Method and device for detecting center hole of wire drawing die

By real-time acquisition and feature extraction of multimodal signals around the wire drawing die, combined with multi-dimensional threshold analysis and multi-objective optimization algorithm, the center hole detection of wire drawing die without shutdown is achieved, and the result of shutdown during the detection process in the prior art is solved Productivity and economic losses have improved the accuracy of detection and the scientific nature of maintenance decisions.

CN118839142BActive Publication Date: 2024-11-26DONGGUAN TOYA ELECTRONIC TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411325340.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-23
Publication Date
2024-11-26
Estimated Expiration
2044-09-23

AI Technical Summary

Technical Problem

The existing wire drawing die center hole detection method requires shutdown operation, resulting in loss of production efficiency and economic benefits, and the inspection process is time-consuming and labor-intensive.

Method used

By synchronously collecting the sound, vibration and temperature signals around the wire drawing die that is undergoing wire drawing operations, using a multi-modal feature fusion network for feature extraction and wear state recognition, combined with multi-dimensional threshold analysis and multi-objective optimization algorithm, real-time online monitoring is achieved and predictive maintenance decisions.

Benefits of technology

It realizes online inspection without downtime, improves production efficiency and inspection accuracy, provides scientific maintenance decisions, reduces maintenance costs and extends equipment life.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118839142B_ABST
    Figure CN118839142B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for detecting the center hole of a wire drawing die, the method comprising synchronously collecting the sound, vibration and temperature signals around the wire drawing die in wire drawing operation; preprocessing and feature extraction of the collected time-aligned multimodal data to obtain a fused multimodal feature vector; performing wear state recognition processing on the multimodal feature vector to obtain the wear state evaluation result of the center hole of the wire drawing die and its uncertainty estimation; performing abnormal state analysis processing on the wear state evaluation result and its uncertainty estimation to obtain an abnormal detection report and early warning information including abnormal type, severity and development trend; performing maintenance strategy decision processing on the abnormal detection report and early warning information to obtain a predictive maintenance decision suggestion including maintenance time, method and specific operation steps. The technical solution of the present invention can detect the center hole of the wire drawing die during the wire drawing process, which is conducive to improving production efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of central hole detection of a wire drawing die, and in particular to a method and a device for detecting the central hole of a wire drawing die. Background Art

[0002] The wire drawing die is a key tool in the production of metal wire. Its core component is a precision-machined tapered center hole. During the wire drawing process, the metal wire is forcefully pulled through this center hole, and the diameter of the wire is reduced by gradually reducing the aperture. During the production process, the quality of the center hole of the wire drawing die directly affects the dimensional accuracy, surface finish and mechanical properties of the finished wire. However, during long-term use, the high-speed moving metal wire will cause continuous wear on the center hole, resulting in an increase in the aperture and deformation of the shape, thus affecting the quality of the wire. Therefore, it is very important to regularly inspect the center hole of the wire drawing die to detect wear problems in a timely manner, ensure product quality, and optimize maintenance plans.

[0003] The existing method for detecting the center hole of wire drawing dies requires the equipment to be shut down before the wire drawing dies can be inspected. During the inspection process, the wire drawing dies are required to be removed from the production line and sent to a special inspection equipment for offline inspection. This process is not only time-consuming and labor-intensive, but also causes the production line to shut down, seriously affecting production efficiency and economic benefits. Especially in large-scale wire rod factories with continuous production, frequent shutdowns for inspection will cause huge economic losses. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problem that during the detection of the center hole of a wire drawing die in the prior art, the machine needs to be shut down for operation.

[0005] A first aspect of the present invention provides a method for detecting a center hole of a wire drawing die, the method comprising:

[0006] The sound, vibration and temperature signals around the wire drawing die during the wire drawing operation are synchronously collected to obtain time-aligned multi-modal data;

[0007] Preprocessing the time-aligned multimodal data, and extracting features from the preprocessed multimodal data using a multimodal feature fusion network to obtain a fused multimodal feature vector;

[0008] Using a plurality of preset models to perform wear state recognition processing on the multimodal feature vector, and obtaining a wear state evaluation result of the center hole of the wire drawing die and an uncertainty estimation thereof;

[0009] A multi-dimensional threshold analysis method is used to perform abnormal state analysis on the wear state assessment result and its uncertainty estimation, so as to obtain an abnormality detection report and early warning information including abnormality type, severity, and development trend;

[0010] A multi-objective optimization algorithm is used to perform maintenance strategy decision processing on the abnormal detection report and warning information to obtain predictive maintenance decision suggestions including maintenance time, method and specific operation steps.

[0011] Optionally, the synchronous collection of sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multimodal data includes:

[0012] The sound signal around the wire drawing die during the wire drawing operation is collected at high frequency to obtain sound data;

[0013] According to the acquisition time point of the sound data, the vibration signal of the surface of the wire drawing die is collected at equal time intervals to obtain vibration data corresponding to the time of the sound data;

[0014] Based on the collection time point of the vibration data, the temperature signal of a preset position on the wire drawing die is collected at a fixed time interval to obtain discrete temperature data;

[0015] The sound data, vibration data and discrete temperature data are time stamp aligned and data interpolated to obtain time-aligned multimodal data with consistent sampling frequency.

[0016] Optionally, preprocessing the time-aligned multimodal data and extracting features from the preprocessed multimodal data using a multimodal feature fusion network to obtain a fused multimodal feature vector includes:

[0017] Performing wavelet transform and singular spectrum analysis on the time-aligned multimodal data to obtain a time-frequency domain feature matrix of each mode;

[0018] Performing adaptive threshold filtering on the time-frequency domain feature matrix to obtain a denoised feature representation;

[0019] The denoised feature representation is input into the corresponding modal branches of the multimodal feature fusion network according to the sound, vibration and temperature modes, each modal branch includes multiple convolutional layers, and each modal feature is independently nonlinearly transformed to obtain modality-specific deep features;

[0020] The cross-modal attention layer of the multimodal feature fusion network is used to interact and assign weights to the modality-specific deep features to obtain fused features that take into account the relationship between the modalities;

[0021] The fused features are subjected to global pooling and full connection layer processing to obtain a fused multimodal feature vector.

[0022] Optionally, the use of multiple preset models to perform wear state recognition processing on the multimodal feature vector to obtain a wear state evaluation result of the center hole of the wire drawing die and its uncertainty estimation includes:

[0023] The multimodal feature vectors are respectively input into three preset models, namely, random forest, support vector machine and deep neural network, and parallel calculation is performed to obtain three sets of independent preliminary evaluation results of wear status;

[0024] Using the Bayesian model averaging method, dynamic weights are assigned to each preset model according to the three sets of independent preliminary wear state assessment results to obtain a weighted fusion wear state assessment result;

[0025] Performing Monte Carlo sampling on the weighted fusion wear state evaluation result to generate a large number of simulation samples and obtain a probability distribution of the wear state;

[0026] The confidence interval of the probability distribution of the wear state is calculated using a kernel density estimation method to obtain an uncertainty estimate of the wear state assessment.

[0027] Optionally, the multidimensional threshold analysis method is used to perform abnormal state analysis processing on the wear state assessment result and its uncertainty estimation to obtain an abnormality detection report and early warning information including abnormality type, severity, and development trend, including:

[0028] Decomposing the wear state evaluation result in multiple dimensions to obtain a wear dynamic feature vector representing the wear degree, wear rate and wear acceleration;

[0029] Performing cluster analysis on the wear dynamic feature vector using a dynamic clustering algorithm to obtain multiple wear state clusters;

[0030] Based on the wear state clusters and uncertainty estimates, a multidimensional decision tree model is constructed to classify and evaluate each wear state cluster to obtain preliminary abnormality type and severity determination results;

[0031] A time series prediction model is used to perform trend analysis on the preliminary abnormality type and severity determination results, and combined with historical data, an abnormality detection report and early warning information including abnormality type, severity, and development trend are obtained.

[0032] Optionally, the dynamic clustering algorithm is used to perform cluster analysis on the wear dynamic feature vector to obtain multiple wear state clusters, including:

[0033] Adaptively normalizing the wear dynamic feature vector to obtain a standardized feature vector;

[0034] Clustering the standardized feature vectors using a density peak clustering algorithm to obtain a first group of wear state clusters;

[0035] Based on the first group of wear state clusters, a spectral clustering algorithm is applied to construct a similarity graph, and feature space mapping is performed to obtain a second group of wear state clusters;

[0036] The second group of wear state clusters are evaluated and screened using the silhouette coefficient and the Davis-Paulding index to obtain a plurality of wear state clusters.

[0037] Optionally, the use of a multi-objective optimization algorithm to perform maintenance strategy decision processing on the abnormal detection report and warning information to obtain a predictive maintenance decision suggestion including maintenance time, method and specific operation steps includes:

[0038] Processing the abnormality detection report and warning information using a pre-built multi-objective optimization model to obtain an initial maintenance strategy candidate set;

[0039] Applying a genetic algorithm to iteratively optimize the initial maintenance strategy candidate set, and obtaining a Pareto optimal solution set through crossover, mutation and selection operations;

[0040] Using the Monte Carlo method to perform random scenario simulation on each solution in the Pareto optimal solution set, the performance of different maintenance strategies is evaluated, and the robustness index of each maintenance strategy is obtained;

[0041] Based on the robustness index and the preset decision maker preference weights, the Pareto optimal solution set is comprehensively evaluated and ranked using the analytic hierarchy process, the optimal maintenance strategy is selected, and a predictive maintenance decision recommendation including maintenance time, method and specific operation steps is obtained.

[0042] A second aspect of the present invention provides a center hole detection device for a wire drawing die, the center hole detection device for a wire drawing die comprising:

[0043] A multimodal data acquisition module is used to synchronously acquire the sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multimodal data;

[0044] A feature extraction processing module is used to preprocess the time-aligned multimodal data and extract features from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector;

[0045] A wear state identification module, used to perform wear state identification processing on the multimodal feature vector using a plurality of preset models, and obtain a wear state evaluation result of the center hole of the wire drawing die and an uncertainty estimation thereof;

[0046] An abnormality analysis and processing module, used to perform abnormality analysis and processing on the wear state assessment result and its uncertainty estimation using a multi-dimensional threshold analysis method, and obtain an abnormality detection report and early warning information including abnormality type, severity and development trend;

[0047] The maintenance decision generation module is used to use a multi-objective optimization algorithm to perform maintenance strategy decision processing on the abnormal detection report and warning information to obtain predictive maintenance decision suggestions including maintenance time, method and specific operation steps.

[0048] The technical solution provided by the embodiments of the present application has at least the following advantages:

[0049] First, the sound, vibration and temperature signals around the wire drawing die in the process of wire drawing are collected synchronously, realizing online real-time monitoring, completely eliminating the need for shutdown detection, and greatly improving production efficiency. Secondly, the collected data is processed and feature extracted through a multimodal feature fusion network, overcoming the limitation of insufficient information of a single sensor, and significantly improving the comprehensiveness and accuracy of detection. Furthermore, the use of multiple preset models for wear state identification not only provides reliable evaluation results, but also quantifies the uncertainty of the results, providing a scientific basis for subsequent decision-making. In addition, the use of a multidimensional threshold analysis method for abnormal state analysis can comprehensively consider multiple aspects of wear, and realize the early detection and trend prediction of potential problems. Finally, the use of a multi-objective optimization algorithm for maintenance strategy decision-making takes into account production efficiency and cost control while considering the equipment status, and realizes the optimization of maintenance plans. This comprehensive and intelligent detection method not only solves the problem of shutdown detection, but also significantly improves the accuracy of detection and the pertinence of maintenance, providing strong technical support for the precise maintenance of wire drawing dies and the continuous optimization of the production process, and ultimately achieves the goal of improving production efficiency, reducing maintenance costs, and extending equipment life. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0051] Figure 1 A schematic diagram of an embodiment of a method for detecting a center hole of a wire drawing die according to an embodiment of the present invention;

[0052] Figure 2 It is a schematic diagram of an embodiment of a center hole detection device for a wire drawing die in an embodiment of the present invention.

[0053] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0056] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, "and / or" in the full text includes three solutions. Taking A and / or B as an example, it includes technical solution A, technical solution B, and technical solution that satisfies both A and B. In addition, the technical solutions between the various embodiments can be combined with each other, which must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0057] An embodiment of the present application provides a method for detecting a center hole of a wire drawing die. Figure 1 A flow chart of a method for detecting a center hole of a wire drawing die provided in one embodiment of the present application.

[0058] See also Figure 1 , synchronously collect the sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multi-modal data;

[0059] In one embodiment of the present invention, the synchronous collection of sound, vibration and temperature signals around a wire drawing die performing a wire drawing operation to obtain time-aligned multimodal data includes: high-frequency collection of sound signals around a wire drawing die performing a wire drawing operation to obtain sound data; based on the collection time point of the sound data, collecting vibration signals on the surface of the wire drawing die at equal time intervals to obtain vibration data corresponding to the sound data time; based on the collection time point of the vibration data, collecting temperature signals at preset positions on the wire drawing die at fixed time intervals to obtain discrete temperature data; timestamp alignment of the sound data, vibration data and discrete temperature data, and data interpolation to obtain time-aligned multimodal data with consistent sampling frequencies.

[0060] Specifically, it is first necessary to perform high-frequency acquisition of sound signals around the drawing die where the wire drawing operation is being performed to obtain sound data. This step can be performed using a high-precision microphone array, which can be strategically arranged around the drawing die. For example, a high-sensitivity microphone can be installed at the wire inlet, wire outlet, and side of the drawing die to form a three-dimensional sound acquisition network. The sampling frequency of these microphones is usually set above 20kHz, which can capture high-frequency sounds that are inaudible to the human ear. These high-frequency sounds often contain rich information about the wear of the wire drawing die. High-frequency acquisition can not only capture instantaneous abnormal sounds, but also record subtle changes in sound intensity, providing rich raw data for subsequent wear status analysis.

[0061] Next, according to the collection time point of the sound data, the vibration signal on the surface of the wire drawing die is collected at equal time intervals to obtain vibration data corresponding to the sound data time. This step can be completed using a high-precision accelerometer. These accelerometers can be directly installed on the outer shell of the wire drawing die. For example, an accelerometer is installed at the inlet, middle and outlet of the wire drawing die to form a three-point vibration monitoring system. The collection frequency of the vibration signal is usually consistent with the sound signal, which can ensure the precise correspondence of the two data in time. The equal time interval collection method ensures the continuity and consistency of the vibration data, which is helpful for subsequent data analysis and processing.

[0062] After obtaining the sound and vibration data, the temperature signals at the preset positions on the wire drawing die are collected at fixed time intervals based on the collection time points of the vibration data to obtain discrete temperature data. Temperature sensors are usually installed at key positions of the wire drawing die, such as the wire inlet, the area where the center hole is located, and the wire outlet. The temperature changes at these positions can directly reflect the friction heat and material deformation heat during the wire drawing process. The collection frequency of temperature data can be lower than that of sound and vibration data, for example, once per second, because the temperature changes relatively slowly. Collection at fixed time intervals can ensure the time consistency of temperature data, which is convenient for subsequent integration and analysis with other data.

[0063] Finally, the collected sound data, vibration data, and discrete temperature data are timestamp aligned and data interpolated to obtain time-aligned multimodal data with consistent sampling frequencies. Timestamp alignment is a key step to ensure that different types of data correspond accurately in the time dimension. For example, the vibration data and temperature data can be aligned to the same time point by linear interpolation based on the acquisition time point of the sound data. For temperature data with a lower sampling frequency, more complex interpolation algorithms, such as cubic spline interpolation, can be used to generate a continuous temperature curve consistent with the frequency of the sound and vibration data. This time alignment and frequency consistency processing of multimodal data lays the foundation for subsequent feature extraction and fusion analysis.

[0064] This multimodal data collection and processing method has several significant advantages. First, by simultaneously collecting sound, vibration and temperature data, it can fully reflect the working status of the wire drawing die, overcoming the limitation of insufficient information from a single sensor. Secondly, high-frequency acquisition and equal-time interval sampling ensure the high quality and continuity of the data, and can capture instantaneous abnormal changes. Furthermore, timestamp alignment and data interpolation processing solve the inconsistency problem of different types of data in time and frequency, providing a reliable data basis for subsequent multimodal feature fusion. Finally, this method realizes real-time monitoring of the wire drawing die under normal working conditions, completely avoiding production interruptions and efficiency losses caused by traditional shutdown detection methods.

[0065] Please continue reading Figure 1 , preprocessing the time-aligned multimodal data, and extracting features from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector;

[0066] In one embodiment of the present invention, the time-aligned multimodal data is preprocessed, and features are extracted from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector, including: performing wavelet transform and singular spectrum analysis on the time-aligned multimodal data to obtain a time-frequency domain feature matrix of each mode; performing adaptive threshold filtering on the time-frequency domain feature matrix to obtain a denoised feature representation; inputting the denoised feature representation into the corresponding modal branches of the multimodal feature fusion network according to sound, vibration and temperature modes, each modal branch comprising multiple convolutional layers, performing independent nonlinear transformations on each modal feature to obtain modality-specific deep features; using the cross-modal attention layer of the multimodal feature fusion network to interact and weight the modality-specific deep features to obtain fused features that take into account the relationship between modalities; performing global pooling and fully connected layer processing on the fused features to obtain a fused multimodal feature vector.

[0067] Specifically, first, these multimodal data are subjected to wavelet transform and singular spectrum analysis to obtain the time-frequency domain feature matrix of each mode. Wavelet transform can analyze signals at different scales and is suitable for processing non-stationary signals, such as the sound and vibration generated during the wire drawing process. For example, for sound signals, wavelet transform can simultaneously capture the continuous low-frequency noise and instantaneous high-frequency anomalies during the wire drawing process. Singular spectrum analysis can extract the main components of the signal and effectively separate trends, periodic components and noise. For temperature data, singular spectrum analysis can identify the main modes of temperature changes, such as periodic temperature fluctuations caused by changes in wire drawing speed.

[0068] Next, the obtained time-frequency domain feature matrix is ​​subjected to adaptive threshold filtering to obtain the denoised feature representation. The advantage of adaptive threshold filtering is that it can dynamically adjust the filter strength according to the local characteristics of the signal, which is particularly important for processing complex and changeable signals in the wire drawing process. For example, when processing vibration signals, adaptive threshold filtering can effectively remove background noise while retaining sudden high-frequency vibrations (which may indicate abnormal wear).

[0069] The denoised feature representations are then input into the corresponding modal branches of the multimodal feature fusion network. Each modal branch contains multiple convolutional layers, which perform independent nonlinear transformations on the features of each modality to obtain modality-specific deep features. The design of this branch structure allows the network to learn specific features of each modality separately. For example, the sound branch may learn specific frequency patterns related to wire drawing die wear, while the vibration branch may identify the time series characteristics of abnormal vibrations, and the temperature branch can capture the long-term trend of temperature changes.

[0070] Next, the cross-modal attention layer of the multimodal feature fusion network interacts and assigns weights to these modality-specific deep features to obtain fused features that take into account the relationship between modalities. The introduction of the cross-modal attention mechanism greatly enhances the network's ability to capture the relationship between different modalities. For example, when an abnormal sound is detected, the attention mechanism may automatically increase the attention to the vibration and temperature data at the corresponding moment, thereby more comprehensively evaluating potential abnormalities.

[0071] Finally, the fused features are processed by global pooling and fully connected layers to obtain a fused multimodal feature vector. The global pooling operation can extract the most significant features, while the fully connected layer further integrates these features to generate a compact and information-rich feature vector. This final feature vector combines information from multiple modes such as sound, vibration, and temperature, providing a strong input for subsequent wear state recognition.

[0072] This multi-step feature extraction and fusion method has several significant advantages. First, the combination of wavelet transform and singular spectrum analysis enables the feature extraction process to capture the time domain and frequency domain characteristics of the signal at the same time, greatly enhancing the analysis ability of complex non-stationary signals. Second, adaptive threshold filtering effectively improves the signal-to-noise ratio of the signal, retaining key information while removing interference. Furthermore, the branch structure design of the multimodal feature fusion network allows specialized processing of the features of each modality, while the cross-modal attention mechanism realizes the intelligent integration of information from different modalities. Finally, the global pooling and fully connected layer processing ensure the compactness and information richness of the final feature vector. This method can not only comprehensively capture the multi-faceted features of the working state of the wire drawing die, but also intelligently integrate these features, providing high-quality input data for subsequent wear state identification and anomaly detection, thereby significantly improving the accuracy and reliability of the center hole detection of the wire drawing die.

[0073] Please continue reading Figure 1 , using a plurality of preset models to perform wear state recognition processing on the multimodal feature vector, and obtaining a wear state evaluation result of the center hole of the wire drawing die and an uncertainty estimation thereof;

[0074] In one embodiment of the present invention, the multimodal feature vector is processed for wear state identification using multiple preset models to obtain a wear state evaluation result of the center hole of the wire drawing die and its uncertainty estimate, including: the multimodal feature vector is input into three preset models of random forest, support vector machine and deep neural network respectively, and parallel calculation is performed to obtain three sets of independent preliminary wear state evaluation results; a Bayesian model averaging method is used to assign dynamic weights to each preset model according to the three sets of independent preliminary wear state evaluation results to obtain a weighted fusion wear state evaluation result; Monte Carlo sampling is performed on the weighted fusion wear state evaluation result to generate a large number of simulated samples to obtain a probability distribution of the wear state; and a kernel density estimation method is used to calculate the confidence interval of the probability distribution of the wear state to obtain an uncertainty estimate of the wear state evaluation.

[0075] Specifically, first, these feature vectors are input into three preset models, namely random forest, support vector machine and deep neural network, for parallel calculation and processing, so as to obtain three independent preliminary evaluation results of wear status. This method makes full use of the advantages of each model. For example, the random forest model may capture subtle changes in the sound characteristics of the drawing die, such as a sudden increase in high-frequency noise, which often indicates increased wear; the support vector machine performs well in processing vibration data and can accurately divide the boundaries between normal working conditions and abnormal vibration conditions; and the deep neural network has advantages in integrating multimodal data and can extract high-level wear characteristics from the complex interaction of temperature, sound and vibration. By using these three models in parallel, the system can evaluate the wear status of the drawing die from multiple angles, improving the comprehensiveness and accuracy of the evaluation.

[0076] Next, the Bayesian model averaging method is used to assign dynamic weights to each preset model based on the three sets of independent preliminary wear status evaluation results, and a weighted fusion wear status evaluation result is obtained. The advantage of the Bayesian model averaging method is that it can dynamically adjust the weights of each model based on historical data and current evaluation results. For example, if the drawing die experiences a sharp temperature change over a period of time, the deep neural network may perform better in this case due to its sensitivity to complex patterns. Then, the Bayesian model averaging method will automatically increase the weight of the deep neural network in the final evaluation. This dynamic adjustment mechanism ensures that the evaluation results can adapt to different working conditions and wear stages, greatly improving the adaptability and accuracy of the evaluation.

[0077] In order to further enhance the reliability of the assessment, Monte Carlo sampling is performed on the weighted fusion wear state assessment results to generate a large number of simulated samples, thereby obtaining the probability distribution of the wear state. The importance of this step lies in its ability to simulate various possible wear conditions, including some extreme conditions that are difficult to observe in actual production. For example, the system can simulate the wear of wire drawing dies under extreme conditions such as high temperature, high speed and heavy load, or simulate the cumulative wear effects after long-term continuous use. By generating thousands or even tens of thousands of simulated samples, the system can not only evaluate the current wear state, but also predict various wear conditions that may occur in the future, providing an important basis for preventive maintenance.

[0078] Finally, the kernel density estimation method is used to calculate the confidence interval of the probability distribution of the wear state, thereby obtaining the uncertainty estimate of the wear state assessment. The advantage of the kernel density estimation method is that it does not require strong assumptions about the data distribution and can more accurately describe complex, non-parametric probability distributions. This is particularly important in the wear assessment of drawing dies, because the wear state often exhibits nonlinear and multi-peak distribution characteristics. By calculating the confidence interval, the system can quantify the uncertainty of the assessment results. For example, in some scenarios (case one), the wear state assessment result of a certain drawing die is 75%, and the 95% confidence interval is 70%-80%. This means that we are 95% sure that the actual wear degree is between 70% and 80%. This narrow interval indicates that the assessment result is very accurate. In this case, the maintenance personnel can be quite sure that the drawing die is already in a wear state of about 75% and make maintenance plans accordingly. In other scenarios (scenario two), the wear state assessment result of another drawing die is also 75%, but the 95% confidence interval is 60%-90%. This wide interval indicates that there is a large uncertainty in the assessment result. Although the most likely degree of wear is still 75%, the actual situation may be much better (as low as 60%) or much worse (as high as 90%) than expected. Faced with this situation, maintenance personnel may need to adopt a more cautious strategy, such as increasing the frequency of inspections or considering more detailed inspections. By comparing these two situations, we can see that even though the average assessment results are the same (both are 75%), the actual situation they reflect and the required response strategies may be very different due to the different confidence intervals. A narrower confidence interval means that the assessment is more precise and decisions can be made based on this result with greater confidence; a wider confidence interval suggests that we need to be more cautious and may need to collect more information or adopt a more conservative maintenance plan. This method not only tells us the estimated degree of wear of the drawing die, but also provides information on the reliability of this estimate, thereby helping to make more informed maintenance decisions.

[0079] This multi-step wear state identification and uncertainty estimation method has several significant advantages in the detection of the center hole of the wire drawing die. First, multi-model parallel processing provides a comprehensive wear state evaluation, which can capture different types of wear characteristics. Second, the application of the Bayesian model averaging method realizes the dynamic adjustment of the model weight, so that the evaluation results can adapt to different working conditions and wear stages. Furthermore, the introduction of Monte Carlo sampling greatly enhances the prediction ability of the system, which can simulate and predict various possible wear conditions. Finally, the calculation of uncertainty through the kernel density estimation method provides key reference information for decision making. This method can not only give an accurate wear state evaluation, but also provide the credibility of the evaluation results, which greatly improves the scientificity and reliability of the maintenance decision of the wire drawing die. Through this comprehensive and accurate wear state evaluation, enterprises can formulate more accurate maintenance plans, effectively reduce maintenance costs, extend equipment life, and significantly improve production efficiency and product quality.

[0080] Please continue reading Figure 1 , using a multi-dimensional threshold analysis method to perform abnormal state analysis and processing on the wear state assessment results and their uncertainty estimation, and obtain an abnormality detection report and early warning information including abnormality type, severity, and development trend;

[0081] In one embodiment of the present invention, the multidimensional threshold analysis method is used to perform abnormal state analysis processing on the wear state assessment results and their uncertainty estimates to obtain abnormal state detection reports and warning information including abnormality types, severity, and development trends, including: multi-dimensional decomposition of the wear state assessment results to obtain wear dynamic feature vectors that characterize wear degree, wear rate, and wear acceleration; cluster analysis of the wear dynamic feature vectors using a dynamic clustering algorithm to obtain multiple wear state clusters; constructing a multidimensional decision tree model based on the wear state clusters and uncertainty estimates, classifying and evaluating each wear state cluster to obtain preliminary abnormality type and severity determination results; using a time series prediction model to perform trend analysis on the preliminary abnormality type and severity determination results, and combining historical data to obtain abnormality detection reports and warning information including abnormality types, severity, and development trends.

[0082] Specifically, the wear state assessment results are first decomposed in multiple dimensions to obtain a wear dynamic feature vector that characterizes the degree of wear, wear rate, and wear acceleration. The importance of this step lies in that it not only takes into account the current degree of wear, but also includes the rate of change and acceleration of wear, so that the dynamic wear state of the wire drawing die can be fully described. For example, for a wire drawing die in use, its degree of wear may be 60%, the wear rate is 0.5% per day, and the wear acceleration is 0.01% / day. These three indicators together constitute a comprehensive wear dynamic feature vector, which provides a rich information basis for subsequent analysis.

[0083] Next, a dynamic clustering algorithm is used to perform cluster analysis on these wear dynamic feature vectors to obtain multiple wear state clusters. The advantage of the dynamic clustering algorithm is that it can adaptively adjust the number and boundaries of clusters to adapt to the diverse wear patterns of wire drawing dies under different working conditions. For example, the system may identify three main wear state clusters: normal wear cluster (low wear, stable rate and acceleration), accelerated wear cluster (significant increase in wear rate or acceleration) and critical wear cluster (high wear, close to the replacement threshold). This clustering result provides a clear classification basis for subsequent anomaly detection and early warning.

[0084] Based on these wear state clusters and the uncertainty estimates obtained previously, the system then constructs a multidimensional decision tree model to classify and evaluate each wear state cluster, thereby obtaining a preliminary abnormality type and severity determination result. The advantage of the multidimensional decision tree model is that it can handle complex nonlinear decision boundaries while providing an explainable decision path. In this process, the system not only considers the characteristics of the wear state cluster, but also incorporates the uncertainty estimate into the decision-making process. For example, for a wire drawing die classified as an accelerated wear cluster, if its uncertainty estimate is low (that is, the evaluation result is more reliable), the system may determine it as "moderate abnormality, requiring close monitoring"; if the uncertainty estimate is high, even if the wear state is similar, the system may more conservatively determine it as "serious abnormality, it is recommended to check as soon as possible."

[0085] Finally, the time series prediction model is used to perform trend analysis on these preliminary abnormality type and severity determination results, and combined with historical data, anomaly detection reports and warning information containing abnormality types, severity and development trends are obtained. The application of the time series prediction model enables the system to not only evaluate the current status, but also predict future development trends. For example, for a wire drawing die that is judged to be "moderately abnormal", the time series model may predict that if the current usage intensity is maintained, the wire drawing die has an 80% probability of entering a "severe abnormal" state in the next 7 days. This predictive analysis greatly enhances the system's early warning capabilities and provides valuable forward-looking information for maintenance decisions.

[0086] This multi-step abnormal state analysis method has several significant advantages in the detection of the center hole of the wire drawing die. First, the multi-dimensional decomposition provides a comprehensive description of the wear state, not only focusing on the current state, but also considering the trend of change, making the analysis more dynamic and forward-looking. Secondly, the application of the dynamic clustering algorithm enables the system to adapt to complex and changeable wear patterns, improving the flexibility and accuracy of the analysis. Furthermore, the multi-dimensional decision tree model enhances the reliability and robustness of the decision by integrating uncertainty estimation when making abnormal judgments. Finally, the introduction of the time series prediction model enables the system to have predictive analysis capabilities, turning from passive response to active prevention. This method can not only accurately identify the current abnormal state, but also predict possible problems in the future, providing comprehensive, accurate and forward-looking decision support for the maintenance and management of wire drawing dies. Through this advanced abnormal state analysis method, enterprises can arrange maintenance plans more accurately, effectively prevent sudden failures, significantly improve equipment utilization and production efficiency, and reduce maintenance costs and quality risks.

[0087] In one embodiment of the present invention, the dynamic clustering algorithm is used to perform cluster analysis on the wear dynamic feature vector to obtain multiple wear state clusters, including: adaptively normalizing the wear dynamic feature vector to obtain a standardized feature vector; clustering the standardized feature vector using a density peak clustering algorithm to obtain a first group of wear state clusters; based on the first group of wear state clusters, a spectral clustering algorithm is used to construct a similarity graph, and feature space mapping is performed to obtain a second group of wear state clusters; the second group of wear state clusters is evaluated and screened using the silhouette coefficient and the Davis-Boulding index to obtain multiple wear state clusters.

[0088] Specifically, the wear dynamic feature vector is first adaptively normalized to obtain a standardized feature vector. The importance of this step is to eliminate the dimensional differences between different features so that each feature has the same weight in subsequent analysis. For example, the degree of wear may be expressed as a percentage ranging from 0 to 100, while the wear rate may be expressed as a percentage change per day, which may be a very small value. Through adaptive normalization, these features of different scales are converted to the same standard range, usually between 0 and 1. This processing not only ensures the comparability of different features, but also adapts to the dynamic changes in data distribution, improving the accuracy and robustness of clustering analysis.

[0089] Next, the standardized feature vectors are clustered using the density peak clustering algorithm to obtain the first set of wear state clusters. The advantage of the density peak clustering algorithm is that it can automatically determine the cluster center and the number of clusters, and is particularly suitable for processing non-spherical distribution data. This feature is particularly important in the analysis of the wear state of drawing dies, because the wear state often presents a complex nonlinear distribution. For example, the algorithm may identify three main wear state clusters: normal wear cluster, accelerated wear cluster, and critical wear cluster, each of which represents a specific wear mode.

[0090] Based on the first group of wear state clusters, the spectral clustering algorithm is applied to construct a similarity graph and perform feature space mapping to obtain the second group of wear state clusters. The introduction of the spectral clustering algorithm further improves the accuracy of clustering, especially when dealing with complex nonlinear separable data. By constructing a similarity graph, the algorithm can capture the local structural relationship between data points, while feature space mapping helps to discover the intrinsic structure of the data. This step may refine or reorganize the previous clustering results. For example, it may further distinguish the two subclusters of "early accelerated wear" and "late accelerated wear" in the accelerated wear cluster.

[0091] Finally, the second group of wear state clusters are evaluated and screened using the silhouette coefficient and Davis-Polding index to obtain the final multiple wear state clusters. The silhouette coefficient measures the compactness within the cluster and the separation between clusters, while the Davis-Polding index evaluates the overall quality of the clustering. Through the comprehensive evaluation of these two indicators, the system is able to screen out the optimal clustering results. For example, if a clustering scheme has a high silhouette coefficient and a low Davis-Polding index, it indicates that this clustering result has both good intra-cluster compactness and obvious inter-cluster distinction, and is therefore selected as the final clustering result.

[0092] This multi-step clustering analysis method has significant advantages in the identification of the wear state of wire drawing dies. First, adaptive normalization ensures fair comparison of different features and improves the accuracy of clustering. Secondly, the application of the density peak clustering algorithm enables the system to automatically adapt to the complex wear pattern distribution. Furthermore, the introduction of the spectral clustering algorithm further optimizes the clustering results and can capture more subtle differences in wear states. Finally, the high quality and reliability of the final clustering results are ensured through the evaluation of the silhouette coefficient and the Davis-Boulding index. This method can not only accurately identify different wear states, but also adapt to the dynamic changes of wear patterns, providing strong data support for the precise maintenance and management of wire drawing dies.

[0093] Please continue reading Figure 1 , a multi-objective optimization algorithm is used to process the abnormal detection report and warning information for maintenance strategy decision-making, and a predictive maintenance decision suggestion including maintenance time, method and specific operation steps is obtained.

[0094] In one embodiment of the present invention, the maintenance strategy decision processing of the abnormal detection report and the warning information using a multi-objective optimization algorithm to obtain a predictive maintenance decision suggestion including maintenance time, method and specific operation steps includes: using a pre-built multi-objective optimization model to process the abnormal detection report and the warning information to obtain an initial maintenance strategy candidate set; applying a genetic algorithm to the initial maintenance strategy candidate set for optimization iteration, and obtaining a Pareto optimal solution set through crossover, mutation and selection operations; using the Monte Carlo method to perform random scenario simulation on each solution in the Pareto optimal solution set, evaluating the performance of different maintenance strategies, and obtaining a robustness index for each maintenance strategy; based on the robustness index and the preset decision maker preference weights, using the hierarchical analysis method to comprehensively evaluate and sort the Pareto optimal solution set, selecting the optimal maintenance strategy, and obtaining a predictive maintenance decision suggestion including maintenance time, method and specific operation steps.

[0095] Specifically, a pre-built multi-objective optimization model is a key component. This model is a mathematical framework that contains three main objective functions: minimizing maintenance costs C(x), maximizing equipment availability A(x), and maximizing production quality Q(x). Among them, C(x) calculates the total cost of various maintenance activities, A(x) evaluates the impact of maintenance activities on equipment operating time, and Q(x) quantifies the improvement effect of maintenance on product quality. Specifically:

[0096] ;

[0097] in, is the cost of each maintenance activity, is the number of times the maintenance activity is performed.

[0098] ;

[0099] in, is the cost of each maintenance activity, is the time of each maintenance activity, and T is the total time.

[0100] ;

[0101] in, is the potential improvement in quality for each maintenance activity, is the number of executions of the maintenance activity, is the improvement rate.

[0102] The resource limits are: ;

[0103] in, are the resources required for each maintenance activity, is the total available resources.

[0104] The time window limits are: ;

[0105] in, and are the earliest and latest execution times of the maintenance activity, respectively.

[0106] The mathematical expression of the model is: .

[0107] The decision variable x of the model represents the execution time and frequency of different maintenance activities. At the same time, the model takes into account constraints such as resource limitations and time windows to ensure that the generated maintenance strategy is feasible in actual operation. In the implementation process, the anomaly detection report and warning information are first converted into model parameters, and then the model is solved using a multi-objective optimization algorithm (such as NSGA-II) to generate a series of non-dominated solutions. These solutions are then converted into specific maintenance strategies, such as "replace the drawing die immediately" or "increase the inspection frequency and replace it within two weeks". These strategies constitute the initial maintenance strategy candidate set, which provides a scientific and comprehensive basis for the subsequent decision-making process, effectively balancing multiple goals such as cost, availability and quality, making the maintenance decision of the drawing die more accurate and efficient.

[0108] Next, the genetic algorithm is applied to this initial set of maintenance strategy candidates for optimization iterations, and the Pareto optimal solution set is obtained through crossover, mutation, and selection operations. The genetic algorithm simulates the biological evolution process, in which each maintenance strategy is regarded as an "individual". The algorithm gradually improves the strategy set through repeated crossover (combining the characteristics of different strategies), mutation (randomly changing certain aspects of the strategy) and selection (retaining strategies with better performance). For example, the two strategies of "replace immediately" and "replace at the next planned downtime" may produce a new strategy of "find the best time to replace within three days" through crossover operations. After many iterations, the algorithm finally obtains a set of Pareto optimal solutions, each of which cannot be fully surpassed by other strategies in a certain goal. The advantage of this method is that it can find the best balance between multiple goals, rather than simply optimizing a single goal.

[0109] Then, the Monte Carlo method is used to simulate random scenarios for each solution in the Pareto optimal solution set, evaluate the performance of different maintenance strategies, and obtain the robustness indicators of each maintenance strategy. The Monte Carlo method evaluates the performance of strategies under various possible situations through a large number of random simulations. For example, for the strategy of "finding the best time to replace within three days", the system simulates multiple scenarios such as different production loads, different wear progression rates, and even possible emergencies (such as temporary equipment failure). Through thousands or even tens of thousands of simulations, the system is able to evaluate the performance of each strategy in the face of uncertainty, thereby deriving the robustness indicators of the strategy. The importance of this step lies in that it takes into account various uncertainties that may be encountered in actual production, ensuring that the selected strategy not only performs well under ideal conditions, but also has good adaptability under various possible situations.

[0110] Finally, based on these robustness indicators and the preset decision-maker preference weights, the AHP is used to comprehensively evaluate and sort the Pareto optimal solution set, select the optimal maintenance strategy, and obtain predictive maintenance decision recommendations including maintenance time, method, and specific operation steps. The AHP is a structured decision-making method that decomposes complex decision-making problems into multiple levels. In this process, the system considers multiple factors, such as the cost-effectiveness ratio of the strategy, the impact on production, and the ease of implementation. Each factor is assigned a weight that reflects the decision-maker's preference. For example, if the decision-maker attaches more importance to cost control, cost-related indicators will be assigned a higher weight. In this way, the system can comprehensively evaluate and sort each strategy based on multiple factors. The optimal strategy finally selected not only performs best in the mathematical model, but also meets actual management needs and decision-making preferences.

[0111] This multi-step maintenance decision-making method has several significant advantages in the detection and maintenance of the center hole of the wire drawing die. First, the use of the multi-objective optimization model ensures that the decision-making process takes into account multiple relevant factors, avoiding the one-sidedness that may be caused by single-objective optimization. Secondly, the application of genetic algorithms enables the system to efficiently search for the optimal solution in a huge strategy space, greatly improving the quality of decision-making. Furthermore, the introduction of the Monte Carlo method enhances the robustness of the decision, so that the selected strategy can cope with various possible situations. Finally, the use of the hierarchical analysis method ensures that the final decision is not only optimal in theory, but also meets the actual management needs. This method can not only generate high-quality maintenance decisions, but also adapt to different decision-making preferences and management styles, providing strong decision-making support for the precise maintenance and efficient management of wire drawing dies. Through this advanced decision-making method, enterprises can maximize production efficiency while ensuring equipment reliability, significantly reduce maintenance costs, improve resource utilization, and ultimately achieve continuous optimization of the production process and maximization of benefits.

[0112] Another embodiment of the present application further provides a central hole detection device for a wire drawing die, Figure 2 The present invention is a structural schematic diagram of a center hole detection device for a wire drawing die. Specifically, the center hole detection device for a wire drawing die includes:

[0113] The multimodal data acquisition module 1 is used to synchronously acquire the sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multimodal data;

[0114] A feature extraction processing module 2 is used to preprocess the time-aligned multimodal data and extract features from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector;

[0115] A wear state identification module 3 is used to perform wear state identification processing on the multimodal feature vector using multiple preset models to obtain a wear state evaluation result of the center hole of the wire drawing die and an uncertainty estimation thereof;

[0116] An abnormality analysis and processing module 4 is used to perform abnormality analysis and processing on the wear state evaluation result and its uncertainty estimation using a multi-dimensional threshold analysis method, and obtain an abnormality detection report and early warning information including abnormality type, severity and development trend;

[0117] The maintenance decision generation module 5 is used to use a multi-objective optimization algorithm to perform maintenance strategy decision processing on the abnormal detection report and warning information to obtain predictive maintenance decision suggestions including maintenance time, method and specific operation steps.

[0118] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A method for detecting the center hole of a wire drawing die, characterized in that: include: The sound, vibration and temperature signals around the wire drawing die during the wire drawing operation are synchronously collected to obtain time-aligned multi-modal data; The method of synchronously collecting the sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multimodal data includes: collecting the sound signals around the wire drawing die during the wire drawing operation at high frequency to obtain sound data; collecting the vibration signals on the surface of the wire drawing die at equal time intervals according to the collection time point of the sound data to obtain vibration data corresponding to the time of the sound data; collecting the temperature signals at preset positions on the wire drawing die at fixed time intervals based on the collection time point of the vibration data to obtain discrete temperature data; aligning the timestamps of the sound data, the vibration data and the discrete temperature data, and performing data interpolation to obtain time-aligned multimodal data with consistent sampling frequencies; The time-aligned multimodal data is preprocessed, and features are extracted from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector; the time-aligned multimodal data is preprocessed, and features are extracted from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector, including: performing wavelet transform and singular spectrum analysis on the time-aligned multimodal data to obtain a time-frequency domain feature matrix of each mode; performing adaptive threshold filtering on the time-frequency domain feature matrix, Obtain a denoised feature representation; input the denoised feature representation into the corresponding modal branches of a multimodal feature fusion network according to the sound, vibration and temperature modes, respectively, each modal branch contains multiple convolutional layers, and performs independent nonlinear transformations on each modal feature to obtain modality-specific deep features; use the cross-modal attention layer of the multimodal feature fusion network to interact and weight the modality-specific deep features to obtain fused features that consider the relationship between modalities; perform global pooling and fully connected layer processing on the fused features to obtain a fused multimodal feature vector; The multimodal feature vector is processed for wear state identification by using multiple preset models to obtain the wear state evaluation result of the center hole of the wire drawing die and its uncertainty estimate; The multimodal feature vector is processed for wear state identification by using multiple preset models to obtain the wear state evaluation result of the center hole of the wire drawing die and its uncertainty estimate, including: inputting the multimodal feature vector into three preset models of random forest, support vector machine and deep neural network respectively, performing parallel calculation processing, and obtaining three sets of independent preliminary wear state evaluation results; using Bayesian model averaging method to assign dynamic weights to each preset model according to the three sets of independent preliminary wear state evaluation results, and obtaining weighted fusion wear state evaluation results; performing Monte Carlo sampling on the weighted fusion wear state evaluation results to generate a large number of simulated samples and obtain the probability distribution of the wear state; using kernel density estimation method to calculate the confidence interval of the probability distribution of the wear state, and obtain the uncertainty estimate of the wear state evaluation; A multi-dimensional threshold analysis method is used to perform abnormal state analysis on the wear state assessment result and its uncertainty estimation, so as to obtain an abnormality detection report and early warning information including abnormality type, severity, and development trend; A maintenance strategy decision is processed on the abnormal detection report and the warning information by using a multi-objective optimization algorithm to obtain a predictive maintenance decision suggestion including maintenance time, method and specific operation steps; The maintenance strategy decision is processed on the abnormal detection report and the warning information by using a multi-objective optimization algorithm to obtain a predictive maintenance decision suggestion including maintenance time, method and specific operation steps, including: processing the abnormal detection report and the warning information by using a pre-built multi-objective optimization model to obtain an initial maintenance strategy candidate set; applying a genetic algorithm to the initial maintenance strategy candidate set for optimization iteration, and obtaining a Pareto optimal solution set through crossover, mutation and selection operations; using a Monte Carlo method to perform random scenario simulation on each solution in the Pareto optimal solution set, evaluating the performance of different maintenance strategies, and obtaining a robustness index of each maintenance strategy; based on the robustness index and the preset decision maker preference weight, using a hierarchical analysis method to comprehensively evaluate and sort the Pareto optimal solution set, selecting the optimal maintenance strategy, and obtaining a predictive maintenance decision suggestion including maintenance time, method and specific operation steps.

2. The method for detecting the center hole of a wire drawing die according to claim 1, characterized in that: The multi-dimensional threshold analysis method is used to perform abnormal state analysis processing on the wear state assessment result and its uncertainty estimation, and obtain an abnormality detection report and early warning information including abnormality type, severity, and development trend, including: Decomposing the wear state evaluation result in multiple dimensions to obtain a wear dynamic feature vector representing the wear degree, wear rate and wear acceleration; Performing cluster analysis on the wear dynamic feature vector using a dynamic clustering algorithm to obtain multiple wear state clusters; Based on the wear state clusters and uncertainty estimates, a multidimensional decision tree model is constructed to classify and evaluate each wear state cluster to obtain preliminary abnormality type and severity determination results; A time series prediction model is used to perform trend analysis on the preliminary abnormality type and severity determination results, and combined with historical data, an abnormality detection report and early warning information including abnormality type, severity, and development trend are obtained.

3. The method for detecting the center hole of a wire drawing die according to claim 2, characterized in that: The dynamic clustering algorithm is used to perform cluster analysis on the wear dynamic feature vector to obtain multiple wear state clusters, including: Adaptively normalizing the wear dynamic feature vector to obtain a standardized feature vector; Clustering the standardized feature vectors using a density peak clustering algorithm to obtain a first group of wear state clusters; Based on the first group of wear state clusters, a spectral clustering algorithm is applied to construct a similarity graph, and feature space mapping is performed to obtain a second group of wear state clusters; The second group of wear state clusters are evaluated and screened using the silhouette coefficient and the Davis-Paulding index to obtain a plurality of wear state clusters.

4. A center hole detection device for a wire drawing die, used to implement the center hole detection method for a wire drawing die as claimed in claim 1, characterized in that: include: A multimodal data acquisition module is used to synchronously acquire the sound, vibration and temperature signals around the wire drawing die during the wire drawing operation to obtain time-aligned multimodal data; A feature extraction processing module is used to preprocess the time-aligned multimodal data and extract features from the preprocessed multimodal data through a multimodal feature fusion network to obtain a fused multimodal feature vector; A wear state identification module, used to perform wear state identification processing on the multimodal feature vector using a plurality of preset models, and obtain a wear state evaluation result of the center hole of the wire drawing die and an uncertainty estimation thereof; An abnormality analysis and processing module, used to perform abnormality analysis and processing on the wear state assessment result and its uncertainty estimation using a multi-dimensional threshold analysis method, and obtain an abnormality detection report and early warning information including abnormality type, severity and development trend; The maintenance decision generation module is used to use a multi-objective optimization algorithm to perform maintenance strategy decision processing on the abnormal detection report and warning information to obtain predictive maintenance decision suggestions including maintenance time, method and specific operation steps.

Citation Information

Patent Citations

  • Ultrasonic machining device and machining method for diamond microporous wire-drawing die

    CN102672552A

  • Middle-deep geothermal coupling phase change heat storage system design optimization method with high calculation efficiency

    CN117556696A

  • Automatic wire drawing die hole drilling machine

    JP2002120138A

  • 3D surface reconstruction with point cloud densification using deep neural networks for autonomous systems and applications

    US20230135088A1