A concrete slump detection method and mixing pump truck based on deep learning

Abstract

The present invention discloses a concrete slump detection method and a mixing pump truck based on deep learning, including: S1: collecting historical concrete slump data and slump process data, and preprocessing; S2: extracting raw material characteristics and mix ratio characteristics, and calculating identification feature vectors to build a historical concrete slump database; S3: calculating the concrete slump sequence in the preprocessed slump process data; S4: extracting the initial concrete slump and concrete slump sequence characteristics, and calculating the first concrete slump; S5: collecting raw material data for this concrete slump detection, calculating its identification feature vector, and using the identification feature vector in the historical concrete slump database, performing a second correction on the first concrete slump to obtain a second concrete slump. The present invention can solve the problems of high uncertainty in traditional concrete detection methods and difficulty in coping with the impact of raw material fluctuations on concrete slump.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete detection, and in particular to a concrete slump detection method based on deep learning and a mixing pump truck. Background Art

[0002] As a key indicator of concrete construction and quality control, concrete slump is of great significance for ensuring construction quality and structural safety. The traditional method of testing concrete slump mainly relies on manual sampling, that is, using a concrete slump cone for on-site testing. This method manually puts concrete into a standard-sized concrete slump cone, pulls out the cone after compaction, and measures the height of the concrete mixture slump to evaluate the concrete slump.

[0003] However, this traditional detection method has the following shortcomings: First, the uncertainty of manual sampling detection of concrete slump is large. Due to the differences in the skill level of the operators, the test environment and the non-uniformity of the concrete itself, the concrete slump test results of the same batch of concrete may have large deviations, affecting the accurate assessment of the construction quality; second, the traditional method is difficult to effectively deal with the impact of raw material fluctuations on concrete slump. In the concrete production process, factors such as the quality, proportion and environmental conditions of raw materials may affect the concrete slump of concrete. When raw materials fluctuate, such as changes in cement quality and aggregate moisture content, the concrete slump of concrete will also change accordingly. However, due to its reliance on manual sampling and manual testing, the traditional detection method cannot accurately reflect such fluctuations, and thus cannot adjust production parameters in time to ensure the quality stability of concrete. Summary of the invention

[0004] In view of this, the present invention aims to provide a concrete slump detection method and a mixing pump truck based on deep learning, so as to solve the problems of high uncertainty in traditional concrete detection methods and difficulty in coping with the impact of raw material fluctuations on concrete slump.

[0005] A concrete slump detection method based on deep learning, comprising:

[0006] S1: Collect historical concrete slump data and slump process data, and perform preprocessing to obtain preprocessed historical concrete slump data and preprocessed slump process data. The specific steps include:

[0007] S11: Collect historical concrete slump data, each piece of historical concrete slump data includes concrete slump, cement strength, average aggregate particle size and mix ratio data; the mix ratio data includes the ratio of cement, water, sand and aggregate in concrete, expressed as weight ratio; the unit of concrete slump is millimeter, the unit of cement strength is MPa, and the unit of average aggregate particle size is millimeter;

[0008] S12: setting the frame rate of the video camera to 30 frames per second, and ensuring that the video camera is located directly in front of the concrete, that is, at the main view angle, to shoot the collapse process of the concrete and obtain collapse process data;

[0009] S13: remove missing values ​​from historical concrete slump data, use the 3σ principle to remove outliers, and then normalize to obtain preprocessed historical concrete slump data; perform contrast enhancement on the slump process data and split it into a series of continuous video frames to obtain preprocessed slump process data;

[0010] S2: Construct a weight gating mechanism, extract raw material characteristics and mix ratio characteristics, calculate identification feature vectors, and build a historical concrete slump database;

[0011] S3: extract the highest slump point of each video frame from the preprocessed slump process data, calculate the instantaneous concrete slump and splice them to obtain the concrete slump sequence;

[0012] S4: extracting the preliminary concrete slump and the concrete slump sequence characteristics, and performing a first correction on the preliminary concrete slump according to the concrete slump sequence characteristics to obtain a first concrete slump;

[0013] S5: collecting raw material data of this concrete slump detection, calculating its identification feature vector, and using the identification feature vector in the historical concrete slump database to perform a second correction on the first concrete slump to obtain a second concrete slump.

[0014] Furthermore, the S2 extracts raw material characteristics and mix ratio characteristics, calculates identification feature vectors, and constructs a historical concrete slump database, including:

[0015] Firstly, the cement strength, average aggregate particle size and concrete slump are spliced ​​to obtain preliminary raw material characteristics. The mix ratio data is spliced ​​with the concrete slump to obtain preliminary mix ratio characteristics. Secondly, a weighted gating mechanism is constructed to calculate the first mix ratio gate, the first raw material gate, the second mix ratio gate and the second raw material gate using the preliminary raw material characteristics and preliminary mix ratio characteristics, and the raw material characteristics and mix ratio characteristics are extracted. Thirdly, the identification feature vector is calculated based on the raw material characteristics and the mix ratio characteristics.

[0016] Furthermore, the S2 further includes:

[0017] S21: Combine cement strength, average aggregate particle size and concrete slump to obtain preliminary raw material characteristics , combine the mix ratio data with the concrete slump to obtain the preliminary mix ratio characteristics ;

[0018] S22: Construct a weight gating mechanism to extract raw material characteristics and mix ratio characteristics. The calculation method is:

[0019] ;

[0020] in, is the first mix ratio gate, For the first raw material gate, is the second mix ratio gate, For the second raw material door, are the first, second, third, and fourth weight matrices respectively, are the first, second, third and fourth bias vectors respectively, is the sigmoid function, is the mix ratio characteristic, For the raw material characteristics, is the vector dot product;

[0021] S23: Combine the raw material characteristics and the mix ratio characteristics to calculate the identification feature vector. The calculation method is:

[0022] ;

[0023] in, is the identification feature vector, For splicing operation;

[0024] The calculation process of steps S21, S22, and S23 is uniformly recorded as:

[0025] ;

[0026] in, Compute operations for identifying eigenvectors;

[0027] S24: All identification feature vectors, average aggregate particle size, cement strength, mix ratio data, and concrete slump are stored in a relational database one by one to obtain a historical concrete slump database.

[0028] In the present invention, step S23 concatenates the raw material characteristics and mix ratio characteristics processed by the weight gating mechanism to form an identification feature vector, which represents the characteristics of the historical case and is used in the subsequent second correction process of the concrete slump.

[0029] The core of step S2 is the weight gating mechanism, which can dynamically adjust the weights of raw material characteristics and mix ratio characteristics through two sets of gating signals (first mix ratio gate, first raw material gate, and second mix ratio gate, second raw material gate).

[0030] The gating signal combines the characteristics of raw materials and mix ratio data, through linear transformation of weight matrix and bias vector, and is activated by sigmoid function; the gating signal assigns a dynamic weight to each feature. These weights are not fixed, but are adjusted in real time according to the characteristics of the input data (such as the type of raw materials, the proportion of the mix ratio, etc.), so that the model can automatically adapt to the changes of different raw materials and mix ratios, thereby more accurately capturing their impact on the slump of concrete.

[0031] The weight gating mechanism plays a role in feature selection and suppression. By calculating the gating signal, the model can automatically identify which features are important for predicting concrete slump and give them greater weights. This feature selection and suppression capability helps the model focus more on those features that are truly valuable for prediction, thereby improving the accuracy of the prediction.

[0032] Compared with traditional methods, the advantage of the weighted gating mechanism is that it can select features based on the characteristics of the data, avoiding the subjectivity of manual feature selection; at the same time, through the dynamic adjustment of the gating signal, the model can have stronger generalization ability when facing different raw materials and mix proportions; it helps to more accurately calculate the identification feature vector and build the database in the subsequent steps, and provide high-quality data support for the subsequent slump correction.

[0033] Furthermore, S3 extracts the highest slump point of each video frame from the pre-processed slump process data, calculates the instantaneous concrete slump and splices them to obtain a concrete slump sequence, including:

[0034] S31: In the pre-processed collapse process data, for each video frame, concrete collapse area identification is performed, and the calculation method is:

[0035] ;

[0036] in, is the image of the concrete collapse area, For the cropping operation, For YOLOv5 model operation, is the collapse process data after preprocessing;

[0037] S32: According to the concrete collapse area image, the highest collapse point of each video frame is extracted, and the calculation method is:

[0038] ;

[0039] in, is the edge contour of concrete collapse. For OpenCV's findContours function, It is the Canny edge detection algorithm;

[0040] Extract the point with the largest ordinate in the concrete collapse edge contour line, and use this point as the highest collapse point of the video frame;

[0041] S33: For the highest collapse point of each video frame, the perspective projection inverse transformation formula is used to calculate the concrete collapse distance as the instantaneous concrete slump of the video frame, and the instantaneous concrete slump of each video frame is spliced ​​to obtain the concrete slump sequence .

[0042] The concrete slump sequence is obtained by extracting the highest slump point of each video frame and calculating the instantaneous concrete slump for splicing. This process accurately records the entire process of concrete slump from the beginning to the end, and can comprehensively and continuously reflect the slump changes of concrete. In the subsequent first correction process of concrete slump, the concrete slump sequence is used as the key input, which makes full use of the dynamic information in the concrete slump process and improves the accuracy of slump calculation.

[0043] Furthermore, the step S4 extracts the preliminary concrete slump and the concrete slump sequence characteristics, and performs a first correction on the preliminary concrete slump according to the concrete slump sequence characteristics to obtain a first concrete slump, including:

[0044] S41: take the last value in the concrete slump sequence as the preliminary concrete slump Distance;

[0045] S42: Extract the concrete slump sequence characteristics, the calculation method is:

[0046] ;

[0047] in, is the first concrete slump sequence characteristic, For the first long short-term memory network operation, is the second concrete slump sequence characteristic, For the second long short-term memory network operation, Operate for multi-layer perceptron;

[0048] S43: According to the concrete slump sequence characteristics, the preliminary concrete slump is corrected for the first time to obtain the first concrete slump, which is calculated as follows:

[0049] ;

[0050] in, First concrete slump, For multi-layer perceptron operations.

[0051] In the present invention, step S42 extracts the features of the concrete slump sequence respectively through two long short-term memory networks (LSTM); LSTM, as a special recurrent neural network, can capture the long-term dependencies in the sequence data; in this step, the first LSTM network focuses on extracting the changing trend of the slump velocity; the second LSTM network focuses on extracting the changes in the slump morphology, providing rich feature information for the subsequent first correction.

[0052] Furthermore, the S5 collects raw material data of this concrete slump detection, calculates its identification feature vector, and uses the identification feature vector in the historical concrete slump database to perform a second correction on the first concrete slump to obtain a second concrete slump, including:

[0053] S51: Collect the raw material data of this concrete slump test, including the cement strength, average aggregate particle size, mix ratio data, and the first concrete slump of this concrete slump test; then splice the cement strength, average aggregate particle size and concrete slump of this concrete slump test to obtain the raw material characteristics of this time , combine the mix ratio data of this test with the concrete slump to obtain the mix ratio characteristics of this test ;

[0054] S52: Calculate the identification feature vector of this concrete slump detection, and the calculation method is:

[0055] ;

[0056] in, is the identification feature vector of this concrete slump detection;

[0057] S53: Calculate the similarity between the identification feature vector of the current concrete slump detection and all identification feature vectors in the historical concrete slump database, and use the identification feature vector with the highest similarity as the modified identification feature vector. The calculation method is:

[0058] ;

[0059] in, is the similarity between the identification feature vector of this concrete slump detection and the i-th identification feature vector in the historical concrete slump database, i is the data index, is the cosine similarity calculation, is the i-th identification feature vector in the historical concrete slump database, To correct the identity eigenvector, To obtain Maximum operate;

[0060] S54: Perform a second correction on the first concrete slump using the correction identification feature vector to obtain a second concrete slump, which is calculated as follows:

[0061] ;

[0062] in, is the slump of the second concrete.

[0063] In the present invention, S5 finds the historical case most similar to this test by calculating the similarity between the identification feature vector of this concrete slump test and the identification feature vector in the historical database; the identification feature vector integrates the key information of raw material characteristics, mix ratio and slump, and is an important representation of concrete slump; cosine similarity, as a commonly used vector similarity measurement method, can accurately reflect the directional consistency of two vectors in multidimensional space; in concrete slump detection, since each dimension of the feature vector has different dimensions and value ranges, cosine similarity is more applicable than measurement methods such as Euclidean distance.

[0064] Most traditional methods only evaluate the slump based on the data of the current test, while step S53 makes full use of the large amount of data in the historical database to make a more accurate correction to the slump of the first concrete based on the most matching historical slump data; in addition, flexible processing of different raw materials, mix ratios and slumps is achieved through similarity calculation, which has stronger adaptability than traditional methods.

[0065] The present invention also discloses a mixing pump truck, comprising:

[0066] Vehicle body;

[0067] A chassis, disposed at the bottom of the vehicle body, for supporting the vehicle body;

[0068] A stirring device, arranged on the chassis, for stirring concrete;

[0069] An arm, arranged on the chassis and located at the rear of the vehicle body;

[0070] A pumping system, which is connected to the boom through a delivery pipeline and is used to pump concrete from the mixing device to a designated location and deliver it to a construction site through the boom;

[0071] a power device, disposed in the vehicle body, for outputting electric power; a power device, disposed in the vehicle body, for outputting electric power;

[0072] A driving motor, wherein the output end of the driving motor is drivingly connected to the stirring device to drive the stirring device to rotate and realize the mixing operation of the concrete;

[0073] A hydraulic device, the hydraulic device is connected to the pumping system and is used to provide power for the pumping system;

[0074] A calculation module, which extracts raw material characteristics and mix ratio characteristics according to historical concrete slump data and slump process data, calculates an identification feature vector, a concrete slump sequence feature and a first concrete slump, and uses the identification feature vector to perform a second correction on the first concrete slump to achieve a second concrete slump obtained by the concrete slump detection method based on deep learning according to the first embodiment of the present invention;

[0075] A control system is electrically connected to the power device and the drive motor to control the operating state of the drive motor.

[0076] It should be noted that, in the present invention, unless otherwise specified, the drive motor may be an integrated motor and reducer.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] (1) In order to solve the problems of high uncertainty in traditional concrete detection methods and difficulty in effectively dealing with the impact of raw material fluctuations on concrete slump, the present invention uses deep learning to combine raw material characteristics with the concrete slump process to correct the concrete slump, thereby improving the accuracy and stability of concrete slump detection.

[0079] (2) The present invention constructs a weight gating mechanism, combines the characteristics of raw materials and mix ratio data, assigns a dynamically adjusted weight to each feature, plays a role in feature selection and suppression, dynamically adjusts the weights of raw material features and mix ratio features, avoids the subjectivity of manual feature selection, and helps subsequent steps to more accurately calculate identification feature vectors and build databases, and provide high-quality data support for subsequent slump correction.

[0080] (3) The present invention constructs a historical concrete slump database, and by calculating the cosine similarity, finds the historical case that is most similar to the current test data, and uses the historical concrete slump to correct the current test result. In addition, the present invention not only considers the single-frame image after the concrete collapse process, but also captures and models the entire collapse process, comprehensively considers the changing characteristics of the collapse speed and collapse morphology, and realizes the comprehensive optimization of the concrete slump test results. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 A schematic diagram of the process of the concrete slump detection method based on deep learning provided by the present invention;

[0082] Figure 2 A schematic diagram of the algorithm flow for calculating the weight gating mechanism provided by the present invention. DETAILED DESCRIPTION

[0083] The present invention is further described below in conjunction with the accompanying drawings, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0084] Embodiment 1: A method for detecting concrete slump based on deep learning, such as Figure 1 As shown, the following steps are included:

[0085] S1: collect historical concrete slump data and slump process data, and perform preprocessing to obtain preprocessed historical concrete slump data and preprocessed slump process data;

[0086] S11: Collect historical concrete slump data, each piece of historical concrete slump data includes concrete slump, cement strength, average aggregate particle size and mix ratio data; the mix ratio data includes the ratio of cement, water, sand and aggregate in concrete, expressed as weight ratio; the unit of the concrete slump is millimeter, the unit of cement strength is MPa, and the unit of average aggregate particle size is millimeter;

[0087] S12: setting the frame rate of the video camera to 30 frames per second, and ensuring that the video camera is located directly in front of the concrete, that is, at the main view angle, to shoot the collapse process of the concrete and obtain collapse process data;

[0088] S13: Remove missing values ​​from the historical concrete slump data, use the 3σ principle to remove outliers, and then normalize to obtain the preprocessed historical concrete slump data; perform contrast enhancement on the slump process data and split it into a series of continuous video frames to obtain the preprocessed slump process data.

[0089] S2: Extract raw material characteristics and mix ratio characteristics, calculate identification feature vectors, and build a historical concrete slump database;

[0090] S21: Combine cement strength, average aggregate particle size and concrete slump to obtain preliminary raw material characteristics , combine the mix ratio data with the concrete slump to obtain the preliminary mix ratio characteristics ;

[0091] S22: Construct a weight gating mechanism to extract raw material characteristics and mix ratio characteristics. The calculation method is:

[0092] ;

[0093] in, is the first mix ratio gate, For the first raw material gate, is the second mix ratio gate, For the second raw material door, are the first, second, third, and fourth weight matrices respectively, are the first, second, third and fourth bias vectors respectively, is the sigmoid function, is the mix ratio characteristic, For the raw material characteristics, is the vector dot product;

[0094] The core of step S2 is the weight gating mechanism, which can dynamically adjust the weights of raw material characteristics and mix ratio characteristics through two sets of gating signals (first mix ratio gate, first raw material gate, and second mix ratio gate, second raw material gate).

[0095] The gating signal combines the characteristics of raw materials and mix ratio data, through linear transformation of weight matrix and bias vector, and is activated by sigmoid function; the gating signal assigns a dynamic weight to each feature. These weights are not fixed, but are adjusted in real time according to the characteristics of the input data (such as the type of raw materials, the proportion of the mix ratio, etc.), so that the model can automatically adapt to the changes of different raw materials and mix ratios, thereby more accurately capturing their impact on the slump of concrete.

[0096] The weight gating mechanism plays a role in feature selection and suppression. By calculating the gating signal, the model can automatically identify which features are important for predicting concrete slump and give them greater weights. This feature selection and suppression capability helps the model focus more on those features that are truly valuable for prediction, thereby improving the accuracy of the prediction.

[0097] Compared with traditional methods, the advantage of the weighted gating mechanism is that it can select features based on the characteristics of the data, avoiding the subjectivity of manual feature selection; at the same time, through the dynamic adjustment of the gating signal, the model can have stronger generalization ability when facing different raw materials and mix proportions; it helps to more accurately calculate the identification feature vector and build the database in the subsequent steps, and provide high-quality data support for the subsequent slump correction.

[0098] S23: Combine the raw material characteristics and the mix ratio characteristics to calculate the identification feature vector. The calculation method is:

[0099] ;

[0100] in, is the identification feature vector, For splicing operation;

[0101] The calculation process of steps S21, S22, and S23 is uniformly recorded as:

[0102] ;

[0103] in, Compute operations for identifying eigenvectors;

[0104] Step S23 concatenates the raw material characteristics and mix ratio characteristics processed by the weight gating mechanism to form an identification feature vector. The identification feature vector represents the characteristics of the historical case and is used in the subsequent second correction process of the concrete slump.

[0105] S24: All identification feature vectors, average aggregate particle size, cement strength, mix ratio data, and concrete slump are stored in a relational database one by one to obtain a historical concrete slump database.

[0106] S3: extract the highest slump point of each video frame from the preprocessed slump process data, calculate the instantaneous concrete slump and splice them to obtain the concrete slump sequence;

[0107] S31: In the pre-processed collapse process data, for each video frame, concrete collapse area identification is performed, and the calculation method is:

[0108] ;

[0109] in, is the image of the concrete collapse area, For the cropping operation, For YOLOv5 model operation, is the collapse process data after preprocessing;

[0110] S32: According to the concrete collapse area image, the highest collapse point of each video frame is extracted, and the calculation method is:

[0111] ;

[0112] in, is the edge contour of concrete collapse. For OpenCV's findContours function, It is the Canny edge detection algorithm;

[0113] Extract the point with the largest ordinate in the concrete collapse edge contour line, and use this point as the highest collapse point of the video frame;

[0114] S33: For the highest collapse point of each video frame, the perspective projection inverse transformation formula is used to calculate the concrete collapse distance as the instantaneous concrete slump of the video frame, and the instantaneous concrete slump of each video frame is spliced ​​to obtain the concrete slump sequence ;

[0115] The concrete slump sequence is obtained by extracting the highest slump point of each video frame and calculating the instantaneous concrete slump for splicing. This process accurately records the entire process of concrete slump from the beginning to the end, and can comprehensively and continuously reflect the slump changes of concrete. In the subsequent first correction process of concrete slump, the concrete slump sequence is used as the key input, which makes full use of the dynamic information in the concrete slump process and improves the accuracy of slump calculation.

[0116] S4: extracting the preliminary concrete slump and the concrete slump sequence characteristics, and performing a first correction on the preliminary concrete slump according to the concrete slump sequence characteristics to obtain a first concrete slump;

[0117] S41: take the last value in the concrete slump sequence as the preliminary concrete slump Distance;

[0118] In particular, in order to deal with the problem that the lumps produced by uneven mixing are located exactly at the highest point of concrete slump, resulting in a low slump measurement, step S41 can be replaced by the following steps:

[0119] Combined with the characteristics of raw materials and mix ratio, the preliminary concrete slump is corrected and calculated as follows:

[0120] ;

[0121] ;

[0122] in, is the initial concrete slump correction, To update the symbol, To obtain a new preliminary concrete slump by subtracting d from the preliminary concrete slump Distance;

[0123] S42: Extract the concrete slump sequence characteristics, the calculation method is:

[0124] ;

[0125] in, is the first concrete slump sequence characteristic, For the first long short-term memory network operation, is the second concrete slump sequence characteristic, For the second long short-term memory network operation, Operate for multi-layer perceptron;

[0126] S43: According to the concrete slump sequence characteristics, the preliminary concrete slump is corrected for the first time to obtain the first concrete slump, which is calculated as follows:

[0127] ;

[0128] in, First concrete slump, For multi-layer perceptron operations.

[0129] Step S42 extracts the features of the concrete slump sequence through two long short-term memory networks (LSTM). As a special recurrent neural network, LSTM can capture long-term dependencies in sequence data. In this step, the first LSTM network focuses on extracting the changing trend of the slump velocity; the second LSTM network focuses on extracting the changes in the slump morphology, providing rich feature information for the subsequent first correction.

[0130] S5: collecting raw material data of this concrete slump test, calculating its identification feature vector, and using the identification feature vector in the historical concrete slump database to perform a second correction on the first concrete slump to obtain a second concrete slump;

[0131] S51: Collect the raw material data of this concrete slump test, including the cement strength, average aggregate particle size, mix ratio data, and the first concrete slump of this concrete slump test; then splice the cement strength, average aggregate particle size and concrete slump of this concrete slump test to obtain the raw material characteristics of this time , combine the mix ratio data of this test with the concrete slump to obtain the mix ratio characteristics of this test ;

[0132] S52: Calculate the identification feature vector of this concrete slump detection, and the calculation method is:

[0133] ;

[0134] in, is the identification feature vector of this concrete slump detection;

[0135] S53: Calculate the similarity between the identification feature vector of the current concrete slump detection and all identification feature vectors in the historical concrete slump database, and use the identification feature vector with the highest similarity as the modified identification feature vector. The calculation method is:

[0136] ;

[0137] in, is the similarity between the identification feature vector of this concrete slump detection and the i-th identification feature vector in the historical concrete slump database, i is the data index, is the cosine similarity calculation, is the i-th identification feature vector in the historical concrete slump database, To correct the identity eigenvector, To obtain Maximum operate;

[0138] S54: Perform a second correction on the first concrete slump using the correction identification feature vector to obtain a second concrete slump, which is calculated as follows:

[0139] ;

[0140] in, is the slump of the second concrete.

[0141] Step S5 finds the historical case most similar to the current test by calculating the similarity between the identification feature vector of the current concrete slump test and the identification feature vector in the historical database; the identification feature vector integrates the key information of the raw material characteristics, mix ratio and slump, and is an important representation of the concrete slump; cosine similarity, as a commonly used vector similarity measurement method, can accurately reflect the directional consistency of two vectors in multidimensional space; in concrete slump detection, since each dimension of the feature vector has different dimensions and value ranges, cosine similarity is more applicable than measurement methods such as Euclidean distance.

[0142] Most traditional methods only evaluate the slump based on the data of this test, while step S53 makes full use of the large amount of data in the historical database to make a more accurate correction to the slump of the first concrete based on the most matching historical slump data; in addition, the similarity calculation realizes flexible processing of different raw materials, mix ratios and slumps, which has stronger adaptability than traditional methods. For details, see Table 1. The concrete slump detection method based on deep learning is used to detect the concrete slump of different raw material data and mix ratio data:

[0143] Table 1 Concrete slump detection method based on deep learning to detect different raw material data

[0144] Concrete slump from mix proportion data

[0145]

[0146] For example: First, video data of a concrete collapse process was collected and a preliminary concrete slump sequence was obtained; at the same time, the raw material data of the concrete sample was also collected, including the average aggregate particle size of 5mm, the cement strength of 42.5MPa, and specific mix ratio data (water: cement: sand: aggregate = 0.45:1:1.5:2.5).

[0147] After preprocessing, the concrete collapse process data was obtained, and the highest collapse point of each video frame was extracted. The instantaneous concrete slump was calculated, and finally the concrete slump sequence [10, 20, 35, 50, 60, 65, 70, 72] was obtained by splicing.

[0148] Next, the preliminary concrete slump (72 mm) and the concrete slump sequence features were extracted, and the sequence features were modeled through the long short-term memory network (LSTM). It was found that the slump speed of the concrete sample was abnormally fast, indicating that there may be some special circumstances that lead to the low slump.

[0149] Based on the features extracted by the LSTM network, a multi-layer perceptron was used to make the first correction to the preliminary concrete slump, and the corrected first concrete slump was 73 mm.

[0150] Then, through similarity calculation, a historical data that is most similar to this test is found in the historical concrete slump database. The raw material values ​​of this data are: average aggregate particle size 4.8mm, cement strength 43MPa, similar mix ratio data (water: cement: sand: aggregate = 0.4:1:1.6:2.4), and the slump measured at that time is 74mm;

[0151] Therefore, a second correction was made to the first concrete slump (73 mm) based on the slump of historical data (74 mm), and finally the second concrete slump was 73.5 mm. This value comprehensively considers the historical similarities and special circumstances of the current test data and is more accurate and reliable.

[0152] Embodiment 2: A mixing pump truck, comprising:

[0153] Vehicle body;

[0154] A chassis, disposed at the bottom of the vehicle body, for supporting the vehicle body;

[0155] A stirring device, arranged on the chassis, for stirring concrete;

[0156] An arm, arranged on the chassis and located at the rear of the vehicle body;

[0157] A pumping system, which is connected to the boom through a delivery pipeline and is used to pump concrete from the mixing device to a designated location and deliver it to a construction site through the boom;

[0158] A power device, disposed in the vehicle body, for outputting electric power;

[0159] A driving motor, wherein the output end of the driving motor is drivingly connected to the stirring device to drive the stirring device to rotate and realize the mixing operation of the concrete;

[0160] A hydraulic device, the hydraulic device is connected to the pumping system and is used to provide power for the pumping system;

[0161] Calculation module: extract raw material characteristics and mix ratio characteristics according to historical concrete slump data and slump process data, calculate identification feature vector, concrete slump sequence characteristics and first concrete slump, and use identification feature vector to make second correction to the first concrete slump to obtain second concrete slump;

[0162] A control system is electrically connected to the power device and the drive motor to control the operating state of the drive motor; and a second concrete slump is obtained according to the concrete slump detection method based on deep learning described in the first embodiment of the present invention to control the rotation speed of the drive motor.

[0163] Specifically, in the present invention, without specific limitation, the driving motor can be an integrated motor and reducer.

[0164] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0165] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in multiple embodiments of the present invention.

[0166] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A concrete slump detection method based on deep learning, characterized in that: The following steps are involved: S1: Collect historical concrete slump data and slump process data, and perform preprocessing to obtain preprocessed historical concrete slump data and preprocessed slump process data. The specific steps include: S11: Collect historical concrete slump data, each piece of historical concrete slump data includes concrete slump, cement strength, average aggregate particle size and mix ratio data; the mix ratio data includes the ratio of cement, water, sand and aggregate in concrete, expressed as weight ratio; the unit of concrete slump is millimeter, the unit of cement strength is MPa, and the unit of average aggregate particle size is millimeter; S12: setting the frame rate of the video camera to 30 frames per second, and ensuring that the video camera is located directly in front of the concrete, that is, at the main view angle, to shoot the collapse process of the concrete and obtain collapse process data; S13: remove missing values ​​from historical concrete slump data, use the 3σ principle to remove outliers, and then normalize to obtain preprocessed historical concrete slump data; perform contrast enhancement on the slump process data and split it into a series of continuous video frames to obtain preprocessed slump process data; S2: Construct a weight gating mechanism, extract raw material characteristics and mix ratio characteristics, calculate identification feature vectors, and build a historical concrete slump database; S3: extract the highest slump point of each video frame from the preprocessed slump process data, calculate the instantaneous concrete slump and splice them to obtain the concrete slump sequence; S4: extracting the preliminary concrete slump and the concrete slump sequence characteristics, and performing a first correction on the preliminary concrete slump according to the concrete slump sequence characteristics to obtain a first concrete slump; S5: collecting raw material data of this concrete slump detection, calculating its identification feature vector, and using the identification feature vector in the historical concrete slump database to perform a second correction on the first concrete slump to obtain a second concrete slump.

2. The method for detecting concrete slump based on deep learning according to claim 1, characterized in that: The S2 includes: Firstly, the cement strength, average aggregate particle size and concrete slump are spliced ​​to obtain preliminary raw material characteristics. The mix ratio data is spliced ​​with the concrete slump to obtain preliminary mix ratio characteristics. Secondly, a weighted gating mechanism is constructed to calculate the first mix ratio gate, the first raw material gate, the second mix ratio gate and the second raw material gate using preliminary raw material characteristics and preliminary mix ratio characteristics, and the raw material characteristics and mix ratio characteristics are extracted. Thirdly, the identification feature vector is calculated based on the raw material characteristics and the mix ratio characteristics.

3. The method for detecting concrete slump based on deep learning according to claim 2, characterized in that: The S2 includes: S21: Combine cement strength, average aggregate particle size and concrete slump to obtain preliminary raw material characteristics , combine the mix ratio data with the concrete slump to obtain the preliminary mix ratio characteristics ; S22: Construct a weight gating mechanism to extract raw material characteristics and mix ratio characteristics. The calculation method is: ； in, is the first mix ratio gate, For the first raw material gate, is the second mix ratio gate, For the second raw material door, are the first, second, third, and fourth weight matrices respectively, are the first, second, third and fourth bias vectors respectively, is the sigmoid function, is the mix ratio characteristic, For the raw material characteristics, is the vector dot product; S23: Combine the raw material characteristics and the mix ratio characteristics to calculate the identification feature vector. The calculation method is: ； in, is the identification feature vector, For splicing operation; The calculation process of steps S21, S22, and S23 is uniformly recorded as: ； in, Compute operations for identifying eigenvectors; S24: All identification feature vectors, average aggregate particle size, cement strength, mix ratio data, and concrete slump are stored in a relational database one by one to obtain a historical concrete slump database.

4. The method for detecting concrete slump based on deep learning according to claim 3, characterized in that: The S3 includes: S31: In the pre-processed collapse process data, for each video frame, concrete collapse area identification is performed, and the calculation method is: ； in, is the image of the concrete collapse area, For the cropping operation, For YOLOv5 model operation, is the collapse process data after preprocessing; S32: According to the concrete collapse area image, the highest collapse point of each video frame is extracted, and the calculation method is: ； in, is the edge contour of concrete collapse. For OpenCV's findContours function, It is the Canny edge detection algorithm; Extract the point with the largest ordinate in the concrete collapse edge contour line, and use this point as the highest collapse point of the video frame; S33: For the highest point of collapse of each video frame, the perspective projection inverse transformation formula is used to calculate the concrete collapse distance as the instantaneous concrete slump of the video frame, and the instantaneous concrete slump of each video frame is spliced ​​to obtain the concrete slump sequence .

5. The method for detecting concrete slump based on deep learning according to claim 4, characterized in that: The S4 includes: S41: take the last value in the concrete slump sequence as the preliminary concrete slump Distance; S42: Extract the concrete slump sequence characteristics, the calculation method is: ； in, is the first concrete slump sequence characteristic, For the first long short-term memory network operation, is the second concrete slump sequence characteristic, For the second long short-term memory network operation, Operate for multi-layer perceptron; S43: According to the concrete slump sequence characteristics, the preliminary concrete slump is corrected for the first time to obtain the first concrete slump, which is calculated as follows: ； in, First concrete slump, For multi-layer perceptron operations.

6. The method for detecting concrete slump based on deep learning according to claim 5, characterized in that: The S5 includes: S51: Collect the raw material data of this concrete slump test, including the cement strength, average aggregate particle size, mix ratio data, and the first concrete slump of this concrete slump test; then splice the cement strength, average aggregate particle size and concrete slump of this concrete slump test to obtain the raw material characteristics of this time , combine the mix ratio data of this test with the concrete slump to obtain the mix ratio characteristics of this test ; S52: Calculate the identification feature vector of this concrete slump detection, and the calculation method is: ； in, is the identification feature vector of this concrete slump detection; S53: Calculate the similarity between the identification feature vector of the current concrete slump detection and all identification feature vectors in the historical concrete slump database, and use the identification feature vector with the highest similarity as the modified identification feature vector. The calculation method is: ； in, is the similarity between the identification feature vector of this concrete slump detection and the i-th identification feature vector in the historical concrete slump database, i is the data index, is the cosine similarity calculation, is the i-th identification feature vector in the historical concrete slump database, To correct the identity eigenvector, To obtain Maximum operate; S54: Perform a second correction on the first concrete slump using the correction identification feature vector to obtain a second concrete slump, which is calculated as follows: ； in, is the slump of the second concrete.

7. A mixing pump truck, characterized in that: include: Vehicle body; A chassis, disposed at the bottom of the vehicle body, for supporting the vehicle body; A stirring device, arranged on the chassis, for stirring concrete; An arm, arranged on the chassis and located at the rear of the vehicle body; A pumping system, which is connected to the boom through a delivery pipeline and is used to pump concrete from the mixing device to a designated location and deliver it to a construction site through the boom; A power device, disposed in the vehicle body, for outputting electric power; A driving motor, wherein the output end of the driving motor is drivingly connected to the stirring device to drive the stirring device to rotate and realize the mixing operation of the concrete; A hydraulic device, the hydraulic device is connected to the pumping system and is used to provide power for the pumping system; A calculation module, extracting raw material characteristics and mix ratio characteristics according to historical concrete slump data and slump process data, and calculating an identification feature vector, a concrete slump sequence feature and a first concrete slump, and performing a second correction on the first concrete slump using the identification feature vector, so as to obtain a second concrete slump according to the concrete slump detection method based on deep learning as described in any one of claims 1 to 6; A control system is electrically connected to the power device and the drive motor to control the operating state of the drive motor.

Images