A method for predicting concrete quality of mixing stations based on digital twins

Abstract

The present invention relates to a method for predicting the quality of concrete at a mixing station based on digital twins, which comprises the following steps: 1) establishing a geometric model of a digital twin of the mixing station; 2) constructing a data model of the digital twin of the mixing station; 3) establishing the digital twin of the mixing station; 4) generating an electronic ledger for concrete production; 5) processing the electronic ledger for concrete production by the digital twin of the mixing station to obtain a virtual image of the production of the mixing station; 6) mining historical data from the electronic ledger for concrete production, training a machine learning model online, and using the trained machine learning model to predict the quality of concrete produced subsequently. This method can improve the visibility of mixing station operators to each production link, thereby facilitating the formulation of a more accurate concrete mix ratio plan, achieving refined management of the production process, and resolving the problems of blind production and extensive management existing in current concrete mixing stations.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins and relates to a method for predicting the quality of concrete in a mixing station based on digital twins. Background Art

[0002] In the current concrete production process, the entire production line produces concrete that meets the project design requirements and meets the pouring strength requirements of the project site. This production line includes not only the production hub, such as the mixing plant, but also the storage and transportation of raw materials and finished concrete, and the corresponding departments for quality inspection. Each department collaborates to complete the concrete production process. To ensure the quality of the delivered concrete, the mixing plant and various departments frequently communicate and cooperate with each other, and conduct multiple cross-checks to ensure the quality of the concrete.

[0003] However, concrete production and quality management are currently handled by separate units within the production chain, resulting in a relatively loose relationship and poor coordination and management of quality issues. For example, concrete mixing plants used in major infrastructure projects are often located in high mountain valleys with limited network coverage and poor communication capabilities. Most work is done manually, and any manual oversight at any stage can result in substandard concrete quality, making traceability extremely difficult.

[0004] Digital twin technology focuses on physical entities in real-world scenarios. By establishing multi-physical field and multi-scale virtual models and using the mapping feedback of information between the two, the virtual model serves as a digital mirror of the physical entity, always maintaining consistency with the state of the physical entity. Through data fusion analysis, artificial intelligence and other means, it achieves the purpose of maintaining and optimizing the physical entity.

[0005] Digital twin technology can achieve management covering the entire production process, effectively improve the utilization of data in the concrete production process, and realize advanced judgment of concrete quality through the introduction of artificial intelligence, providing a basis for quality management and proportion optimization.

[0006] In view of the above technical defects of the existing technology, there is an urgent need to develop a concrete quality prediction method for mixing stations based on digital twins. Summary of the Invention

[0007] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the quality of concrete in a mixing station based on digital twins to solve the problems existing in the current production process in terms of concrete quality assurance.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] A method for predicting concrete quality of a mixing station based on digital twins, characterized by comprising the following steps:

[0010] 1) Obtain the attribute parameters of each physical entity of the mixing station and establish the geometric model of the digital twin of the mixing station;

[0011] 2) Real-time collection of production status data of the mixing station during operation and construction of a data model of the mixing station’s digital twin;

[0012] 3) Combining the data model with the geometric model to establish a digital twin of the mixing station;

[0013] 4) Acquire the production record data of the mixing station in real time, associate the production record data with the production status data according to the production batch, obtain detailed process record data of each production batch of concrete, store the detailed process record data, and generate an electronic ledger of concrete production;

[0014] 5) The digital twin of the mixing station processes the concrete production electronic ledger to obtain a production virtual image of the mixing station;

[0015] 6) Mining historical data from the concrete production electronic ledger, training a machine learning model online, and using the trained machine learning model to predict the quality of concrete produced subsequently.

[0016] Preferably, the processing of the concrete production electronic ledger by the digital twin of the mixing station also includes: determining whether there is any abnormality in the working status of the mixing station reflected in the concrete production electronic ledger, and if it is shown that the working status is abnormal, issuing a warning of the abnormal condition.

[0017] Preferably, the step 1) of obtaining the attribute parameters of each physical entity of the mixing station and establishing the geometric model of the digital twin of the mixing station specifically includes:

[0018] 1.1) Divide the mixing station into several sub-physical entities according to their functions, including the material storage system, metering system, conveying system, liquid supply system, pneumatic system, mixing system, main building, control room and dust removal system;

[0019] 1.2) Obtain the attribute parameters of the mixing station and construct a geometric model of the digital twin of the mixing station based on the attribute parameters. The attribute parameters include the appearance shape, size, internal structure, spatial position, posture and assembly relationship of each sub-physical entity.

[0020] Preferably, the production record data of the mixing station obtained in real time in step 4) includes raw material ratio information, raw material monitoring information, feeding error, concrete strength grade, slump and production volume.

[0021] Preferably, in step 6), mining historical data from the concrete production electronic ledger, training a machine learning model online, and using the trained machine learning model to predict the quality of subsequently produced concrete specifically includes:

[0022] 6.1) Constructing a data set using the production record data in the concrete production electronic ledger, and analyzing the variables that can significantly affect the various strength indicators of concrete using principal component analysis, and determining them as the main variables of the various strength indicators of concrete;

[0023] 6.2) Using an incremental machine learning algorithm to establish machine learning models for each strength index of concrete, and using the primary variables of each strength index of concrete as independent variables to train the machine learning models for each strength index of concrete;

[0024] 6.3) Use the trained machine learning model of various concrete strength indicators to predict the quality of subsequent concrete production.

[0025] Preferably, the various strength indicators of the concrete include compressive strength, tensile strength, frost resistance and impermeability.

[0026] Preferably, after the concrete production electronic ledger stores the detailed process record data of each production batch of concrete, the new main variables of the various strength indicators of concrete are extracted from the new detailed process record data, and the new main variables of the various strength indicators of concrete are used to train the machine learning model of the various strength indicators of concrete, thereby completing the parameter update of the machine learning model of the various strength indicators of concrete.

[0027] Preferably, the incremental machine learning algorithm is a Hoeffding tree algorithm, and the machine learning model of various strength indicators of concrete is based on the Hoeffding tree model by adding a sliding window. At the same time, the Hoeffding tree model is set to use the data in the sliding window to train a backup subtree in the background of each node. When the node splitting gain of the backup subtree is significantly greater than that of the current subtree, the node splitting is completed.

[0028] Preferably, the digital twin-based mixing station concrete quality prediction method further includes displaying the production virtual image and the prediction results of the subsequent concrete quality through the user front end.

[0029] Compared with the prior art, the concrete quality prediction method for a mixing plant based on digital twins of the present invention has one or more of the following beneficial technical effects:

[0030] 1. The present invention processes the electronic ledger data of the mixing station operation and the collected current status data based on the digital twin of the mixing station to obtain virtual image data of the mixing station, which can be fed back to the user front end and directly displayed to the operators at the concrete production site. This helps the operators to grasp the production status of the mixing station in real time, so that abnormal conditions generated in the mixing station are reflected through the collected status data, which facilitates the operators to discover faults in a timely manner.

[0031] 2. The present invention uses detailed data stored in the electronic ledger data of the mixing station operation and an incremental machine learning method to train the concrete strength prediction model online. The prediction model is closely integrated with the digital twin. By predicting the strength indicators of freshly mixed concrete, the operator is assisted in comparing and judging the quality of the concrete. The material ratio scheme can also be continuously optimized based on the error between the theoretical mixing strength and the predicted strength. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the concrete quality prediction method of a mixing station based on digital twin of the present invention.

[0033] Figure 2 It is a flow chart of the Hoffding tree algorithm adopted by the present invention.

[0034] Figure 3 Taking the prediction of concrete compressive strength as an example, the comparison chart of the prediction index results of the Hoffding tree algorithm before and after 300 iterations is shown. DETAILED DESCRIPTION

[0035] The present invention will be further described below with reference to the accompanying drawings and examples, and the contents of the examples are not intended to limit the scope of protection of the present invention.

[0036] In response to the problems existing in concrete quality assurance in the current production process, the present invention provides a concrete quality prediction method for a mixing station based on digital twins, which can improve the visibility of mixing station operators on each production link, thereby helping to formulate a more accurate concrete mix ratio plan, realize the refined management of the production process, and solve the problems of blind production and extensive management in the current concrete mixing stations.

[0037] Figure 1 The flowchart of the method for predicting the quality of concrete of a mixing station based on digital twins of the present invention is shown. Figure 1 As shown, the concrete quality prediction method of a mixing station based on digital twin of the present invention includes the following steps:

[0038] 1. Obtain the attribute parameters of each physical entity of the mixing station and establish the geometric model of the digital twin of the mixing station.

[0039] Specifically, the geometric model of the mixing station digital twin is established as follows;

[0040] 1. The concrete mixing station is divided into several sub-physical entities according to their functions. These sub-physical entities include the storage system, metering system, conveying system, liquid supply system, pneumatic system, mixing system, main building, control room, and dust removal system. They are used to complete various tasks such as storage, metering, conveying, mixing, discharging, and control of concrete raw materials.

[0041] 2. Obtain the attribute parameters of the mixing station and construct a geometric model of the mixing station digital twin based on the attribute parameters. The attribute parameters include the appearance, size, internal structure, spatial position, posture and assembly relationship of each sub-physical entity.

[0042] The property parameters of the mixing station can be obtained from the design documents of the mixing station, or can be obtained through measurement.

[0043] 2. Collect the production status data of the mixing station in real time during operation and build a data model of the digital twin of the mixing station.

[0044] In the present invention, various sensors arranged in various sub-physical entities of the mixing station can be used to obtain production status data of the concrete mixing station during operation. With the production status data, a data model of the digital twin of the mixing station can be constructed.

[0045] 3. Combine the data model with the geometric model to establish a digital twin of the mixing station.

[0046] According to the preset mapping relationship, the collected production status data is substituted into the geometric model, and the status of the geometric model is dynamically updated to obtain the digital twin of the mixing station.

[0047] 4. Acquire the production record data of the mixing station in real time, associate the production record data with the production status data according to the production batch, obtain detailed process record data of each production batch of concrete, store the detailed process record data, and generate an electronic ledger of concrete production.

[0048] In the present invention, the real-time production record data of the mixing station includes raw material ratio information, raw material monitoring information, feeding error, concrete strength grade, slump and production volume, etc. The production capacity of the mixing station can be obtained through the production record data.

[0049] 5. The digital twin of the mixing station processes the concrete production electronic ledger to obtain a virtual production image of the mixing station.

[0050] The concrete production electronic ledger is input into the digital twin of the mixing station in real time, and the production status and production records of the digital twin of the mixing station are updated in real time, so that a virtual production image of the mixing station can be obtained.

[0051] Of course, the processing of the concrete production electronic ledger by the digital twin of the mixing station also includes: determining whether there is any abnormality in the working status of the mixing station reflected in the concrete production electronic ledger, and if it is shown that the working status is abnormal, issuing a warning of the abnormal situation.

[0052] In this invention, the operating status of the concrete mixing plant can be determined based on pre-set rules. For example, by checking whether the air pressure and concrete arch pressure are within a reasonable range, the normal operation of the air compressor can be determined; by testing the liquid flow rate, the unobstructed operation of the water supply system and the admixture system can be determined. This determination can record and issue alerts for any abnormal operating conditions of the concrete mixing plant's sub-physical entities.

[0053] 6. Mining historical data from the concrete production electronic ledger, training a machine learning model online, and using the trained machine learning model to predict the quality of subsequently produced concrete.

[0054] In the present invention, mining historical data from the concrete production electronic ledger, training a machine learning model online, and using the trained machine learning model to predict the quality of subsequently produced concrete specifically include:

[0055] 1. A data set was constructed using the production record data in the concrete production electronic ledger. The variables that significantly affected the various strength indicators of concrete were analyzed using principal component analysis, and the variables were identified as the main variables of the various strength indicators of concrete.

[0056] Among them, the various strength indicators of the concrete include compressive strength, tensile strength, frost resistance and impermeability.

[0057] Through the principal component analysis method, variables that can significantly affect the compressive strength, tensile strength, frost resistance, impermeability, etc. of concrete can be obtained from the data set constructed using the production record data in the concrete production electronic ledger, and used as the main variables.

[0058] That is, each strength index of the concrete, such as compressive strength, tensile strength, frost resistance or impermeability, is used as a prediction target, and the principal component analysis method is used to analyze which data in the data set has a significant impact on the prediction target, and use it as the main variable of the prediction target.

[0059] Specifically, a data set is constructed using existing records in the concrete production electronic ledger, and the variance contribution rate of each principal component is calculated using the principal component analysis method, and the principal component with the highest contribution rate is selected as the main variable.

[0060] Extract the existing records in the concrete production electronic ledger at the current moment, and use the variables therein to construct a sample matrix. The constructed sample matrix is ​​a set of n p-dimensional vectors. The sample matrix elements are standardized as shown in formula (1):

[0061]

[0062] Among them, x ij represents the i-th value in the j-th feature, represents the average value of the jth feature, Get the normalized matrix Z.

[0063] For the standardized matrix Z, the correlation coefficient matrix is ​​calculated as shown in formula (2):

[0064]

[0065] in

[0066] Solve the characteristic equation of the sample correlation matrix R |R-λI p |=0, p characteristic roots are obtained, and m principal components whose principal component information utilization rate reaches the set index are determined.

[0067] 2. An incremental machine learning algorithm is used to establish machine learning models for each strength index of concrete as prediction models for each strength index. The main variables of each strength index of concrete are used as independent variables to train the machine learning models of each strength index of concrete (that is, the prediction models of each strength index).

[0068] In this invention, an incremental machine learning algorithm, the Hoeffding tree algorithm, is used to establish machine learning models for various concrete strength indicators, namely, compressive strength machine learning models, tensile strength machine learning models, frost resistance machine learning models, and anti-permeability strength machine learning models. These models serve as prediction models for compressive strength, tensile strength, frost resistance, and anti-permeability strength. Prediction models for these strength indicators are trained based on existing data, namely, the primary variables for each concrete strength indicator.

[0069] The Hoeffding tree model uses the Hoeffding bound to determine the optimal splitting attributes on nodes in relatively small-scale training data. The expression of the Hoeffding bound is shown in formula (3):

[0070]

[0071] Consider a real-valued random variable r with a range of R. Suppose that n independent observations have been made on this variable and their mean is calculated. According to the Hoefding inequality, under the premise of setting the probability size to 1-δ, the mean calculation result of the variable is at least

[0072] The Hoeffding bound has the property of being independent of the probability distribution that generates the observations. The cost of this property is that its bound is more conservative than the bound that depends on the distribution, and more observations are required to achieve the same δ and ε. Let G(X i ) is the calculation method of information gain, where X a and X b are the two attributes with the largest information gain, and the information gain difference between the two attributes is expressed as As shown in formula (4):

[0073]

[0074] If the information gain difference between two attributes is greater than the Hoeffding bound, that is Then it can be proved that X a The confidence level of the attribute with the largest information gain is 1-δ, so we can choose X a As a split point, the leaf node is turned into a branch node.

[0075] In the Hoeffding tree model, information gain is used to select the best split attributes on the root node and internal nodes. Information gain IG is the difference between the information entropy H(D) and the corresponding conditional entropy H(D|X). The information entropy and conditional entropy are calculated as shown in Equations (5), (6), and (7):

[0076]

[0077]

[0078] IG=H(D)-H(D|X) (7)

[0079] When constructing the Hoeffding tree model, after observing m independent objects at a leaf node, let X a The attribute with the highest IG value IG(X a ), X b The second highest IG value attribute IG(X b ), then through IG(X a ) minus IG(X b ) can get a new variable ΔIG. If ΔIG is greater than ε, then select X aAs a splitting attribute, if ΔIG is not obvious, it takes a long time to determine the best splitting attribute. When the present invention applies the Hoffding tree model to process the regression problem, the variance reduction function is selected to replace the information gain function in the classification tree, as shown in equations (8) and (9), let S be the data in the node, the total amount of data is N, select the attribute A of the data and use h A The boundary can divide the current node into two parts, where S R and S L They are the data in two parts, and the data volume is N R and N L , that is, S=S L +S R , N=N L +N R .

[0080]

[0081]

[0082] The Hoeffding tree algorithm is based on the assumption that the data stream it analyzes and processes is distributed in a stationary manner. The Hoeffding tree model itself lacks a design to update itself for outdated samples, resulting in its inability to address sample data feature shifts in the data stream. The present invention adds a sliding window structure to the Hoeffding tree model, enabling it to continuously update the data range of interest. The Hoeffding tree model also uses the data within the sliding window to train a backup subtree in the background of each node. When the node splitting gain of the backup subtree is significantly greater than that of the current subtree (for example, three times that of the current subtree), the node is split. Figure 2 Shown is a flow chart of the Hoffding tree algorithm of the present invention.

[0083] In the improved Hoffding tree model of the present invention, each internal node in the tree has a list of replacement subtrees. The subtree starts training when concept drift occurs in the data stream, that is, when it is found that the information gain of a certain other attribute on the node is better than the current attribute gain, a replacement subtree is generated, and the new attribute gain difference satisfies and The improved Hoeffding tree model can replace the subtrees at the same time, which can protect the model from the influence of the earlier outdated data while keeping the prediction model lightweight. Taking the concrete compressive strength prediction model as an example, the prediction index results of the improved model before and after 300 iterations are compared. Figure 3 shown.

[0084] In summary, the Hoeffding tree model is a classic incremental online learning algorithm. Since online learning differs from offline learning in that it uses batch data to adjust model parameters, making the model more susceptible to instability, the Hoeffding tree algorithm uses a Hoeffding bound to ensure that node samples can approximate the overall distribution with arbitrary precision. The Hoeffding tree itself is used for classification problems. This invention, by modifying the original algorithm's information gain function, makes the modified Hoeffding tree model applicable to regression problems such as concrete strength index prediction.

[0085] At the same time, the Hoeffding tree model itself does not have a pruning design, which means that the algorithm can only process smoothly distributed data streams. When the data stream changes dynamically, the Hoeffding tree model cannot adjust the outdated parts in the tree structure. The present invention mainly improves this algorithm in two aspects. First, a concept drift detector (sliding window structure) is added to the data stream, so that the model can only consider the recent data contained in the sliding window, while the previous model needs to consider all historical data on the node. The improvement makes the model immune to the influence of outdated data. Secondly, when the gain of the current attribute in the node decreases, the model starts to train a backup subtree for the corresponding node in the background. When the gain of the attribute selected by the backup subtree is significantly higher than the gain of the current attribute, the subtree is replaced. This is very similar to the process of splitting a new node, but the requirements are more stringent, thereby effectively preventing the generation of too many replacement subtrees. The splitting and growth method of the replacement subtree is the same as that of the decision tree. Each replacement subtree will obtain the data stream obtained by its corresponding node for subtree training. Through the above improvements, the adaptability of the prediction model is substantially enhanced, the lightweight of the model is maintained, and better prediction results are achieved in tests using existing data, such as concrete compressive strength.

[0086] In addition, in the present invention, preferably, after the concrete production electronic ledger stores the detailed process record data of each production batch of concrete, the new main variables of the various strength indicators of concrete are extracted from the new detailed process record data, and the new main variables of the various strength indicators of concrete are used to train the machine learning model of the various strength indicators of concrete, thereby completing the parameter update of the machine learning model of the various strength indicators of concrete.

[0087] Therefore, each time the concrete production electronic ledger completes the record of concrete production, the principal component feature vector is extracted from the new record as a new sample, and the new sample is used to train the prediction model to complete the parameter update of the model, so that the model is more in line with the actual situation.

[0088] 3. Use the trained machine learning model of various concrete strength indicators to predict the quality of subsequent concrete production.

[0089] After establishing and training the machine learning models for various strength indicators of concrete, the main variables of the various strength indicators of concrete can be extracted from the concrete production electronic ledger generated in real time during the production process, and input into the corresponding machine learning model to obtain the concrete quality prediction results of the mixing station.

[0090] To implement the digital twin-based concrete quality prediction method of the present invention, a corresponding digital twin-based concrete quality prediction system for the mixing station is required. The prediction system includes a data acquisition terminal, a server terminal, a user front-end, and a back-end feedback terminal.

[0091] The data acquisition end includes sensors and related intelligent terminal devices distributed in each sub-physical entity of the concrete mixing station. The data acquired by the data acquisition end is transmitted to the server end in real time. The server end completes the real-time mapping of the data in the digital twin according to the pre-established mapping relationship of the digital twin, warns of the abnormal production status of the mixing station, and sends an abnormal signal to the mixing station main console by the background feedback end. The server end also associates the production status data with the production record data and saves them in the concrete production electronic ledger, and extracts the data in the electronic ledger for training the prediction model. The operator can view the current production status of the mixing station and the prediction results of the strength index of the concrete in the user front end, thereby solving the problems of blind production and extensive management in traditional concrete production to a certain extent.

[0092] The above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications based on the above description are possible. It is not possible to enumerate all embodiments here. Any obvious variations or modifications arising from the technical solutions of the present invention remain within the scope of protection of the present invention.

Claims

1. A method for predicting concrete quality of a mixing station based on digital twin, characterized in that: The following steps are involved: 1) Obtain the attribute parameters of each physical entity of the mixing station and establish the geometric model of the mixing station digital twin, which specifically includes: 1.1) Divide the mixing station into several physical sub-entities according to their functions, including the material storage system, metering system, conveying system, liquid supply system, pneumatic system, mixing system, main building, control room and dust removal system; 1.2) Obtaining the attribute parameters of the mixing plant and constructing a geometric model of the mixing plant digital twin based on the attribute parameters. The attribute parameters include the appearance, size, internal structure, spatial position, posture and assembly relationship of each sub-physical entity; 2) Real-time collection of production status data of the mixing station during operation to build a data model of the mixing station’s digital twin; 3) combining the data model with the geometric model to establish a digital twin of the mixing station; 4) Real-time acquisition of production record data from the mixing station, associating the production record data with the production status data according to production batches, obtaining detailed process record data for each production batch of concrete, storing the detailed process record data, and generating an electronic ledger for concrete production. The real-time acquired production record data from the mixing station includes raw material ratio information, raw material monitoring information, feeding error, concrete strength grade, slump, and production volume; 5) The digital twin of the mixing station processes the concrete production electronic ledger to obtain a production virtual image of the mixing station; 6) Mining historical data from the concrete production electronic ledger, training a machine learning model online, and using the trained machine learning model to predict the quality of subsequently produced concrete, specifically including: 6.1) Constructing a dataset using the production record data in the concrete production electronic ledger, and using principal component analysis to analyze the variables that significantly affect each strength index of concrete, identifying them as the primary variables for each strength index of concrete; 6.2) Using an incremental machine learning algorithm to establish machine learning models for each concrete strength indicator, and using the primary variables of each concrete strength indicator as independent variables to train the machine learning models for each concrete strength indicator, the incremental machine learning algorithm is a Hoeffding tree algorithm. The machine learning models for each concrete strength indicator are based on the Hoeffding tree model with a sliding window added. The Hoeffding tree model is configured to train a backup subtree using data within the sliding window in the background of each node. When the node splitting gain of the backup subtree is significantly greater than that of the current subtree, the node is split. 6.3) Use the trained machine learning model of various concrete strength indicators to predict the quality of subsequent concrete production.

2. The method for predicting concrete quality of a mixing station based on digital twin according to claim 1 is characterized in that: The processing of the concrete production electronic ledger by the digital twin of the mixing station also includes: determining whether there is any abnormality in the working status of the mixing station reflected in the concrete production electronic ledger, and if it is shown that the working status is abnormal, issuing a warning of the abnormal condition.

3. The method for predicting concrete quality of a mixing station based on digital twin according to claim 2 is characterized in that: The various strength indicators of the concrete include compressive strength, tensile strength, frost resistance and impermeability.

4. The method for predicting concrete quality of a mixing station based on digital twin according to claim 3 is characterized in that: After the concrete production electronic ledger stores detailed process record data of each production batch of concrete, new main variables of various strength indicators of concrete are extracted from the new detailed process record data, and the new main variables of various strength indicators of concrete are used to train the machine learning model of various strength indicators of concrete, thereby completing the parameter update of the machine learning model of various strength indicators of concrete.

5. The method for predicting concrete quality of a mixing station based on digital twin according to any one of claims 1 to 4, characterized in that: It further includes displaying the production virtual image and the prediction results of the subsequent concrete quality through the user front end.

Images