A method, apparatus, equipment and medium for predicting defects in pultruded products based on multi-parameter collaborative fusion

Abstract

This application provides a method, apparatus, equipment, and medium for predicting defects in pultruded products based on multi-parameter collaborative fusion. This application relates to the field of pultrusion process inspection technology. The method includes: collecting operational status data during the pultrusion product manufacturing process; extracting multi-dimensional features from the operational status data within a time window and performing normalization processing to form a feature set for the time window; converting the feature set into feature vectors, inputting them into a pre-established defect prediction model, calculating the cosine similarity between the real-time feature vector and the feature vectors of historical defect cases, determining the target defect type and the probability of occurrence of the target defect type based on a similarity grading threshold; and constructing defect early warning information based on the target defect type and the probability of occurrence of the target defect type. This technical solution, through multi-parameter fusion, significantly improves the accuracy of defect prediction, reduces the false alarm rate, and the early warning information can clearly guide process adjustments.

Description

Technical Field

[0001] This application relates to the field of pultrusion process inspection technology, and in particular to a method, apparatus, equipment and medium for predicting defects in pultruded products based on multi-parameter collaborative fusion. Background Technology

[0002] Pultruded products have a wide range of applications, including insulating strips, ladder rungs, and more. However, defects such as voids, delamination, and poor fiber impregnation can directly affect product performance and lifespan. Therefore, quality control during the manufacturing process is of paramount importance.

[0003] Traditional defect detection methods largely rely on offline inspection after product molding, such as X-rays, ultrasonic scanning, and manual visual inspection. This not only suffers from significant time lag, but also means that by the time defects are discovered, a large number of scrap products may have already been lost, significantly increasing production costs. Therefore, there is an urgent need to solve the technical problem of detection lag, which is a core issue of concern to those skilled in the art. Application content

[0004] This application provides a method, apparatus, equipment, and medium for predicting defects in pultruded products based on multi-parameter collaborative fusion, aiming to solve the technical problem of lagging traditional detection in pultruded product manufacturing.

[0005] In a first aspect, embodiments of this application provide a method for predicting defects in pultruded products based on multi-parameter collaborative fusion, the method comprising: During the manufacturing process of pultruded products, operational status data is collected; wherein, the operational status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature; Based on a preset time window, multi-dimensional features are extracted from the running status data within the time window, and the extracted multi-dimensional features are normalized to form a feature set for the time window. The feature set is transformed into a feature vector, which is input into a pre-established defect prediction model. The cosine similarity between the real-time feature vector and the feature vector of historical defect cases stored in the defect prediction model is calculated. Based on the preset similarity grading threshold, the target defect type and the probability of occurrence of the target defect type are determined. Defect warning information is constructed based on the target defect type and the probability of occurrence of the target defect type, and the defect warning information is output.

[0006] In one feasible embodiment, multi-dimensional feature extraction is performed on the runtime status data within the time window, including: Calculate at least two of the mean, maximum, minimum, standard deviation, and range of each running status data within the time window to obtain multi-dimensional numerical features; A linear fitting algorithm is used to calculate the rate of change of each running state data within the time window, and multi-dimensional trend characteristics are obtained.

[0007] In one feasible embodiment, multi-dimensional feature extraction of the running status data within the time window further includes: The Pearson correlation coefficient algorithm is used to calculate the correlation coefficient between the rate of change of each operating status data within the time window, thus obtaining multi-dimensional correlation characteristics.

[0008] In one feasible embodiment, the training process of the defect prediction model includes: Collect a preset number of historical production data for pultruded products. The historical data includes normal production case data and defect case data. The historical data includes raw operating status data and corresponding defect labels or normal labels. The historical data is subjected to feature extraction and normalization processing to extract the historical feature set and transform it into a historical feature vector to form a training feature dataset; The SVM algorithm is used to train a support vector machine model with the training feature dataset as input and the defect label or normal label as the output target. The trained support vector machine model, along with its corresponding historical feature vectors and defect labels, are stored to form a defect prediction model.

[0009] In one feasible embodiment, the similarity grading threshold includes: a first threshold and a second threshold, wherein the first threshold is less than the second threshold; Based on a preset similarity grading threshold, the target defect type and its probability of occurrence are determined, including: If the cosine similarity is greater than or equal to the second threshold, then the corresponding target defect type and its probability of occurrence are output. If the cosine similarity is between the first threshold and the second threshold, the suspected defect type and its probability of occurrence are output, and key monitoring parameters are prompted; wherein, the key monitoring parameters are determined based on a pre-established table of the correlation between defect types and key parameters; If the cosine similarity is less than or equal to the first threshold, then the output is "no defect risk".

[0010] In one feasible embodiment, the method further includes: If the target defect type is poor fiber impregnation, an adjustment prompt message for adjusting the traction speed is generated; If the target defect type is an internal void defect, then adjustment prompts will be generated to adjust the resin tank temperature, mold inlet temperature, mold middle section temperature and mold outlet temperature. If the target defect type is a delamination defect, an adjustment prompt message for adjusting the yarn roller stress is generated.

[0011] In one feasible embodiment, the method further includes: The validity of the operation status data is verified. If there is missing operation status data or the data exceeds a preset reasonable range, linear interpolation is performed using the first preset number of adjacent valid data to obtain the completion result of the operation status data. If data for the same operating status is missing or exceeds a preset reasonable range within a second preset number of consecutive time windows, a sensor fault message will be generated.

[0012] Secondly, embodiments of this application provide a defect prediction device for pultruded products based on multi-parameter collaborative fusion, the device comprising: The operation status data acquisition module is used to collect operation status data during the manufacturing process of pultruded products; wherein, the operation status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature. The feature extraction module is used to extract multi-dimensional features from the running status data within a time window based on a preset duration, and to normalize the extracted multi-dimensional features to form a feature set for the time window. The target defect type prediction module is used to convert the feature set into a feature vector, input it into a pre-established defect prediction model, calculate the cosine similarity between the real-time feature vector and the feature vector of historical defect cases stored in the defect prediction model, and determine the target defect type and the probability of occurrence of the target defect type according to a preset similarity grading threshold. The defect warning information generation module is used to construct defect warning information based on the target defect type and the probability of occurrence of the target defect type, and output the defect warning information.

[0013] Thirdly, embodiments of this application provide a pultrusion product defect prediction device based on multi-parameter collaborative fusion. The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement any of the pultrusion product defect prediction methods based on multi-parameter collaborative fusion in the above embodiments.

[0014] Fourthly, embodiments of this application provide a computer storage medium storing computer program instructions. When the computer program instructions are executed by a processor, they implement any of the multi-parameter collaborative fusion-based pultrusion product defect prediction methods described in the above embodiments.

[0015] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements any of the multi-parameter collaborative fusion-based pultrusion product defect prediction methods described in the above embodiments.

[0016] This application presents a multi-parameter collaborative fusion-based method, apparatus, equipment, and medium for predicting defects in pultruded products. During the pultrusion product manufacturing process, operational status data is collected. This operational status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multiple temperature data segments. The multiple temperature data segments include at least two of the following: resin tank temperature, die inlet temperature, die mid-section temperature, and die outlet temperature. Based on a preset time window, multi-dimensional features are extracted from the operational status data within the time window, and the extracted multi-dimensional features are normalized to form a feature set for the time window. This feature set is converted into a feature vector, input into a pre-established defect prediction model, and the cosine similarity between the real-time feature vector and the historical defect case feature vectors stored in the defect prediction model is calculated. Based on a preset similarity grading threshold, the target defect type and its probability of occurrence are determined. Defect warning information is constructed based on the target defect type and its probability of occurrence, and the defect warning information is output. This solution can accurately predict defects through multi-parameter fusion, meeting the real-time requirements of defect handling and solving problems such as high scrap rates and increased material costs caused by delayed defect handling. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for predicting defects in pultruded products based on multi-parameter collaborative fusion, as provided in an embodiment of this application. Figure 2 This is a schematic diagram of a multi-parameter-based defect prediction process for pultruded products provided in one embodiment of the application. Figure 3 This is a schematic diagram of the structure of a pultrusion product defect prediction device based on multi-parameter collaborative fusion provided in an embodiment of this application; Figure 4This is a schematic diagram of the hardware structure of the pultrusion product defect prediction device based on multi-parameter collaborative fusion provided in the embodiments of this application. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit it. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] To address the problems in existing technologies, this application provides a method, apparatus, equipment, and medium for predicting defects in pultruded products based on multi-parameter collaborative fusion. This technical solution deploys corresponding sensors at key locations to collect operational status data such as roller stress, traction speed, vibration, and multiple temperature ranges. Within a specific time window, statistical, trend, and correlation features are extracted and normalized to form a feature set. This feature set is then transformed into a feature vector and input into a support vector machine model trained on historical data. Through cosine similarity comparison and preset threshold grading, early warning information such as defect type and probability of occurrence is output, achieving early warning and precise control of defects in pultruded products during the production process.

[0022] The following section first introduces the defect prediction method for pultruded products based on multi-parameter collaborative fusion provided in the embodiments of this application.

[0023] Figure 1 This is a flowchart illustrating a method for predicting defects in pultruded products based on multi-parameter collaborative fusion, as provided in an embodiment of this application. Figure 1 As shown, the method may include the following steps: S101, During the manufacturing process of pultruded products, operational status data is collected; wherein, the operational status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature. Pultruded products are composite material products made by combining fiber materials and resin materials using the pultrusion process. They can be various products produced using the pultrusion process, such as insulating strips, ladder rungs, and building support components.

[0024] Operational status data can be various quantitative data that reflect the equipment operation status, material processing status, and process execution status throughout the entire pultrusion product manufacturing process. Specifically, it can be at least two of the following: roller stress, traction speed, equipment vibration data, and multiple temperature data, used to comprehensively depict the real-time situation of the production process.

[0025] Roller stress is data related to the tension experienced by the rollers in pultrusion equipment during the conveying of fiber materials. It can be the feedback value of the tensile force generated on the rollers during fiber material conveying, and directly reflects the stress state of the fiber material during conveying.

[0026] Traction speed is the rate at which the traction mechanism in a pultrusion machine moves the composite material being molded. It can be the distance the traction mechanism transports material per unit time and is a key process parameter that affects the molding efficiency and quality of the product.

[0027] Equipment vibration data refers to the vibration signals generated by key components such as the die support and traction mechanism during the operation of pultrusion equipment. These data can include characteristic parameters such as vibration amplitude and frequency, and can reflect the operational stability of the equipment.

[0028] Multi-segment temperature data refers to the quantified temperature data at multiple key process locations during the pultrusion product manufacturing process. It can be at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature. It directly affects the resin curing effect and product molding quality.

[0029] The resin bath temperature is the temperature data of the environment in which the tank containing the resin material used to impregnate the fiber is located. It can be a key temperature parameter required for the resin to maintain its fluidity and wettability, and determines the bonding effect between the resin and the fiber.

[0030] The die inlet temperature is the temperature data at the inlet position of the pultrusion die. It can be the preheating temperature of the material at the contact position before entering the die, laying the temperature foundation for subsequent molding in the die.

[0031] The temperature of the middle section of the mold is the temperature data of the middle part of the pultrusion mold. It can be the temperature of the core reaction area during the material forming process inside the mold, which directly affects the progress and effect of the resin curing reaction.

[0032] The mold exit temperature is the temperature data at the exit position of the pultrusion mold. It can be the temperature when the molded product leaves the mold, reflecting the final temperature state of the product after solidification.

[0033] This solution utilizes dedicated sensors deployed at key locations. Specifically, a high-precision tension sensor can be deployed on the roller frame to collect roller stress at a sampling frequency of 10Hz. The traction mechanism uses a built-in encoder to collect traction speed data at a sampling frequency of 5Hz. Vibration sensors are installed on the mold support and traction mechanism to collect equipment vibration data at a sampling frequency of 20Hz. Two thermocouple sensors are deployed at the resin tank, mold inlet, mold middle section, and mold outlet to collect temperature data at the corresponding locations, with a sampling frequency of 5Hz. Recommended sensor models can be selected. For example, the HBMT12 can be used for the tension sensor, the PCB352C65 for the vibration sensor, and the Omega K-type thermocouple sensor.

[0034] S102, Based on a preset time window, multi-dimensional features are extracted from the running status data within the time window, and the extracted multi-dimensional features are normalized to form a feature set of the time window. The preset time window can be a fixed time interval set in advance to divide the continuous running state data. It can be a reasonable duration such as 10s or 15s, used to divide the continuous data stream into data segments that can be analyzed independently.

[0035] Multi-dimensional feature extraction can be a process of mining different dimensions and types of feature information from operational status data within a time window. It can extract various types of features such as statistical features, trend features, and correlation features to comprehensively capture the process change patterns behind the data.

[0036] This solution can convert the extracted multi-dimensional feature values ​​into a standardized processing method with a unified numerical range. It can map the feature values ​​to the 0-1 range to eliminate the influence of differences in the units and numerical ranges of different features.

[0037] A feature set can be a collection of various effective features that have been extracted and normalized through multi-dimensional feature extraction. It can be a structured data set composed of statistical feature values, trend feature values, correlation feature values, etc., which provides a foundation for subsequent model input.

[0038] Based on a preset time window, the continuously collected operational status data can be segmented according to a pre-set fixed time interval (such as 10 seconds). This solution allows the data processing system to automatically cut the data stream according to the set duration, ensuring that the data within each time window has independence and analytical value.

[0039] In this solution, a system that deploys data processing algorithms can perform in-depth analysis of the operational status data within each time window. This can involve using algorithms to calculate statistical indicators of the data, fit data change trends, analyze the correlation between parameters, and extract key information related to defects from the data.

[0040] Normalization can be an operation in which the data processing system uses a preset standardization algorithm to perform numerical transformation on the extracted multi-dimensional features, or it can be to uniformly map feature values ​​of different magnitudes to the 0-1 range so that all types of features have equal weight in model training and prediction.

[0041] In the process of integrating the normalized features according to a preset format, statistical features, trend features, and correlation features can be arranged in a fixed order to form a complete feature set corresponding to each time window, which is convenient for subsequent conversion into feature vectors.

[0042] S103, the feature set is converted into a feature vector, input into a pre-established defect prediction model, the cosine similarity between the real-time feature vector and the historical defect case feature vector stored in the defect prediction model is calculated, and the target defect type and the probability of occurrence of the target defect type are determined according to the preset similarity grading threshold. A feature vector can be a one-dimensional numerical vector that transforms structured data in a feature set into a vector according to a preset rule. For example, it can be a vector formed by arranging the normalized values ​​of each feature in the feature set in order, which facilitates mathematical operations and comparisons by the model.

[0043] Defect prediction models can be mathematical models trained based on historical production data to predict defects in pultruded products, such as Support Vector Machine (SVM) models. These models include feature vectors of historical defect cases, defect labels, and corresponding mapping relationships, enabling comparative analysis between input feature vectors and historical cases.

[0044] Real-time feature vectors can be feature vectors obtained by transforming the feature set of the current time window. They can be numerical vectors that reflect the current production process status and are used to compare similarity with feature vectors of historical cases.

[0045] The feature vector of historical defect cases can be a feature vector obtained by transforming the feature set corresponding to the cases of defects that occurred in the historical production process. It can contain a numerical vector of the operating status characteristics when the historical defect occurred and is stored in the defect prediction model as a comparison benchmark.

[0046] Cosine similarity is a mathematical metric that measures the degree of similarity between two feature vectors. It can be represented by calculating the cosine of the angle between the two vectors, with a value ranging from 0 to 1. The closer the value is to 1, the more similar the production states of the two vectors are.

[0047] The preset similarity grading threshold can be a pre-set numerical standard for dividing cosine similarity levels, such as multiple thresholds determined based on historical data verification, used to distinguish the defect risk levels corresponding to different degrees of similarity.

[0048] The target defect type can be a type of defect that may occur in the current production process, determined by similarity comparison. For example, common defect types of pultruded products such as internal voids, delamination, and poor fiber impregnation can provide a clear direction for process adjustment.

[0049] The probability of occurrence of a target defect type can be estimated based on similarity comparison results, such as the probability value calculated by the frequency of defect occurrence corresponding to similar feature vectors in historical cases, providing a quantitative basis for risk assessment.

[0050] This solution can perform format conversion operations on the feature set according to preset vector conversion rules. For example, it can arrange the normalized values ​​of each feature in the feature set in a fixed order to form a one-dimensional real-time feature vector. The process of inputting the converted real-time feature vector into the preset mathematical model can be achieved by sending the feature vector to the server or terminal device where the defect prediction model is deployed through a data interface, providing input data for the model calculation.

[0051] The cosine similarity between the real-time feature vector and the historical defect case feature vectors stored in the defect prediction model is calculated. This process involves iterating through all historical defect case feature vectors and calculating the cosine similarity with the real-time feature vector for each, resulting in multiple similarity values. These similarity values ​​are then compared to a grading threshold to match corresponding historical defect cases, thereby determining the current potential target defect type and its probability of occurrence.

[0052] S104, construct defect warning information based on the target defect type and the probability of occurrence of the target defect type, and output the defect warning information.

[0053] Defect warning information can be a cautionary message containing key information related to defects. It can be structured information composed of target defect type, probability of occurrence, warning level, etc., and can provide operators with clear defect prompts.

[0054] This solution can display text information on the production monitoring terminal screen, issue alarms via audio-visual devices, and send prompt messages to designated terminals. Specifically, it can integrate target defect types, occurrence probabilities, and related supplementary information according to a preset information format. For example, information such as "poor fiber impregnation, occurrence probability 92%, warning time 30s" can be combined according to a fixed template to form clear and easy-to-understand defect warning information. Furthermore, the warning information can be displayed in real time on the monitoring screen in the production workshop, and corresponding alarm levels can be issued via the equipment's audio-visual alarms, or warning notifications can be pushed to the operator's mobile app, ensuring that operators receive warning information promptly.

[0055] The technical solution provided in this embodiment achieves early warning and precise control of defects in pultruded products through a complete process of multi-dimensional operational status data collection, feature extraction and normalization, model comparison analysis, and early warning information output. By comparing cosine similarity and judging graded thresholds, it improves the accuracy of predicting various defects and can provide early warning of defects, allowing sufficient time for process adjustments and effectively reducing scrap rate. At the same time, it is adaptable to the production of pultruded products of different materials and specifications, which can significantly improve production efficiency and quality control level.

[0056] In one feasible embodiment, multi-dimensional feature extraction is performed on the runtime status data within the time window, including: The average value is the arithmetic mean of all collected values ​​of a certain operating state within a time window. It can be the average level of the parameter over a period of time, reflecting the overall operating state of the parameter.

[0057] The maximum value is the highest value among the data collected for a specific operating state within a time window. It can be the peak value of the parameter during this period, reflecting the parameter's extreme operating state. The minimum value is the lowest value among the data collected for a specific operating state within a time window. It can be the lowest value of the parameter during this period, and together with the maximum value, they reflect the parameter's fluctuation range. The standard deviation is a quantitative indicator of the degree to which the data collected for a specific operating state deviates from the average value within a time window. It can be a value reflecting the dispersion of the parameter data, demonstrating the stability of the parameter's operation. The range is the difference between the maximum and minimum values ​​of the data for a specific operating state within a time window. It can be an indicator that directly reflects the amplitude of the parameter data fluctuation, intuitively showing the parameter's range of change.

[0058] Multidimensional numerical features are feature sets composed of multiple statistical indicators, which can be feature groups consisting of at least two of the following: mean, maximum, minimum, standard deviation, and range. They characterize the statistical properties of parameters from different perspectives. Specifically, mathematical algorithms that describe the linear relationship between variables by constructing linear equations can be used, such as the least squares method or other commonly used linear fitting algorithms, to fit the changing trend of running state data over time.

[0059] The rate of change is the rate at which data of a certain operating state changes over time within a time window. It can be the amount of change in the parameter value per unit time, reflecting the dynamic trend of parameter change.

[0060] Multidimensional trend features are feature sets composed of the rates of change of multiple parameters. They can be feature groups composed of the rates of change of data in different operating states, which depict the dynamic change patterns of each parameter.

[0061] This solution utilizes pre-defined statistical algorithms to calculate various operational status data within each time window. For example, the algorithm iterates through all data points within the window, calculates selected statistical indicators, and integrates the results to form multi-dimensional numerical features. It also employs linear fitting algorithms, such as the least squares method, to perform trend analysis on various operational status data within each time window. This involves using time as the independent variable and operational status data as the dependent variable, fitting a linear equation, calculating the rate of change of the data using the equation's slope, and integrating the rates of change of each parameter to form multi-dimensional trend features.

[0062] This technical solution achieves in-depth mining of operational status data by extracting multi-dimensional numerical and trend features. Numerical features reflect the static statistical characteristics of parameters, while trend features depict the dynamic changes in parameters. The combination of both makes feature extraction more comprehensive, avoiding the problem that a single feature cannot fully reflect the production status. Furthermore, through the calculation of specific statistical indicators and the application of linear fitting algorithms, richer feature inputs are provided for subsequent models, effectively improving the input quality of the defect prediction model and laying the foundation for improving prediction accuracy.

[0063] In one feasible embodiment, multi-dimensional feature extraction of the running status data within the time window further includes: The Pearson correlation coefficient algorithm is used to calculate the correlation coefficient between the rate of change of each operating status data within the time window, thus obtaining multi-dimensional correlation characteristics.

[0064] The Pearson correlation coefficient algorithm is a statistical algorithm used to measure the degree of linear correlation between two variables. It can be an algorithm that characterizes the degree of correlation between variables by calculating the ratio of the product of covariance and standard deviation. The value ranges from -1 to 1, and the larger the absolute value, the higher the degree of correlation.

[0065] The correlation coefficient is a quantitative indicator calculated using the Pearson correlation coefficient algorithm. It can be a numerical value that reflects the degree of linear correlation between the rate of change of two operating status data, and reflects the correlation of the changing trends of different parameters.

[0066] Multidimensional correlation features are a set of features composed of correlation coefficients between multiple parameters. They can be feature groups composed of correlation coefficients between the change rates of data in different operating states, which characterize the collaborative change relationship between parameters.

[0067] This solution can perform correlation analysis on the rate of change of various operational status data within each time window by deploying the Pearson correlation coefficient algorithm. It can calculate the correlation coefficient between the rate of change of any two parameters one by one, integrate all correlation coefficients to form multi-dimensional correlation features, and explore the synergistic relationship between parameters.

[0068] This technical solution, based on numerical and trend features, adds multi-dimensional correlation feature extraction to compensate for the deficiency of single-parameter features in reflecting the synergistic relationship between parameters. By quantifying the correlation between the rates of change of different parameters using the Pearson correlation coefficient algorithm, it can capture the synergistic change patterns between parameters, and this synergistic relationship is often closely related to the formation of defects. The addition of correlation features makes the feature system more complete, further enriching the feature dimensions of the model input, helping the model to more accurately identify defect types related to parameter synergy, and improving the comprehensiveness and accuracy of defect prediction.

[0069] In one feasible embodiment, the training process of the defect prediction model includes: Collect a preset number of historical production data for pultruded products. The historical data includes normal production case data and defect case data. The historical data includes raw operating status data and corresponding defect labels or normal labels. The historical data is processed by a feature extraction and normalization method to extract a set of historical features and convert them into historical feature vectors to form a training feature dataset. The SVM algorithm is used to train a support vector machine model with the training feature dataset as input and the defect label or normal label as the output target. The trained support vector machine model, along with its corresponding historical feature vectors and defect labels, are stored to form a defect prediction model.

[0070] The preset amount of historical data on pultruded product production refers to a certain scale of pultruded product production process data collected in advance, such as 1,000 sets of complete case data including normal production and defective production, to provide a sufficient sample basis for model training.

[0071] Normal production case data refers to production data in which no defects occur during the production of pultruded products. For example, it can be case data containing complete raw data of the operating status and corresponding normal labels, which is used by the model to learn the characteristics of normal production status.

[0072] Defect case data refers to production data that shows various defects during the production of pultruded products. For example, it can be case data that includes complete raw data of the operating status and corresponding defect labels, such as internal voids, delamination, poor fiber impregnation, etc., which are used by the model to learn the characteristics of defective production status.

[0073] Raw operational data is raw data collected directly during the production process without any processing. It can be raw collected values ​​such as roller stress, traction speed, equipment vibration data, and multi-segment temperature data, which are the basic data for feature extraction.

[0074] Defect labels are markers used to identify the types of defects in defect case data. They can be text or numerical markers corresponding to specific defect types and are used by the model to establish a mapping relationship between features and defect types.

[0075] Normal labels are tags used to identify normal production case data. They can be text or numerical tags that correspond to the normal production status and are used by the model to distinguish between normal and defective production statuses.

[0076] Historical feature sets are feature sets obtained after feature extraction and normalization of historical data. For example, they can be structured data sets composed of statistical features, trend features, and correlation features of historical data, and have the same feature dimensions and format as real-time feature sets.

[0077] Historical feature vectors are one-dimensional numerical vectors obtained by transforming a set of historical features. For example, they can be vectors formed by arranging the normalized values ​​of each feature in the historical feature set in order, and are used as input data in model training.

[0078] The training feature dataset is a dataset consisting of historical feature vectors and their corresponding labels. It can be a structured set of training samples used by the model to learn the mapping relationship between features and labels.

[0079] Support Vector Machine (SVM) models are models that can establish a mapping relationship between input feature vectors and output labels, and are used to classify and predict unknown samples.

[0080] This solution involves collecting data from historical databases of production equipment and production record archives. This process includes filtering for cases containing complete operational status raw data and corresponding labels, such as defect labels or normal labels, ensuring the collected data scale and quality meet the model training requirements. Then, the historical data is processed using the same feature extraction and normalization algorithms as real-time data processing. This involves first extracting multi-dimensional features from the historical data, normalizing them to form a historical feature set, converting this set into historical feature vectors, and finally combining them with the corresponding labels to form the training feature dataset. This solution can divide the training feature dataset into training and validation sets. The model is trained using the training set as input and the labels as the output target. Model parameters, such as penalty coefficients and kernel function parameters, are optimized through methods like 5-fold cross-validation to improve the model's classification accuracy.

[0081] This technical solution ensures the reliability and effectiveness of the defect prediction model by collecting large-scale historical data, standardizing feature processing, and using the SVM algorithm for training and model storage. Training based on historical case data, combined with parameter optimization through 5-fold cross-validation, improves the model's training accuracy and demonstrates good generalization ability and classification performance.

[0082] In one feasible embodiment, the similarity grading threshold includes: a first threshold and a second threshold, wherein the first threshold is less than the second threshold; Based on a preset similarity grading threshold, the target defect type and its probability of occurrence are determined, including: If the cosine similarity is greater than or equal to the second threshold, then the corresponding target defect type and its probability of occurrence are output. If the cosine similarity is between the first threshold and the second threshold, the suspected defect type and its probability of occurrence are output, and key monitoring parameters are prompted; wherein, the key monitoring parameters are determined based on a pre-established table of the correlation between defect types and key parameters; If the cosine similarity is less than or equal to the first threshold, then the output is "no defect risk".

[0083] The first threshold is a smaller value among the similarity grading thresholds, which can be a critical value used to distinguish between low and medium similarity, such as 0.7. The second threshold is a larger value among the similarity grading thresholds, which can be a critical value used to distinguish between medium and high similarity, such as 0.85.

[0084] Suspected defect type refers to the type of defect that is identified when the similarity between the real-time feature vector and the feature vector of historical defect cases is at a moderate level.

[0085] Monitoring key parameters are operational status parameters that are associated with suspected defect types and have a significant impact on defect formation. These parameters can be determined based on a pre-established table of the correlation between defect types and key parameters, such as mold temperature parameters related to internal void defects.

[0086] In this scheme, the calculated cosine similarity can be compared with a second threshold before execution. For example, if the similarity reaches or exceeds the second threshold, the current production status is determined to be highly similar to a corresponding historical defect case, and the defect type and probability of occurrence for that historical case are directly output. If the cosine similarity is compared with both the first and second thresholds, and the similarity is between the two thresholds, the current production status is determined to be moderately similar to a historical defect case, and the suspected defect type and its corresponding probability range are output. Simultaneously, based on a pre-set correlation table, operators are prompted to focus on monitoring key parameters related to the suspected defect. If the cosine similarity is compared with the first threshold, and the similarity is lower than or equal to the first threshold, the current production status is determined to be significantly different from all historical defect cases, and the risk of a defect is extremely low, resulting in the conclusion that there is no significant risk of a defect at present.

[0087] This technical solution establishes a clear similarity grading standard by setting a first and second threshold, making defect prediction results more accurate and hierarchical. Different similarity levels correspond to different output results: high similarity directly outputs a clear defect type and probability of occurrence; medium similarity outputs suspected defects and monitoring prompts; and low similarity outputs no defect risk. This ensures accurate early warning of high-risk defects while avoiding false alarms and missed alarms. Simultaneously, the prompts for monitoring key parameters provide operators with clear focus areas, facilitating targeted monitoring and further improving the effectiveness and practicality of defect management. This solves the problems of vague prediction results and lack of clear guidance in existing technologies.

[0088] In one feasible embodiment, the method further includes: If the target defect type is poor fiber impregnation, an adjustment prompt message for adjusting the traction speed is generated; If the target defect type is an internal void defect, then adjustment prompts will be generated to adjust the resin tank temperature, mold inlet temperature, mold middle section temperature and mold outlet temperature. If the target defect type is a delamination defect, an adjustment prompt message for adjusting the yarn roller stress is generated.

[0089] The adjustment prompts for traction speed are process adjustment suggestions for defects caused by poor fiber impregnation. They can be text messages that clearly instruct operators to speed up or slow down the traction speed in order to improve the resin's impregnation effect on the fibers.

[0090] The adjustment prompts for resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature are process adjustment suggestions for the generation of internal void defects. They can be textual messages that clearly instruct operators to raise or lower the relevant temperature parameters to optimize the resin curing process and reduce internal voids.

[0091] The adjustment prompts for adjusting the yarn roller stress are process adjustment suggestions for the generation of delamination defects. They can be textual messages that clearly instruct operators to increase or decrease the yarn roller stress in order to enhance the bonding force between fiber layers and prevent delamination.

[0092] When the defect type is determined to be poor fiber impregnation, a prompt message such as "It is recommended to adjust the traction speed from the current X m / min to Y m / min" can be generated based on preset correlation rules between defects and process adjustments, guiding operators to optimize the process. When the defect is determined to be internal voids, specific adjustment prompts such as "It is recommended to increase the resin bath temperature by 5℃ and the mold middle section temperature by 3℃" can be generated to help operators reduce internal voids by optimizing temperature parameters. When the defect is determined to be delamination, a prompt message such as "It is recommended to adjust the yarn roller stress from the current XN to YN" can be generated, guiding operators to enhance the interlayer bonding force of fibers by adjusting the yarn roller stress.

[0093] This technical solution generates clear process adjustment prompts for different defect types, achieving a seamless connection between defect prediction and process optimization. Operators do not need to rely on complex professional knowledge and extensive experience; they can simply adjust key process parameters such as traction speed, temperature parameters, and roller stress based on the system's output prompts. This effectively solves the problem in existing technologies where prediction results cannot directly guide process adjustments. Through precise process adjustment suggestions, production status can be quickly improved, the probability of defects can be reduced, and production efficiency and product quality can be further enhanced, increasing the method's practicality and operability.

[0094] In one feasible embodiment, the method further includes: The validity of the operation status data is verified. If there is missing operation status data or the data exceeds a preset reasonable range, linear interpolation is performed using the first preset number of adjacent valid data to obtain the completion result of the operation status data. If data for the same operating status is missing or exceeds a preset reasonable range within a second preset number of consecutive time windows, a sensor fault message will be generated.

[0095] Validity verification is the process of verifying the reasonableness of collected operational status data. It can be a verification operation to check whether the data is missing or exceeds a preset reasonable range, ensuring the integrity and reliability of the data.

[0096] Missing operational status data refers to the situation where data at certain time points is not collected during the data acquisition process due to sensor failure, transmission interruption, or other reasons. This could mean that no valid data was collected for a certain parameter at a certain time point or within a certain period of time.

[0097] The preset reasonable range is the normal value range of each operating state data set in advance according to the production process requirements of pultruded products and the range of equipment operating parameters. It can be a numerical range determined based on the statistical analysis of historical normal production data. For example, the reasonable range of yarn roller stress is 0-500N, and the reasonable range of die temperature is 0-300℃.

[0098] The first preset number of adjacent valid data points refers to a certain number of normal data points adjacent to the missing or abnormal data points. These can be 5, 10, or other preset number of valid collected data points before the missing data point, used for data completion.

[0099] Linear interpolation can fill in missing data by calculating the linear fit value of the adjacent valid data before and after the missing data point, thus ensuring the continuity of the data.

[0100] The completion result is a numerical value calculated using a linear interpolation method to fill in missing or outlier data. It can be a reasonable value that is consistent with the trend of adjacent valid data, so as to maintain the integrity of the data sequence.

[0101] The second preset number of consecutive time windows refers to consecutive time windows of a number equal to the second preset value. These can be 3, 5, or other preset number of consecutive time windows, used to determine whether the data anomaly is caused by a persistent fault.

[0102] Sensor fault warning messages are alerts generated when abnormal data is determined to be caused by sensor malfunction. They can be textual messages that clearly indicate a potential malfunction of a certain type of sensor at a specific location, serving as a reminder for maintenance personnel to perform timely repairs.

[0103] Each data point is checked individually for missing values ​​or values ​​exceeding a preset reasonable range, such as roller stress >500N or <0N. If an anomaly is found, the first preset number of valid data points adjacent to the missing data are selected, and a linear interpolation algorithm is used to calculate and complete the value, filling in the missing or replacing the abnormal data to ensure the integrity of the data sequence. If the same parameter is found to have missing or abnormal data within a second preset number of time windows, it is determined that the sensor corresponding to that parameter, such as the thermocouple sensor at the mold inlet, may be faulty. A prompt message "Mold inlet temperature sensor suspected of being faulty, please repair promptly" is generated to notify maintenance personnel for handling.

[0104] This technical solution ensures the integrity and reliability of operational status data through data validity verification and completion mechanisms, avoiding deviations in defect prediction results caused by missing or abnormal data, and improving the robustness of the method. By using linear interpolation to complete abnormal data within a single time window, the continuity and validity of the data are maintained, guaranteeing the accuracy of feature extraction and model prediction. Furthermore, for abnormal data across multiple consecutive time windows, sensor fault alerts are generated, enabling timely detection of equipment faults, preventing escalation, and reducing production quality issues caused by sensor malfunctions. This solution not only addresses the impact of data anomalies on prediction results but also provides timely warnings of sensor faults, further improving the applicability and stability of the entire defect prediction system.

[0105] To enable those skilled in the art to more clearly understand the technical solutions provided in this application, this application also provides a preferred embodiment. Figure 2 This is a schematic diagram illustrating a multi-parameter-based defect prediction process for pultruded products, provided in one embodiment of the application. For example... Figure 2 As shown, the specific implementation method is as follows: Step 1: Accurately collect multi-dimensional operational status data; For the entire pultrusion product manufacturing process, dedicated sensors are deployed at key locations to ensure data integrity and reliability.

[0106] Specifically, an HBMT12 high-precision tension sensor is deployed on the wire roller frame to collect the wire roller stress at a sampling frequency of 10Hz; The traction mechanism collects the traction speed via a built-in encoder at a sampling frequency of 5Hz. Install PCB 352C65 vibration sensors at the mold support and traction mechanism to collect equipment vibration data at a sampling frequency of 20Hz; Two OmegaK thermocouple sensors are deployed at the resin tank, mold inlet, mold middle section, and mold outlet to collect temperature data at the corresponding locations. The sampling frequency is 5Hz, which comprehensively captures core operating status data such as roller stress, traction speed, equipment vibration, and multi-segment temperature.

[0107] Step 2, data preprocessing and multi-dimensional feature extraction; First, a fixed time window of 10 seconds is set, and the continuously collected running status data is divided into segments according to the window to ensure that each segment of data has independent analytical value.

[0108] Then, multi-dimensional feature extraction is performed: First, calculate the statistical characteristics of the data within each window, such as the mean, maximum, minimum, standard deviation, and range. Second, linear fitting algorithms such as the least squares method are used to calculate the slope of change of each data point to obtain trend characteristics; Third, the correlation coefficient between the rates of change of different parameters is calculated using the Pearson correlation coefficient algorithm to uncover correlation features; Finally, the min-max normalization algorithm is used to map the extracted multi-dimensional feature values ​​to the 0-1 interval, eliminating the difference in units and forming a feature set with a unified dimension.

[0109] Step 3: Defect prediction model construction and real-time prediction; First, collect 1000 or more sets of historical production data for pultruded products, covering normal production cases and various defect cases. Each set of data includes complete raw data of the operating status and corresponding defect or normal labels. The historical data undergoes the same feature extraction and normalization processing as the real-time data, transforming it into historical feature vectors to form a training feature dataset.

[0110] Based on the SVM algorithm, the training feature dataset is used as input and the labels are used as output targets. The model parameters are optimized by 5-fold cross-validation to complete the model training. The trained model, historical feature vectors and corresponding labels are stored to form a defect prediction model.

[0111] During real-time prediction, the feature set of the current window is transformed into a real-time feature vector, which is then input into the defect prediction model. The cosine similarity algorithm is used to calculate the similarity with the feature vectors of historical defect cases.

[0112] The preset first threshold is 0.7, and the second threshold is 0.85.

[0113] If the similarity is ≥0.85, output the corresponding defect type and probability of occurrence; If the similarity is between 0.7 and 0.85, output the suspected defect type, risk level, and key parameters that need to be monitored; If the similarity is less than 0.7, output "No obvious defect risk at present".

[0114] Step 4: Defect warning, process adjustment and data verification.

[0115] Based on the prediction results, defect early warning information is constructed, which includes defect type, occurrence probability, warning time and risk level, and is output through production monitoring terminal display screen, audible and visual alarm or designated terminal push.

[0116] Generate targeted process adjustment prompts for different defect types: If fiber impregnation is poor, it indicates a need to adjust the traction speed. When internal void defects are detected, it is recommended to adjust the temperature of the resin tank and each section of the mold. When a delamination defect is detected, the prompt is to adjust the wire roller stress.

[0117] Simultaneously, the validity of the collected operating status data is verified. If there is missing data or the data exceeds the preset reasonable range (such as roller stress >500N or <0N, mold temperature >300℃ or <0℃), linear interpolation is performed to complete the data using the five adjacent valid data. If the same data is abnormal within three consecutive time windows, a sensor fault prompt message is generated to notify maintenance personnel to carry out timely repairs.

[0118] This technical solution overcomes the lag of traditional detection by employing multi-parameter collaborative acquisition, deep feature extraction, and accurate model prediction, achieving defect early warning 10-40 seconds in advance, allowing ample time for process adjustments. It transforms previously isolated equipment monitoring data into core resources for defect prediction, increasing data utilization to over 85%. Through multi-parameter fusion and historical case comparison, the accuracy rate for various defect predictions exceeds 93%, with a false alarm rate below 2.5%, far superior to single-parameter threshold warning methods. The warning information is clear, and the process adjustment suggestions are specific, reducing reliance on human experience and enhancing practicality. Furthermore, it can flexibly adapt to the production of pultruded products of different materials and specifications, requiring only adjustments to historical case data in the model, demonstrating broad adaptability, effectively reducing scrap rates, and significantly improving production efficiency and quality control.

[0119] Figure 3 This is a schematic diagram of a pultrusion product defect prediction device based on multi-parameter collaborative fusion provided in an embodiment of this application. Figure 3 As shown, the pultrusion product defect prediction device based on multi-parameter collaborative fusion includes: The operation status data acquisition module 310 is used to collect operation status data during the manufacturing process of pultruded products; wherein, the operation status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature. The feature extraction module 320 is used to extract multi-dimensional features from the running status data within a time window based on a preset duration, and to normalize the extracted multi-dimensional features to form a feature set for the time window. The target defect type prediction module 330 is used to convert the feature set into a feature vector, input it into a pre-established defect prediction model, calculate the cosine similarity between the real-time feature vector and the feature vector of historical defect cases stored in the defect prediction model, and determine the target defect type and the probability of occurrence of the target defect type according to a preset similarity grading threshold. The defect warning information generation module 340 is used to construct defect warning information based on the target defect type and the probability of occurrence of the target defect type, and output the defect warning information.

[0120] The pultrusion product defect prediction device based on multi-parameter collaborative fusion provided in this embodiment has functional modules and beneficial effects corresponding to the pultrusion product defect prediction method embodiment based on multi-parameter collaborative fusion. To avoid repetition, it will not be described again here.

[0121] Figure 4 This is a schematic diagram of the hardware structure of a pultrusion product defect prediction device based on multi-parameter collaborative fusion provided in an embodiment of this application. The pultrusion product defect prediction device based on multi-parameter collaborative fusion may include a processor 401 and a memory 402 storing computer program instructions.

[0122] Specifically, the processor 401 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0123] Memory 402 may include mass storage for data or instructions. For example, and not limitingly, memory 402 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 402 may include removable or non-removable (or fixed) media, or memory 402 may be non-volatile solid-state memory. Memory 402 may be internal or external to the integrated gateway disaster recovery device.

[0124] In one instance, memory 402 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0125] The processor 401 reads and executes computer program instructions stored in the memory 402 to achieve... Figure 1 The embodiment shown illustrates a method for predicting defects in pultruded products based on multi-parameter collaborative fusion.

[0126] Furthermore, in conjunction with the multi-parameter collaborative fusion-based defect prediction method for pultruded products described in the above embodiments, this invention can be implemented using a computer storage medium. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the multi-parameter collaborative fusion-based defect prediction methods for pultruded products described in the above embodiments.

[0127] This application also provides a computer program product, including a computer program that, when executed by a processor, implements any of the multi-parameter collaborative fusion-based pultrusion product defect prediction methods described in the above embodiments.

[0128] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0129] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (ROM), flash memory, erasable read-only memory (EROM), floppy disks, compact disc read-only memory (CD-ROM), optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0130] It should also be noted that the exemplary embodiments mentioned in this invention describe methods or systems based on a series of steps or apparatus. However, this invention is not limited to the order of the steps described above; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0131] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to create a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0132] The above description is merely a specific embodiment of the present invention. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A method for predicting defects in pultruded products based on multi-parameter collaborative fusion, characterized in that, include: During the manufacturing process of pultruded products, operational status data is collected; wherein, the operational status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature; Based on a preset time window, multi-dimensional features are extracted from the running status data within the time window, and the extracted multi-dimensional features are normalized to form a feature set for the time window. The feature set is transformed into a feature vector, which is input into a pre-established defect prediction model. The cosine similarity between the real-time feature vector and the feature vector of historical defect cases stored in the defect prediction model is calculated. Based on the preset similarity grading threshold, the target defect type and the probability of occurrence of the target defect type are determined. Defect warning information is constructed based on the target defect type and the probability of occurrence of the target defect type, and the defect warning information is output.

2. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 1, characterized in that, Multi-dimensional feature extraction is performed on the runtime status data within the time window, including: Calculate at least two of the mean, maximum, minimum, standard deviation, and range of each running status data within the time window to obtain multi-dimensional numerical features; A linear fitting algorithm is used to calculate the rate of change of each running state data within the time window, and multi-dimensional trend characteristics are obtained.

3. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 2, characterized in that, Multi-dimensional feature extraction of operational status data within a time window also includes: The Pearson correlation coefficient algorithm is used to calculate the correlation coefficient between the rate of change of each operating status data within the time window, thus obtaining multi-dimensional correlation characteristics.

4. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 1, characterized in that, The training process of the defect prediction model includes: Collect a preset number of historical production data for pultruded products. The historical data includes normal production case data and defect case data. The historical data includes raw operating status data and corresponding defect labels or normal labels. The historical data is subjected to feature extraction and normalization processing to extract the historical feature set and transform it into a historical feature vector to form a training feature dataset; The SVM algorithm is used to train a support vector machine model with the training feature dataset as input and the defect label or normal label as the output target. The trained support vector machine model, along with its corresponding historical feature vectors and defect labels, are stored to form a defect prediction model.

5. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 1, characterized in that, The similarity grading threshold includes: a first threshold and a second threshold, wherein the first threshold is less than the second threshold; Based on a preset similarity grading threshold, the target defect type and its probability of occurrence are determined, including: If the cosine similarity is greater than or equal to the second threshold, then the corresponding target defect type and its probability of occurrence are output. If the cosine similarity is between the first threshold and the second threshold, the suspected defect type and its probability of occurrence are output, and key monitoring parameters are prompted; wherein, the key monitoring parameters are determined based on a pre-established table of the correlation between defect types and key parameters; If the cosine similarity is less than or equal to the first threshold, then the output is "no defect risk".

6. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 1, characterized in that, The method further includes: If the target defect type is poor fiber impregnation, an adjustment prompt message for adjusting the traction speed is generated; If the target defect type is an internal void defect, then adjustment prompts will be generated to adjust the resin tank temperature, mold inlet temperature, mold middle section temperature and mold outlet temperature. If the target defect type is a delamination defect, an adjustment prompt message for adjusting the yarn roller stress is generated.

7. The method for predicting defects in pultruded products based on multi-parameter collaborative fusion according to claim 1, characterized in that, The method further includes: The validity of the operation status data is verified. If there is missing operation status data or the data exceeds a preset reasonable range, linear interpolation is performed using the first preset number of adjacent valid data to obtain the completion result of the operation status data. If data for the same operating status is missing or exceeds a preset reasonable range within a second preset number of consecutive time windows, a sensor fault message will be generated.

8. A defect prediction device for pultruded products based on multi-parameter collaborative fusion, characterized in that, include: The operation status data acquisition module is used to collect operation status data during the manufacturing process of pultruded products; wherein, the operation status data includes at least two of the following: roller stress, traction speed, equipment vibration data, and multi-segment temperature data, wherein the multi-segment temperature data includes at least two of the following: resin tank temperature, mold inlet temperature, mold middle section temperature, and mold outlet temperature. The feature extraction module is used to extract multi-dimensional features from the running status data within a time window based on a preset duration, and to normalize the extracted multi-dimensional features to form a feature set for the time window. The target defect type prediction module is used to convert the feature set into a feature vector, input it into a pre-established defect prediction model, calculate the cosine similarity between the real-time feature vector and the feature vector of historical defect cases stored in the defect prediction model, and determine the target defect type and the probability of occurrence of the target defect type according to a preset similarity grading threshold. The defect warning information generation module is used to construct defect warning information based on the target defect type and the probability of occurrence of the target defect type, and output the defect warning information.

9. A defect prediction device for pultruded products based on multi-parameter collaborative fusion, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the pultrusion product defect prediction method based on multi-parameter collaborative fusion as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the pultrusion product defect prediction method based on multi-parameter collaborative fusion as described in any one of claims 1-7.

Images