threshold is set to 50 and the high threshold is set to 150, the image edge is extracted, the gray difference value of adjacent pixels is calculated to determine whether the pixel is a boundary point and the edge position, the connectivity of adjacent pixel points is judged, and the boundary connected point set is aggregated, the coordinates of the points are recorded as a data set, and the crack profile data is generated;

[0057] Based on the boundary connected point set, open operation and closed operation are performed, a 3x3 matrix is used to screen the region, the size and shape ratio of the region are judged, the noise point set is removed, the simplification error tolerance is set to 0.01, the boundary points are sequentially tracked to form a polyline contour, the tolerance is set to 0.05, the path gap is closed, and the adjacent crack contours are merged to label the bricks and gravel, and crack contour data is generated.

[0058] The specific steps for establishing the crack geometric feature parameters are:

[0059] Based on the crack contour data, the concrete boundary is scanned and isolated line segments are removed, the connected main path is retained, the convolutional neural network is used for pixel skeleton tracking in the brick area and line extraction, the endpoint coordinates and intersection point coordinates are recorded, and skeleton node data is generated;

[0060] Based on the skeleton node data, the pixel gray of the steel bar path is detected and the peak value position is identified, the corresponding coordinates are recorded, the straight line connectivity between the endpoints is calculated to obtain the path direction angle and establish the direction sequence, and path direction data is generated;

[0061] Based on the path direction data, the number of gravel intersection points is counted and numbered with the endpoint set, integrated into an integrated structure, the direction sequence and endpoint number are jointly stored and a multi-parameter set is formed, and the crack geometric feature parameters are established;

[0062] Based on the crack contour data, the concrete boundary is scanned using a 3x3 window, isolated line segments are removed, the connected main path is retained, the convolutional neural network is used for pixel skeleton tracking in the brick area, the convolution kernel size is set to 3x3 for feature extraction, and the pooling window size is set to 2x2 for maximum pooling. The extracted features are extracted from the feature map, the threshold is set to 0.5 to convert the pixels in the image to skeleton data, the endpoint coordinates and intersection point coordinates of each skeleton node are recorded, and skeleton node data is generated;

[0063] Based on the skeleton node data, the pixel gray value of the steel bar path is detected by gray scale analysis, the gray scale peak value in the path is identified, the corresponding coordinates are recorded, the connectivity between the path endpoints is calculated, the error tolerance is set to 0.01, the path direction angle is calculated, the direction range is set to -180° to 180°, the calculated angle data and path information are recorded and stored together, and path direction data is generated;

[0064] Based on path direction data, the number of intersections in the sand and gravel area is counted, and the minimum number of intersections is set to 5. The endpoint set and the intersection set are traversed, and the data are uniformly numbered and merged. The endpoints and intersections are uniformly numbered and a dataset is generated. The direction sequence and endpoint numbers are jointly stored and integrated into a unified multi-parameter set, which provides basic data for the subsequent generation of crack geometric feature parameters, thus forming crack geometric feature parameters.

[0065] The specific process of the convolutional neural network is as follows: First, the acquired crack contour data is input into the input layer of the convolutional neural network. The input tensor dimension is set to the number of image length, width, and channels. In the feature extraction stage, edge features are extracted through multi-layer convolution operations. Zero padding is used to keep the image size unchanged. After each layer of convolution, a batch normalization layer is connected to stabilize the gradient distribution. The rectified linear unit activation function is connected to retain non-linear features. The feature map is compressed and the main crack area is highlighted. In the deep feature aggregation stage, high-order structural information is extracted by multi-layer convolution superposition. The correspondence between shallow boundary features and deep abstract features is retained through skip connections. In the decoding stage, the convolution features are deconvolved layer by layer to restore the original size and output a skeleton pixel probability map. The probability map is binarized according to a set threshold to form a skeleton path. Finally, lines are detected and extracted in the skeleton path, and the pixel coordinates of the path endpoints and intersections are recorded to generate skeleton node data.

[0066] Convolutional neural networks, according to the formula:

[0067]

[0068] in: For nodes Confidence values ​​of skeleton nodes For nodes pixel skeleton probability at that location For nodes The mean probability within a 3×3 neighborhood. For nodes Euclidean distance to the crack boundary For the maximum allowable distance, For nodes Predicted principal orientation angle For nodes The direction angle in the path direction data. For pixel probability weights, The weight is the neighborhood mean. For boundary distance weights, Weights for directional consistency;

[0069] Execution process: For the skeleton node set, first obtain the pixel probability of each node. Then the probability mean value of the 3x3 neighborhood around the node is calculated The pixel distance of the node to the crack boundary is measured The maximum boundary value is normalized The direction angle is predicted by the structure tensor The direction angle is read from the path direction data The difference between the two is calculated and substituted into the cosine similarity The four parts are multiplied by the weight .

[0070] The specific steps for obtaining the life distribution interval result are as follows:

[0071] Based on the crack geometric feature parameters, the concrete load cycle number and brick humidity sequence are read and the time index row and column are generated. The crack length and width values are compared and written in the corresponding row and column positions. The matrix is numbered and stored in segments. The time corresponding matrix is generated.

[0072] Based on the time corresponding matrix, the steel sequence and sand sequence are inserted column by column to generate a multi-dimensional matrix. After extracting the upper and lower limit boundaries, they are written into the mapping table. The boundary interval is calibrated and the parameters are summarized. The life interval parameters are outputted, and the life distribution interval result is obtained.

[0073] Based on the crack geometric feature parameters, the concrete load cycle number and brick humidity sequence are read. The time step is set to 1 hour. The time index row and column are generated using a two-dimensional matrix structure. In each time step, the crack length value range 0 to 500 mm and the width value range 0 to 50 mm are compared. The corresponding values are written in the row and column positions of the matrix. The row data is numbered from 0001 to 9999. The matrix is stored in segments. The size of each segment is set to 1000 records. The segmented data is stored in independent data blocks. The time corresponding matrix is generated.

[0074] Based on the time corresponding matrix, the steel sequence and sand sequence are inserted column by column in the matrix. The steel sequence value range is set to 0 to 200 MPa, and the sand sequence value range is set to 0 to 80 MPa. After the insertion is completed, a multi-dimensional matrix is generated. The upper and lower limit boundary values are extracted by scanning the column data. The upper and lower limit boundaries are based on the minimum and maximum values. After the extraction is completed, the boundary values are written into the mapping table. The fields of the mapping table include time index number, crack length interval, crack width interval, steel value interval, and sand value interval. After the mapping table is generated, the interval calibration method is used. The tolerance threshold is set to 0.01 to adjust the upper and lower limit intervals to avoid boundary repetition and loss. The life interval parameters are generated by summarizing the parameters, and the life distribution interval result is outputted.

[0075] The specific steps for outputting the comprehensive grade determination value are as follows:

[0076] Based on the life distribution interval result, the long short-term memory network is used to process the time series data, the long-term dependence feature is extracted, the brick stability value is arranged in ascending order and the index position is recorded, the sandstone strength value is aligned and sorted synchronously, the same interval value is classified into a set and the set number is generated, and the sorted set data is obtained;

[0077] Based on the sorted set data, the sum of each set value is calculated and the weight value is recorded, the weight value is accumulated item by item according to the set number sequence, and the accumulated sequence is generated, the correspondence between the accumulated sequence and the set number is stored, and the accumulated score data is generated;

[0078] Based on the accumulated score data, the integral value is compared with the interval critical value item by item, and the difference value position is recorded, the comparison result is classified and the grade number index table is established, the output grade judgment sequence of the grade number is confirmed, and the comprehensive grade judgment value is obtained;

[0079] Based on the life distribution interval result, the long short-term memory network is used to process the time series data, the long-term dependence feature is extracted, the brick stability value is arranged in ascending order and the index position is recorded, the sandstone strength value is aligned and sorted synchronously, the same interval value is classified into a set and the set number is generated, and the sorted set data is obtained;

[0080] Based on the sorted set data, the weighted accumulation method is used, the weight value range is set to 0.10 to 1.00, the step is set to 0.10, the weight is assigned to the set according to the ascending order of the set number, the default weight 0.50 is used when the minimum sample size threshold is less than 5, the sum of each set value is calculated with four decimal places rounding accuracy, the weight value and the set number are recorded, the weight value is accumulated item by item according to the set number sequence to generate the accumulated sequence, the position index starts from 0, the correspondence between the accumulated sequence and the set number is stored as a key-value structure, the key is the set number and the value is the cumulative value, and the accumulated score data is generated;

[0081] Based on the accumulated score data, the interval critical value table is set to 0.25, 0.50, 0.75, 0.90, the difference tolerance is set to 0.05, the integral value is compared with the interval critical value item by item, the difference value position index is recorded, the integral value less than 0.25 is classified into grade number 1, the integral value between 0.25 and less than 0.50 is classified into grade number 2, the integral value between 0.50 and less than 0.75 is classified into grade number 3, the integral value between 0.75 and less than 0.90 is classified into grade number 4, the integral value not less than 0.90 is classified into grade number 5, the comparison result is classified and written into the grade number index table field including number index integral value difference position, the output grade judgment sequence is output, and the comprehensive grade judgment value is obtained.

[0082] The long short-term memory network performs a process of forming an input sequence by combining a brick stability sequence arranged in time and a sandstone strength sequence in the life distribution interval result together with a time index, sending the input sequence into a memory unit composed of an input gate, a forget gate and an output gate by time step, calculating a gating weight according to the current input and the hidden state of the last time step in the time step, selectively retaining and clearing the content of the memory unit of the last time step by using the forget gate, writing the information of the current time step into the memory unit by using the input gate to form an updated memory, and generating the hidden state of the current time step based on the updated memory by the output gate, splicing the intermediate hidden state and the terminal hidden state in time sequence to generate a time feature sequence for sorting set construction after completing the full sequence traversal, arranging the brick stability values in ascending order according to the time feature sequence and recording the original index position, aligning the sandstone strength values according to the same time index and synchronously sorting, grouping the values at the same interval boundary into a set and generating a set number, and outputting the sorted set data;

[0083] The long short-term memory network is according to the formula:

[0084]

[0085] Among them: is an independent weighted time score scalar, is a stability term weight coefficient, is a wet-carrying term weight coefficient, is a length term weight coefficient, is a width term weight coefficient, is a sequence length, is a time step index, is a time step brick stability value, is a stability lower bound value, is a stability upper bound value, is a wet-carrying ratio coefficient, is a time step load cycle number value, is a load upper bound value, is a time step humidity value, is a humidity upper bound value, is a time step crack length value, is a crack length upper bound value, is a time step crack width value, is a crack width upper bound value;

[0086] Execution process: assemble the life distribution interval results into a sequence according to time index, set the sequence length and index by time step extract and and and and five types of numerical values, through the stability boundary and and calculate on each get the stability normalized sequence, use and and and in turn after the ratio normalization to and linear proportioning to get the moisture carrying sequence, specify and and and respectively after the ratio normalization to get the length sequence and width sequence, take the time average of the above four sequences in the range of to , according to the weight coefficient and and and linear weighting and get as an independent score output, the determination of weight and proportioning can be selected in a grid way under the constraint premise, set step size and fixed as well as enumeration , under the given verification criteria to select the parameter combination, finally represented by .

[0087] The specific steps of the construction material grade classification result are:

[0088] Based on the comprehensive grade judgment value, extract the concrete batch number and steel bar batch number one by one and establish the number sequence, compare the number and grade demarcation point range one by one and record the qualified and unqualified state, and write into the corresponding number position, arrange the sorting sequence according to the number order and state value, generate the number classification data;

[0089] Based on the number classification data, extract the brick batch number and sandstone batch number and establish an index table, generate a mapping set after comparing the index table and the number state, merge the sets in turn to form a unified mapping and output the grade classification sequence, and construct the material grade classification result;

[0090] At the comprehensive grade determination value, the concrete batch number and the steel bar batch number are extracted item by item, the number sequence is established, the numbers are compared with the grade boundary point range one by one, the comparison standard is set to a numerical interval of 0 to 100, the interval boundary points are 20, 40, 60, 80, 100, when the determination value corresponding to the number is in 0 to less than 20, it is recorded as unqualified, in 20 to less than 40, it is recorded as qualified grade 1, in 40 to less than 60, it is recorded as qualified grade 2, in 60 to less than 80, it is recorded as qualified grade 3, in 80 to 100, it is recorded as qualified grade 4, the qualified state of each number is written into the number state table, and the sorting sequence is generated in ascending order of the number combined with the state value, and the number classification data is output;

[0091] Based on the number classification data, the brick batch number and the sandstone batch number are extracted item by item to establish an index table, the index table fields include batch number, state value, grade number, when the batch number state value is unqualified, it is recorded as grade 0, when the state value is qualified grade 1, it is recorded as grade 1, and so on to grade 4, a mapping set is generated using the index table, which contains four types of data: concrete number, steel number, brick number, and sandstone number, and multiple mapping sets are merged in ascending order of batch number to output the grade classification sequence and construct the material grade classification result.

[0092] Please refer to Figure 2 , the building engineering material detection system based on artificial intelligence, the system comprises:

[0093] Crack recognition module: based on the surface image data collected by the sensor, the gray values of the concrete and the steel bar are read, the boundary area is determined by the pixel gray difference, the crack area is screened combined with the connectivity, the brick and sandstone boundary points are recorded, and the crack contour data is generated;

[0094] Skeleton modeling module: based on the crack contour data, the skeleton lines are extracted by scanning the concrete and brick area, the path end points and intersection points are marked, the steel path peak value is extracted and the strike angle is calculated, the number of sandstone intersection points is counted, and the crack geometric feature parameters are established;

[0095] Life evaluation module: based on the crack geometric feature parameters, the concrete load cycle number and the brick humidity sequence are called, the crack length and width values are written into the matrix, the mapping table is established by combining the steel and sandstone sequence, the life interval parameters are output, and the life distribution interval result is obtained;

[0096] Grade determination module: based on the life distribution interval result, the brick stability and the sandstone strength are sorted and layered, the set weight is recorded and accumulated item by item to form the integral value, the integral value is compared with the interval critical value, and the comprehensive grade determination value is output;

[0097] Batch classification module: based on the comprehensive grade judgment value, the concrete and steel bar batch number is classified and sorted, the grade boundary point is compared to judge the qualified state, the brick and gravel batch index corresponding table is established, the mapping set is output, and the material grade classification result is constructed.

[0098] The above is only a preferred embodiment of the present application, and does not limit the present application in other forms. Any person skilled in the art can use the disclosed technical content to make changes or modifications as equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. A method for testing building materials based on artificial intelligence, characterized in that, Includes the following steps: S1: Acquire surface image data of building materials through sensors, read gray values ​​of concrete and steel bars, extract and determine edges by boundary pixels and gray value differences, compare connectivity to filter crack areas, record boundary point sets of bricks and sand and gravel, and generate crack contour data. S2: Based on the crack contour data, extract the concrete and brick skeleton, record the path endpoints and intersections, extract the peak value and orientation angle of the steel reinforcement path, count the number of sand and gravel intersections and integrate them into a set, and establish the crack geometric feature parameters. S3: Based on the crack geometric feature parameters, call the concrete load cycle number and brick humidity sequence, compare the time series with the crack length and width to establish a correspondence, combine the steel reinforcement and sand and gravel sequence matrix, record the upper and lower limit boundaries and establish a mapping table, output the life interval parameters, and obtain the life distribution interval result. S4: Based on the life distribution interval results, sort the brick stability and sand and gravel strength, classify them into sets according to their numerical values, record the corresponding weight values ​​of the sets, accumulate the weights of each layer to obtain the integral value, compare the integral value with the interval threshold value, and output the comprehensive level judgment value. S5: Based on the comprehensive grade judgment value, classify and sort the batch numbers of concrete and steel bars, judge the qualification of the number after comparing the grade boundary point, establish a batch index correspondence table of bricks and sand and gravel, output the mapping set and complete the grade classification, and construct the material grade classification result.

2. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The crack profile data specifically comprises a data set consisting of concrete crack boundaries, rebar crack boundaries, brick crack boundaries, and aggregate crack boundaries. The crack geometric feature parameters include concrete skeleton endpoints, brick skeleton intersections, rebar path direction angles, and the number of aggregate intersections. The lifespan distribution range results specifically comprise a distribution set consisting of concrete lifespan ranges, brick lifespan ranges, rebar lifespan ranges, and aggregate lifespan ranges. The comprehensive grade determination value includes brick grade values, aggregate grade values, concrete grade values, and rebar grade values. The material grade classification results specifically comprise a classification set consisting of concrete batch classification, rebar batch classification, brick batch classification, and aggregate batch classification.

3. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The specific steps for generating the crack profile data are as follows: The system acquires surface image data of building materials through sensors, reads the gray values ​​of concrete and steel bars and records their coordinates, calculates the gray value difference between adjacent pixels to determine the edge, checks the continuity of boundary pixels and aggregates them to generate a set of boundary connected points. Based on the set of boundary connected points, the size and shape ratio of the region are determined and noise sets are removed. The boundary sequence is traced to form a polyline profile and the gap is closed. After merging adjacent profiles, bricks and gravel are labeled to generate crack profile data.

4. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The specific steps for establishing the geometric characteristic parameters of the crack are as follows: Based on the crack contour data, the concrete boundary is scanned and isolated line segments are removed, while the connected main path is retained. A convolutional neural network is used to perform pixel skeleton tracking in the brick area and extract lines. The endpoint coordinates and intersection coordinates are recorded to generate skeleton node data. Based on the skeleton node data, the pixel grayscale of the rebar path is detected and the peak position is identified. The corresponding coordinates are recorded, the straight line connection between the endpoints is calculated to obtain the path direction angle and establish a direction sequence to generate path direction data. Based on the path direction data, the number of sand and gravel intersections is counted and uniformly numbered with the endpoint set, integrated into an integrated structure, and the direction sequence and endpoint number are jointly stored to form a multi-parameter set, establishing crack geometric feature parameters.

5. The method for detecting building materials based on artificial intelligence according to claim 4, characterized in that, The specific process of the convolutional neural network is as follows: First, the acquired crack contour data is input into the input layer of the convolutional neural network. The input tensor dimension is set to the number of image width and height channels. In the feature extraction stage, edge features are extracted through multi-layer convolution operations. Zero padding is used to keep the image size unchanged. After each layer of convolution, a batch normalization layer is connected to stabilize the gradient distribution. A rectified linear unit activation function is connected to retain nonlinear features. The feature map is compressed and the main crack area is highlighted. In the deep feature aggregation stage, high-order structural information is extracted by superimposing multiple layers of convolution. The correspondence between shallow boundary features and deep abstract features is retained through skip connections. In the decoding stage, the convolutional features are deconvolved layer by layer to restore the original size and output a skeleton pixel probability map. The probability map is binarized according to a set threshold to form a skeleton path. Finally, lines are detected and extracted in the skeleton path, and the pixel coordinates of the path endpoints and intersections are recorded to generate skeleton node data.

6. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The specific steps to obtain the lifetime distribution interval result are as follows: Based on the crack geometric feature parameters, the number of concrete load cycles and the brick humidity sequence are read and a time index row and column are generated. The crack length and width values ​​are compared and written into the corresponding row and column positions. The matrix is ​​numbered and stored in segments to generate a time-corresponding matrix. Based on the time correspondence matrix, the steel bar sequence and sand and gravel sequence are inserted column by column to generate a multi-dimensional matrix. After extracting the upper and lower limit boundaries, they are written into the mapping table. The boundary intervals are calibrated and the parameters are summarized. The life interval parameters are output to obtain the life distribution interval results.

7. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The specific steps for outputting the comprehensive rating value are as follows: Based on the lifespan distribution interval results, the time series data is processed using a long short-term memory network to extract long-term dependency features. The brick stability values ​​are arranged in ascending order and the index positions are recorded. The sand and gravel strength values ​​are aligned and then sorted synchronously. Values ​​in the same interval are grouped into a set and a set number is generated to obtain sorted set data. Based on the sorted set data, calculate the sum of values ​​for each set and record the weight values. Accumulate the weight values ​​item by item according to the set number order and generate an accumulation sequence. Store the correspondence between the accumulation sequence and the set number to generate accumulated integral data. Based on the accumulated integral data, the integral value is compared with the interval threshold value item by item and the position of the difference is recorded. The comparison results are classified and a level number index table is established. The output level judgment sequence of the level number is confirmed and the comprehensive level judgment value is obtained.

8. The method for detecting building materials based on artificial intelligence according to claim 7, characterized in that, The execution process of the Long Short-Term Memory (LSTM) network is as follows: The brick stability sequence and sand and gravel strength sequence, arranged by time from the lifetime distribution interval results, are combined with time indices to form an input sequence. This input sequence is fed into a memory unit consisting of an input gate, a forget gate, and an output gate, time-step by time. At each time step, the gating weights are calculated based on the current input and the hidden state of the previous time step. The forget gate selectively retains and clears the contents of the memory unit from the previous time step. The input gate writes the current time step information into the memory unit to form an updated memory. The output gate then generates the hidden state of the current time step based on the updated memory. After completing the full sequence traversal, the intermediate and final hidden states are concatenated in chronological order to generate a time feature sequence for constructing the sorted set. The brick stability values ​​are sorted in ascending order according to the time feature sequence, and the original index positions are recorded. The sand and gravel strength values ​​are aligned and sorted synchronously according to the same time index. Values ​​at the same interval boundary are grouped into a set and a set number is generated. The sorted set data is then output.

9. The method for detecting building materials based on artificial intelligence according to claim 1, characterized in that, The specific steps for constructing the material grade classification results are as follows: Based on the comprehensive grade judgment value, the batch numbers of concrete and steel reinforcement are extracted one by one and a number sequence is established. The number is compared with the grade boundary point range one by one and the qualified and unqualified status is recorded and written into the corresponding number position. The sorting sequence is generated by arranging the number order and status value to generate number classification data. Based on the numbering and classification data, the batch numbers of bricks and sand and gravel are extracted and an index table is established. After comparing the index table with the number status, a mapping set is generated. The sets are then merged to form a unified mapping and output a grade classification sequence to construct the material grade classification result.

10. An artificial intelligence-based building materials testing system, characterized in that, The method for detecting building materials based on artificial intelligence according to any one of claims 1-9, wherein the system comprises: Crack identification module: Based on surface image data collected by sensors, read the gray values ​​of concrete and steel bars, determine the boundary area by pixel gray value difference, filter crack areas by connectivity, record the boundary points of bricks and sand and gravel, and generate crack contour data. Skeleton modeling module: Based on the crack contour data, scan the concrete and brick areas to extract skeleton lines, mark path endpoints and intersections, extract the peak value of the steel reinforcement path and calculate the direction angle, count the number of sand and gravel intersections, and establish crack geometric feature parameters. Life assessment module: Based on the crack geometric feature parameters, it calls the concrete load cycle number and brick humidity sequence, writes the crack length and width values ​​into a matrix, combines the steel reinforcement and sand and gravel sequences to establish a mapping table, outputs life interval parameters, and obtains the life distribution interval results; The grade determination module: Based on the life distribution interval results, the stability of bricks and the strength of sand and gravel are sorted and layered, the set weights are recorded and accumulated item by item to form an integral value, the integral value is compared with the interval critical value, and the comprehensive grade determination value is output. Batch classification module: Based on the comprehensive grade judgment value, classify and sort the batch numbers of concrete and steel bars, compare the grade boundary points to determine the qualified status, establish a batch index correspondence table of bricks and sand and gravel, output the mapping set, and construct the material grade classification result.

Images