Method and system for managing superalloy fastener dies, computing device, and media

Abstract

The application discloses a high-temperature alloy fastener mold management method and system, a computing device and a medium. Based on the process of machining high-temperature alloy fasteners by a cold header mold, multi-dimensional working condition data of the cold header mold are synchronously collected. The multi-dimensional working condition data include acoustic emission signals, impact stress signals, temperature signals and machining timing data. The multi-dimensional working condition data are input into a wear prediction model to output a predicted wear degree of the cold header mold. The wear prediction model is built based on a machine learning algorithm. According to the predicted wear degree of the cold header mold, corresponding control strategies are executed for the process of machining high-temperature alloy fasteners by the cold header mold. In this way, by inputting the multi-dimensional working condition data into the wear prediction model to predict the wear degree of the mold and by real-time control of the mold operation based on the prediction result, the wear state of the mold can be accurately identified and graded, and the service life of the mold and the machining precision of the high-temperature alloy fasteners are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, in particular to a management method and system of a high-temperature alloy fastener mold, a computing device and a medium. BACKGROUND

[0002] High-temperature alloy is a metal material with nickel, cobalt, iron and other elements as matrix, adding chromium, molybdenum, tungsten, aluminum, titanium and other alloy elements, which can maintain high strength, oxidation resistance and creep resistance under the action of high temperature environment above 600℃ and certain stress. Cold upsetting is a forging process that forms a shaped part by applying high pressure impact force to a metal blank through a mold at room temperature. Due to the poor plasticity, fast work hardening and weak flowability of high-temperature alloy, the mold bears large load and intense friction during the cold upsetting process of forming high-temperature alloy fastener, which easily causes mold fatigue and wear and even early failure.

[0003] In the prior art, the maintenance of high-temperature alloy fastener mold mainly relies on experience to set a fixed replacement period or manual regular inspection, which makes it difficult to accurately predict and timely control the predicted wear degree of the mold, thereby affecting the process stability of cold upsetting, the consistency of high-temperature alloy fastener quality and the service life of the mold. Therefore, how to monitor the wear state of the mold in real time and actively control it during the cold upsetting process of high-temperature alloy fastener has become a technical problem to be solved. SUMMARY

[0004] The purpose of the present application is to provide a management method and system of a high-temperature alloy fastener mold, a computing device and a computer readable storage medium, which can predict the wear degree of the mold by inputting multi-dimensional working condition data into a wear prediction model, and can real-time control the mold operation based on the prediction result, so as to accurately identify the wear state of the mold and perform hierarchical control in time, thereby improving the service life of the mold and the processing precision of the high-temperature alloy fastener.

[0005] To achieve the above purpose:

[0006] In a first aspect, the embodiments of the present application provide a management method of a high-temperature alloy fastener mold, comprising the following steps:

[0007] The multi-dimensional working condition data of the cold header die is input into a wear prediction model to output a predicted wear degree of the cold header die; the wear prediction model is built based on a machine learning algorithm;

[0008] The multi-dimensional working condition data of the cold header die is input into a wear prediction model to output a predicted wear degree of the cold header die; the wear prediction model is built based on a machine learning algorithm;

[0009] According to the predicted wear degree of the cold header die, a corresponding control strategy is executed for the process of machining the high-temperature alloy fastener by using the cold header die.

[0010] In an embodiment, before the multi-dimensional working condition data is input into the wear prediction model, the method comprises:

[0011] Based on the historical multi-dimensional working condition data and the wear degree label of the cold header die, sample data is constructed;

[0012] The wear prediction model is built based on a machine learning model; the machine learning model comprises at least one of a random forest model, a support vector machine model and a convolutional neural network model;

[0013] The wear prediction model is trained by using the sample data.

[0014] In an embodiment, before the multi-dimensional working condition data is input into the wear prediction model, the method further comprises: for the wear prediction model based on a random forest or a support vector machine, digital preprocessing and feature construction are performed according to the physical characteristics of the multi-dimensional working condition data to obtain a numerical feature vector.

[0015] In an embodiment, the multi-dimensional working condition data is input into the wear prediction model, comprising:

[0016] The multi-dimensional working condition data is respectively intercepted into sliding window segments of different time scales;

[0017] The sliding window segments of each time scale are fused and input into the wear prediction model; for the wear prediction model built based on random forest or support vector machine, the fusion processing includes splicing the sliding window segments of each time scale to form a multi-scale feature vector; for the wear prediction model built based on convolutional neural network, the fusion processing includes constructing a signal tensor based on the sliding window segments of each time scale.

[0018] In an embodiment, the corresponding control strategy for the process of machining high-temperature alloy fasteners by the cold header die according to the predicted wear degree of the cold header die includes at least one of the following:

[0019] If the cold header die is slightly worn, the oil injection frequency of the lubrication unit of the cold header die is increased;

[0020] If the cold header die is moderately worn, the ram speed of the stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range;

[0021] If the cold header die is severely worn, a die replacement warning is triggered.

[0022] In an embodiment, before the ram speed of the stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range, the method comprises:

[0023] A digital twin model is constructed based on the physical parameters of the cold header die and the material properties of the high-temperature alloy;

[0024] The cold heading process parameters to be adjusted are input into the digital twin model to obtain a simulation result of the high-temperature alloy fastener machined by the cold header die; the cold heading process parameters to be adjusted include at least one of the adjusted ram speed and pre-tightening force of the stamping pressure module;

[0025] If the simulation result meets the quality requirements of the high-temperature alloy fastener, the stamping pressure module is adjusted according to the cold heading process parameters to be adjusted;

[0026] If the simulation result does not meet the quality requirements of the high-temperature alloy fastener, the cold heading process parameters to be adjusted are iteratively optimized until the simulation result meets the quality requirements of the high-temperature alloy fastener.

[0027] In an embodiment, after the corresponding control strategy for the process of machining high-temperature alloy fasteners by the cold header die according to the predicted wear degree of the cold header die is executed, the method comprises:

[0028] determine a die wear rate index based on historical operation data of the cold header die; the historical operation data at least includes a historical predicted wear degree, historical processing timer data and a historical control strategy;

[0029] determine whether the cold header die currently stays in a rapid deterioration period or an abnormal fluctuation period according to the die wear rate index;

[0030] if the cold header die currently stays in the rapid deterioration period, strengthen the execution mode of the control strategy;

[0031] if the cold header die currently stays in the abnormal fluctuation period, increase the sampling frequency of the current state data of the cold header die, and / or, simulate and verify the current wear degree of the cold header die.

[0032] In a second aspect, the embodiments of the present application provide a management system of a high-temperature alloy fastener die, the system comprising a collection module, a prediction module and a control module; wherein,

[0033] the collection module synchronously collects multi-dimensional working condition data of the cold header die during the process of processing the high-temperature alloy fastener by the cold header die; the multi-dimensional working condition data includes acoustic emission signals, impact stress signals, temperature signals and processing timer data, the acoustic emission signals are collected based on acoustic emission sensors, the acoustic emission sensors are installed in a vibration conduction area of an associated structure of the cold header die in a non-contact manner; the impact stress signals are collected based on impact stress sensors, the impact stress sensors are arranged on a fixed structure on a load transmission path of the cold header die; the temperature signals are collected based on temperature sensors, the temperature sensors are embedded thermocouples or non-contact infrared temperature measuring heads; the collection time stamps of the acoustic emission sensors, the impact stress sensors and the temperature sensors are synchronized based on a processing timer;

[0034] the prediction module inputs the multi-dimensional working condition data into a wear prediction model to output a predicted wear degree of the cold header die; the wear prediction model is built based on a machine learning algorithm;

[0035] the control module executes a corresponding control strategy on the process of processing the high-temperature alloy fastener by the cold header die according to the predicted wear degree of the cold header die.

[0036] In a third aspect, the embodiments of the present application provide a computing device, specifically comprising:

[0037] a processor;

[0038] a memory for storing processor-executable instructions;

[0039] The processor is configured to execute instructions for performing the management method of the high-temperature alloy fastener mold according to the first aspect.

[0040] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, and the computer readable storage medium stores a computer program. When the instructions in the computer readable storage medium are executed by a processor of a computing device, the computing device can implement the management method of the high-temperature alloy fastener mold according to the first aspect.

[0041] The management method of the high-temperature alloy fastener mold, the computing device and the computer readable storage medium provided by the embodiments of the present application, the management method of the high-temperature alloy fastener mold comprises: based on the process of machining the high-temperature alloy fastener by the cold header mold, synchronously collecting multi-dimensional working condition data of the cold header mold; the multi-dimensional working condition data comprises acoustic emission signals, impact stress signals, temperature signals and machining timing data, the acoustic emission signals are collected based on acoustic emission sensors, the acoustic emission sensors are installed in a vibration conduction area of an associated structure of the cold header mold in a non-contact manner; the impact stress signals are collected based on impact stress sensors, the impact stress sensors are arranged on a fixed structure on a load transmission path of the cold header mold; the temperature signals are collected based on temperature sensors, the temperature sensors are embedded thermocouples or non-contact infrared temperature measuring heads; the collection time stamps of the acoustic emission sensors, the impact stress sensors and the temperature sensors are synchronized based on a machining timer; the multi-dimensional working condition data is input into a wear prediction model to output a predicted wear degree of the cold header mold; the wear prediction model is built based on a machine learning algorithm; and corresponding control strategies are executed for the process of machining the high-temperature alloy fastener by the cold header mold according to the predicted wear degree of the cold header mold. In this way, by inputting the multi-dimensional working condition data into the wear prediction model to predict the wear degree of the mold, and based on the prediction result, the mold operation is controlled in real time, the wear state of the mold can be accurately identified and graded in time, and the service life of the mold and the machining precision of the high-temperature alloy fastener are improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart of the management method of the high-temperature alloy fastener mold provided by the embodiments of the present application is shown.

[0043] Figure 2 The structural diagram of the management system of the high-temperature alloy fastener mold provided by the embodiments of the present application is shown.

[0044] Figure 3 The structural diagram of the computing device provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0045] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description herein relates to the drawings, in which the same numbers represent the same or similar elements, throughout several figures. The following exemplary embodiments are described in order to provide a thorough understanding of the present application. It is understood that the application is not limited to the embodiments described and as will be obvious those skilled in the art, can make modifications without departing from the scope of the present application. The following exemplary embodiments described herein are examples of apparatuses and methods consistent with some aspects of the present application as set forth in the appended claims.

[0046] It should be noted that, as used in this document, the terms "includes," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements is not limited to those elements, but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a... " does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that includes the element. Also, components having the same name in different embodiments can have the same meaning or different meanings, as will be apparent from their description in the context of the particular embodiment.

[0047] It should be understood that although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a particular order or hierarchy among the information. These terms are used only to distinguish one category of information from another. For example, a first information can be termed a second information, and similarly, a second information can be termed a first information without departing from the scope of this document. As used herein, the term "if' can be interpreted to mean "when" or "upon" or "in response to determining" taking into account the context in which the term is used. Also, as used herein, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context indicates otherwise. It will be further understood that the terms "comprises," "comprising," "includes," "including," and the like, when used in this document, specify the presence of stated features, steps, operations, elements, components, items, and / or groups but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, items, and / or groups thereof. As used herein, the term "or" and "and / or" is intended to mean an inclusive-or, or is intended to mean any one of several alternatives, or an any combination of alternatives. Thus, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C." Only when changes of subject matter are inherently mutually exclusive, do they appear in different embodiments.

[0048] It should be understood that although each step in the flowchart in the embodiments of the present application is shown in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless explicitly stated herein, the execution of these steps is not strictly limited in sequence, and they can be executed in other sequences. Moreover, at least part of the steps in the figure can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be alternately executed with at least part of other steps or sub-steps or stages of other steps.

[0049] It should be noted that in this paper, step codes such as S1, S2, etc. are used, the purpose of which is to more clearly and briefly express the corresponding content, and does not constitute a substantial limitation on the order. In specific implementation, a person skilled in the art may perform S2 before performing S1, etc., but these should be within the scope of protection of the present application.

[0050] It should be understood that the specific embodiments described herein are merely used to explain the present application and do not limit the present application.

[0051] In the following description, the suffix such as "module", "component" or "unit" used to represent an element is only for the convenience of the description of the present application, and has no specific meaning. Therefore, "module", "component" or "unit" can be used interchangeably.

[0052] Reference Figure 1 To cope with the above challenges, the embodiments of the present application provide a management method for a high-temperature alloy fastener mold. In the embodiments, the management method for the high-temperature alloy fastener mold is applied to a computing device. The management method for the high-temperature alloy fastener mold provided by the embodiments includes the following steps:

[0053] Step S1: During the process of machining the high-temperature alloy fastener by the cold header mold, multi-dimensional working condition data of the cold header mold are synchronously collected; the multi-dimensional working condition data include acoustic emission signals, impact stress signals and temperature signals; the acoustic emission signals are collected based on an acoustic emission sensor, the acoustic emission sensor is installed in a vibration conduction area of an associated structure of the cold header mold in a non-contact manner; the impact stress signals are collected based on an impact stress sensor, the impact stress sensor is arranged on a fixed structure on a load transmission path of the cold header mold; the temperature signals are collected based on a temperature sensor, the temperature sensor is an embedded thermocouple or a non-contact infrared temperature measuring head; the collection time stamps of the acoustic emission sensor, the impact stress sensor and the temperature sensor are synchronized based on a machining timer.

[0054] It can be understood that during the process of machining high-temperature alloy fasteners, the high-temperature alloy material is deformed plastically in the mold cavity, and a closed machining space is formed between the mold and the material. At the same time, the cold heading machine mold bears high-frequency and high-intensity impact load applied by the stamping pressure module. Therefore, the cold heading machine mold is prone to fatigue wear under the combined action of severe friction, high impact load and heat accumulation, and the wear will intensify over time, and the single-dimensional monitoring data of the cold heading machine mold cannot fully and accurately represent the wear evolution process. The present application can capture the state changes of the mold during the working process from different dimensions by collecting multi-dimensional working condition data such as acoustic emission signals, impact stress signals and temperature signals, avoiding the limitations of using single-dimensional working condition data to evaluate the wear of the mold, and improving the accuracy and comprehensiveness of the wear monitoring.

[0055] Optionally, the acoustic emission signal can be used to detect the micro-damage or micro-cracks inside the mold that are difficult to detect with the naked eye. The acoustic emission signal can be a high-frequency elastic wave, wherein the high-frequency elastic wave is a stress wave with high frequency (100 kHz-1 MHz) propagating in an elastic medium. By collecting the acoustic emission signal sequence distributed along the time axis, the development and change process of the micro-damage or micro-cracks inside the mold can be represented, thereby improving the sensitivity and interpretability of the wear state of the mold.

[0056] Optionally, the impact stress signal is used to quantify the instantaneous load borne by the mold during the cold heading stamping process, such as the maximum stress value, the impact loading rate, the stress variation coefficient, etc. By collecting the impact stress signal sequence distributed along the time axis, the mechanical response of the mold under high-speed impact can be reflected, such as stress concentration, stress anomaly, etc.

[0057] Optionally, the temperature signal is used to record the thermal energy distribution of the mold surface or near-surface during the cold heading process, so as to obtain the instantaneous temperature rise and thermal fatigue accumulation effect of the mold. High temperature may cause material performance degradation, and local overheating may cause cracks. The time sequence analysis of the temperature signal can determine the thermal parameters of the mold, such as temperature rise rate, temperature fluctuation range, temperature peak value, local hot spot, etc., which are used to assist in evaluating the coupling effect of thermal stress and mechanical stress of the mold, such as thermal fatigue accumulation and local annealing phenomenon, etc.

[0058] In this way, the embodiments of the present application combine the physical mechanisms of mold wear caused by cold heading machining, analyze physical phenomena such as friction contact, thermal fatigue, impact damage and micro-crack propagation, and select multi-dimensional working condition data that can directly represent these mechanisms, thereby improving the response sensitivity and interpretability of the wear behavior.

[0059] Optionally, the acoustic emission signal is collected based on an acoustic emission sensor. The acoustic emission sensor is arranged in the mold area and is preferably mounted on the vibration conduction area of the associated structure of the cold header mold, such as the external metal area of the mold clamping structure or the die holder, by indirect coupling. In this way, compared with the arrangement of being directly embedded in the mold cavity, the non-contact mounting not only avoids interference with the machining area of the mold, but also effectively transmits the acoustic emission signal through the rigid structure path. The acoustic emission sensor can be selected as a high-sensitivity wide-band model, and combined with a high-performance preamplifier, to ensure that the mold outside still has sufficient signal-to-noise ratio, and meet the collection requirements of acoustic emission events such as micro-cracks and friction slip.

[0060] Optionally, the impact stress signal is collected based on an impact stress sensor. The impact stress sensor is arranged on the fixed structure on the load transmission path of the cold header mold, such as the punch connecting structure or the mold mounting base position. In this way, the impact stress sensor can be ensured to be on the load transmission path, and the high-speed moving parts such as the punch are avoided, and the original structure of the machining equipment does not need to be changed, which is convenient for integration and maintenance in industrial field. Specifically, the impact stress sensor can be integrated below the mold mounting base or on the side of the punch impact direction by clamping or flange mounting, to collect the stress change of the mold at the moment of forming the high-temperature alloy fastener in real time.

[0061] Optionally, the temperature signal is collected based on a temperature sensor. According to the on-site process, an embedded thermocouple for directly measuring the temperature signal can be selected, or a non-contact infrared temperature measuring head for indirectly measuring the temperature signal can be selected. The embedded thermocouple can be insulated and protected and arranged in a pre-set small hole on the side of the mold fixing block or the die holder, to realize high-speed response to the temperature signal. The non-contact infrared temperature measuring head can be fixed on the open window of the mold area, to monitor the surface temperature of the forming area in real time through the infrared window, and is suitable for high-speed and non-contact measurement scene.

[0062] Optionally, the synchronous acquisition of multi-dimensional working condition data is realized by a processing timer. The processing timer is used to provide long-term use working condition information of the mold, and cumulative processing time, beat number, average beat interval, beat fluctuation coefficient and other indexes of the mold can be extracted. Specifically, the processing timer is embedded in the control system, and signals are output to each sensor by accessing the cold header PLC (Programmable Logic Controller) to synchronize the time stamp. To realize multi-sensor collaborative acquisition, a unified data acquisition platform is constructed, and based on a high-performance multi-channel data acquisition module, all channels share a unified clock source. The simultaneous start of the acoustic emission, stress and temperature channels can be controlled by a main trigger signal, and the collected data is automatically bound to a high-precision time stamp to ensure the time consistency and event comparability between different physical quantities. For sensors with inconsistent sampling frequencies, the time stamp can be aligned by downsampling high-frequency signals or upsampling low-frequency signals through interpolation. For example, if the sampling interval of the temperature sensor is 1 second and the sampling rate of the acoustic emission sensor is 1 MHz, the temperature data is distributed to the corresponding acoustic emission signal time window through an interpolation algorithm.

[0063] The above-mentioned embodiments fully consider the particularity of the closed nature, compact structure and impact environment of the cold header mold, and effectively realize the high-quality collaborative acquisition of multi-dimensional working condition data by reasonably avoiding the layout area of the mold inner cavity, using external rigid structure transmission, embedding or non-contact measurement technology. In addition, the data acquisition of each sensor is coordinated by a unified time reference signal or a main control trigger mechanism, the high-precision time stamp binding of multi-dimensional working condition data is realized, and the consistency of multi-dimensional working condition data in time sequence is ensured. In this way, the physical correlation and time comparability between multi-modal working condition characteristics are significantly enhanced, providing a stable data foundation for subsequent feature extraction, prediction modeling and dynamic control.

[0064] Step S2: inputting the multi-dimensional working condition data into the wear prediction model to output the predicted wear degree of the cold header mold; the wear prediction model is built based on a machine learning algorithm.

[0065] Optionally, a machine learning method such as Random Forest (RF), Support Vector Machine (SVM) or Convolutional Neural Network (CNN) can be used to establish the wear prediction model. The Random Forest is composed of multiple decision trees, each of which independently predicts the sample, and the final prediction result is determined by the decision results of all decision trees. The randomness of the Random Forest is reflected in the two aspects of randomly selecting samples to construct decision trees and randomly selecting features to split each node. The Support Vector Machine is a binary classification model that maximizes the classification interval by finding a hyperplane, thereby effectively segmenting the data. The Convolutional Neural Network is a feedforward neural network containing convolutional computation and having a deep structure, which automatically extracts local features of the input data through the hierarchical structure of convolutional layers, pooling layers and fully connected layers, gradually compresses information, reduces redundancy and improves generalization ability.

[0066] Optionally, for the wear prediction model established based on a deep learning model such as a Convolutional Neural Network, the original signal segment can be input into the pattern prediction model in the form of a tensor to automatically extract time series features for judgment.

[0067] In an embodiment, before the multi-dimensional working condition data is input into the wear prediction model, it includes: for the wear prediction model based on the Random Forest or the Support Vector Machine, according to the physical characteristics of the multi-dimensional working condition data, digital preprocessing and feature construction are performed to obtain a numerical feature vector with time series alignment and physical interpretability.

[0068] Optionally, for the wear prediction model established based on the Random Forest or the Support Vector Machine, before the multi-dimensional working condition data is input into the wear prediction model, since the model cannot directly process the original time series signal of the multi-dimensional working condition data, digital preprocessing and feature construction can be performed according to the physical characteristics of different monitoring data to extract a numerical feature vector with time series alignment and physical interpretability, and the constructed structured feature vector is input into the wear prediction model.

[0069] For acoustic emission signals, the collected high-frequency waveform signals are first denoised, filtered and windowed, and a sliding window or a synchronous window division strategy based on an impact cycle can be selected to extract the signal segment corresponding to a single forming process. Then, typical features are extracted from the time domain, frequency domain and statistical dimensions, including time domain features such as peak amplitude, root mean square value, energy envelope area, signal duration, zero-crossing rate, frequency domain features such as main frequency, spectral centroid and frequency band energy ratio, and statistical features reflecting signal distribution characteristics such as skewness and kurtosis. The above features can effectively reveal the friction, slip and micro-damage occurring between the die and the blank during the forming process.

[0070] For the impact stress signal, window segmentation and filtering smoothing are performed based on each impact cycle to extract the maximum stress value, impact loading rate, impact duration, stress variation coefficient, and impact contrast, which are used to describe the stress state and variation law of the mold during high-speed stamping and to identify stress abnormalities or overload working conditions.

[0071] For the temperature signal, first-order difference and detrend filtering operations are performed on the continuous low-frequency variation curve of the temperature signal to extract the peak temperature, average temperature, temperature rise rate, temperature fluctuation range, and temperature variation coefficient, which reflect the thermal load level and thermal fatigue trend of the mold during continuous forming. For the case of using non-contact infrared temperature measurement, the periodic temperature peak and average value sequences can be extracted according to the forming beat matching time period.

[0072] The processing timer data provides long-term use condition information of the mold, and directly extracts cumulative processing time, beat number, average beat interval, and beat fluctuation coefficient. These variables serve as time dimension inputs for wear evolution and can be combined with the features of acoustic emission signals, impact stress signals, and temperature signals to jointly model the wear prediction model, enhancing the sensitivity of the wear prediction model to the life change trend of the mold.

[0073] Optionally, real-time collected multi-dimensional working condition data such as acoustic emission signals, impact stress signals, and temperature signals are input into a wear prediction model based on machine learning to dynamically predict the wear degree of the cold header mold. The wear degree can be represented by wear grades such as mild wear, moderate wear, and severe wear. The wear degree can also be represented by wear speed, such as change frequency or change amplitude per unit time.

[0074] Optionally, in actual cold heading production, multi-dimensional working condition data are collected in real time and input into the trained wear prediction model to output the wear grade of the current mold, and the prediction result is fed back to the control system. For the case of multiple molds existing in the same device, working condition data of each mold can be collected independently and input into the same wear prediction model for judgment. In addition, multiple sets of multi-dimensional working condition data of different molds can be processed in parallel, and the multi-dimensional working condition data of different molds can be distinguished by identifying the mold number or station number to achieve parallel monitoring and unified prediction of multiple mold states.

[0075] In an embodiment, before inputting the multi-dimensional working condition data into the wear prediction model, the management method of the high-temperature alloy fastener mold comprises:

[0076] constructing sample data based on historical multi-dimensional working condition data and wear degree labels of the cold header mold;

[0077] The wear prediction model is built based on a machine learning model, and the machine learning model includes at least one of a random forest model, a support vector machine model, and a convolutional neural network model.

[0078] The wear prediction model is trained using sample data.

[0079] Optionally, before the multi-dimensional working condition data is input into the wear prediction model, the wear prediction model needs to be constructed and trained. First, sample data for training the wear prediction model is constructed. Historical multi-dimensional working condition data of the cold header die is obtained, such as acoustic emission signals, impact stress signals, and temperature signals. For each piece of multi-dimensional working condition data, a wear degree label matched therewith is determined.

[0080] The wear degree label is determined based on the actual wear state of the die, and the actual wear state of the die can be determined by evaluating the die by non-destructive detection means. For example, the forming surface of the die can be non-contact scanned by a high-precision laser three-dimensional profiler, the spatial topography of the die is reconstructed according to the three-dimensional profile scanning result of the die, the wear depth, profile deviation, feature edge degradation rate and other parameters of the die are extracted, and the geometric wear degree of the die in the forming area is quantitatively reflected. For another example, the surface hardness of several key areas of the die can be detected by a portable Vickers or Rockwell hardness tester, and the performance degradation degree of the die caused by heat accumulation or material fatigue is evaluated by comparing the current collected hardness data with the initial hardness value of the die. For another example, the current die state can be comprehensively evaluated and wear grade determined by experienced quality or equipment experts in combination with the die use cycle recorded by the processing timer, the number of forming beats, and the auxiliary features such as the typical working condition characteristics and abnormal event information (such as lubrication abnormality, impact overload, etc.) collected in the historical stage.

[0081] Based on the determination of the actual wear state of the die, the wear grading rules are formulated, and exemplarily, the die wear grade can be divided according to the following rules:

[0082] ① Normal: the die profile deviation does not exceed 10% of the design tolerance, the hardness decreases by less than 5%, and there is no obvious working condition abnormality;

[0083] ② Mild wear: the die has local profile deformation, the hardness decreases by 5%-10%, and the beat or lubrication fluctuation is small;

[0084] ③ Moderate wear: the die has regional wear depression, the hardness decreases by more than 10%, the beat fluctuation increases or there is lubrication response abnormality;

[0085] ④ Severe wear: the die has significant geometric deformation or collapse angle signs, the hardness is significantly reduced or part of the area fails to be detected, accompanied by multiple working condition abnormality records or ineffective near-end lubrication enhancement.

[0086] Then, based on the constructed sample data, a machine learning method such as random forest, support vector machine or convolutional neural network is used to establish a wear prediction model. For the wear prediction model established by random forest or support vector machine, feature selection needs to be performed on the multi-dimensional working condition data before the model is trained. For the case of using a deep learning model such as convolutional neural network to establish a wear prediction model, the deep learning model can automatically extract discriminative features from the multi-dimensional working condition data, without the need for feature selection.

[0087] Optionally, when performing feature selection on the multi-dimensional working condition data, the feature variables that are most discriminative for the wear state are identified and retained, and redundant or noise features are removed, thereby improving the learning efficiency and prediction accuracy of the model, and enhancing the interpretability of the model. Preferably, the feature selection can be combined with physical mechanisms, for example, the friction contact, thermal fatigue, micro-crack propagation, etc. mechanism is used to guide the design and selection of signal features, so that the extracted features have clear physical meaning, which helps to enhance the stability and interpretability of the model.

[0088] Optionally, for the acoustic emission signal, the peak amplitude, kurtosis, spectral centroid, etc. can be selected as the model input. The peak amplitude represents the energy transmission intensity at the impact moment, reflecting the energy level released during micro-crack propagation or material failure. The kurtosis is used to measure the sharpness of the signal, and high kurtosis often corresponds to a sudden and strong impact or instantaneous crack propagation. The spectral centroid is used to describe the concentration position of the signal energy in the frequency spectrum, and high frequency shift often indicates micro-crack growth or severe friction.

[0089] Optionally, for the impact stress signal, the maximum stress value, impact loading rate, stress variation coefficient, etc. can be selected as the model input. The maximum stress value and impact loading rate are used to represent the structural response intensity and loading rate during the forming impact process, and indirectly reveal the stress concentration degree of the die. The stress variation coefficient is used to quantify the stress fluctuation anomaly during the forming process, indicating that the die has fatigue risk.

[0090] Optionally, for the temperature signal, the temperature rise rate, peak temperature, temperature fluctuation range, etc. related to thermal load can be selected as the model input to reflect the thermal fatigue accumulation and local annealing phenomenon. At the same time, combined with the cumulative processing time and beat number collected by the processing timer, a beat variation coefficient is constructed to represent the time variable of the long-term use state, providing time sequence support for identifying the wear trend.

[0091] Finally, the sample data is used to train the wear prediction model. The sample data is divided into multiple subsets, and the wear prediction model is trained and cross-validated multiple times based on multiple sample data subsets to evaluate the stability and generalization ability of the model on different data, preventing overfitting. At the same time, based on the difference between the prediction results output by the model and the wear degree label, the recognition effect of each wear level is analyzed, and for the wear levels that are easily confused and the misjudgment samples, the distribution and causes of the prediction error are further explored. Combined with the evaluation conclusion, feedback to the feature construction, sample processing or model structure design link based on the error-driven closed-loop optimization mechanism. For example, for misjudgment samples concentrated in a particular working condition, further analyze whether the misjudgment prediction result is caused by sensor noise data under that particular working condition. If so, pre-process the sensor noise before inputting the sample data into the model. After multiple rounds of iterative training and adjustment, a stable and accurate wear prediction model is finally obtained.

[0092] In an embodiment, the multi-dimensional working condition data is input into the wear prediction model, including:

[0093] The multi-dimensional working condition data is respectively intercepted into sliding window segments of different time scales;

[0094] The sliding window segments of each time scale are fused and input into the wear prediction model. For a wear prediction model based on random forest or support vector machine, the fusion processing includes splicing the sliding window segments of each time scale to form a multi-scale feature vector. For a wear prediction model based on a convolutional neural network, the fusion processing includes constructing a signal tensor based on the sliding window segments of each time scale.

[0095] Optionally, a fixed scale (length) sliding time window can be set to intercept the real-time collected multi-dimensional working condition data. For example, a 5-second time window is set to intercept the multi-dimensional working condition data to generate continuous features or signal tensors for continuous dynamic monitoring of the mold wear state. The fixed length sliding time window can smooth out short-term random noise fluctuations (such as sensor transient interference or environmental impulse noise) by averaging or trending the multi-dimensional working condition data in the time dimension, thereby highlighting the persistent change characteristics of the wear state and enhancing the model's ability to identify the true degradation trend, effectively suppressing the interference of noise on the prediction results.

[0096] Optionally, different time length sliding window segments can be cut from the multi-dimensional working condition data at each prediction. Different time length data segments reflect different characteristics in the mold wear process. For example, a short time window (such as 5 seconds) can capture short-term dynamic characteristics of high-frequency changes such as impact stress transients, a medium time window (such as 20 seconds) can extract medium-term change patterns such as temperature periodic fluctuations, and a long time window (such as 60 seconds) can represent the long-term trend evolution of the overall wear degradation of the mold. It should be understood that for a wear prediction model based on a random forest or a support vector machine, the data features in each time window can be spliced into a fusion feature vector and input into the model; for a wear prediction model based on a convolutional neural network or other deep learning model, the signal segments of multiple time scales can be directly input into the model in tensor form, and the neural network can automatically complete the extraction and fusion of multi-level time sequence features, and combine the attention mechanism to enhance the response of the model to key time scale information, to realize deep joint modeling of wear evolution features at different time scales. In this way, multi-scale time sequence fusion prediction captures short-term fluctuations, medium-term changes and long-term trends at the same time, enhances the sensitivity of the model to early minor changes in wear state, and thus improves the recognition accuracy of the evolution process of the mold from "normal" to "mild", "moderate" and "severe".

[0097] Step S3: According to the predicted wear degree of the cold header die, a corresponding control strategy is executed for the process of machining the superalloy fastener by the cold header die.

[0098] Based on real-time multi-dimensional working condition data such as acoustic emission, impact stress and temperature signals, and combined with the wear prediction model, the wear grade of the cold header die is output. When the wear grade is mild, moderate or severe, a graded and differentiated control strategy is automatically triggered to cope with different degrees of mold wear state. In this way, a closed-loop control from wear state perception to dynamic intervention of the cold header die is realized, the service life of the mold is prolonged, and the machining accuracy of the superalloy fastener is improved.

[0099] In an embodiment, according to the predicted wear degree of the cold header die, a corresponding control strategy is executed for the process of machining the superalloy fastener by the cold header die, including at least one of the following:

[0100] If the cold header die is mildly worn, the oil injection frequency of the lubrication unit of the cold header die is increased;

[0101] If the cold header die is moderately worn, the ram speed of the stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range;

[0102] If the cold header die is severely worn, a mold replacement warning is triggered.

[0103] Optionally, the prediction result (i.e. the current die wear level) output by the wear prediction model is used to control the lubrication unit, the stamping pressure module and the warning module in linkage, so as to realize automatic adjustment and dynamic optimization control of the die parameters. Specifically, if the prediction result is normal, no adjustment is made; if the prediction result is light wear, the lubrication unit is linked to increase the oil injection frequency, so as to reduce the friction and heat accumulation between the die and the blank; if the prediction result is moderate wear, the stamping pressure module is linked to fine-tune the punch speed and the pre-tightening force, such as reducing the punch speed of the stamping pressure module of the cold header die and / or increasing the pre-tightening force of the stamping pressure module, so as to relieve the local impact load of the die. If the prediction result is severe wear, the warning module is linked to trigger the die change warning prompt.

[0104] Optionally, in order to prevent forming defects caused by parameter adjustment, a limit control and dynamic feedback mechanism can be used to make a slight adjustment only within an experience-determined safe window, so as to ensure that the die life is guaranteed while the forming precision and structural integrity of the high-temperature alloy screw are not affected. It can be understood that the safe window refers to setting the allowed range of parameter adjustment (such as the upper and lower limits of the lubrication frequency, the stamping speed and the pre-tightening force) in combination with the process experience and physical constraints, so as to ensure that the adjustment can both relieve the die wear and not cause forming defects or structural damage. For example, the range of the adjusted parameters is set to be less than 10% of the original parameters.

[0105] In an embodiment, before reducing the punch speed of the stamping pressure module of the cold header die and / or increasing the pre-tightening force of the stamping pressure module within a preset range, the management method of the high-temperature alloy fastener die comprises:

[0106] constructing a digital twin model based on the physical parameters of the cold header die and the material properties of the high-temperature alloy;

[0107] inputting the cold heading processing parameters to be adjusted into the digital twin model to obtain a simulation result of the high-temperature alloy fastener processed by the cold header die; the cold heading processing parameters to be adjusted include at least one of the punch speed and the pre-tightening force of the stamping pressure module to be adjusted;

[0108] if the simulation result meets the quality requirements of the high-temperature alloy fastener, adjusting the stamping pressure module according to the cold heading processing parameters to be adjusted;

[0109] if the simulation result does not meet the quality requirements of the high-temperature alloy fastener, iteratively optimizing the cold heading processing parameters to be adjusted until the simulation result meets the quality requirements of the high-temperature alloy fastener.

[0110] Optionally, the punch speed or pre-tightening force of the stamping pressure module can be virtually verified by the digital twin simulation module before adjustment, to ensure the stability of product quality after parameter adjustment. Specifically, based on the physical parameters of the cold header die (such as die material properties, geometric structure, surface hardness) and the material properties of the high-temperature alloy fastener (such as deformation resistance, thermal conductivity, forming limit), a digital twin model is constructed. The cold heading process parameters to be adjusted (punch speed and / or pre-tightening force of the stamping pressure module) are input into the digital twin model, and the forming process after adjustment is simulated and controlled, and the quality indicators of the high-temperature alloy fastener are evaluated, such as head thickness tolerance, thread forming accuracy, surface integrity, etc. If the simulation result meets the quality requirement, the stamping pressure module is adjusted according to the cold heading process parameters to be adjusted. That is, after virtual verification using the digital twin simulation model, the stamping pressure module is adjusted, so as to ensure that the wear of the cold header die is reduced without affecting the product quality. If the simulation result shows that the forming quality does not meet the standard, the cold heading process parameters are iteratively adjusted and the simulation test is continued in the digital twin model until the quality of the fastener meets the standard, and then the stamping pressure module is adjusted. In this way, through the prediction and verification capability of the digital twin model, both safety and effectiveness are ensured, and the risk of die damage or batch waste caused by blind adjustment is avoided.

[0111] In an embodiment, after the cold header die processes the high-temperature alloy fastener according to the predicted wear degree of the cold header die, the management method of the high-temperature alloy fastener die comprises:

[0112] determining a die wear rate index based on historical operation data of the cold header die; the historical operation data at least includes historical predicted wear degree, historical processing timer data and historical control strategy;

[0113] According to the die wear rate index, it is judged whether the current wear change stage of the cold header die is in the rapid deterioration period or the abnormal fluctuation period;

[0114] If the cold header die is currently in the rapid deterioration period, the execution mode of the control strategy is strengthened;

[0115] If the cold header die is currently in the abnormal fluctuation period, the sampling frequency of the current state data of the cold header die is increased, and / or the current wear degree of the cold header die is simulated and verified.

[0116] Optionally, based on historical operation data of the mold, an abrasion speed index such as abrasion change frequency or abrasion change amplitude is constructed to quantify the evolution law of mold abrasion. The abrasion change amplitude refers to the difference between the abrasion grades of adjacent two predictions. The abrasion change frequency refers to the time required for the adjacent two predictions corresponding to the processing period, such as the time required from mild abrasion to moderate abrasion. According to the mold abrasion speed index, it is determined whether the mold of the cold header is in a rapid deterioration period or an abnormal fluctuation period. Whether it is in a rapid deterioration period or an abnormal fluctuation period can be determined by time sequence matching of the real-time collected mold state data with the historical data, using a clustering algorithm or dynamic time warping. If the mold is in a rapid deterioration period, the execution intensity of the regulation and control strategy needs to be strengthened, such as increasing the lubrication frequency to the upper limit, reducing the stamping speed, increasing the pre-tightening force, etc., and shortening the feedback period of parameter adjustment, such as changing from monitoring every hour to monitoring every 10 minutes, to avoid aggravating the abrasion. If the mold is in an abnormal fluctuation period, the sampling frequency of the state data needs to be increased, such as from collecting once every 60 seconds to collecting once every 30 seconds, and the prediction abrasion degree function of the digital twin model can be integrated to predict the abrasion degree of the cold header mold, so as to simulate and verify the current abrasion degree of the cold header mold, and ensure the robustness of the regulation and control strategy. In this way, by calling the historical prediction results of the current mold, the processing timer data and the historical linkage adjustment records, a joint decision based on the current state of the mold and the historical evolution trend is realized, and the forward-looking and stability of the control strategy are improved.

[0117] Optionally, in a specific embodiment, the abrasion prediction model constructed based on random forest or support vector machine realizes the steps of the management method of the high-temperature alloy fastener mold as follows:

[0118] S11. Multi-source working condition data acquisition.

[0119] In this embodiment, according to the characteristics of the cold header mold position closure and severe impact, the layout mode of the acoustic emission, impact stress and temperature sensors is structurally adapted and optimized.

[0120] The acoustic emission sensor arranged in the mold area is preferably installed in the external metal area of the mold clamping structure or the mold seat by indirect coupling, rather than directly embedded in the mold cavity. This way not only avoids interference with the mold processing area, but also effectively transmits high-frequency elastic wave signals through the rigid structure path. The sensor selects a high-sensitivity wide-band model, combined with a high-performance preamplifier, to ensure that there is enough signal-to-noise ratio outside the mold to meet the collection needs of acoustic emission events such as micro-cracks and friction slip.

[0121] The impact stress sensor is arranged at the position of the punch connecting structure or the die mounting base to ensure being on the load transmission path while avoiding the high-speed moving components. The impact stress sensor can be integrated below the base or on the side of the impact direction by clamping or flange mounting to collect the stress change at the instant of forming in real time. This arrangement does not need to change the equipment body structure and is convenient for industrial site integration and maintenance.

[0122] The arrangement of the die temperature sensor is selected according to the on-site process, such as an embedded thermocouple or a non-contact infrared temperature measuring head. The former is embedded with a high-speed response thermocouple through a small hole preset on the side of the die fixing block or die seat and is subjected to insulation protection treatment. The latter is fixed on the open window of the die area to monitor the surface temperature of the forming area in real time through the infrared window, which is suitable for high-speed and non-contact measurement scenes.

[0123] In addition, the processing timer is embedded in the control system to automatically record the impact cycle and the total processing time of each time through the output signal of the cold header PLC (Programmable Logic Controller), and synchronize the time stamp with the sensor system. To realize the cooperative collection of multiple sensors, a unified data acquisition platform is constructed based on a high-performance multi-channel data acquisition module, and all channels share a unified clock source. The simultaneous start of the acoustic emission, stress and temperature channels can be controlled through the main trigger signal (such as the limit position encoder output of the punch), and the collected data is automatically bound with high-precision time stamps to ensure the time consistency and event comparability between different physical quantities.

[0124] This embodiment fully considers the closedness, compactness and particularity of the impact environment of the die in the cold header, and effectively realizes the high-quality cooperative collection of “temperature-force-sound” multi-source working condition data by reasonably avoiding the arrangement area in the die cavity, using external rigid structure conduction, embedding or non-contact measurement technology.

[0125] S12. Signal pre-processing and feature extraction

[0126] In this embodiment, based on the “temperature-force-sound” multi-source working condition data obtained by synchronous collection in step S11, a structured artificial feature extraction method is adopted to construct an input feature matrix for traditional machine learning models such as random forest or support vector machine. Since such models cannot directly process original time series signals, digital pre-processing and feature construction need to be performed for the signal characteristics of different physical quantities to extract numerical feature indicators with time alignment and physical interpretability.

[0127] For the acoustic emission signal, the collected high-frequency waveform signal is first denoised, filtered, and windowed. The sliding window or the synchronous window division strategy based on the impact cycle is selected to extract the signal segment corresponding to the single forming process. Then, typical features are extracted from the time domain, frequency domain, and statistical dimension, including peak amplitude, root mean square value, energy envelope area, signal duration, zero-crossing rate, and other time-domain features, dominant frequency, spectral centroid, frequency band energy ratio, and other frequency-domain features, and skewness, kurtosis, and other statistical features reflecting signal distribution characteristics. The above features can effectively reveal the friction, slip, and micro-damage behavior between the die and the blank during the forming process.

[0128] For the impact stress signal, window division and filter smoothing processing are performed based on each impact cycle to extract features such as maximum stress value, impact loading rate, impact duration, stress variation coefficient, and impact contrast, which are used to describe the stress state and variation law of the die during high-speed stamping and identify stress abnormalities or overload working conditions.

[0129] The die temperature signal is a continuous low-frequency change curve. In the preprocessing process, first-order difference, detrend filtering, and other operations are performed to extract features such as peak temperature, average temperature, temperature rise rate, temperature fluctuation range, and temperature variation coefficient, which reflect the thermal load level and thermal fatigue trend of the die during continuous forming. For the case of using non-contact infrared temperature measurement, periodic temperature peak and average value sequences can be extracted according to the forming beat matching time period.

[0130] The processing timer data provides long-term use condition information of the die. Directly extracted are cumulative processing time, beat number, average beat interval, beat fluctuation coefficient, and other indicators. These variables serve as time dimension inputs for wear evolution and can be combined with acoustic-force-temperature features for modeling to enhance the sensitivity of the prediction model to die life variation trends.

[0131] S13. Constructing a training sample library

[0132] In this embodiment, after the preprocessing and feature extraction of the "temperature-force-acoustic" multi-source signal are completed, a structured feature vector is constructed, and the actual wear state of the die is labeled to form a standardized training sample set that can be used for model training.

[0133] To ensure the reliability of the training data and the repeatability of the sampling process, while avoiding any structural damage to the mold, the embodiment adopts a full non-destructive detection method to evaluate the wear grade of the mold, which specifically includes: ① Three-dimensional profile scanning detection: Through a high-precision laser three-dimensional profiler, the forming surface of the mold is non-contact scanned, the spatial topography is reconstructed, the wear depth, profile deviation, feature edge degradation rate and other parameters are extracted, and the geometric wear degree of the mold in the forming area is quantitatively reflected. ② Surface hardness detection: Use a portable Vickers or Rockwell hardness tester to perform point hardness testing on several key areas of the mold surface, collect the current hardness data, and compare it with the initial hardness value of the mold to evaluate the performance degradation caused by heat accumulation or material fatigue. ③ Auxiliary feature comparison and expert evaluation: Combined with the mold usage cycle recorded by the processing timer, the number of forming beats, and the typical working condition features and abnormal event information collected in the historical stage (such as lubrication abnormalities, impact overload, etc.), an experienced quality or equipment expert comprehensively evaluates and determines the wear grade of the current mold.

[0134] Based on the above data, wear grading rules are developed. For example, the mold wear grades are divided as follows:

[0135] ① Normal: Profile deviation does not exceed 10% of the design tolerance, hardness decreases by less than 5%, and there are no obvious working condition abnormalities;

[0136] ② Mild wear: Local profile deformation occurs, hardness decreases by 5%-10%, and beat or lubrication fluctuation is small;

[0137] ③ Moderate wear: There are regional wear depressions, hardness decreases by more than 10%, beat fluctuation increases or there are lubrication response abnormalities;

[0138] ④ Severe wear: Significant geometric deformation or collapse angle signs, hardness is significantly reduced or part of the area fails to detect, accompanied by multiple working condition abnormal records or ineffective near-end lubrication enhancement.

[0139] S14. Training wear prediction model

[0140] After completing the sample data construction, the embodiment adopts a random forest or support vector machine to establish a mold wear prediction model, which is trained based on the extracted structured "temperature-force-sound" multi-source feature vector. The model is suitable for structured input, has good classification ability and interpretability, and can accurately distinguish the wear grade of the mold. The model training process includes feature optimization, cross-validation, model evaluation and closed-loop optimization, etc.

[0141] Firstly, in the feature preparation stage, the embodiment emphasizes the feature optimization combined with physical mechanism knowledge, which is different from the traditional pure data-driven feature selection method. Specifically, instead of relying only on statistical correlation or redundancy for variable screening, the embodiment conducts systematic screening on multi-source signal features based on physical cognition of the mold wear evolution process. The core idea of this strategy is to optimize those feature indicators that can reflect typical physical mechanism processes such as friction contact, thermal fatigue, impact damage, and micro-crack propagation, thereby improving the sensitivity and interpretability of the model to wear behavior from the source.

[0142] For example, for acoustic emission signals, features such as peak amplitude, kurtosis, and spectral centroid are selected. Peak amplitude represents the intensity of energy transfer at the moment of impact, reflecting the strength of micro-crack energy release. Kurtosis measures the sharpness of the signal, and high kurtosis often corresponds to sudden and intense impacts or instantaneous crack propagation. Spectral centroid is used to describe the concentration position of signal energy in the frequency spectrum, and high frequency shift often indicates micro-crack growth or severe friction. For impact stress signals, features such as maximum stress value, impact loading rate, and stress coefficient of variation are retained. Maximum stress value and impact loading rate are used to represent the structural response strength and loading rate during the forming impact process, indirectly revealing the stress concentration degree of the mold; a higher stress coefficient of variation indicates abnormal stress fluctuation during the forming process, suggesting that the mold is at risk of fatigue. In the temperature signal, attention is paid to thermal load-related features such as temperature rise rate, peak temperature, and temperature fluctuation range to reflect thermal fatigue accumulation and local annealing phenomena; at the same time, by combining the cumulative processing time and beat number collected by the processing timer, a beat coefficient of variation is constructed to represent the long-term usage state of time variables, providing time sequence support for identifying wear trends.

[0143] On the basis of the above, a feature matrix is constructed as the input of the model, and a random forest or support vector machine is selected as the prediction model of the wear state. During training, the sample data is divided and evaluated multiple times through cross-validation and other methods to ensure that the model has good robustness and generalization ability under different data combinations.

[0144] For the support vector machine model, by selecting appropriate kernel functions (such as radial basis functions, polynomial kernels, etc.), the input data is mapped to a high-dimensional space, thereby processing non-linear samples that are difficult to separate in the original space; at the same time, the penalty parameter and kernel function parameter are optimized to improve the accuracy and robustness of the classification boundary. For the random forest model, the number of decision trees and the maximum depth of each tree (i.e., the longest path from the root node to the leaf node) are adjusted to balance the learning ability and generalization ability of the model, improve the discrimination ability of complex wear features, and effectively prevent overfitting.

[0145] After training, the model's classification results at each wear level are evaluated in detail, and misclassified samples are traced back to identify issues such as feature noise, overlapping distributions, or insufficient sample representativeness that lead to errors. Based on the evaluation results, feedback is fed back to feature construction, sample distribution, and model structure design, forming an error-driven closed-loop optimization mechanism. Through multiple rounds of iterative adjustments and training, a set of stable, accurate, and physically interpretable wear prediction models is finally obtained, which can be used for subsequent mold wear condition assessment tasks in cold heading equipment.

[0146] Step S15: Predict mold wear condition online.

[0147] In this embodiment, a structured feature vector is constructed based on real-time acquired multi-source signals, and the dynamic wear status is determined by machine learning models such as random forest or support vector machine.

[0148] By deploying acoustic emission sensors, impact stress sensors, temperature sensors, and processing timers, multi-source operating condition data are continuously collected during the mold's service life. To improve the timeliness and stability of the prediction results, the signal sequence is processed online with a fixed-length sliding time window (e.g., 5 seconds) to continuously generate feature or signal tensor input models, thereby achieving continuous dynamic monitoring of the mold's wear state.

[0149] S151: Online Feature Extraction and Structured Vector Construction

[0150] Within each sliding window, three main types of signals (acoustic emission, impact stress, and temperature) are processed to extract key physical features reflecting the mold's condition, as detailed below:

[0151] (1) Acoustic emission signal characteristics: Let the acoustic emission signal within the sliding window be... The sampling frequency is .

[0152] Peak amplitude:

[0153] ;

[0154] in, For the first i The amplitude of each sampling point represents the intensity of energy released at the moment of impact.

[0155] kurtosis:

[0156]

[0157] in, , These are the mean and standard deviation of the acoustic emission signal samples within the sliding window, respectively, used to measure the sharpness of the impact waveform.

[0158] Spectral Centroid:

[0159]

[0160] where, represents the Fourier amplitude of the signal at frequency ; is the total number of frequency components, reflecting the energy concentration position.

[0161] (2) Shock stress signal features: Let the shock stress signal be ;

[0162] Maximum stress value:

[0163]

[0164] Impact loading speed:

[0165]

[0166] where, is the sampling time interval.

[0167] Stress variation coefficient:

[0168]

[0169] where, and are the mean and standard deviation of the sampled values of the shock stress signal within the sliding window, respectively.

[0170] (3) Temperature signal features: Let the temperature signal be

[0171] Temperature rise rate:

[0172]

[0173] Peak temperature:

[0174]

[0175] Temperature fluctuation range:

[0176]

[0177] (4) Time features (based on machining timer):

[0178] Machining timer collects beat data: where, represents the i th beat interval.

[0179] Beat variation coefficient:

[0180]

[0181] where, is the mean of the beat interval, is the standard deviation of the beat interval, used to reflect the stability changes in the long-term use. Finally, the vector feature is:

[0182]

[0183] where, subscript T denotes the time scale corresponding to the current sliding window.

[0184] S152: Prediction model construction

[0185] For using random forest or support vector machine model, the feature vectors extracted by multi-scale sliding window are fused:

[0186]

[0187] where, denotes the fused multi-scale feature vector, specifically, is a high-dimensional vector obtained by splicing the feature vectors extracted from the sliding time windows from different time scales (5 seconds, 20 seconds, 60 seconds), which is used as the final input of the machine learning model. denotes the feature vector extracted from the recent 5-second sliding window; denotes the feature vector extracted from the recent 20-second sliding window; denotes the feature vector extracted from the recent 60-second sliding window; denotes the vector splicing operation.

[0188] The random forest prediction function can be expressed as:

[0189]

[0190] where, is the predicted wear level label, such as "normal", "mild wear", "moderate wear", "severe wear".

[0191] The support vector machine adopts a Gaussian kernel function:

[0192]

[0193] where, denotes the fused feature vector of the i th support vector in the training set. denotes the bandwidth parameter of the kernel function. denotes the exponential function.

[0194] The decision function of the support vector machine can be expressed as:

[0195]

[0196] wherein, N is the number of support vectors, is the Lagrange multiplier corresponding to the support vector, is the label of the support vector, b is the bias term.

[0197] According to the decision function or comparing the scores of each class in the classification problem, the system finally outputs the wear state prediction result , realizing the discrimination of the current wear state of the mold.

[0198] S153: Multi-mold parallel detection

[0199] Considering that the same device is often configured with multiple molds, the system supports independent collection of working condition data for each mold and parallel processing. In the data collection stage, each mold is bound with a unique number , and the system packs the mold number and signal data to form a key-value pair data stream . Each mold signal sequence independently completes the sliding window division, feature extraction, and state judgment.

[0200] All mold signals can be uniformly input into the same wear prediction model, and each group of data is processed in turn:

[0201]

[0202] wherein, is the prediction result of the k th mold, Y is the total number of molds, and the mold number is retained in the output structure, realizing the parallel monitoring and unified maintenance strategy support for the wear states of multiple molds.

[0203] Step S16: The intelligent decision layer module adjusts and controls according to the prediction result

[0204] On the high-temperature alloy screw cold upsetting forming production line, the system deploys a wear prediction model, an intelligent decision layer module, a lubrication unit, a stamping pressure module, a warning module, and a digital twin simulation module. During the system operation, by collecting multi-source working condition data, the trained wear prediction model is used to judge the current wear state (i.e., the wear grade) of the mold in real time.

[0205] When the system runs to a certain stage, the wear prediction model outputs that the current wear grade of the mold is “mild wear”, and the intelligent decision layer module links the lubrication unit to improve the oil injection frequency.

[0206] When the system runs to another stage, the wear prediction model outputs that the current wear level of the die is "moderate wear". The intelligent decision-making layer module accordingly starts the linkage control logic, and the stamping pressure module slightly reduces the punch speed and simultaneously increases the pre-tightening force (within the experience-determined safe window) to relieve the local impact load of the die and prevent the fatigue crack from accelerating expansion.

[0207] To further ensure that parameter adjustment will not cause forming defects, the system calls the digital twin simulation module to virtually verify the stamping parameter change scheme before adjustment. The simulation result shows that the forming precision and the structural integrity of the screw head remain within the allowable error range, and the system confirms that the adjustment is executable, and then completes the control instruction issuance to the stamping pressure module.

[0208] The historical data storage module records the historical prediction results, processing timer data, and historical linkage adjustment corresponding to the die. The intelligent decision-making layer module calls the data of the historical data storage module to construct the "wear speed" index (based on the frequency and amplitude of wear level transition per unit time) to determine that the current state of the die is in the rapid deterioration period. Accordingly, the intelligent decision-making layer module strengthens the intervention measures: further increases the lubrication frequency, and further reduces the punch speed and fine-tunes the pre-tightening force after verification by the digital twin simulation module (within the experience-determined safe window).

[0209] After a period of time, the system monitors that the wear level continuously jumps between "mild wear" and "moderate wear", and the trend index remains high, and the intelligent decision-making layer module identifies it as an "abnormal fluctuation period", encrypts the data sampling frequency, and calls the auxiliary prediction sub-module of the digital twin simulation module to independently estimate the wear state of the current die for comparison with the output results of the wear state prediction model. The auxiliary prediction can be constructed based on historical evolution rules, physical driving models, or experience databases, and uses different reasoning paths than the wear state prediction model to ensure the independence and reliability of the verification results.

[0210] If the simulation result is basically consistent with the output of the wear prediction model, the established linkage adjustment strategy is continued to be executed; if there is a significant deviation between the two, the intelligent decision-making layer module suspends the linkage adjustment action, and immediately links the warning module to trigger an abnormal prompt, reminding the operation and maintenance personnel to jointly evaluate and correct the wear prediction model and the digital twin simulation module.

[0211] If the wear prediction model and the digital twin simulation module are corrected, after a period of time, the wear prediction level increases to "severe wear", and the intelligent decision-making layer module automatically links the warning module to trigger a die change warning prompt to guide the on-site personnel to implement the die replacement operation.

[0212] Optionally, in yet another specific embodiment, the wear prediction model based on convolutional neural network is constructed to implement the management method of the superalloy fastener mold as follows:

[0213] S111: Multi-source working condition data acquisition. (For reference, the step S11 in the random forest model is adopted)

[0214] S112: Signal preprocessing and feature extraction

[0215] Traditional manual feature extraction is not required.

[0216] In this embodiment, the original acoustic emission signal, impact stress signal and temperature signal are directly input to the CNN model in the form of a sliding window sequence after necessary noise reduction and normalization processing. The model automatically extracts key features in the time series through multiple convolution kernels.

[0217] S113: Constructing a training sample library. (For reference, the step S13 in the random forest model is adopted)

[0218] S114: Training the wear prediction model (CNN model)

[0219] In this embodiment, a one-dimensional convolutional neural network is selected as the prediction model of the mold wear state. Compared with traditional machine learning models, the CNN model can automatically extract deep local pattern information from time series data, and is particularly suitable for processing sensor signals with time dependence, such as acoustic emission signals, impact stress and mold temperature sequences, etc.

[0220] The training process first performs time synchronization and normalization processing on the multi-source sensor data, then extracts fixed-length time series fragments in the form of a sliding window to construct a signal tensor structure. Each sample contains data from 3-4 sensor channels (acoustic emission, stress, temperature and processing time).

[0221] The CNN model is composed of multiple convolution layers, pooling layers and nonlinear activation functions, which learn time series features of different scales layer by layer. The convolution kernel slides on each input channel to extract feature information within the local time window.

[0222] In the training phase, the CNN model uses a labeled sample library for supervised learning, that is, each input sample corresponds to a known wear grade label (for example, 0 = normal, 1 = mild wear, 2 = moderate wear, 3 = severe wear). The output layer of the model is a multi-classification Softmax structure, which can map a set of real numbers output by the network to a probability distribution, used to represent the possibility of the sample belonging to each wear grade.

[0223] During the training process, the multi-class cross-entropy loss function is used as the objective function to measure the difference between the predicted probability distribution of the model and the true label. This loss function can punish the wrong prediction of the class and promote the model to optimize in a more accurate direction. (See the specific loss function later)

[0224] The update of model parameters uses the Adam adaptive gradient optimizer, which is an optimization algorithm that combines the momentum method and adaptive learning rate mechanism. It can automatically adjust the update step size according to the historical gradient of each parameter, improving the training efficiency and stability. In order to improve the generalization ability of the model and prevent overfitting, the cross-validation strategy is introduced during the training process, that is, the data set is divided several times to evaluate the performance of the model on different data; At the same time, the early stopping mechanism is set, which automatically terminates the training when the performance of the validation set stops improving, so as to obtain a more robust model performance.

[0225] S115: Online prediction of mold wear state

[0226] For the case of using convolutional neural network (CNN) as the prediction model, this embodiment directly uses the original multi-source working condition signals collected in the sliding window as the input, without performing manual feature extraction and structured splicing operation. The system constructs a signal tensor with the original signal sequence in a fixed time scale (such as 5 seconds), and automatically extracts key features and classifies wear grades through the network.

[0227] Model input: Let the number of signal sampling points corresponding to the current sliding window length be S , and the total number of collected sensor channels be C , then the signal tensor constructed at each time can be represented as:

[0228]

[0229] Among them, S represents the number of time steps in the sliding window (such as 5 seconds x 2048 Hz = 10240), C represents the number of channels, including acoustic emission, impact stress, mold temperature and metronome signal channels, that is C =4, represents the time sequence of each channel signal.

[0230] The CNN model is composed of a number of one-dimensional convolutional layers, activation functions, pooling layers, fully connected layers and Softmax output layers, which are used to automatically extract patterns and features in time series signals.

[0231] One-dimensional convolutional layer, signal tensor , feature extraction through multiple one-dimensional convolution kernels:

[0232]

[0233] where, represents the output of the th convolutional channel in the th layer at time t ; k represents the length of the convolution kernel; represents the convolution kernel weight, the th kernel weight at position from the input channel to the output channel in the th layer; represents the signal value of the t th channel at time ; represents the activation function; represents the bias term of the th output channel in the th layer.

[0234] If multiple convolutional layers are stacked, the output of the th layer will be used as the input of the next convolutional layer.

[0235] Pooling layer: used to compress the time dimension and enhance robustness. Take max pooling as an example:

[0236]

[0237] where, represents the value of the th channel at position t after pooling; represents the size of the pooling window; the maximum value is selected from using a sliding window method.

[0238] Fully connected layer: after several convolution + pooling layers, the output tensor is obtained, which is flattened into a one-dimensional vector , and then input into the fully connected layer to calculate the score of each class:

[0239]

[0240] where, represents the weight vector of the k th class, represents the bias of the k th class, and k corresponds to the original score of the th class.

[0241] Softmax output layer (for classification): normalize the scores of all classes into a probability distribution, output the predicted probability of each class:

[0242] wherein, represents the probability of being predicted as the k th class; Q represents the total number of classes (this embodiment has 4 wear grades, that is, Q =4); represents the probability distribution output by the model.

[0243] During training, the embodiment can use the loss function:

[0244]

[0245] wherein, represents the true label.

[0246] S116: The intelligent decision-making layer module adjusts and controls according to the prediction result (which can be referred to as the case of step S16 in the random forest model).

[0247] In summary, the management method of the superalloy fastener mold provided in the above embodiment includes: during the process of machining the superalloy fastener by the cold header mold, synchronously collecting multi-dimensional working condition data of the cold header mold; the multi-dimensional working condition data includes acoustic emission signals, impact stress signals, temperature signals, and machining timing data; the acoustic emission signals are collected based on an acoustic emission sensor, the acoustic emission sensor is installed in a vibration conduction area of an associated structure of the cold header mold in a non-contact manner; the impact stress signals are collected based on an impact stress sensor, the impact stress sensor is arranged on a fixed structure on a load transmission path of the cold header mold; the temperature signals are collected based on a temperature sensor, the temperature sensor is an embedded thermocouple or a non-contact infrared temperature measuring head; the collection time stamps of the acoustic emission sensor, the impact stress sensor, and the temperature sensor are synchronized based on a machining timer; the multi-dimensional working condition data is input into a wear prediction model to output a predicted wear degree of the cold header mold; the wear prediction model is built based on a machine learning algorithm; and corresponding control strategies are executed for the process of machining the superalloy fastener by the cold header mold according to the predicted wear degree of the cold header mold. In this way, by inputting the multi-dimensional working condition data into the wear prediction model to predict the wear degree of the mold, the mold operation is real-time controlled based on the prediction result, the wear state of the mold can be accurately identified and graded controlled in a timely manner, and the service life of the mold and the machining precision of the superalloy fastener are improved.

[0248] Based on the same inventive concept as the foregoing embodiments, the embodiments of the present application provide a management system 20 of a superalloy fastener mold, which comprises an acquisition module 21, a prediction module 22, and a control module 23; wherein,

[0249] The acquisition module 21 synchronously acquires multi-dimensional working condition data of the cold header die during the process of machining the high-temperature alloy fastener, the multi-dimensional working condition data including acoustic emission signals, impact stress signals, temperature signals and machining timing data, the acoustic emission signals being acquired based on an acoustic emission sensor, the acoustic emission sensor being installed in a vibration conduction area of an associated structure of the cold header die in a non-contact manner, the impact stress signals being acquired based on an impact stress sensor, the impact stress sensor being arranged on a fixed structure on a load transmission path of the cold header die, the temperature signals being acquired based on a temperature sensor, the temperature sensor being an embedded thermocouple or a non-contact infrared temperature measuring head, and the acquisition time stamps of the acoustic emission sensor, the impact stress sensor and the temperature sensor being synchronized based on a machining timer;

[0250] The prediction module 22 inputs the multi-dimensional working condition data into a wear prediction model to output a predicted wear degree of the cold header die, the wear prediction model being built based on a machine learning algorithm;

[0251] The regulation module 23 executes a corresponding regulation strategy for the process of machining the high-temperature alloy fastener by the cold header die according to the predicted wear degree of the cold header die.

[0252] In an embodiment, before the multi-dimensional working condition data is input into the wear prediction model, the prediction module 22 is further configured to:

[0253] construct sample data based on historical multi-dimensional working condition data and wear degree labels of the cold header die;

[0254] build the wear prediction model based on a machine learning model, the machine learning model including at least one of a random forest model, a support vector machine model and a convolutional neural network model;

[0255] train the wear prediction model by using the sample data.

[0256] In an embodiment, before the multi-dimensional working condition data is input into the wear prediction model, the prediction module 22 is further configured to, for the wear prediction model based on a random forest or a support vector machine, perform digital preprocessing and feature construction according to physical characteristics of the multi-dimensional working condition data to obtain a numerical feature vector.

[0257] In an embodiment, the inputting of the multi-dimensional working condition data into the wear prediction model includes:

[0258] respectively cutting sliding window segments of different time scales from the multi-dimensional working condition data;

[0259] The sliding window segments of each time scale are fused and input into the wear prediction model; for the wear prediction model built based on random forest or support vector machine, the fusion processing includes splicing the sliding window segments of each time scale to form a multi-scale feature vector; for the wear prediction model built based on convolutional neural network, the fusion processing includes constructing a signal tensor based on the sliding window segments of each time scale.

[0260] In an embodiment, according to the predicted wear degree of the cold header die, a corresponding regulation strategy is executed for the process of machining the superalloy fastener by the cold header die, including at least one of the following:

[0261] If the cold header die is slightly worn, the oil injection frequency of the lubrication unit of the cold header die is increased;

[0262] If the cold header die is moderately worn, the ram speed of the stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range;

[0263] If the cold header die is severely worn, a die replacement warning is triggered.

[0264] In an embodiment, before the corresponding regulation strategy is executed for the process of machining the superalloy fastener by the cold header die according to the predicted wear degree of the cold header die, the prediction module 22 is further configured to:

[0265] construct a digital twin model based on the physical parameters of the cold header die and the material properties of the superalloy;

[0266] input the adjusted cold heading process parameters into the digital twin model to obtain a simulation result of the superalloy fastener machined by the cold header die; the adjusted cold heading process parameters include at least one of the adjusted ram speed and pre-tightening force of the stamping pressure module;

[0267] If the simulation result meets the quality requirements of the superalloy fastener, the stamping pressure module is regulated according to the adjusted cold heading process parameters;

[0268] If the simulation result does not meet the quality requirements of the superalloy fastener, the adjusted cold heading process parameters are iteratively optimized until the simulation result meets the quality requirements of the superalloy fastener.

[0269] In an embodiment, after the corresponding regulation strategy is executed for the process of machining the superalloy fastener by the cold header die according to the predicted wear degree of the cold header die, the regulation module 23 is further configured to:

[0270] determine a die wear rate index based on historical operation data of the cold header die; the historical operation data at least includes a historical predicted wear degree, historical processing timer data and a historical control strategy;

[0271] determine whether the cold header die is currently in a rapid deterioration period or an abnormal fluctuation period according to the die wear rate index;

[0272] if the cold header die is currently in the rapid deterioration period, intensify the execution mode of the control strategy;

[0273] if the cold header die is currently in the abnormal fluctuation period, increase the sampling frequency of the current state data of the cold header die, and / or simulate and verify the current wear degree of the cold header die.

[0274] The specific implementation of the management system of the superalloy fastener die can refer to the related description of the management method of the superalloy fastener die, and will not be repeated here.

[0275] Based on the same inventive concept as the foregoing embodiments, the embodiments of the present application provide a computing device, as shown in Figure 3 , the computing device comprises a processor 410 and a memory 411 storing a computer program; wherein, Figure 3 The processor 410 in the figure is not used to refer to the number of processors 410 being one, but is only used to refer to the positional relationship of the processor 410 relative to other devices. In actual application, the number of processors 410 can be one or more; similarly, Figure 3 The memory 411 in the figure also has the same meaning, that is, it is only used to refer to the positional relationship of the memory 411 relative to other devices. In actual application, the number of memories 411 can be one or more. When the processor 410 runs the computer program, the management method of the superalloy fastener die is realized.

[0276] The computing device can further comprise at least one network interface 412. The various components in the computing device are coupled together through a bus system 413. It can be understood that the bus system 413 is used to realize the connection and communication between the components. The bus system 413 includes not only a data bus, but also a power bus, a control bus and a status signal bus. However, for the purpose of clear illustration, all kinds of buses are marked as the bus system 413 in Figure 3 .

[0277] The memory 411 can be a volatile memory or a nonvolatile memory, and can include both a volatile and a nonvolatile memory. The nonvolatile memory can be a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a ferroelectric random access memory (FRAM), a flash memory, a magnetic storage memory, an optical disc, or a Compact Disc Read-Only Memory (CD-ROM). The magnetic storage memory can be a magnetic disk memory or a magnetic tape memory. The volatile memory can be a Random Access Memory (RAM) used as an external cache. By way of example, but not limitation, many forms of RAM can be used, such as a Static Random Access Memory (SRAM), a Synchronous Static Random Access Memory (SSRAM), a Dynamic Random Access Memory (DRAM), a Synchronous Dynamic Random Access Memory (SDRAM), a Double Data Rate Synchronous Dynamic Random Access Memory (DDR SDRAM), an Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), a Sync Link Dynamic Random Access Memory (SLDRAM), a Direct Rambus Random Access Memory (DRRAM).The memory 411 described in the embodiments of the present application is intended to include, but is not limited to, these and any other suitable type of memory.

[0278] The memory 411 in the embodiments of the present application is used to store various types of data to support the operation of the computing device. Examples of the data include: any computer programs for operating on the computing device, such as an operating system and application programs; contact data; phonebook data; messages; pictures; videos; and the like. The operating system contains various system programs, such as a framework layer, a core library layer, a driver layer, and the like, for implementing various basic services and processing hardware-based tasks. The application programs can contain various application programs, such as a media player, a browser, and the like, for implementing various application services. Here, the program for implementing the method of the embodiments of the present application can be contained in the application programs.

[0279] Based on the same inventive concept as the foregoing embodiments, the present embodiment also provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer readable storage medium can be a ferromagnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, a compact disc read-only memory (CD-ROM), or the like. The computer readable storage medium can also be various devices including one or any combination of the above memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, and the like. The computer program stored in the computer readable storage medium is run by a processor to implement the management method of the superalloy fastener mold applied to the computing device. Figure 1 The specific step flow implemented by the computer program executed by the processor will not be described here again in the description of the illustrated embodiments.

[0280] Each technical feature of the above-described embodiments can be combined arbitrarily, and to make the description concise, each technical feature in the above-described embodiments is not described in all possible combinations, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present disclosure.

[0281] In this document, the terms "comprise", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0282] The above description is only specific embodiments of the application, but the protection scope of the application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered by the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A method for managing high-temperature alloy fastener molds, characterized in that, The method comprises the following steps: Synchronous acquisition of multi-dimensional working condition data of the cold header die during the process of machining the high-temperature alloy fastener based on the cold header die; The multi-dimensional working condition data comprises an acoustic emission signal, an impact stress signal and a temperature signal, the acoustic emission signal is acquired based on an acoustic emission sensor, and the acoustic emission sensor is installed in a vibration conduction area of an associated structure of the cold header die in a non-contact manner; The impact stress signal is acquired based on an impact stress sensor, and the impact stress sensor is arranged on a fixed structure in a load transmission path of the cold header die; the temperature signal is acquired based on a temperature sensor, and the temperature sensor is an embedded thermocouple or a non-contact infrared temperature measuring head; The acquisition time stamps of the acoustic emission sensor, the impact stress sensor and the temperature sensor are synchronized based on a processing timer; The multi-dimensional working condition data is input into a wear prediction model to output a predicted wear degree of the cold header die; the wear prediction model is built based on a machine learning algorithm; If the cold header die is slightly worn, the oil injection frequency of a lubrication unit of the cold header die is increased; If the cold header die is moderately worn, the punch speed of a stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range; If the cold header die is severely worn, a die replacement warning is triggered; Before the punch speed of the stamping pressure module of the cold header die is reduced and / or the pre-tightening force of the stamping pressure module is increased within a preset range, the method comprises: A digital twin model is constructed based on physical parameters of the cold header die and material properties of the high-temperature alloy; Adjusted cold heading processing parameters are input into the digital twin model to obtain a simulation result of the high-temperature alloy fastener machined by the cold header die; the adjusted cold heading processing parameters comprise at least one of an adjusted punch speed and an adjusted pre-tightening force of the stamping pressure module; If the simulation result meets the quality requirements of the high-temperature alloy fastener, the stamping pressure module is adjusted according to the adjusted cold heading processing parameters; If the simulation result does not meet the quality requirements of the high-temperature alloy fastener, the adjusted cold heading processing parameters are iteratively optimized until the simulation result meets the quality requirements of the high-temperature alloy fastener.

2. The method of claim 1, wherein, Before the multi-dimensional working condition data is input into the wear prediction model, the method comprises: Sample data is constructed based on historical multi-dimensional working condition data and wear degree labels of the cold header die; The wear prediction model is built based on a machine learning model; the machine learning model comprises at least one of a random forest model, a support vector machine model and a convolutional neural network model; The wear prediction model is trained using the sample data.

3. The method of claim 2, wherein, Before the multi-dimensional working condition data is input into the wear prediction model, the method further comprises, for the wear prediction model based on random forest or support vector machine, digitally pre-processing and feature construction according to physical properties of the multi-dimensional working condition data to obtain a numerical feature vector.

4. The method of claim 2, wherein, The method comprises the following steps: The multi-dimensional working condition data is input into a wear prediction model, comprising: The multi-dimensional working condition data is respectively intercepted into sliding window segments of different time scales; 5. The method according to claim 1 or 4, characterized in that, The sliding window segments of each time scale are fused and input into the wear prediction model; for the wear prediction model based on random forest or support vector machine, the fusion processing comprises splicing the sliding window segments of each time scale to form a multi-scale feature vector; for the wear prediction model based on convolutional neural network, the fusion processing comprises constructing a signal tensor based on the sliding window segments of each time scale. After the corresponding control strategy is executed on the process of machining the high-temperature alloy fastener by the cold header die according to the predicted wear degree of the cold header die, the method comprises the following steps: A die wear speed index is determined based on historical operation data of the cold header die; the historical operation data at least comprises historical predicted wear degree, historical machining timer data and historical control strategy; Whether the cold header die is currently in a rapid deterioration period or an abnormal fluctuation period is judged according to the die wear speed index; If the cold header die is currently in the rapid deterioration period, the execution mode of the control strategy is strengthened; 6. A management system for superalloy fastener dies, characterized by, If the cold header die is currently in the abnormal fluctuation period, the sampling frequency of the current state data of the cold header die is increased, and / or the current wear degree of the cold header die is simulated and verified. The system comprises a collection module, a prediction module and a control module; wherein, The collection module synchronously collects multi-dimensional working condition data of the cold header die in the process of machining the high-temperature alloy fastener; the multi-dimensional working condition data comprises acoustic emission signals, impact stress signals, temperature signals and machining timer data; the acoustic emission signals are collected based on an acoustic emission sensor; the acoustic emission sensor is installed in a vibration conduction area of an associated structure of the cold header die in a non-contact manner; the impact stress signals are collected based on an impact stress sensor; the impact stress sensor is arranged on a fixed structure in a load transmission path of the cold header die; the temperature signals are collected based on a temperature sensor; the temperature sensor is an embedded thermocouple or a non-contact infrared temperature measuring head; the collection time stamps of the acoustic emission sensor, the impact stress sensor and the temperature sensor are synchronized based on a machining timer; The prediction module inputs the multi-dimensional working condition data into a wear prediction model to output the predicted wear degree of the cold header die; the wear prediction model is built based on a machine learning algorithm; The control module, if the cold header die is slightly worn, increases the oil injection frequency of the lubrication unit of the cold header die; if the cold header die is moderately worn, reduces the punch speed of the stamping pressure module of the cold header die and / or increases the pre-tightening force of the stamping pressure module within a preset range; if the cold header die is severely worn, triggers a die replacement warning. The control module is further configured to: before reducing the punch speed of the stamping pressure module of the cold header die and / or increasing the pre-tightening force of the stamping pressure module within a preset range, construct a digital twin model based on the physical parameters of the cold header die and the material properties of the superalloy; input the adjusted cold heading process parameters into the digital twin model to obtain a simulation result of the superalloy fastener processed by the cold header die; the adjusted cold heading process parameters include at least one of the adjusted punch speed and the pre-tightening force of the stamping pressure module; if the simulation result meets the quality requirements of the superalloy fastener, the stamping pressure module is adjusted according to the adjusted cold heading process parameters; if the simulation result does not meet the quality requirements of the superalloy fastener, the adjusted cold heading process parameters are iteratively optimized until the simulation result meets the quality requirements of the superalloy fastener.

7. A computing device, comprising: Comprise: A processor and a memory for storing executable instructions; wherein the processor is configured to execute the instructions to implement the management method of the superalloy fastener die according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, When the instructions in the computer readable storage medium are executed by the processor, the management method of the superalloy fastener die according to any one of claims 1 to 5 is implemented.

Images