Pre-stirring concrete quality prediction method driven by multi-dimensional label-free data
By adopting a multi-dimensional label-free data-driven method in concrete quality prediction and using unsupervised machine learning and statistical models to build a hybrid model, the problem of dependence on label data in the prior art is solved, and accurate concrete quality prediction in the absence of sufficient label data is achieved.
Patent Information
- Application Number
- CN202411974266.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-06-10
AI Technical Summary
The prior art relies on a large amount of labeled data in concrete quality prediction, making it difficult to achieve accurate predictions in the absence of sufficient labeled data.
A multi-dimensional label-free data-driven method is used to construct a mixed model through unsupervised machine learning anomaly detection algorithm and statistical model, generate ready-mixed concrete quality labels, and train them in combination with deep learning models to obtain a concrete quality prediction model before mixing.
In the absence of sufficient label data, the concrete quality can be accurately identified and the workingability and mechanical properties of the concrete can be predicted in advance, which improves the accuracy and robustness of the prediction and reduces the quality risks in the production process.
Smart Images
Figure CN120126618A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of concrete quality prediction, and in particular to a method for predicting concrete quality before mixing driven by multi-dimensional unlabeled data. Background Art
[0002] As the largest building material, the quality of concrete plays a fundamental and decisive role in the quality of building structures. However, the quality control of ready-mixed concrete is lagging and strongly dependent on experience. The control process usually occurs after the mixing is completed and cannot be synchronized with the actual production needs. Therefore, a technical means is urgently needed to accurately judge the quality of concrete before mixing is completed, so as to assist in the optimization and adjustment of the production process.
[0003] Artificial intelligence technology can provide intelligent solutions to complex problems with objectivity and consistency through training on a large amount of actual data. In terms of industrial quality inspection, artificial intelligence technology has been widely used, such as real-time monitoring of various parameters in the production process through analysis of sensor data, image recognition and other means, predicting equipment failures in advance, reducing downtime, etc. For the concrete industry, artificial intelligence technology can help achieve accurate prediction and real-time monitoring of concrete quality, identify potential problems in advance, optimize production processes, reduce production costs, and ensure the performance and service life of concrete.
[0004] 202310557712.2 discloses a method for predicting the quality of concrete in a mixing plant based on digital twins. This method trains a machine learning model through the electronic ledger data of the mixing plant production to predict the quality of concrete produced in the subsequent production. However, this method has a strong dependence on labeled data, which is contrary to the fact that there are few labels for the actual quality data of the concrete ready-mix plant, making it difficult to obtain an accurate training model. Summary of the invention
[0005] The present disclosure provides a multi-dimensional unlabeled data-driven method for predicting the quality of concrete before mixing, which aims to solve the limitation of lack of sufficient labeled data in concrete quality prediction, and provide an effective method for predicting the quality of concrete before mixing. The method first constructs a hybrid model based on an unsupervised machine learning anomaly detection algorithm and a statistical model to generate ready-mixed concrete quality labels. Then, these labels are combined with the original data before mixing to form a data set of pre-mixing data-labels. Finally, the data set is trained using a deep learning model to obtain an efficient model for predicting the quality of concrete before mixing. The method can accurately predict the quality of concrete by analyzing the multi-dimensional data before mixing, thereby effectively controlling the quality of concrete before production, avoiding cost waste caused by later adjustments, and improving production efficiency.
[0006] The multi-dimensional unlabeled data driven method for predicting the quality of concrete before mixing provided by the present disclosure mainly includes the following methods:
[0007] S1. Acquire data before and during mixing of ready-mixed concrete, wherein the data before mixing includes industrial control data of the mixing station and raw material sensor data, and the data during mixing is arranged sensor data;
[0008] S2, perform data analysis, data preprocessing and feature engineering on the collected multidimensional data;
[0009] S3. Use the data before stirring to perform unsupervised machine learning anomaly detection training and establish an unsupervised anomaly recognition model before stirring; perform statistical classification on the data during stirring and establish a mathematical statistical model for anomaly recognition during stirring; compare the anomaly recognition results of the two models and conduct a preliminary feasibility analysis;
[0010] S4. After feasibility analysis, unsupervised machine learning anomaly detection training is performed using the stirring data to obtain a corresponding unsupervised anomaly recognition model, and the model is coupled with a mathematical statistical model for anomaly recognition during stirring to obtain a hybrid model for anomaly recognition during stirring;
[0011] S5. Use real experimental data to verify the accuracy of the mixing model for abnormality identification during mixing;
[0012] S6, using the hybrid model for identifying abnormalities during mixing to generate labels for the quality of ready-mixed concrete, combining the processed pre-mixing data and labels to form a data set, and performing deep learning network model training to obtain a pre-mixing concrete quality prediction model.
[0013] S7, the data before the actual production of the ready-mixed concrete plant is imported into the pre-mixing concrete quality prediction model, so as to determine whether the workability of the ready-mixed concrete obtained in this production meets the standard.
[0014] Compared with the prior art, the beneficial effects of the present invention are:
[0015] (1) Unlabeled data driven: This method does not rely on traditional labeled data. Through anomaly detection and statistical analysis based on unsupervised learning, it can accurately identify the quality of concrete in the absence of sufficient labeled data. This breaks the reliance of traditional methods on a large amount of labeled data and improves the breadth and adaptability of application scenarios.
[0016] (2) Predicting concrete quality in advance: This method can predict whether the workability and mechanical properties of concrete are qualified by analyzing the data before mixing and the abnormal conditions during mixing before the concrete mixing is completed;
[0017] (3) Fusion of multidimensional data sources: This method comprehensively improves the accuracy and robustness of prediction by fusing multidimensional data sources (such as industrial control data, raw material parameters, sensor data, etc.) and combining unsupervised anomaly detection and statistical models. It can effectively cope with complex production environments and variable raw material characteristics.
[0018] (4) Hybrid model improves anomaly recognition accuracy: By combining the unsupervised anomaly recognition model with statistical methods (such as confidence interval analysis), a hybrid model is constructed for anomaly detection. This integrated model approach enhances the accuracy of model anomaly recognition.
[0019] (5) It can detect potential problems earlier in the production process, help optimize the production process in real time, avoid the lag of traditional methods that rely on post-mixing detection, improve production efficiency and product quality, and reduce quality risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] The above and other objects, features and advantages of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present disclosure.
[0021] Figure 1 is a flow chart of an exemplary embodiment according to the present disclosure. DETAILED DESCRIPTION
[0022] The preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the preferred embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.
[0023] The present disclosure provides a method for predicting the quality of concrete before mixing driven by multi-dimensional unlabeled data. In an exemplary embodiment, the process is as shown in the attached figure. Figure 1 As shown, it mainly includes the following steps:
[0024] 1. Obtaining data before and during mixing of ready-mixed concrete, wherein the data before mixing includes industrial control data of the mixing station and raw material sensor data, and the data during mixing is arranged sensor data;
[0025] 2. Perform data analysis, data preprocessing and feature engineering on the collected multidimensional data;
[0026] 3. Use the data before stirring to conduct unsupervised machine learning anomaly detection training, and perform statistical classification on the data during stirring. Establish an unsupervised machine learning anomaly recognition model before stirring and a mathematical statistical model for anomaly recognition during stirring. Then compare the anomaly recognition results of the two models and conduct a preliminary feasibility analysis.
[0027] 4. After feasibility analysis, unsupervised machine learning anomaly detection training is performed using stirring data to obtain the corresponding unsupervised anomaly recognition model. This model is coupled with the mathematical statistical model for anomaly recognition during stirring to obtain a hybrid model for anomaly recognition during stirring;
[0028] 5. By comparing with a small amount of real experimental data, the accuracy of the hybrid model for identifying abnormalities during mixing is verified and the model is optimized;
[0029] 6. Use the mixing anomaly recognition hybrid model to generate labels for the quality of ready-mixed concrete. Combine the processed pre-mixing data and labels to form a data set, and perform deep learning training to obtain a prediction model for the quality of mixing concrete.
[0030] 7. Import the actual production data of the ready-mixed concrete plant into the pre-mixing concrete quality prediction model to obtain the result of whether the workability of the ready-mixed concrete obtained in this production meets the standards.
[0031] The data of step 1 include: obtaining the mix ratio, designed mechanical properties and working performance of ready-mixed concrete from the industrial control system; obtaining the actual parameters of the raw materials used for the batch of concrete from the constructed sensors, such as the actual sand moisture content; and obtaining the vibration data, temperature data, current data, alkalinity data, real-time moisture content, etc. during the mixing of ready-mixed concrete.
[0032] The step 2 comprises the following steps:
[0033] 2-1. Perform signal processing on the three-dimensional vibration data, including vibration signal preprocessing and vibration signal analysis. The preprocessing technology first extracts the vibration signal during the stirring process according to the threshold, and then performs filtering, noise reduction and other processing on the extracted signal. Then, statistical analysis is performed on the processed signal to extract vibration characteristic parameters;
[0034] 2-2. Perform feature processing on the temperature data, extract the outlet temperature of the ready-mixed concrete machine and the ambient temperature, and obtain the temperature difference data of each data;
[0035] 2-3. Integrate the processed data into a database, and conduct in-depth analysis and processing of the database through box plots, heat distribution maps, kernel density distribution curves, multicollinearity, principal component analysis and other techniques;
[0036] 2-4. Use feature engineering to extract the most representative features from the processed database.
[0037] Preferably, the unsupervised machine learning anomaly detection training in step 3 includes but is not limited to tree-based iForest, density-based DBScan model and classification-based One-Class SVM.
[0038] Specifically, the step 3 includes the following steps:
[0039] 3-1. Based on the central limit theorem, the average value of a random variable tends to be normally distributed. By using the normal distribution characteristics of each variable X in the mixing, a confidence interval is constructed and statistical inference is performed. The confidence interval formula is:
[0040] μ-n×σ≤X≤μ+n×σ
[0041] Where μ is the mean of variable X, σ is the standard deviation of variable X, n = 1, 2, 3…;
[0042] If X is within the above confidence interval, the concrete quality is calibrated as "qualified", otherwise it is "unqualified", thereby obtaining the statistical calibration result of the concrete quality during mixing, and the result is used as the target label for subsequent feasibility verification;
[0043] 3-2. Use the pre-mixing data to train multiple unsupervised machine learning anomaly detection models to obtain the pre-mixing unsupervised anomaly recognition model.
[0044] The model 1 method is as follows:
[0045] Calculate the anomaly score s for each sample:
[0046]
[0047] where E(h(i)) is the average of h(x) from a set of isolated trees and C(n) is the average path length, calculated as:
[0048]
[0049] Here, H(i) is the harmonic number and c(n) is the average value of h(x) for a given n.
[0050] h(x)=e+c(Tsize)
[0051] Among them, h(x) represents the path length of point x, e is the number of edges that point x traverses from the root node until it terminates at an external node, and C(T.size) is a correction value.
[0052] The specific method of model 2 is as follows:
[0053] N ε (p)={q∈D|dist(p,q)≤ε}
[0054] Among them, N ε (p) is the ε-domain of point p; the ε-domain refers to the area of a given object within the radius ε; the dataset D is a given set of objects; p is a core point, q is a point in the dataset D and is within the domain of point p;
[0055] Model 1 is used to calculate the sample anomaly score s, and model 2 is used to determine whether the sample is a core point or a noise point, thereby identifying whether the quality of each concrete sample is abnormal;
[0056] 3-3. The statistical calibration results of concrete quality during mixing are used as target labels, and compared with the concrete quality detection results of the unsupervised anomaly recognition model before mixing. The accuracy of the model is quantified through evaluation indicators such as accuracy, recall, and F1-Score, so as to determine the feasibility of pre-mixing concrete quality prediction based on unsupervised machine learning anomaly detection;
[0057] The calculation methods of the evaluation indicators are:
[0058]
[0059] Among them, TP is the true positive example, that is, the number of samples that are actually qualified; TN is the true negative example, that is, the number of samples that are actually unqualified; FP is the false positive example, that is, the number of samples that are predicted to be qualified but are actually unqualified; FN is the false negative example, that is, the number of samples that are predicted to be unqualified but are actually qualified; Precision refers to the proportion of all samples classified as positive that are actually positive; β is the adjustment parameter of F-Score, which is used to balance the importance between Precision and Recall;
[0060] If the concrete quality detection result evaluation indicators of the unsupervised anomaly recognition model before mixing are all higher than the set threshold value, the feasibility is determined to be passed, confirming the feasibility of the prediction of concrete quality before mixing based on unsupervised machine learning anomaly detection.
[0061] Furthermore, the step 6 specifically includes:
[0062] The database data is input into the mixing anomaly recognition hybrid model to generate labels, and the label result of whether the workability of each ready-mixed concrete is qualified is obtained;
[0063] Match the acquired labels with the pre-mixing data to obtain a concrete pre-mixing database with labels;
[0064] The labeled database is used to train the deep learning network model to obtain a concrete quality prediction model before mixing.
[0065] Preferably, the deep learning network model in step 6 includes: a classification model, an anomaly recognition model, etc.
[0066] The quality in step 6 includes: concrete workability, mechanical properties, etc.
[0067] The pre-mixing concrete quality prediction model obtained in this embodiment can perceive the concrete quality in advance based on the pre-mixing data, which can effectively reduce the quality risk in the production process, improve production efficiency, reduce production costs, and optimize the production process.
[0068] The above technical scheme is only an exemplary embodiment of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the application methods and principles disclosed in the present invention, and it is not limited to the method described in the above specific embodiment of the present invention. Therefore, the method described above is only preferred and does not have a restrictive meaning.
Claims
1. A multi-dimensional unlabeled data driven method for predicting the quality of concrete before mixing, comprising the following steps: S1. Acquire data before and during mixing of ready-mixed concrete, wherein the data before mixing includes: mixing station industrial control data and raw material sensor data, and the data during mixing includes: arranged sensor data; S2, perform data analysis, data preprocessing and feature engineering on the collected multidimensional data; S3. Perform unsupervised anomaly detection training on the data before stirring to establish an unsupervised anomaly recognition model before stirring; perform statistical classification on the data during stirring to establish a mathematical statistical model for anomaly recognition during stirring; compare the anomaly recognition results of the two models to conduct a preliminary feasibility analysis; S4. After feasibility analysis, unsupervised anomaly detection training is performed using the stirring data to obtain a corresponding unsupervised anomaly recognition model, and the model is coupled with a mathematical statistical model for abnormality recognition during stirring to obtain a hybrid model for abnormality recognition during stirring; S5. Use real experimental data to verify the accuracy of the mixing model for abnormality identification during mixing; S6, using the hybrid model for identifying abnormalities during mixing to generate labels for the quality of ready-mixed concrete, combining the processed pre-mixing data and labels to form a data set for deep learning network model training to obtain a pre-mixing concrete quality prediction model.
2. The method according to claim 1, characterized in that The data of step S1 includes: Data before mixing: Ready-mixed concrete mix proportions, time data, design mechanical properties and workability obtained from the industrial control system; The actual parameters of the raw materials used in this batch of concrete obtained from the installed sensors include: actual sand moisture content data; Stirring data: Vibration data, temperature data, current data, alkalinity data, and real-time moisture content during mixing of ready-mixed concrete.
3. The method according to claim 1 or 2, characterized in that: The step S2 specifically includes: S21, performing vibration signal preprocessing and vibration signal analysis on the three-dimensional vibration data, including: Extract the vibration signal during the stirring process according to the threshold; Filter and reduce noise on the extracted signal; Conduct statistical analysis on the processed signals and extract vibration characteristic parameters; S22, performing feature processing on the temperature data, extracting the outlet temperature of the ready-mixed concrete machine and the ambient temperature and obtaining temperature difference data for each piece of data; S23, integrating the processed data into a database, and analyzing the database through one or more of a box plot, a heat distribution diagram, a kernel density distribution curve, multicollinearity, and a principal component analysis; S24, feature engineering is used to extract representative features from the processed database.
4. The method according to claim 1, characterized in that In step S3 and step S4, the unsupervised machine learning anomaly detection model includes: any one of a tree-based iForest, a density-based DBScan model, and a classification-based One-Class SVM.
5. The method according to claim 4, characterized in that The step S3 specifically includes: S31, establish a mathematical statistical model for abnormal identification during stirring: Through the normal distribution characteristics of each variable X in stirring, a confidence interval is constructed and statistical inference is performed: The confidence interval is: μ-n×σ≤X≤μ+n×σ Where μ is the mean of variable X, σ is the standard deviation of variable X, n = 1, 2, 3…; If X is within the above confidence interval, the concrete quality is calibrated as "qualified", otherwise it is "unqualified". Thus, the statistical calibration result of the concrete quality during mixing is obtained, and the result is used as the target label for subsequent feasibility verification; S32, using the pre-stirring data to train multiple unsupervised machine learning anomaly detection models to obtain a pre-stirring unsupervised anomaly recognition model: The specific model 1 method is as follows: Calculate the anomaly score s for each sample: where E(h(i)) is the average value of h(x) from a set of isolated trees and C(n) is the average path length, calculated as: Where H(i) is the harmonic number, c(n) is the average value of h(x) for a given n; h(x)=e+c(Tsize) Among them, h(x) represents the path length of point x, e is the number of edges that point x traverses from the root node until it ends at an external node, and C(T.size) is a correction value; The specific method of model 2 is as follows: N ε (p)={q∈D∣dist(p,q)≤ε} Among them, N ε (p) is the ε-domain of point p; the ε-domain refers to the area of a given object within the radius ε; the dataset D is a given set of objects; p is a core point, q is a point in the dataset D and is within the ε-domain of point p; Model 1 is used to calculate the sample anomaly score s, and model 2 is used to determine whether the sample is a core point or a noise point, thereby identifying whether the quality of each concrete sample is abnormal; S33, using the statistical calibration results of the concrete quality during mixing as the target label, and comparing it with the concrete quality detection results of the unsupervised anomaly recognition model before mixing, and quantifying the accuracy of the model through evaluation indicators, so as to determine the feasibility of the prediction of concrete quality before mixing based on unsupervised machine learning anomaly detection; The calculation methods of the evaluation indicators are: Among them, TP is the true positive example, that is, the number of samples that are actually qualified; TN is the true negative example, that is, the number of samples that are actually unqualified; FP is the false positive example, that is, the number of samples that are predicted to be qualified but are actually unqualified; FN is the false negative example, that is, the number of samples that are predicted to be unqualified but are actually qualified; Precision refers to the proportion of all samples classified as positive that are actually positive; β is the adjustment parameter of F-Score, which is used to balance the importance between Precision and Recall; If the concrete quality detection result evaluation indicators of the unsupervised anomaly recognition model before mixing are all higher than the set threshold value, the feasibility is determined to be passed, confirming the feasibility of the prediction of concrete quality before mixing based on unsupervised machine learning anomaly detection.
6. The method according to claim 1, characterized in that The step S6 specifically includes: The database data is input into the mixing anomaly recognition hybrid model to generate labels, and the label result of whether the workability of each ready-mixed concrete is qualified is obtained; Match the acquired labels with the pre-mixing data to obtain a concrete pre-mixing database with labels; The deep learning network model is trained using a labeled database to obtain a concrete quality prediction model before mixing.
7. The method according to claim 1 or 6, characterized in that: The deep learning network model in step S6 includes: a classification model and an anomaly recognition model.
8. The method according to claim 1 or 6, characterized in that: The quality of the ready-mixed concrete in step S6 includes: concrete workability and mechanical properties.
9. The method according to claim 1, characterized in that: The following steps are also included: S7, importing the actual pre-production data of the ready-mixed concrete plant into the pre-mixing concrete quality prediction model to determine whether the workability of the ready-mixed concrete produced this time meets the standard.