CN117630344B - Method for detecting slump range of concrete on line in real time - Google Patents
Method for detecting slump range of concrete on line in real time Download PDFInfo
- Publication number
- CN117630344B CN117630344B CN202410105937.9A CN202410105937A CN117630344B CN 117630344 B CN117630344 B CN 117630344B CN 202410105937 A CN202410105937 A CN 202410105937A CN 117630344 B CN117630344 B CN 117630344B
- Authority
- CN
- China
- Prior art keywords
- concrete
- slump
- neural network
- unit
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000001514 detection method Methods 0.000 claims abstract description 43
- 238000003062 neural network model Methods 0.000 claims abstract description 14
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 238000013145 classification model Methods 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims description 28
- 238000003756 stirring Methods 0.000 claims description 17
- 239000000463 material Substances 0.000 claims description 15
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 14
- 238000011176 pooling Methods 0.000 claims description 13
- 238000009826 distribution Methods 0.000 claims description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000007599 discharging Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 210000004205 output neuron Anatomy 0.000 claims description 5
- 238000005070 sampling Methods 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 3
- 230000002194 synthesizing effect Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000011049 filling Methods 0.000 claims description 2
- 239000004816 latex Substances 0.000 claims description 2
- 229920000126 latex Polymers 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 description 14
- 238000012360 testing method Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 10
- 238000011068 loading method Methods 0.000 description 8
- 238000011897 real-time detection Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000013135 deep learning Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000000007 visual effect Effects 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000011179 visual inspection Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000036571 hydration Effects 0.000 description 1
- 238000006703 hydration reaction Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 239000012779 reinforcing material Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/38—Concrete; Lime; Mortar; Gypsum; Bricks; Ceramics; Glass
- G01N33/383—Concrete or cement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Chemical & Material Sciences (AREA)
- Image Analysis (AREA)
- Food Science & Technology (AREA)
- Medicinal Chemistry (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Ceramic Engineering (AREA)
Abstract
The invention discloses a method for detecting the slump range of concrete on line in real time, which relates to the field of concrete slump detection and comprises the following steps: s1, acquiring a complete concrete falling video in a hopper in real time through a camera unit; s2, drawing frames of concrete falling videos in different slump ranges in an equidistant mode to construct a plurality of data sets matched with the different slump ranges; s3, constructing a classification model for each data set image by adopting a neural network so that the slump range of the concrete can be detected and classified; and S4, deploying the neural network model constructed in the S3 in an upper computer so as to identify the slump range of the concrete image transmitted by the camera unit in real time through the neural network model. The invention provides an application method for detecting the slump range of concrete on line in real time, which can detect the slump range of concrete in real time, provide data feedback in time and avoid the timeliness problem in the traditional method.
Description
Technical Field
The invention relates to the field of concrete slump detection. More particularly, the invention relates to a method for real-time on-line detection of the slump range of concrete.
Background
Slump is an index of the working performance of concrete and is also directly related to the quality of concrete. Proper slump helps to ensure that the concrete is easy to construct, fully encapsulates the reinforcing material, and improves structural quality and durability. Slump is an important parameter related to the quality of concrete.
The detection of existing concrete slump is typically determined by conventional slump tests. Conventional slump test methods are ASTM-C143/C143M cone slump test and ASTM-C1611/C1611M-V type slump test. The method has the defects of long time and poor timeliness, and cannot meet the requirement of high-speed production. Also relying on operator skill, these methods require a trained operator to perform the test, and thus the results may be affected by the operator experience and skill. These methods are typically off-line tests, requiring time to prepare the sample and perform the test, and do not provide immediate feedback. Each test requires preparation of a concrete sample, which can lead to waste of resources. And because of the specificity of the concrete, the slump test of the traditional concrete is easily influenced by factors such as water content, cement hydration, temperature and the like, so that the quick and accurate performance is required to reduce the interference of timeliness problems on test results and ensure the accurate assessment of the fluidity and plasticity of the concrete.
While other monitoring methods mainly include:
1. liquid limit gauge testing, which is commonly used in laboratory environments, is not suitable for real-time job site monitoring, and limits the sample size, which may be limited in size and geometry and thus not suitable for all concrete types.
2. Slope basin visual inspection is a method of estimating slump based on visual inspection of the speed of concrete flowing on a slope. The results may be affected by subjective judgment by the operator.
3. And (5) detecting a vibration meter. Vibration meters are commonly used to measure the fluidity of concrete, rather than directly measuring slump. The slump of concrete is a combination of various indexes, not just fluidity.
4. Based on the machine vision and the deep learning method, the currently proposed method is to judge the slump of the concrete by using a single frame image, and the image requirement is a shot image of the concrete in a standing state. In addition, in the embodiment, the stirred concrete needs to be sampled and stood for shooting detection. This embodiment does not integrate the test into the concrete production process, but rather adds to the concrete production process. The sampling and standing process also requires a certain time and cost, and relatively expensive cameras and specialized software, has high requirements on image quality, and increases the concrete production process and cost. Moreover, only the detection of a single frame image can obtain the characteristics of concrete at the surface part, and the slump detection result of the same batch cannot be completely represented, so that the accuracy of the detection result may be problematic.
5. Acoustic wave measuring devices, methods based on acoustic wave measurement are affected by the concrete material and texture and therefore may not be accurate enough in some cases. Acoustic wave measurement devices are expensive.
6. The method of laser scanners generally requires that the concrete surface be smooth, unsuitable for rough or porous concrete surfaces, and costly.
7. Current parameters detect slump and some methods attempt to indirectly infer slump of concrete using current parameters as the concrete is being stirred. The method has limited precision, and the current parameter method depends on the stirring quantity of concrete, so that the method can have differences under different stirring quantity, proportion and raw materials, and the accuracy is influenced. No direct visual information can be provided. Concrete slump cannot be intuitively reflected, and the current parameter method generally requires complex analysis to obtain slump information.
8. A multi-parameter fusion method. And fusing the motor current parameter and the parameter of the vibration measuring device during concrete stirring, and comprehensively analyzing to obtain the slump of the concrete. This is an indirect measurement that does not directly reflect the state of concrete slump. And is greatly influenced by the proportion of concrete and raw materials.
9. The prior commercial concrete station relies on experience and visual observation of engineers and laboratory technicians to judge the slump by observing the state of concrete in multiple aspects of the buffer hopper, but manual observation is easily influenced by subjective judgment of operators, so that inconsistency and reliability of results are reduced. The culture has higher experienced labor cost and longer period and also lacks stability.
10. The existing concrete slump detection method based on remarkable target detection is characterized in that a single frame image is taken into consideration in a concrete standing state, judgment is carried out only through texture features of the single frame image, the judgment stability of the single frame image is not high, the single frame image cannot reflect the flowability of concrete, and the flowability is an important index for reflecting the slump of the concrete.
Therefore, the existing concrete slump detection technology has advantages and disadvantages, and the actual application cannot be satisfied.
Disclosure of Invention
It is an object of the present invention to address at least the above problems and/or disadvantages and to provide at least the advantages described below.
To achieve these objects and other advantages and in accordance with the purpose of the invention, there is provided a method of detecting a slump range of concrete on line in real time, comprising:
s1, acquiring a complete concrete falling video in a hopper in real time through a camera unit;
s2, drawing frames of concrete falling videos in different slump ranges in an equidistant mode to construct a plurality of data sets matched with the different slump ranges;
s3, constructing a classification model for each data set image by adopting a neural network so that the slump range of the concrete can be detected and classified;
and S4, deploying the neural network model constructed in the S3 in an upper computer so as to identify the slump range of the concrete image transmitted by the camera unit in real time through the neural network model.
Preferably, in S2, the different slump ranges include: three grades of 50mm-100mm, 110mm-150mm and 160mm-210mm.
Preferably, in S3, the classification model construction is based on multi-stream ResNet34-Late
And the ConvFusion architecture combines a multi-flow ResNet34 network and a Late Conv Fusion layer to realize the extraction of concrete texture characteristics and fluidity characteristics.
In S3, the method for constructing the classification model includes:
s31, extracting features of three process images of a concrete bucket-in image, a bucket-in stacking image and a bucket-out image;
s32, setting a weight distribution model for the prediction results of the three process images;
s33, synthesizing the detection results of the three process images to obtain a detection result which can finally reflect the slump range of the concrete.
Preferably, in S31, for each process image, three independent convolutional neural network flows are established for respective processing using a multi-flow convolutional neural network flow architecture;
in the multi-stream convolutional neural network stream architecture, resNet34 is adopted as a basic network, a convolutional kernel with a convolutional layer of 1:7x7 is set, the stride is 2, the filling is 3, and an output channel 64 is formed;
setting a maximum pooling layer 1:3x3 of pooling cores, wherein the stride is 2;
setting 3 residual blocks, wherein each residual block comprises two convolution kernels of 3x3, and each residual block output channel is respectively: 64. 256, 614;
for each stream, adding a multi-scale convolutional layer after ResNet34, and at the end of each stream, applying a global averaging pooling layer;
the Late Conv Fusion layer splices the three features after the global average pooling of the flows, and then outputs the feature vector through a convolution kernel with the size of 1x1 and the channel number of 256.
Preferably, in S32, the weight distribution model is implemented by using a full-connection layer, and the feature vector output by the latex fusion is used as the input of the full-connection layer, and the number of output neurons of the full-connection layer is set to be 3 so as to correspond to three categories, and then the full-connection layer predicts the concrete slump category by adopting the following formula:
Y= {softmax}(W[f(x)]+ B)
wherein W [ f (x) ] represents matrix multiplication, f (x) is a eigenvector output by Late Conv Fusion, W is a weight matrix of the full-connection layer, B is a bias vector of the full-connection layer, softmax is an activation function, and Y is a classification class probability.
The device for detecting the slump range of the concrete on line in real time is further included, and the device comprises:
a concrete stirring unit communicated with the plurality of charging barrels;
the assembly unit is arranged at the discharging side of the concrete stirring unit;
the material guiding unit is arranged between the concrete stirring unit and the assembling unit;
the shooting unit is arranged between the concrete stirring unit and the material guiding unit and used for sampling the slump of the concrete in real time;
the upper computer is in communication connection with the camera unit;
wherein, the guide unit includes:
a material transfer conduit mated with the feed side of the assembly unit;
the hopper is arranged on the material conveying pipe and is at a preset distance with the discharging side of the concrete stirring unit.
The invention at least comprises the following beneficial effects: the device for detecting the concrete slump range in real time provided by the invention can greatly improve the automation degree and the production quality of concrete production, improve the accuracy of concrete slump detection and bring innovative solutions to the concrete industry, and has the following specific effects:
firstly, the slump range of the concrete can be detected in real time, data feedback can be provided in time, and the timeliness problem in the traditional method is avoided.
Secondly, the slump identification accuracy is improved by introducing a deep learning technology, and meanwhile, the automatic detection is realized, and the manual intervention is reduced.
Thirdly, the neural network model based on the training of the big data set has the learning capability of deep learning, can be better suitable for different concrete conditions, and provides data-driven production decision support.
Fourth, compared with the method relying on experience and manual visual inspection, the device reduces labor cost and improves production efficiency.
Fifthly, the invention can carry out deep data analysis through real-time detection and data recording, and provides references and guidance for continuous improvement and optimization of the production process.
Sixthly, multi-frame image detection is performed by using a multi-stream network. And obtaining a final detection result by integrating the detection results of the images of the three processes through feature extraction of the concrete bucket-in image, the bucket-in accumulation image and the bucket-out image and a weight distribution model of the prediction results of the three process images. The characteristics of time dimension are added to a certain extent, the characteristics of concrete fluidity are considered in multi-process and multi-frame image detection, and the detection is more comprehensive.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a schematic diagram of the system components of the device for detecting the slump range of concrete on line in real time according to the invention;
FIG. 2 is a schematic diagram of classification of a data set sample according to the present invention;
FIG. 3 is a schematic diagram of a network architecture conforming to classification tasks according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of the real-time detection and feedback flow of the present invention when applied.
Detailed Description
The present invention is described in further detail below with reference to the drawings to enable those skilled in the art to practice the invention by referring to the description.
In order to solve the problems of lack of timeliness and poor subjective strong stability of manual visual observation in the traditional slump test, and the existing method based on machine vision and deep learning similar to the method of the invention, which is only in a standing state, has the problems of poor stability and incomplete detection of single-frame image detection in a single state, and provides higher automation and accuracy of concrete production, further improves the production efficiency, ensures the production quality, and provides references and instructions for improvement and optimization of production processes.
Referring to fig. 1, an apparatus for real-time on-line detection of a slump range of concrete, comprising:
a concrete mixing unit 2 communicating with the plurality of cartridges 1;
the assembling unit 3 is arranged at the discharging side of the concrete stirring unit;
the material guiding unit is arranged between the concrete stirring unit and the assembling unit;
the shooting unit 4 is arranged between the concrete stirring unit and the material guiding unit and used for sampling the concrete slump in real time;
an upper computer 5 connected with the camera unit in a communication way;
wherein, the guide unit includes:
a material transfer conduit 6 cooperating with the feed side of the assembly unit;
the hopper 7 is arranged on the material conveying pipe and has a preset distance with the discharging side of the concrete stirring unit, and in actual operation, the material guiding unit is fixed through the concrete loading platform, namely, the material guiding unit can bear the weight and pressure in the concrete loading process by designing a firm concrete loading platform; a firm metal concrete hopper is arranged on the platform, so that when concrete enters the platform from the stirrer, the hopper can bear the weight and pressure of concrete loading and smoothly load the concrete through the hopper;
and further, a high-resolution camera unit (also called a camera) is arranged on the top or the side surface of the funnel to capture concrete falling and loading processes (including bucket in, bucket stacking and bucket out), so that the camera can shoot at high resolution to capture details. In practical application, the LED lamp light source can be used for arranging the window shade to avoid the interference of sunlight, and the camera can be a sea-health PTZ infrared camera, so that the shooting quality is improved.
Examples:
when the device for detecting the slump range of concrete on line in real time is applied, the device mainly comprises:
video data recording: the camera device is arranged to record video continuously to capture the flow accumulation and hopper discharging process of the concrete in the hopper device. The recorded video data will be stored for later processing. In the recorded video data, a conventional slump test is performed to obtain a slump value of concrete. An association between concrete slump and video data is established. The slump value measured will be the label of the data. The tags are associated with corresponding video data to form a data set with slump tags. This dataset will be used to train the neural network model. After the neural network is built, training the neural network model by using the manufactured labeled data set, so as to obtain a model file with weight. This process ensures an efficient correlation between recorded video data and experimentally measured slump values. By the method, the model can learn the visual characteristics of the concrete in different slump ranges, so that the slump ranges are classified. The data driven method provides a reliable solution for monitoring and classifying concrete slump.
And (3) data set preparation: extracting image frames from recorded videos, classifying video frame data acquired in different slump ranges, and labeling each frame of image according to time sequence; recorded is a complete video of the concrete drop, with a complete drop at about 50s, with the segment of interest in the middle 30s. The interesting segments are subjected to equidistant frame extraction, and the image change of the similar frames is very tiny, so that the frame extraction is carried out once every 5 s. The above operations are performed on video clips of different slump ranges to obtain 6728 original data sets. According to the traditional slump test and the production requirement of a commercial concrete station, dividing a concrete data set into three slump grades according to different slump ranges, wherein the slump grades are sequentially 50mm-
100mm, 110mm-150mm, 160mm-210mm. A dataset was created containing images of the concrete for these three slump ranges, as shown in figure 2.
Building a neural network image classification model: selecting proper deep learning architecture, and selecting Resnet as basic network. A network architecture designed to meet the classification task is shown in fig. 3. The architecture of multi-stream ResNet34-LateConvFusion is employed. Combining a multi-stream ResNet34 network with LateConv
And a Fusion layer for extracting texture characteristics and fluidity characteristics of the concrete. And training the neural network by using the manufactured data set, generating a model file so that the model file can detect and classify the slump range of the concrete, and classifying and detecting the input concrete image. Specifically, the classification result of each process image is judged by extracting the characteristics of the concrete bucket-in image, the bucket-in accumulation image and the bucket-out image. And (3) carrying out weight distribution on the predicted results of the three process images, wherein the weight can be changed and trained, the classification detection results of the stacked images in the bucket have the greatest influence on the final classification detection results, and the classification detection results of the other two processes have less influence on the final classification detection results. The classification detection results of the three processes are Y (1), Y (2) and Y (3), and the final classification detection result is Y. Y=a×y (1) +b×y (2) +c×y (3), wherein a, b, c are weights, and specific values of a, b, c can be obtained by training. Satisfying a+b+c=1, which means that the sum of weights is 1, ensures that the final classification result is determined jointly by the image detection classifications of the three processes. The weight is set in such a way that the classification detection results of the three images are all considered. In the training process, the weight can be adjusted according to the requirements of actual problems, so that the adaptability of the model under different conditions is ensured. The method for synthesizing the detection results of the three process images provides a scientific and flexible solution for the classification task of the concrete slump range. Thus, the final detection result is obtained by combining the detection results of the images in the three processes. The specific network design is as follows:
multi-stream convolutional neural network stream architecture: for each process image, three independent convolutional neural network flows are established, and the images of the concrete bucket-in, bucket-in stacking and bucket-out processes are respectively processed.
ResNet34 serves as the base network. Convolution layer 1:7x7, stride of 2, fill of 3, output channel 64. Maximum pooling layer 1. 3x3 pooling cores, stride 2. Residual block 1: comprising two 3x3 convolution kernels, output channel 64. Residual block 2: comprising two 3x3 convolution kernel output channels 128. Residual block 3: comprising two 3x3 convolution kernels, output channel 256. Residual block 4: comprising two 3x3 convolution kernels, output channel 512. Multi-scale convolution layer: for each stream, a multi-scale convolutional layer is added after ResNet 34. The convolution kernel of the multi-scale convolution layer 1:1x1 outputs a channel 64. The convolution kernel of multi-scale convolution layer 2:3x3 outputs a channel 128. The convolution kernel of multi-scale convolution layer 3:5x5 outputs channel 256. Global average pooling layer: at the end of each stream, a global averaging pooling layer is applied.
Deployment detection device: training is carried out by using the manufactured data set and the built network model, and the trained neural network model is deployed into the device, so that a real-time detection function is realized.
Real-time detection and feedback: the camera device transmits video data to the neural network model, the model identifies the slump range of concrete, and real-time feedback and record data are provided through an OPC communication protocol to help an operator to adjust.
The Late Conv Fusion layer (Chinese name: post-convolution Fusion layer) first performs global average pooling processing on the features of the three streams, and then splices the processed features together. The layer is then processed through a convolution kernel of size 1x1, channel number 256, and finally outputs the feature vector.
The multi-flow ResNet34-LateConvFusion architecture describes the forward propagation of the network with mathematical expressions as follows:
1. concrete fill flow (Concrete Pouring Flow):
input image: x {1};
convolution layer: y {1} = { Conv } (x {1}, W {1 }) +b {1};
residual block: z {1} = { ResBlock } (y {1}, W {2};
multi-scale convolution layer: g {1} (x {1 }) = { MultiScaleConv } (z {1}, W {3 }), g {1} (x {1 })
The result of the multi-scale convolution operation on the intermediate feature map z {1} is shown. The multi-scale convolution layer convolves the input feature map with a convolution kernel W {3} to generate a final output feature map g {1} (x {1 }).
2. Build-up flow in bucket (Bucket Stacking Flow):
input image: x {2};
convolution layer: y {2} = { Conv } (x {2}, W {4 }) +b {2};
residual block: z {2} = { ResBlock } (y {2}, W {5 };
multi-scale convolution layer: g {2} (x {2 }) = { MultiScaleConv } (z {2}, W {6 };
3. outlet bucket flow (Concrete Pouring Out Flow):
input image: x {3};
convolution layer: y {3} = { Conv } (x {3}, W {7 }) +b {3};
residual block: z {3} = { ResBlock } (y {3}, W {8 };
multi-scale convolution layer: g {3} (x {3 }) = { MultiScaleConv } (z {3}, W {9 });
4. late Conv Fusion layer:
fusing the feature map: f (x) =g {1} (x {1 }) +g {2} (x {2 }) +g {3} (x {3 })
W { i } in the above formula represents a convolution kernel parameter, b { i } represents a bias parameter, { Conv } represents a convolution operation, { Resblock } represents a residual block, { MultiScaleConv } represents a multi-scale convolution operation. The result of the multi-scale convolution operation on the intermediate feature map z { i } is shown. The multi-scale convolution layer convolves the input feature map with a convolution kernel W { i } to generate a final output feature map g { i } (x { i }). And finally, obtaining a fusion feature map in a feature superposition mode.
Full connectivity layer and classification: the feature vector output by LateConvFusion is input into a full-connection layer, the number of output neurons of the final full-connection layer is 3, and a softmax activation function is used for three categories, so that the output is ensured to be a probability distribution. The characteristic vector output by Late Conv Fusion is f (x), the weight matrix of the full connection layer is W, the bias vector is B, the weight matrix W and the bias vector B can continuously and automatically update the numerical values along with the back propagation of the neural network until the loss function of the neural network does not have obvious change, and the weight matrix W and the bias vector B can be determined. The output Y of the fully connected layer can be calculated by the following formula:
Y= {softmax}(W[f(x)]+ B)
where W [ f (x) ] represents a matrix multiplication operation to obtain an intermediate vector, then adding the bias vector B, and finally activating by a softmax function to ensure that the output is a probability distribution. The number of output neurons is 3, corresponding to three categories. The Softmax function maps each element of the output vector to a probability value between 0 and 1, ensuring that the sum of all probability values is 1, representing a probability distribution. And the classification type probability can be obtained through the Softmax function, and then the concrete slump of the corresponding type can be obtained. By means of the softmax function a probability distribution is obtained, wherein the probability of each category represents the confidence of the model for that category. And finally, selecting the category with the highest probability as the classification prediction of the network. This design ensures that the output is a probability distribution and that the softmax function maps each element in the vector to a probability value between 0 and 1. Since the number of output neurons is 3, the application of the softmax function also ensures that the sum of all probability values is 1, corresponding to three categories, forming an effective probability distribution.
Deployment detection device: deploying the trained neural network model into the device for real-time detection; the device is arranged to enable the device to continuously detect the concrete loading process.
Real-time detection and feedback: when the concrete loading process starts, the camera device captures and transmits video data to the neural network model; the model analyzes the data and identifies the slump range of the concrete; information feedback is performed on the concrete batching system through the communication protocol and data exchange standard of OPC to provide real-time feedback and record data for operators and the system to adjust accordingly, and the flow is shown in figure 4.
Maintenance and calibration: and (3) regularly maintaining the camera device and the neural network model, ensuring normal operation, and calibrating the device to adapt to different concrete types and conditions.
The embodiment enables the device to detect the slump range in real time during the loading process of the concrete and provides qualitative analysis on the fluidity characteristics of the concrete. This helps to adjust the subsequent concrete processing in time to ensure consistency in quality and performance.
The above is merely illustrative of a preferred embodiment, but is not limited thereto. In practicing the present invention, appropriate substitutions and/or modifications may be made according to the needs of the user.
The number of equipment and the scale of processing described herein are intended to simplify the description of the present invention. Applications, modifications and variations of the present invention will be readily apparent to those skilled in the art.
Although embodiments of the invention have been disclosed above, they are not limited to the use listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Additional modifications will readily occur to those skilled in the art. Therefore, the invention is not to be limited to the specific details and illustrations shown and described herein, without departing from the general concepts defined in the claims and their equivalents.
Claims (4)
1. A method for real-time on-line detection of the slump range of concrete, comprising:
s1, acquiring a complete concrete falling video in a hopper in real time through a camera unit;
s2, drawing frames of concrete falling videos in different slump ranges in an equidistant mode to construct a plurality of data sets matched with the different slump ranges;
s3, constructing a classification model for each data set image by adopting a neural network so that the slump range of the concrete can be detected and classified;
s4, deploying the neural network model constructed in the S3 in an upper computer to identify the slump range of the concrete image transmitted by the camera unit in real time through the neural network model;
in S3, the classification model construction is based on a multi-flow ResNet34-LateConvfusion architecture, and the multi-flow ResNet34 network and LateConv Fusion layer are combined to realize extraction of concrete texture characteristics and fluidity characteristics;
in S3, the method for constructing the classification model includes:
s31, extracting features of three process images of a concrete bucket-in image, a bucket-in stacking image and a bucket-out image;
s32, setting a weight distribution model for the prediction results of the three process images;
s33, synthesizing detection results of the three process images to obtain a detection result which can finally reflect the slump range of the concrete;
in S32, the weight distribution model is implemented by using a full-connection layer, the feature vector output by the latex fusion is used as the input of the full-connection layer, and the number of output neurons of the full-connection layer is set to be 3 to correspond to three categories, and then the full-connection layer predicts the concrete slump category by adopting the following method:
Y= {softmax}(W[f(x)]+ B)
wherein W [ f (x) ] represents matrix multiplication, f (x) is a eigenvector output by Late Conv Fusion, W is a weight matrix of the full-connection layer, B is a bias vector of the full-connection layer, softmax is an activation function, and Y is a classification class probability.
2. The method for real-time on-line detection of a concrete slump range according to claim 1, wherein in S2, the different slump ranges include: three grades of 50mm-100mm, 110mm-150mm, 160mm-210mm.
3. The method for real-time on-line detection of a concrete slump range according to claim 1, wherein in S31, for each process image, three independent convolutional neural network flows are established for respective processing using a multi-flow convolutional neural network flow architecture;
in the multi-stream convolutional neural network stream architecture, resNet34 is adopted as a basic network, a convolutional kernel with a convolutional layer of 1:7x7 is set, the stride is 2, the filling is 3, and an output channel 64 is formed;
setting a maximum pooling layer 1:3x3 of pooling cores, wherein the stride is 2;
setting 3 residual blocks, wherein each residual block comprises two convolution kernels of 3x3, and each residual block output channel is respectively: 64. 256, 614;
for each stream, adding a multi-scale convolutional layer after ResNet34, and at the end of each stream, applying a global averaging pooling layer;
the Late Conv Fusion layer splices the three features after the global average pooling of the flows, and then outputs the feature vector through a convolution kernel with the size of 1x1 and the channel number of 256.
4. The method for real-time on-line detection of a concrete slump range according to claim 1, further comprising means for real-time on-line detection of a concrete slump range, said means comprising:
a concrete stirring unit communicated with the plurality of charging barrels;
the assembly unit is arranged at the discharging side of the concrete stirring unit;
the material guiding unit is arranged between the concrete stirring unit and the assembling unit;
the shooting unit is arranged between the concrete stirring unit and the material guiding unit and used for sampling the slump of the concrete in real time;
the upper computer is in communication connection with the camera unit;
wherein, the guide unit includes:
a material transfer conduit mated with the feed side of the assembly unit;
the hopper is arranged on the material conveying pipe and is at a preset distance with the discharging side of the concrete stirring unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410105937.9A CN117630344B (en) | 2024-01-25 | 2024-01-25 | Method for detecting slump range of concrete on line in real time |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410105937.9A CN117630344B (en) | 2024-01-25 | 2024-01-25 | Method for detecting slump range of concrete on line in real time |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117630344A CN117630344A (en) | 2024-03-01 |
CN117630344B true CN117630344B (en) | 2024-04-05 |
Family
ID=90027268
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410105937.9A Active CN117630344B (en) | 2024-01-25 | 2024-01-25 | Method for detecting slump range of concrete on line in real time |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117630344B (en) |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111178319A (en) * | 2020-01-06 | 2020-05-19 | 山西大学 | Video behavior identification method based on compression reward and punishment mechanism |
CN111611847A (en) * | 2020-04-01 | 2020-09-01 | 杭州电子科技大学 | Video motion detection method based on scale attention hole convolution network |
CN112130216A (en) * | 2020-08-19 | 2020-12-25 | 中国地质大学(武汉) | Geological advanced fine forecasting method based on convolutional neural network multi-geophysical prospecting method coupling |
CN113642482A (en) * | 2021-08-18 | 2021-11-12 | 西北工业大学 | Video character relation analysis method based on video space-time context |
CN113902725A (en) * | 2021-10-19 | 2022-01-07 | 中国联合网络通信集团有限公司 | Slump measuring method, device, equipment and storage medium |
CN114002415A (en) * | 2021-10-29 | 2022-02-01 | 商车云(北京)科技有限公司 | Visual artificial intelligence real-time detection equipment and method for concrete slump |
CN114547249A (en) * | 2022-02-24 | 2022-05-27 | 济南融瓴科技发展有限公司 | Vehicle retrieval method based on natural language and visual features |
CN114708635A (en) * | 2022-03-31 | 2022-07-05 | 长沙理工大学 | Face video tampering detection method based on MesoNet and LSTM |
CN114842365A (en) * | 2022-07-04 | 2022-08-02 | 中国科学院地理科学与资源研究所 | Unmanned aerial vehicle aerial photography target detection and identification method and system |
CN116071682A (en) * | 2023-02-01 | 2023-05-05 | 成都睿瞳科技有限责任公司 | Elevator door opening and closing detection method and system based on neural network |
CN219302461U (en) * | 2023-01-05 | 2023-07-04 | 河南新融建设工程有限公司 | Concrete slump detection device |
CN117191517A (en) * | 2023-09-08 | 2023-12-08 | 江苏大学 | Concrete laboratory full-automatic sample preparation device and use and modification optimization method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20220314492A1 (en) * | 2021-04-05 | 2022-10-06 | Sysdyne Technologies LLC | Concrete mixer truck drum rotation measurement using camera |
-
2024
- 2024-01-25 CN CN202410105937.9A patent/CN117630344B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111178319A (en) * | 2020-01-06 | 2020-05-19 | 山西大学 | Video behavior identification method based on compression reward and punishment mechanism |
CN111611847A (en) * | 2020-04-01 | 2020-09-01 | 杭州电子科技大学 | Video motion detection method based on scale attention hole convolution network |
CN112130216A (en) * | 2020-08-19 | 2020-12-25 | 中国地质大学(武汉) | Geological advanced fine forecasting method based on convolutional neural network multi-geophysical prospecting method coupling |
CN113642482A (en) * | 2021-08-18 | 2021-11-12 | 西北工业大学 | Video character relation analysis method based on video space-time context |
CN113902725A (en) * | 2021-10-19 | 2022-01-07 | 中国联合网络通信集团有限公司 | Slump measuring method, device, equipment and storage medium |
CN114002415A (en) * | 2021-10-29 | 2022-02-01 | 商车云(北京)科技有限公司 | Visual artificial intelligence real-time detection equipment and method for concrete slump |
CN114547249A (en) * | 2022-02-24 | 2022-05-27 | 济南融瓴科技发展有限公司 | Vehicle retrieval method based on natural language and visual features |
CN114708635A (en) * | 2022-03-31 | 2022-07-05 | 长沙理工大学 | Face video tampering detection method based on MesoNet and LSTM |
CN114842365A (en) * | 2022-07-04 | 2022-08-02 | 中国科学院地理科学与资源研究所 | Unmanned aerial vehicle aerial photography target detection and identification method and system |
CN219302461U (en) * | 2023-01-05 | 2023-07-04 | 河南新融建设工程有限公司 | Concrete slump detection device |
CN116071682A (en) * | 2023-02-01 | 2023-05-05 | 成都睿瞳科技有限责任公司 | Elevator door opening and closing detection method and system based on neural network |
CN117191517A (en) * | 2023-09-08 | 2023-12-08 | 江苏大学 | Concrete laboratory full-automatic sample preparation device and use and modification optimization method |
Non-Patent Citations (3)
Title |
---|
Numerical Detection of Concrete Slump by Fusion of Target Segmentation and Image Classification Network;Xiao Gao等;Journal of Physics: Conference Series;20231231;1-6 * |
混凝土塌落度预测算法研究;李明珠;中国优秀博硕士学位论文全文数据库(硕士)工程科技I辑;20220315;20-30 * |
路面滑模摊铺平整度施工技术;罗亮;价值工程;20101231;29(1);207 * |
Also Published As
Publication number | Publication date |
---|---|
CN117630344A (en) | 2024-03-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5898221B2 (en) | Computer-aided assignment of evaluations for digital samples of manufactured web products | |
CN105051781A (en) | Machine-vision system and method for remote quality inspection of a product | |
CN110009617A (en) | A kind of shaft forgings online quality control method | |
US20230003674A1 (en) | System and method for identifying lithology based on images and xrf mineral inversion | |
CN115184359A (en) | Surface defect detection system and method capable of automatically adjusting parameters | |
CN111896549A (en) | Building crack monitoring system and method based on machine learning | |
WO2024001467A1 (en) | Fatigue crack propagation rate test method and device based on deep learning | |
CN116465315B (en) | Automatic screen quality detection method and system | |
CN110440761A (en) | A kind of processing method of unmanned plane aerophotogrammetry data | |
CN117630344B (en) | Method for detecting slump range of concrete on line in real time | |
CN115797811A (en) | Agricultural product detection method and system based on vision | |
CN111830070A (en) | Automatic defect identification and judgment system and method based on edge calculation | |
Ponick et al. | Image-based deep learning for rheology determination of bingham fluids | |
CN112381190B (en) | Cable force testing method based on mobile phone image recognition | |
CN110935646A (en) | Full-automatic crab grading system based on image recognition | |
CN116843615B (en) | Lead frame intelligent total inspection method based on flexible light path | |
CN115019252B (en) | Concrete quality detection method and device and monitoring equipment | |
CN116091506A (en) | Machine vision defect quality inspection method based on YOLOV5 | |
CN115946109A (en) | Method and device for monitoring motion process of space manipulator | |
CN115375961A (en) | Product accessory identification and weight assessment method, device and related equipment | |
CN113269751A (en) | Pavement disease identification and repair method, system, equipment and medium based on convolutional neural network and 3D printing repair technology | |
CN115032206B (en) | Intelligent bridge robot detection method | |
RU2003136180A (en) | METHOD FOR SELF-ORGANIZING IMPLEMENTATION OF AGRICULTURAL PRODUCTION TECHNOLOGY | |
TWI494549B (en) | A luminance inspecting method for backlight modules based on multiple kernel support vector regression and apparatus thereof | |
CN116662931A (en) | Roadbed compaction quality evaluation method and device based on multi-mode compound vision |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |