A cantilever die crack monitoring method and system for hot forming of a bearing inner race

Abstract

The application discloses a cantilever die crack monitoring method and system for bearing inner ring hot forming, relates to the technical field of cantilever die crack monitoring, and comprises the following steps: embedding a high-temperature-resistant fiber bragg grating sensor array in a stress concentration area inside a cantilever die, and configuring acoustic emission sensors, an infrared thermal imager and vibration sensors outside the die to form a multi-source synchronous acquisition system; obtaining a fiber bragg grating wavelength sequence, an acoustic emission event stream, a surface temperature time sequence image and a vibration time sequence signal in a continuous multiple hot forming cycle process through the multi-source synchronous acquisition system; calculating the slope and the second-order change rate of the change of the core strain with the number of hot forming cycles based on the fiber bragg grating wavelength sequence, calculating the variation coefficient of the event interval and the energy release burst index in a unit cycle in combination with the acoustic emission event stream, and generating a time sequence active feature set reflecting the crack evolution dynamic characteristics; and the multi-source synchronous monitoring system is constructed by fusing the high-temperature-resistant fiber bragg grating, acoustic emission, infrared thermal imaging and vibration sensing.

Description

Technical Field

[0001] This invention relates to the field of cantilever die head crack monitoring technology, and in particular to a cantilever die head crack monitoring method and system for hot forming of bearing inner rings. Background Technology

[0002] Cantilever die head crack monitoring technology refers to a real-time sensing, dynamic evaluation, and proactive intervention system used in the thermoforming process of bearing inner rings to address the problem of fatigue cracks easily generated in cantilever dies under high temperature, high pressure, and alternating loads. This technology utilizes multiple sensors deployed inside and outside the die head to collect its mechanical, acoustic, thermal, and vibration response signals during continuous thermoforming cycles. Combined with data fusion, intelligent algorithms, and digital twin simulation, it achieves high-precision, closed-loop monitoring of the entire process of microcrack initiation, propagation, and remaining lifespan within the die head. The aim is to prevent sudden fractures, ensure forming quality, extend die life, and avoid unplanned downtime.

[0003] Traditional methods often employ static thresholds or simple statistics, which fail to capture the nonlinear acceleration behavior and sudden energy release during the crack initiation and propagation process. This leads to underreporting of early microcracks or false alarms of normal fluctuations. Furthermore, existing risk assessments are mostly based on expert rules or univariate trend extrapolation, which makes it difficult to handle the complex coupling relationships between heterogeneous data such as strain, acoustic emission, and temperature. Moreover, the models have not been trained on a large number of failure samples, resulting in poor adaptability to new working conditions or new material molds. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] This invention provides a crack monitoring method for cantilever die heads used in the thermoforming of bearing inner rings. It solves the problems of traditional methods that mostly use static thresholds or simple statistics, which cannot capture the nonlinear acceleration behavior and sudden energy release during the crack initiation and propagation process, resulting in missed early microcracks or false alarms of normal fluctuations. In addition, existing risk assessments are mostly based on expert rules or univariate trend extrapolation, which are difficult to handle the complex coupling relationship between heterogeneous data such as strain, acoustic emission, and temperature. Furthermore, the models have not been trained with large-scale failure samples, resulting in poor adaptability to new working conditions or new material dies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] A method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings, comprising:

[0008] A high-temperature resistant fiber optic grating sensor array is embedded in the stress concentration area inside the cantilever head, and an acoustic emission sensor, an infrared thermal imager, and a vibration sensor are configured on the outside of the head to form a multi-source synchronous acquisition system.

[0009] The fiber grating wavelength sequence, acoustic emission event stream, surface temperature time-series image and vibration time-series signal are acquired through a multi-source synchronous acquisition system during multiple consecutive thermoforming cycles.

[0010] Based on the fiber grating wavelength sequence, the slope and second-order rate of change of core strain with the number of thermoforming cycles are calculated. Combined with the acoustic emission event flow, the coefficient of variation of the event interval within a unit cycle and the burst index of energy release are calculated to generate a time-series active feature set that reflects the dynamic characteristics of crack evolution.

[0011] The temporal active feature set and the local micro-strain field corrected by thermo-coupling are jointly input into a pre-trained multimodal fusion neural network model, and the crack risk discrimination result is output.

[0012] When the crack risk assessment result exceeds the preset threshold, the time-series active feature set and the current process parameters are injected into the digital twin corresponding to the physical model head, driving multi-physics field coupling simulation and dynamically adjusting the damage evolution rate weight to calculate the remaining safe service life.

[0013] A mold head health index is generated based on the remaining safe service life, and a graded early warning signal is triggered or a load adjustment command is sent to the thermoforming equipment based on the health index.

[0014] Preferably, a high-temperature resistant fiber Bragg grating sensor array is embedded in the stress concentration area inside the cantilever mold head, and an acoustic emission sensor, an infrared thermal imager, and a vibration sensor are configured outside the mold head to form a multi-source synchronous acquisition system. The specific steps are as follows:

[0015] During the die manufacturing stage, the root of the cantilever and the transition fillet area are identified as the area of ​​maximum principal stress concentration through finite element static simulation. Micro-hole channels with a diameter of 0.5 mm are drilled along the axial direction in the area, penetrating to the core of the die.

[0016] A fiber optic grating sensor with a nickel-based alloy coating is implanted in a series in a micro-pore channel. The pores are filled with alumina ceramic slurry and sintered at high temperature to form a rigid connection between the sensor and the mold substrate, ensuring effective strain transmission.

[0017] A piezoelectric acoustic emission sensor with a resonant frequency of 150 kHz is attached to the outer surface of the mold head near the root of the cantilever. An infrared thermal imager with a resolution of 640×480 is installed 1.2 meters above the mold head with the lens facing the working surface of the mold head. A triaxial MEMS vibration accelerometer is fixed on the mold head support flange.

[0018] The fiber optic demodulator, acoustic emission preamplifier, infrared thermal imager controller, and vibration acquisition module are all connected to the time synchronization controller. The IEEE 1588 precision time protocol is used to achieve microsecond-level synchronous triggering, thus forming a multi-source synchronous acquisition system.

[0019] Preferably, the step of acquiring fiber grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals through a multi-source synchronous acquisition system during multiple consecutive thermoforming cycles includes the following steps:

[0020] In the During each thermoforming cycle, the fiber optic grating demodulator records the center wavelength of each grating unit at a sampling rate of 20 kHz, forming a wavelength sequence. ;

[0021] The acoustic emission sensor continuously monitors elastic wave signals, and records the moment the instantaneous amplitude exceeds a threshold of 45 dB. With energy value Generate acoustic emission event stream { },in For the first Total number of loop events;

[0022] The infrared thermal imager outputs a surface temperature matrix at a rate of 25 frames per second, extracts the average temperature of a 20×20 pixel block in the central area of ​​the working surface of the mold head, and forms a single-point temperature time series. ;

[0023] The triaxial signal output from the vibration accelerometer is low-pass filtered and stored at a sampling rate of 5 kHz, with the radial component being extracted. As the main vibration characteristic;

[0024] After aligning the above four types of data by timestamp, representative values ​​from the steady-state phase at the end of each cycle are extracted for feature calculation.

[0025] Preferably, the calculation of the slope and second-order rate of change of core strain with the number of thermoforming cycles based on the fiber grating wavelength sequence, combined with the calculation of the coefficient of variation of the event interval per unit cycle and the burst index of energy release by the acoustic emission event flow, generates a time-series active feature set reflecting the dynamic characteristics of crack evolution. The specific steps are as follows:

[0026] From the Extracting the steady-state center wavelength from a cyclic fiber grating wavelength sequence The original strain was calculated using the strain-wavelength linear relationship. ,in The initial wavelength in the cold state. The strain sensitivity coefficient;

[0027] Using the first Cyclic infrared temperature data Combined with the coefficient of linear expansion of the material and reference temperature Calculate the additional strain due to thermal expansion ;

[0028] The values ​​of the local micro-strain field at the key points were obtained after thermo-coupling correction. The expression is:

[0029] ;

[0030] Constructing strain sequences Calculate the first-order difference The expression is:

[0031] ;

[0032] Calculate the second-order difference The expression is:

[0033] ;

[0034] For the A cyclic acoustic emission event stream, calculating the time interval between adjacent events. Calculate the mean with standard deviation Define the event interval variation coefficient as follows:

[0035] ;

[0036] Accumulate event energy over time. The cycle is divided into 5 equal-length sub-intervals, and the slope of each segment is fitted. Calculate the standard deviation With the maximum value Define the burst index of energy release, expressed as:

[0037] ;

[0038] Will , , , Form a four-dimensional vector As the first The temporal activity characteristics of cycles;

[0039] all This forms a temporally active feature set, which is used as input for the next step of the neural network.

[0040] Preferably, the steps for inputting the temporal active feature set and the thermo-coupled corrected local micro-strain field into a pre-trained multimodal fusion neural network model to output crack risk discrimination results are as follows:

[0041] The first Temporal active feature vector of the cycle Timing-aligned thermo-coupled local micro-strain field Concatenate into a five-dimensional input vector ;

[0042] The multimodal fusion neural network model contains five input nodes, which correspond to... The five components;

[0043] The first hidden layer contains 10 neurons and uses the ReLU activation function;

[0044] The second hidden layer contains 5 neurons;

[0045] The output layer is a single neuron, using the Sigmoid activation function;

[0046] The model was trained on a historical model failure dataset, and FocalLoss was used as the loss function to address the imbalance between positive and negative samples.

[0047] Will Input the trained model and output the crack risk probability value. ;

[0048] Set the discrimination threshold ,like If the output crack risk assessment result is high risk, the subsequent digital twin simulation process will be triggered.

[0049] Preferably, when the crack risk assessment result exceeds a preset threshold, the time-series active feature set and current process parameters are injected into the digital twin corresponding to the physical model head to drive multi-physics coupling simulation and dynamically adjust the damage evolution rate weight to calculate the remaining safe service life. The specific steps are as follows:

[0050] During the system initialization phase, a finite element digital twin is established based on the three-dimensional geometric model and material properties of the mold head, and the basic equation for Paris crack propagation is pre-set.

[0051] When the crack risk assessment result is high risk, extract the current loop. Temporal activity characteristics and And current process parameters: extrusion pressure Heating temperature Feed rate ;

[0052] Calculate the crack activity product factor The expression is:

[0053] ;

[0054] Introducing adaptive damage evolution rate weights :

[0055] ;

[0056] in, The highest historical activity factor, To adjust the gain constant;

[0057] Will Introducing the Paris equation, we obtain the modified crack propagation rate model, expressed as:

[0058] ;

[0059] In the digital twin, with the current crack length Using the initial value, numerically integrate the differential equation until the crack length is reached. Reaching the critical failure length The cumulative cycle increment is the remaining safe service life. .

[0060] Preferably, the steps of generating a die head health index based on the remaining safe service life and triggering a graded early warning signal or sending a load adjustment command to the thermoforming equipment based on the health index are as follows:

[0061] Obtaining Remaining Safe Service Life from Digital Twins Read the number of service cycles from the system counter. ;

[0062] Calculate the health index of the model head The expression is:

[0063] ;

[0064] Preset Level 1 Warning Threshold Compared with the level 2 warning threshold ;

[0065] like and If the condition is not met, a Level 1 warning will be displayed on the monitoring terminal: the health status of the mold head has deteriorated, and it is recommended to replace it during the next planned shutdown window;

[0066] like In addition to displaying a level-two warning: the die head is on the verge of failure and immediate intervention is required, an analog signal is simultaneously sent to the PLC controller of the thermoforming equipment.

[0067] After receiving the signal, the PLC reduces the output current of the extrusion roller servo motor, thereby reducing the radial pressure acting on the cantilever end of the die head. ,in, This is the pressure regulation coefficient. Rated working pressure;

[0068] The pressure adjustment continues until the current batch is completed, after which the system locks the die head and prohibits further use.

[0069] A crack monitoring system for a cantilever die head used in the thermoforming of bearing inner rings includes:

[0070] Multi-source sensor integration module, cyclic signal acquisition module, time-series active feature generation module, crack risk intelligent judgment module, digital twin life prediction module and health index early warning and control module;

[0071] The multi-source sensing integrated module is used to embed a high-temperature resistant fiber optic grating sensor array in the stress concentration area inside the cantilever head, and to configure an acoustic emission sensor, an infrared thermal imager and a vibration sensor on the outside of the head, and to build a multi-source synchronous acquisition system through a time synchronization controller.

[0072] The cyclic signal acquisition module is used to simultaneously acquire fiber grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals during multiple consecutive thermoforming cycles, and to store the four types of data in alignment with the cycle number and timestamp.

[0073] The time-series active feature generation module is used to calculate the slope and second-order rate of change of core strain with the number of thermoforming cycles based on the fiber grating wavelength sequence, and to calculate the coefficient of variation of event interval within a unit cycle and the burst index of energy release in combination with the acoustic emission event flow, so as to generate a time-series active feature set that reflects the dynamic characteristics of crack evolution.

[0074] The crack risk intelligent discrimination module is used to input the temporal active feature set and the local micro-strain field corrected by thermo-coupling into a pre-trained multimodal fusion neural network model and output the crack risk discrimination result.

[0075] The digital twin life prediction module is used to inject the time-series active feature set and current process parameters into the digital twin corresponding to the physical mold when the crack risk judgment result exceeds the preset threshold, drive the thermo-mechanical-fatigue multiphysics coupling simulation, dynamically correct the Paris crack propagation equation through adaptive damage evolution rate weight, and calculate the remaining safe service life.

[0076] The health index early warning and control module is used to generate a mold head health index based on the remaining safe service life and the number of service cycles, and to trigger a graded early warning signal or send a radial load adjustment command to the thermoforming equipment according to the range of the health index, so as to realize active maintenance of the mold head and collaborative optimization of working conditions.

[0077] The beneficial effects of this invention are as follows: By constructing a multi-source synchronous monitoring system integrating high-temperature resistant fiber optic gratings, acoustic emission, infrared thermal imaging, and vibration sensing, and introducing high-order temporal dynamic features such as strain change slope and second-order rate of change, event interval variation coefficient, and energy release burst index, an adaptive digital twin life prediction mechanism with crack evolution activity as the core is established. This realizes a closed-loop monitoring chain from static threshold alarm to dynamic evolution perception, intelligent risk judgment, accurate life prediction, and health index-driven active regulation. This not only improves the sensitivity of early microcrack identification and the accuracy of remaining life prediction in cantilever die heads, but also effectively avoids unplanned downtime and extends the safe service life of the die head through health index-based early warning and real-time load feedback adjustment, ensuring the continuity, safety, and product quality stability of the bearing inner ring thermoforming process. Attached Figure Description

[0078] Figure 1 This is a flowchart of a cantilever die head crack monitoring method used in the embodiment for thermoforming of bearing inner rings.

[0079] Figure 2 This is a schematic diagram of a cantilever die head crack monitoring system used for thermoforming of the bearing inner ring in an embodiment. Detailed Implementation

[0080] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0081] Example, refer to Figure 1 and Figure 2 A method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings, comprising:

[0082] S1. A high-temperature resistant fiber optic grating sensor array is embedded in the stress concentration area inside the cantilever head, and an acoustic emission sensor, an infrared thermal imager, and a vibration sensor are configured on the outside of the head to form a multi-source synchronous acquisition system.

[0083] During the mold manufacturing stage, finite element static simulation was used to identify the area at the root of the cantilever and the transition fillet region as the area of ​​maximum principal stress concentration. Micro-hole channels with a diameter of 0.5 mm were drilled along the axial direction in this area, penetrating to the core of the mold. Fiber Bragg grating sensors with a nickel-based alloy coating were implanted in series into the micro-hole channels. Alumina ceramic slurry was used to fill the pores and sintered at high temperature to form a rigid connection between the sensor and the mold substrate, ensuring effective strain transmission. A piezoelectric acoustic emission sensor with a resonant frequency of 150 kHz was attached to the outer surface of the mold near the root of the cantilever. An infrared thermal imager with a resolution of 640×480 was installed 1.2 meters above the mold, with the lens facing the working surface of the mold. A triaxial MEMS vibration accelerometer was fixed on the mold support flange. The fiber Bragg grating demodulator, acoustic emission preamplifier, infrared thermal imager controller, and vibration acquisition module were all connected to a time synchronization controller. The IEEE 1588 precision time protocol was used to achieve microsecond-level synchronous triggering, forming a multi-source synchronous acquisition system.

[0084] By accurately locating stress concentration areas based on finite element simulation and embedding high-temperature resistant fiber grating arrays during the die manufacturing stage, combined with the coordinated layout of external acoustic emission, infrared and vibration sensors, not only was in-situ sensing of multi-physics information of the die core and surface achieved, but also microsecond-level time synchronization was achieved through the IEEE 1588 protocol. This effectively solved the problem of asynchronous mismatch of multi-source signals under high temperature and strong vibration conditions, laying a reliable data foundation for subsequent high-precision crack evolution analysis.

[0085] S2. Acquire fiber optic grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals through a multi-source synchronous acquisition system during multiple consecutive thermoforming cycles.

[0086] In the During each thermoforming cycle, the fiber optic grating demodulator records the center wavelength of each grating unit at a sampling rate of 20 kHz, forming a wavelength sequence. The acoustic emission sensor continuously monitors elastic wave signals, and records the moment the instantaneous amplitude exceeds a threshold of 45 dB. With energy value Generate acoustic emission event stream { },in For the first Total number of loop events; the infrared thermal imager outputs a surface temperature matrix at a rate of 25 frames per second, extracts the average temperature of a 20×20 pixel block in the central area of ​​the working surface of the mold head, and forms a single-point temperature time series. The triaxial signal output from the vibration accelerometer is low-pass filtered and stored at a sampling rate of 5 kHz, with the radial component being extracted. As the main vibration characteristics, the above four types of data are aligned by timestamp, and representative values ​​of the steady-state stage at the end of each cycle are extracted for feature calculation.

[0087] By synchronously acquiring four types of sensor signals at a high sampling rate in each thermoforming cycle and focusing on extracting representative data in the steady-state phase, the influence of transient interference noise is avoided, and strict alignment of data from different modalities in the time dimension is ensured. This improves the reliability and consistency of subsequent multi-source feature fusion and provides high-quality input for constructing a time-series active feature set.

[0088] S3. Based on the fiber grating wavelength sequence, calculate the slope and second-order rate of change of core strain with the number of thermoforming cycles. Combine the acoustic emission event flow to calculate the coefficient of variation of event interval within a unit cycle and the burst index of energy release, and generate a time-series active feature set that reflects the dynamic characteristics of crack evolution.

[0089] From the Extracting the steady-state center wavelength from a cyclic fiber grating wavelength sequence The original strain was calculated using the strain-wavelength linear relationship. ,in The initial wavelength in the cold state. The strain sensitivity coefficient is used; using the first Cyclic infrared temperature data Combined with the coefficient of linear expansion of the material and reference temperature Calculate the additional strain due to thermal expansion The values ​​of the local micro-strain field at the key points, corrected by thermo-coupling, were obtained. The expression is:

[0090] ;

[0091] Constructing strain sequences Calculate the first-order difference The expression is:

[0092] ;

[0093] Calculate the second-order difference The expression is:

[0094] ;

[0095] For the first A cyclic acoustic emission event stream, calculating the time interval between adjacent events. Calculate the mean with standard deviation Define the event interval variation coefficient as follows:

[0096] ;

[0097] Accumulate event energy over time. The cycle is divided into 5 equal-length sub-intervals, and the slope of each segment is fitted. Calculate the standard deviation With the maximum value Define the burst index of energy release, expressed as:

[0098] ;

[0099] Will , , , Form a four-dimensional vector As the first The temporal activity characteristics of cycles; all This forms a temporally active feature set, which is used as input for the next step of the neural network.

[0100] The proposed first and second strain rates can sensitively capture the nonlinear accelerated deformation behavior in the early stage of crack initiation, while the event interval variation coefficient and energy release burst index quantify the intermittency and burstiness of crack propagation from an acoustic perspective. The temporal active feature set formed by the integration of the four factors breaks through the limitations of traditional single strain or acoustic emission threshold criteria, and effectively enhances the characterization ability of early microcrack dynamic evolution.

[0101] S4. Input the time-series active feature set and the local micro-strain field corrected by thermo-coupling into the pre-trained multimodal fusion neural network model, and output the crack risk discrimination result.

[0102] The first Temporal active feature vector of the cycle Timing-aligned thermo-coupled local micro-strain field Concatenate into a five-dimensional input vector The multimodal fusion neural network model contains five input nodes, each corresponding to... The system consists of five components; the first hidden layer contains 10 neurons and uses the ReLU activation function; the second hidden layer contains 5 neurons; and the output layer is a single neuron and uses the Sigmoid activation function.

[0103] The model was trained on a historical model failure dataset, and FocalLoss was used as the loss function to address the imbalance between positive and negative samples. Input the trained model and output the crack risk probability value. Set the discrimination threshold ,like If the output crack risk assessment result is high risk, the subsequent digital twin simulation process will be triggered.

[0104] By inputting the thermo-coupled corrected local micro-strain field and temporally active features into a pre-trained multimodal fusion neural network, and utilizing the Focal Loss optimization model's learning ability on a small number of high-risk samples, the crack risk discrimination results achieve both high sensitivity and low false alarm rate. This intelligent discrimination mechanism effectively bridges the semantic gap between physical sensing and damage state, providing a precise basis for decision-making on whether to initiate high-cost digital twin simulation.

[0105] S5. When the crack risk assessment result exceeds the preset threshold, the time-series active feature set and the current process parameters are injected into the digital twin corresponding to the physical model head, driving multi-physics field coupling simulation and dynamically adjusting the damage evolution rate weight to calculate the remaining safe service life.

[0106] During the system initialization phase, a finite element digital twin is established based on the three-dimensional geometric model and material properties of the mandrel, and the Paris crack propagation basic equation is pre-set; when the crack risk assessment result is high risk, the current loop is extracted. Temporal activity characteristics and And current process parameters: extrusion pressure Heating temperature Feed rate ; Calculate the crack activity product factor The expression is:

[0107] ;

[0108] Introducing adaptive damage evolution rate weights :

[0109] ;

[0110] in, The highest historical activity factor, To adjust the gain constant.

[0111] Will Introducing the Paris equation, we obtain the modified crack propagation rate model, expressed as:

[0112] ;

[0113] In the digital twin, with the current crack length Using the initial value, numerically integrate the differential equation until the crack length is reached. Reaching the critical failure length The cumulative cycle increment is the remaining safe service life. .

[0114] By introducing an activity product factor driven by both strain acceleration and energy burst index. Based on this, the damage evolution rate weights in the Paris equation are dynamically adjusted. This enables the digital twin to adaptively reflect the actual activity level of crack propagation, overcoming the shortcomings of traditional constant material parameter models with large prediction deviations under varying working conditions, and improving the engineering practicality and accuracy of remaining life calculation.

[0115] S6. Generate a mold head health index based on the remaining safe service life, and trigger a graded early warning signal or send a load adjustment command to the thermoforming equipment based on the health index.

[0116] Obtaining Remaining Safe Service Life from Digital Twins Read the number of service cycles from the system counter. ; Calculate the health index of the model head The expression is:

[0117] ;

[0118] Preset Level 1 Warning Threshold Compared with the level 2 warning threshold .

[0119] like and The monitoring terminal will then display a Level 1 warning: the health status of the mold head has deteriorated, and it is recommended to replace it during the next planned shutdown window; if In addition to displaying a level-two warning: the die head is on the verge of failure and immediate intervention is required, an analog signal is simultaneously sent to the PLC controller of the thermoforming equipment. After receiving the signal, the PLC reduces the output current of the extrusion roller servo motor, thereby reducing the radial pressure acting on the cantilever end of the die head. ,in, This is the pressure regulation coefficient. The rated working pressure is maintained until the current batch is completed, after which the system locks the die head and prohibits further use.

[0120] The hierarchical early warning and closed-loop control mechanism based on the dimensionless health index H not only quantifies the health status of the module with intuitive numerical values, but also realizes the leap from passive alarm to active intervention: when the health index is lower than the secondary threshold, the system automatically reduces the equipment load to delay crack propagation and ensure the safe completion of the current batch. At the same time, the forced replacement strategy eliminates the risk of exceeding the service life, taking into account both production continuity and equipment safety.

[0121] This embodiment also provides a cantilever die head crack monitoring system for hot forming of bearing inner rings, including:

[0122] Multi-source sensor integration module, cyclic signal acquisition module, time-series active feature generation module, crack risk intelligent judgment module, digital twin life prediction module and health index early warning and control module;

[0123] A multi-source sensing integrated module is used to embed a high-temperature resistant fiber optic grating sensor array in the stress concentration area inside the cantilever head, and to configure an acoustic emission sensor, an infrared thermal imager and a vibration sensor on the outside of the head, and to build a multi-source synchronous acquisition system through a time synchronization controller;

[0124] The cyclic signal acquisition module is used to simultaneously acquire fiber grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals during multiple consecutive thermoforming cycles, and to store the four types of data in alignment with the cycle number and timestamp.

[0125] The temporal active feature generation module is used to calculate the slope and second-order rate of change of core strain with the number of thermoforming cycles based on fiber grating wavelength sequence, and to calculate the coefficient of variation of event interval and energy release burst index within a unit cycle by combining acoustic emission event flow, so as to generate a temporal active feature set that reflects the dynamic characteristics of crack evolution.

[0126] The crack risk intelligent discrimination module is used to input the temporal active feature set and the local micro-strain field corrected by thermo-coupling into a pre-trained multimodal fusion neural network model and output the crack risk discrimination result.

[0127] The digital twin life prediction module is used to inject the time-series active feature set and current process parameters into the digital twin corresponding to the physical model when the crack risk judgment result exceeds the preset threshold. This drives the thermo-mechanical-fatigue multiphysics coupling simulation and dynamically corrects the Paris crack propagation equation through adaptive damage evolution rate weights to calculate the remaining safe service life.

[0128] The health index early warning and control module is used to generate a mold head health index based on the remaining safe service life and the number of service cycles already completed. It also triggers graded early warning signals or sends radial load adjustment commands to the thermoforming equipment based on the range of the health index, thereby realizing active mold head maintenance and collaborative optimization of working conditions.

[0129] By constructing a multi-source synchronous monitoring system integrating high-temperature resistant fiber optic gratings, acoustic emission, infrared thermal imaging, and vibration sensing, and introducing high-order temporal dynamic characteristics such as strain change slope and second-order rate of change, event interval variation coefficient, and energy release burst index, an adaptive digital twin life prediction mechanism with crack evolution activity as the core is established. This realizes a closed-loop monitoring chain from static threshold alarm to dynamic evolution perception, intelligent risk judgment, accurate life prediction, and health index-driven active control. This not only improves the sensitivity of early microcrack identification and the accuracy of remaining life prediction in cantilever die heads, but also effectively avoids unplanned downtime and extends the safe service life of the die head through health index-based early warning and real-time load feedback adjustment, ensuring the continuity, safety, and product quality stability of the bearing inner ring thermoforming process.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings, characterized in that: include: A high-temperature resistant fiber optic grating sensor array is embedded in the stress concentration area inside the cantilever head, and an acoustic emission sensor, an infrared thermal imager, and a vibration sensor are configured on the outside of the head to form a multi-source synchronous acquisition system. The fiber grating wavelength sequence, acoustic emission event stream, surface temperature time-series image and vibration time-series signal are acquired through a multi-source synchronous acquisition system during multiple consecutive thermoforming cycles. Based on the fiber grating wavelength sequence, the slope and second-order rate of change of core strain with the number of thermoforming cycles are calculated. Combined with the acoustic emission event flow, the coefficient of variation of the event interval within a unit cycle and the burst index of energy release are calculated to generate a time-series active feature set that reflects the dynamic characteristics of crack evolution. The temporal active feature set and the local micro-strain field corrected by thermo-coupling are jointly input into a pre-trained multimodal fusion neural network model, and the crack risk discrimination result is output. When the crack risk assessment result exceeds the preset threshold, the time-series active feature set and the current process parameters are injected into the digital twin corresponding to the physical model head, driving multi-physics field coupling simulation and dynamically adjusting the damage evolution rate weight to calculate the remaining safe service life. A mold head health index is generated based on the remaining safe service life, and a graded early warning signal is triggered or a load adjustment command is sent to the thermoforming equipment based on the health index.

2. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 1, characterized in that: The specific steps are as follows: A high-temperature resistant fiber Bragg grating sensor array is embedded in the stress concentration area inside the cantilever mold head, and an acoustic emission sensor, an infrared thermal imager, and a vibration sensor are configured outside the mold head to form a multi-source synchronous acquisition system. During the die manufacturing stage, the root of the cantilever and the transition fillet area are identified as the area of ​​maximum principal stress concentration through finite element static simulation. Micro-hole channels with a diameter of 0.5 mm are drilled along the axial direction in the area, penetrating to the core of the die. A fiber optic grating sensor with a nickel-based alloy coating is implanted in a series in a micro-pore channel. The pores are filled with alumina ceramic slurry and sintered at high temperature to form a rigid connection between the sensor and the mold substrate, ensuring effective strain transmission. A piezoelectric acoustic emission sensor with a resonant frequency of 150 kHz is attached to the outer surface of the mold head near the root of the cantilever. An infrared thermal imager with a resolution of 640×480 is installed 1.2 meters above the mold head with the lens facing the working surface of the mold head. A triaxial MEMS vibration accelerometer is fixed on the mold head support flange. The fiber optic demodulator, acoustic emission preamplifier, infrared thermal imager controller, and vibration acquisition module are all connected to the time synchronization controller. The IEEE 1588 precision time protocol is used to achieve microsecond-level synchronous triggering, thus forming a multi-source synchronous acquisition system.

3. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 2, characterized in that: The specific steps for acquiring fiber grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals through a multi-source synchronous acquisition system during multiple consecutive thermoforming cycles are as follows: In the During each thermoforming cycle, the fiber optic grating demodulator records the center wavelength of each grating unit at a sampling rate of 20 kHz, forming a wavelength sequence. ; The acoustic emission sensor continuously monitors elastic wave signals, and records the moment the instantaneous amplitude exceeds a threshold of 45 dB. With energy value Generate acoustic emission event stream { },in For the first Total number of loop events; The infrared thermal imager outputs a surface temperature matrix at a rate of 25 frames per second, extracts the average temperature of a 20×20 pixel block in the central area of ​​the working surface of the mold head, and forms a single-point temperature time series. ; The triaxial signal output from the vibration accelerometer is low-pass filtered and stored at a sampling rate of 5 kHz, with the radial component being extracted. As the main vibration characteristic; After aligning the above four types of data by timestamp, representative values ​​from the steady-state phase at the end of each cycle are extracted for feature calculation.

4. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 3, characterized in that: The method involves calculating the slope and second-order rate of change of core strain with the number of thermoforming cycles based on fiber grating wavelength sequences, and combining this with the acoustic emission event flow to calculate the coefficient of variation of the event interval per unit cycle and the burst index of energy release, thereby generating a temporal active feature set reflecting the dynamic characteristics of crack evolution. The specific steps are as follows: From the Extracting the steady-state center wavelength from a cyclic fiber grating wavelength sequence The original strain was calculated using the strain-wavelength linear relationship. ,in The initial wavelength in the cold state. This is the strain sensitivity coefficient; Using the first Cyclic infrared temperature data Combined with the coefficient of linear expansion of the material and reference temperature Calculate the additional strain due to thermal expansion ; The values ​​of the local micro-strain field at the key points were obtained after thermo-coupling correction. The expression is: ； Constructing strain sequences Calculate the first-order difference The expression is: ; Calculate the second-order difference The expression is: ; For the first A cyclic acoustic emission event stream, calculating the time interval between adjacent events. Calculate the mean with standard deviation Define the event interval variation coefficient as follows: ； Accumulate event energy over time. The cycle is divided into 5 equal-length sub-intervals, and the slope of each segment is fitted. Calculate the standard deviation With the maximum value Define the burst index of energy release, expressed as: ； Will , , , Form a four-dimensional vector As the first The temporal activity characteristics of cycles; all This forms a temporally active feature set, which is used as input for the next step of the neural network.

5. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 4, characterized in that: The steps for inputting the temporally active feature set and the thermo-coupled corrected local micro-strain field into a pre-trained multimodal fusion neural network model to output crack risk discrimination results are as follows: The first Temporal active feature vector of the cycle Timing-aligned thermo-coupled local micro-strain field Concatenate into a five-dimensional input vector ; The multimodal fusion neural network model contains five input nodes, which correspond to... The five components; The first hidden layer contains 10 neurons and uses the ReLU activation function; The second hidden layer contains 5 neurons; The output layer is a single neuron, using the Sigmoid activation function; The model was trained on a historical model failure dataset, and FocalLoss was used as the loss function to address the imbalance between positive and negative samples. Will Input the trained model and output the crack risk probability value. ; Set the discrimination threshold ,like If the output crack risk assessment result is high risk, the subsequent digital twin simulation process will be triggered.

6. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 5, characterized in that: When the crack risk assessment result exceeds a preset threshold, the time-series active feature set and current process parameters are injected into the digital twin corresponding to the physical model head to drive multi-physics coupling simulation and dynamically adjust the damage evolution rate weight to calculate the remaining safe service life. The specific steps are as follows: During the system initialization phase, a finite element digital twin is established based on the three-dimensional geometric model and material properties of the mold head, and the basic equation for Paris crack propagation is pre-set. When the crack risk assessment result is high risk, extract the current loop. Temporal activity characteristics and And current process parameters: extrusion pressure Heating temperature Feed rate ; Calculate the crack activity product factor The expression is: ； Introducing adaptive damage evolution rate weights : ; in, The highest historical activity factor, To adjust the gain constant; Will Introducing the Paris equation, we obtain the modified crack propagation rate model, expressed as: ； In the digital twin, with the current crack length Using the initial value, numerically integrate the differential equation until the crack length is reached. Reaching the critical failure length The cumulative cycle increment is the remaining safe service life. .

7. The method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings as described in claim 6, characterized in that: The specific steps for generating a mold head health index based on the remaining safe service life, and triggering a graded early warning signal or sending a load adjustment command to the thermoforming equipment based on the health index are as follows: Obtaining Remaining Safe Service Life from Digital Twins Read the number of service cycles from the system counter. ; Calculate the health index of the model head The expression is: ； Preset Level 1 Warning Threshold Compared with the level 2 warning threshold ; like and If the condition is not met, a Level 1 warning will be displayed on the monitoring terminal: the health status of the mold head has deteriorated, and it is recommended to replace it during the next planned shutdown window; like In addition to displaying a level-two warning: the die head is on the verge of failure and immediate intervention is required, an analog signal is simultaneously sent to the PLC controller of the thermoforming equipment. After receiving the signal, the PLC reduces the output current of the extrusion roller servo motor, thereby reducing the radial pressure acting on the cantilever end of the die head. ,in, This is the pressure regulation coefficient. Rated working pressure; The pressure adjustment continues until the current batch is completed, after which the system locks the die head and prohibits further use.

8. A monitoring system based on the method for monitoring cracks in a cantilever die head used in the thermoforming of bearing inner rings according to any one of claims 1-7, characterized in that: include: Multi-source sensor integration module, cyclic signal acquisition module, time-series active feature generation module, crack risk intelligent judgment module, digital twin life prediction module and health index early warning and control module; The multi-source sensing integrated module is used to embed a high-temperature resistant fiber optic grating sensor array in the stress concentration area inside the cantilever head, and to configure an acoustic emission sensor, an infrared thermal imager and a vibration sensor on the outside of the head, and to build a multi-source synchronous acquisition system through a time synchronization controller. The cyclic signal acquisition module is used to simultaneously acquire fiber grating wavelength sequences, acoustic emission event streams, surface temperature time-series images, and vibration time-series signals during multiple consecutive thermoforming cycles, and to store the four types of data in alignment with the cycle number and timestamp. The time-series active feature generation module is used to calculate the slope and second-order rate of change of core strain with the number of thermoforming cycles based on the fiber grating wavelength sequence, and to calculate the coefficient of variation of event interval within a unit cycle and the burst index of energy release in combination with the acoustic emission event flow, so as to generate a time-series active feature set that reflects the dynamic characteristics of crack evolution. The crack risk intelligent discrimination module is used to input the temporal active feature set and the local micro-strain field corrected by thermo-coupling into a pre-trained multimodal fusion neural network model and output the crack risk discrimination result. The digital twin life prediction module is used to inject the time-series active feature set and current process parameters into the digital twin corresponding to the physical mold when the crack risk judgment result exceeds the preset threshold, drive the thermo-mechanical-fatigue multiphysics coupling simulation, dynamically correct the Paris crack propagation equation through adaptive damage evolution rate weight, and calculate the remaining safe service life. The health index early warning and control module is used to generate a mold head health index based on the remaining safe service life and the number of service cycles, and to trigger a graded early warning signal or send a radial load adjustment command to the thermoforming equipment according to the range of the health index, so as to realize active maintenance of the mold head and collaborative optimization of working conditions.

Images