Self-adaptive control method and system for spin riveting gap of rotary fastener

Abstract

The invention belongs to the technical field of spin riveting control, and discloses a self-adaptive control method and system for a spin riveting gap of a rotary fastener. Multi-dimensional key parameters of the fastener are comprehensively captured in a collection and calibration stage, and cold rolling physical characteristics of spin riveting machining are combined; constructing three core models of dynamic prediction of the spin riveting gap, gap springback prediction and interference factor association, and then realizing real-time linkage of parameters, gaps and interference factors through an adaptive adjustment algorithm; aiming at various dynamic interferences such as material parameter difference, tool abrasion and temperature fluctuation, accurate identification and targeted adjustment can be carried out, the problem that a gap is too large or too small due to fixed parameters is avoided, and the bonding strength, the coaxiality and the anti-fatigue performance of the rotary fastener are improved; a cooperative detection system is formed by a displacement sensor, a torque sensor and a laser gap sensor, and dynamic changes of gaps in the spin riveting process are accurately captured through double verification of indirect detection and direct detection.

Description

Technical Field

[0001] This invention belongs to the field of riveting control technology, specifically a method and system for adaptive control of riveting gap of rotary fasteners. Background Technology

[0002] Rotary fasteners are widely used in construction, machinery, aerospace, and other fields. The riveting clearance, as a core control parameter in riveting, directly affects the joint strength, coaxiality, and fatigue resistance after riveting. Currently, most rotary fastener riveting processes employ conventional metal riveting assembly methods, but existing riveting clearance control methods suffer from the following technical problems: Existing riveting clearance control methods mostly adopt the mode of preset fixed process parameters. Typically, parameters such as the riveting head speed, pressure, and feed rate are set in advance according to the fastener model, and the parameters remain unchanged throughout the processing. This does not fully consider the impact of the machining error of the rotating fastener itself, the difference in material physical parameters, and the dynamic interference during the riveting process (such as changes in riveting temperature and tooling wear) on the clearance.

[0003] Some improvement solutions attempt to adjust parameters through single torque detection, but these solutions suffer from problems such as excessively long transmission chains and large detection errors. Furthermore, they fail to predict and compensate for the springback amount after riveting, ultimately resulting in excessively large or small gaps after riveting. Excessively large gaps can easily cause fasteners to loosen or fall off, while excessively small gaps can lead to fastener deformation and cracks, and in severe cases, damage to the riveting equipment.

[0004] Existing control methods rely on a single detection method, mostly using torque detection to adjust parameters. This not only results in large detection errors but also makes it difficult to accurately capture the dynamic changes in clearance during riveting, affecting clearance control accuracy and ultimately leading to a low product qualification rate. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive control method and system for the riveting gap of rotary fasteners, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control method for the riveting clearance of a rotary fastener, comprising the following specific steps: Preferably, in the acquisition and calibration stage, the rotating fastener to be riveted is sent to the pre-processing station, and the multi-dimensional parameter synchronous acquisition system is started. Through the laser rangefinder, hardness tester, vision recognition module and elastic modulus tester, the fastener's special key parameters are acquired synchronously, including the fastener hole diameter form and position tolerance, wall thickness deviation value, riveting surface roughness, material hardness and elastic modulus fluctuation value. At the same time, the spatial three-dimensional coordinates of the fastener riveting reference surface are obtained through the visual positioning algorithm to complete the riveting positioning reference calibration. The data synchronization acquisition mechanism adopts a hardware-triggered synchronization method. Each detection device is triggered to acquire data through a unified clock signal, and the acquisition frequency is uniformly set to 200Hz to ensure that the acquisition time difference of different parameters is ≤5ms. For slowly changing parameters such as the roughness of the riveted surface, the acquisition frequency can be dynamically reduced to 50Hz. At the same time, the acquisition results are temporarily stored through a data caching mechanism and synchronously written to the database after all parameters are acquired, avoiding data mismatch caused by the acquisition delay of a single parameter.

[0007] The reference riveting clearance of the swivel fastener is calibrated, and a database linking fastener multi-dimensional parameters, reference clearance, and load level is established. An automatic fastener model recognition function is embedded to achieve integrated adaptation of pre-detection of fasteners to be riveted, automatic calibration of reference clearance, and load level matching. Initial wear data of tooling and reference values ​​of ambient temperature are collected and entered into the database.

[0008] Preferably, in the model construction stage, the multi-dimensional special parameters of the rotary fastener, the initial tooling data and the environmental benchmark value collected in the acquisition and calibration stage are combined with the cold rolling physical characteristics of the rotary fastener riveting process to construct a dynamic prediction model for the riveting gap. The core input variables are the hardness, elastic modulus fluctuation value and wall thickness deviation value of the fastener material, and the benchmark gap calibrated in the acquisition and calibration stage is used as the target variable. The correlation coefficients of the riveting head rotation speed, feed pressure and holding time are introduced. Combined with the load level weight of the rotary fastener, the model is trained by the LSTM neural network algorithm to optimize the model convergence speed and prediction accuracy, so as to realize the advance prediction of the gap change during the riveting process. A gap springback prediction sub-model was constructed. Based on the elastic recovery characteristics of the rotating fastener material and combined with historical riveting springback data, the springback amount of the gap after riveting unloading was predicted. The springback amount, tooling wear attenuation coefficient, and ambient temperature fluctuation coefficient were included in the model correction term. The Monte Carlo method was used to optimize the springback prediction accuracy. Construct a sub-model of interference factors to clarify the correspondence between tooling wear, temperature fluctuation, material parameter deviation and clearance interference, so as to realize the early prediction of the degree of influence of interference factors on clearance.

[0009] The specific association rules are as follows: ① For every 0.05mm increase in tooling wear, the clearance interference increases by 0.006-0.008mm; ② For every ±5℃ deviation of the ambient temperature from the reference value, the gap interference changes by ±0.005mm; ③ For every ±20 HV deviation of the material hardness from the standard value, the gap interference changes by ±0.004 mm; ④ For every ±5 GPa fluctuation in elastic modulus, the gap disturbance changes by ±0.003 mm. The above correlation was verified by fitting historical data and embedded into the sub-model.

[0010] Preferably, the initial matching stage combines the database matching results from the calibration stage to perform the matching of initial riveting parameters and equipment initialization. Through the dynamic prediction model, gap springback prediction sub-model, and interference factor correlation sub-model constructed in the model building stage, the actual parameters, load level, and database matching results of the fastener to be riveted are linked to automatically match the initial riveting parameters, including riveting head rotation speed, feed speed, riveting pressure, riveting angle, and initial holding time. The initial parameter matching follows three principles: differentiated adaptation, load adaptation, and interference prediction adaptation. For fasteners with high hardness, high elastic modulus, and high load level, a higher riveting pressure and a moderate rotation speed are matched; for fasteners with large wall thickness deviations, the feed speed is dynamically adjusted; and for cases where slight interference is predicted, the initial parameters are fine-tuned in advance. The six-degree-of-freedom positioning mechanism aligns the riveting head with the fastener riveting reference surface, initializes the displacement sensor and torque sensor of the riveting equipment, and simultaneously starts the sensor zero-point calibration module to ensure sensor detection accuracy. An initial parameter pre-verification step is added to verify the adaptability of the initial parameters through small-scale trial riveting. If the pre-verification gap deviation exceeds the allowable range, the system automatically returns to the model to re-match the parameters.

[0011] Preferably, during the detection and identification phase, the riveting equipment is activated, and the riveting head performs riveting operations according to the initial parameters matched in the initial matching phase. Simultaneously, a multi-sensor collaborative detection system is activated to achieve triple verification of indirect detection, direct detection, and interference detection. The displacement sensor detects the feed displacement of the riveting head in real time, indirectly reflecting the change in gap. The torque sensor detects the riveting torque during the riveting process in real time, supplementing the detection of gap status and simultaneously collecting torque fluctuation data to assist in judging interference factors. The laser gap sensor directly detects the real-time gap value of the fastener riveting surface, achieving direct gap detection. The data from the three sensors are fused in real time to complete the dual detection verification of gap and interference detection verification. The laser gap sensor is installed on the side of the riveting surface, at a vertical distance of 50-80mm from the reference surface, with a detection spot diameter ≤0.5mm; the displacement sensor and torque sensor are integrated into the output end of the riveting head, at a distance ≤10mm from the riveting surface; the collaborative triggering adopts a laser sensor priority triggering mechanism, with the laser sensor triggering once every 10ms for direct detection, and simultaneously triggering the displacement and torque sensors for indirect detection. The data transmission priority is laser data > displacement data > torque data, ensuring that the core detection data is given priority for parameter adjustment.

[0012] The Kalman filter algorithm is used to filter out detection noise caused by tooling vibration, temperature change, and material parameter fluctuation. The interference factor correlation sub-model in the linkage model construction stage is linked to identify the main interference factors affecting the gap in real time, and output the interference signal strength, interference type and impact on the gap in real time; the detected interference information is synchronously fed back to the parameter adjustment module.

[0013] Preferably, the parameter adjustment stage compares the real-time gap value and gap change detected in the detection and identification stage with the reference gap calibrated in the acquisition and calibration stage, and calculates the gap deviation value. The dynamic prediction model, gap rebound prediction sub-model, interference factor correlation sub-model, and interference identification results in the linkage model construction stage are all implemented using a PID adaptive adjustment algorithm to adjust the riveting parameters in real time. This achieves closed-loop control of gap detection, deviation calculation, interference identification, and parameter adjustment. When the real-time gap is greater than the reference gap, the riveting pressure is automatically increased, the feed speed is reduced, the riveting head speed is finely adjusted, the initial pressure holding time is extended, the riveting deformation is accelerated, and the gap is reduced. When the real-time gap is less than the reference gap, the riveting pressure is reduced, the feed speed is increased, and the riveting angle is reduced. For interference factors identified during the detection and identification phase, targeted adjustments are made. For example, if the material hardness is too high and the gap shrinks slowly, the riveting angle and pressure are increased; if the tooling wear is aggravated and the gap deviation is caused, the displacement of the riveting head is dynamically compensated; if the temperature fluctuation causes the gap to be abnormal, the holding time is finely adjusted. The adjustment interval is 10ms, which realizes real-time linkage between riveting parameters and gap changes and interference factors, and the adjustment logic is adapted to the structural characteristics and load requirements of the rotary fastener.

[0014] Preferably, in the springback verification stage, when the real-time gap value detected in the detection and identification stage is close to the reference gap, the riveting equipment is controlled to reduce the riveting pressure and enter the unloading stage. By predicting the rebound amount of the gap rebound sub-model during the model construction stage, and associating it with the material elastic recovery characteristics of the fastener to be riveted, the holding time and unloading speed of the riveting head are automatically adjusted to compensate for the rebound of the gap, and the real-time gap is adjusted to a preset value that is smaller than the reference gap in advance. After unloading, the final gap after riveting is detected a second time using a laser gap sensor. Combined with the final torque value detected by the torque sensor, a dual judgment of gap qualification is completed to determine whether the gap value is within the design allowable error range. If it is not qualified, the springback amount and interference factors are linked to return to the parameter adjustment stage to readjust the riveting parameters and perform a second fine-tuning riveting until the gap is qualified. If it is qualified, the entire process riveting data of the fastener is recorded and entered into the associated database. The springback compensation mechanism of gradient pressure holding and gradient unloading realizes the prediction and compensation of springback amount.

[0015] Preferably, the learning and optimization stage is based on the full-process riveting data of the aforementioned six stages, and performs self-learning optimization of the model and parameters. The full-process riveting data of each rotary fastener is completely recorded, including fastener multi-dimensional parameters, load level, initial riveting parameters in the initial matching stage, real-time adjustment parameters in the parameter adjustment stage, gap deviation value, springback amount, type and influence of interference factors, final gap detection result, tooling wear real-time data, and ambient temperature fluctuation data. Reinforcement learning algorithms are used to analyze the correlation between fastener multi-dimensional parameters, riveting parameters, interference factors, clearance control effect and load adaptability. The parameters of the dynamic prediction model, clearance springback prediction sub-model and interference factor correlation sub-model in the model building stage are optimized. The parameter matching relationship in the fastener multi-dimensional parameters, reference clearance and load level correlation database in the data acquisition and calibration stage is corrected. A riveting parameter optimization library is constructed to form a special parameter adaptation scheme based on the riveting data of rotating fasteners of different models and load levels. A model iteration is initiated every 50 sets of grid riveting data, or an emergency iteration is triggered when the average deviation of fastener gaps in 10 consecutive sets is greater than 0.01 mm. The convergence criterion is that the deviation between the model's predicted gap change and the actual value is ≤0.008 mm, and the deviation shows no decreasing trend after 3 consecutive iterations. The current optimization cycle is then stopped, and the optimization parameters are solidified into the riveting parameter optimization library.

[0016] A tooling wear early warning module was added to predict the timing of tooling replacement based on the recorded tooling wear data and the amount of clearance interference. The adaptive control system of riveting assembly was improved to upgrade the riveting clearance control.

[0017] The present invention also provides an adaptive control system for the riveting clearance of rotary fasteners, based on the above method, comprising: The data acquisition and calibration module is equipped with a laser rangefinder, hardness tester, vision recognition module and elastic modulus tester. It completes the acquisition of multi-dimensional parameters of the fastener to be riveted, the positioning and calibration of the riveting reference surface, the calibration of the reference riveting gap, the establishment of a database linking fastener multi-dimensional parameters, reference gap and load level, and the input of tooling initial wear data and ambient temperature reference value. The model prediction and initial matching module embeds an LSTM neural network and Monte Carlo method, runs a dynamic prediction model for riveting gap, a gap springback prediction sub-model and an interference factor correlation sub-model to complete gap and interference prediction; it links the database matching results to automatically adapt the initial riveting parameters, and uses a six-degree-of-freedom positioning mechanism to align the riveting head and complete sensor zero-point calibration and initial parameter pre-verification. The real-time detection and interference identification module relies on a collaborative detection system composed of displacement sensors, torque sensors, and laser gap sensors to achieve dual detection of gaps, both indirect and direct. It uses a Kalman filter algorithm to filter detection noise, identifies interference factors such as material hardness fluctuations and increased tooling wear in real time, outputs the type and magnitude of interference, and feeds it back to the parameter adjustment and springback verification module. The parameter adjustment and springback verification module adopts a PID adaptive adjustment algorithm, links detection data and prediction model, and fine-tunes riveting parameters in real time to achieve closed-loop control; it controls the riveting equipment to maintain pressure and unload gradient to complete springback compensation, and performs gap qualification judgment through laser gap sensor and torque sensor. If it fails to meet the qualification, it returns to readjustment. The self-learning optimization module uses reinforcement learning algorithms to record riveting data throughout the entire process, optimizes the matching relationship between the parameters of each prediction model and the database, builds a riveting parameter optimization library, and adds a tooling wear early warning function.

[0018] The beneficial effects of this invention are as follows: 1. This invention comprehensively captures multi-dimensional key parameters of fasteners during the calibration phase. Combining the cold rolling physical characteristics of riveting, it constructs three core models: dynamic prediction of riveting gap, prediction of gap springback, and correlation of interference factors. Then, through an adaptive adjustment algorithm, it achieves real-time linkage between parameters, gap, and interference factors. For various dynamic interferences such as material parameter differences, tooling wear, and temperature fluctuations, it can accurately identify and perform targeted adjustments to avoid problems of excessively large or small gaps caused by fixed parameters. This ensures that the gap after riveting is always within a reasonable error range, improves the bonding strength, coaxiality, and fatigue resistance of the rotating fastener, and fully adapts to high-load usage requirements.

[0019] 2. This invention employs a collaborative detection system composed of a displacement sensor, a torque sensor, and a laser gap sensor. Through dual verification of indirect and direct detection, it accurately captures the dynamic changes in the gap during the riveting process. Combined with a filtering algorithm, it effectively filters out various detection noises. At the same time, it optimizes the design of the detection transmission link to reduce detection errors. Furthermore, it adds a dual judgment mechanism of initial parameter pre-verification and gap qualification to further eliminate potential influences such as unreasonable initial parameters and detection deviations, providing accurate and reliable data support for parameter adjustment and ensuring the stability of riveting quality from the detection level.

[0020] 3. This invention accurately predicts the amount of springback caused by the elastic recovery of materials through a gap springback prediction sub-model, and completes targeted springback compensation by combining gradient pressure holding and gradient unloading mechanisms, filling the gap in existing technologies that do not consider the impact of springback. In the learning and optimization stage, relying on the full-process riveting data, an adaptation algorithm is used to continuously optimize the matching relationship between model parameters and database, build a dedicated riveting parameter optimization library, and add a tooling wear warning function to predict the timing of tooling replacement in advance. It ensures the gap accuracy of single riveting, and flexibly adapts to different models and load levels of rotary fasteners through self-learning capabilities, reducing the parameter debugging cost of subsequent production, reducing quality risks caused by excessive tooling wear, and improving product qualification rate and large-scale production efficiency. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the adaptive control process for the riveting clearance of the rotary fastener according to the present invention. Figure 2 This is a flowchart of the model construction stage of the present invention; Figure 3 This is a flowchart of the parameter adjustment and rebound verification process of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figures 1 to 3 As shown, this embodiment of the invention provides an adaptive control method for the riveting clearance of a rotary fastener, including the following specific steps: In the acquisition and calibration stage, the swivel fasteners to be riveted (including the base fastener and the rivet pin, and swivel fasteners adapted for scaffolding) are sent to the pre-processing station, and a multi-dimensional parameter synchronous acquisition system is started. Through a laser rangefinder, hardness tester, vision recognition module and elastic modulus tester, the measurement accuracy of the laser rangefinder is ≤ ±0.01mm. Key parameters of the fastener are acquired synchronously, including the fastener hole diameter form and position tolerance, wall thickness deviation, riveting surface roughness, material hardness, and elastic modulus fluctuation value. The HV detection range is 80-350. At the same time, the spatial three-dimensional coordinates of the fastener riveting reference surface are obtained through a visual positioning algorithm to complete the riveting positioning reference calibration. The visual positioning algorithm adopts a feature point matching and edge detection fusion scheme to extract three key feature points: the center of the riveting reference surface, the midpoint of the edge, and the center of the positioning hole. The coordinate iteration optimization eliminates the influence of perspective distortion, and the positioning accuracy is ≤ ±0.008mm. For different types of fasteners, the algorithm has a built-in feature point template library and automatically matches the corresponding template to complete the rapid positioning, with an adaptation efficiency of ≥10 pieces / minute.

[0024] Based on the scaffolding load-bearing design requirements and riveting assembly process specifications of swivel couplers, the reference riveting gap of the swivel couplers is calibrated, that is, the target gap value to be achieved after riveting, with an allowable deviation range of ±0.02mm. A database linking multi-dimensional parameters of the couplers, reference gap, and load level is established, and an automatic coupler model recognition function is embedded to achieve integrated adaptation of pre-inspection of couplers to be riveted, automatic calibration of reference gap, and load level matching. Initial wear data of tooling and reference values ​​of ambient temperature are collected and entered into the database.

[0025] The associated database uses a structured storage format, and the core fields include fastener model, hole diameter tolerance, wall thickness deviation, material hardness, elastic modulus fluctuation, reference gap, load rating, tooling number, ambient temperature reference value, and data acquisition time. After every 100 sets of fastener riveting operations are completed, an incremental update is automatically triggered, writing new riveting parameters, gap control effects, and interference factor data into the corresponding fields; when the cumulative data for the same type of fastener reaches 500 sets, data clustering optimization is automatically performed, outliers are removed, and the baseline parameter matching relationship for that type of fastener is updated.

[0026] In the model construction phase, the multi-dimensional special parameters of the rotary fastener, initial tooling data, and environmental benchmark values ​​collected in the data acquisition and calibration phase are combined with the cold rolling physical characteristics of the rotary fastener riveting process to construct a dynamic prediction model for the riveting gap. The core input variables are the hardness, elastic modulus fluctuation value, and wall thickness deviation of the fastener material, and the benchmark gap calibrated in the data acquisition and calibration phase is used as the target variable. The correlation coefficients of the riveting head rotation speed, feed pressure, and holding time are introduced. Combined with the load level weight of the rotary fastener, the model is trained through the LSTM neural network algorithm to optimize the model convergence speed and prediction accuracy, so as to realize the advance prediction of the gap change during the riveting process with a prediction error ≤0.015mm. The LSTM neural network model contains three hidden layers, with 64, 32, and 16 neurons in each layer, respectively. The input layer dimension matches the number of core input variables, being 3-dimensional, corresponding to material hardness, elastic modulus fluctuation value, and wall thickness deviation value; the output layer is 1-dimensional, corresponding to the predicted value of gap change. The model training steps are as follows: ① Extract 5,000 sets of historical sample data from the associated database, including fastener core parameters, riveting process parameters, and actual clearance changes, and divide them into training set, validation set, and test set in a ratio of 7:2:1; ② Set the optimizer to Adam, the initial learning rate to 0.001, and adopt a learning rate decay strategy, which decays to 0.9 times the original value every 100 rounds; ③ The loss function is the mean squared error, and the number of iterations is set to 500 rounds. Training stops when the loss on the validation set does not decrease for 20 consecutive rounds. ④ After training is completed, the prediction error is verified through the test set to ensure that the prediction error is ≤0.015mm.

[0027] The correlation logic of the physical properties of cold rolling is as follows: extract the plastic deformation law, contact stress distribution characteristics and metal flow direction of the material during riveting; use the correlation between plastic deformation and riveting pressure and rotation speed as the model constraint condition; use the contact stress threshold (corresponding to 80% of the yield strength of the fastener material) as the upper limit of parameter adjustment to ensure that the model prediction results conform to the physical boundary of actual processing and avoid fastener damage caused by parameters exceeding the material's tolerance range.

[0028] A gap springback prediction sub-model was constructed. Based on the elastic recovery characteristics of the rotating fastener material and combined with historical riveting springback data, the springback amount of the gap after riveting unloading was predicted. The springback amount, tooling wear attenuation coefficient, and ambient temperature fluctuation coefficient were included in the model correction term. The Monte Carlo method was used to optimize the springback prediction accuracy. Construct a sub-model of interference factors to clarify the correspondence between tooling wear, temperature fluctuation, material parameter deviation and clearance interference, so as to realize the early prediction of the degree of influence of interference factors on clearance.

[0029] The specific application logic of the Monte Carlo method is as follows: ① Based on historical rebound data, determine the probability distribution type of rebound amount (normal distribution), and extract the mean μ and standard deviation σ; ② Combining the tooling wear attenuation coefficient (range 0.95-1.0) and the ambient temperature fluctuation coefficient (range 0.9-1.1), the sampling variables are set as rebound amount, wear coefficient, and temperature coefficient, and the sampling number is 1000 times; ③ Calculate the theoretical rebound compensation amount for each sampling result, remove outliers and take the average value as the final predicted rebound amount. Outliers refer to data that deviate from the mean by 3σ. ④ The model inputs are material elastic recovery characteristic parameters, real-time tooling wear data, and ambient temperature data. The output is the optimized rebound prediction value, which is then fused with the basic prediction results of the sub-model to complete the accuracy correction.

[0030] The initial matching phase combines the database matching results from the calibration phase to perform initial riveting parameter matching and equipment initialization. Through the dynamic prediction model, gap rebound prediction sub-model, and interference factor correlation sub-model constructed in the model building phase, the actual parameters, load level, and database matching results of the fastener to be riveted are linked to automatically match the initial riveting parameters, including riveting head speed, feed speed, riveting pressure, riveting angle, and initial holding time. The adaptation range for riveting head speed is 800-1500 rpm, and the adaptation range for riveting pressure is 5-25 MPa. Initial parameter matching follows three principles: differentiated adaptation, load adaptation, and interference prediction adaptation. For fasteners with high hardness, high elastic modulus, and high load level, higher riveting pressure (18-25 MPa) and moderate speed (1000-1200 rpm) are matched. For fasteners with large wall thickness deviations, the feed speed is dynamically adjusted to avoid gap deviations caused by unreasonable initial parameters. For cases where slight interference is predicted, the initial parameters are fine-tuned in advance to achieve interference prediction. The scenario adaptation design of the reinforcement learning algorithm is as follows: ① The intelligent agent is the decision-making unit for riveting parameter optimization, and the environment is the entire riveting process, including fastener characteristics, interference factors, and gap status; ②The state space includes fastener multi-dimensional parameters, real-time riveting parameters, types and influence of interference factors, gap deviation value, and springback amount; ③ The action space is the adjustment range of the riveting parameters, corresponding to the riveting head speed ±50rpm, feed speed ±0.1mm / s, riveting pressure ±0.5MPa, riveting angle ±1°, and holding time ±0.1s; ④ Reward function design: +10 when the gap is acceptable and the deviation is ≤0.01mm, +5 when the deviation is between 0.01-0.02mm, -8 when unacceptable, and -10 when tooling wear exceeds the threshold. ⑤ The algorithm input is the riveting data of the entire process, and the output is the optimized model parameter adjustment amount and riveting parameter adaptation scheme. The algorithm is updated once every 100 sets of riveting data.

[0031] The specific criteria for determining the triple adaptation principle are as follows: ① The load levels are divided into three levels: low (≤10MPa), medium (10-20MPa), and high (>20MPa), with corresponding basic riveting pressure values ​​of 5-10MPa, 10-18MPa, and 18-25MPa, respectively; ② When the material hardness HV≤150, the riveting speed should be 1200-1500 rpm; when HV is between 150-300, the speed should be 1000-1200 rpm; when HV>300, the speed should be 800-1000 rpm; when the wall thickness deviation is >0.03mm, the feed rate should be reduced by 20%-30%. ③ When the predicted interference factor is ≤0.005mm, the initial parameter fine-tuning range is ≤5%; when the interference factor is between 0.005-0.01mm, the fine-tuning range is 5%-10%; when the interference factor is >0.01mm, the fine-tuning range is 10%-15%.

[0032] The six-degree-of-freedom positioning mechanism aligns the riveting head with the fastener riveting reference surface, with an alignment deviation of ≤0.01mm. The displacement sensor and torque sensor of the riveting equipment are initialized. Both the displacement sensor and torque sensor are installed at the output end of the riveting head, shortening the transmission chain and reducing the detection error to ±0.005MPa. The sensor zero-point calibration module is started simultaneously to ensure the sensor detection accuracy. An initial parameter pre-verification step is added. By performing small-scale riveting trials with a riveting stroke ≤0.5mm, the adaptability of the initial parameters is verified. If the pre-verification gap deviation exceeds the allowable range, the system automatically returns to the model to re-match the parameters.

[0033] The specific process for initial parameter pre-verification is as follows: ① Trial riveting parameters: riveting stroke is 0.3-0.5mm, holding time is 0.5s, and riveting speed is 80% of the initial matching value; ② Judgment threshold: When the gap deviation after trial riveting is ≤ ±0.015mm and the torque fluctuation is ≤ ±0.005MPa, the pre-verification is deemed qualified; ③ If the deviation is between 0.015-0.025mm, the initial parameters will be automatically fine-tuned (by 5%-8%) and the riveting will be retried; if the deviation is greater than 0.025mm, the system will return to the model prediction and initial matching module to rematch the initial parameters. A maximum of 3 riveting attempts are allowed. If all 3 attempts fail, the equipment will be stopped for inspection.

[0034] In the detection and identification phase, the riveting equipment is activated, and the riveting head performs riveting operations according to the initial parameters matched in the initial matching phase. Simultaneously, a multi-sensor collaborative detection system is activated to achieve triple verification of indirect detection, direct detection, and interference detection. The displacement sensor detects the feed displacement of the riveting head in real time, indirectly reflecting the amount of gap change. The torque sensor detects the riveting torque in real time during the riveting process. The torque change is positively correlated with the gap change, supplementing the detection of the gap status and simultaneously collecting torque fluctuation data to assist in judging interference factors. The laser gap sensor directly detects the real-time gap value of the fastener riveting surface, realizing direct gap detection. The data from the three sensors are fused in real time to complete the dual detection and verification of the gap and the interference detection verification. The data fusion adopts a direct detection priority combined with indirect detection auxiliary correction rules. The direct detection data of the laser gap sensor is used as the core basis. When the detection frequency is ≥100Hz and the data fluctuation is ≤±0.008mm, the real-time gap value is directly output. When the laser sensor is interfered with, that is, the detection frequency is <80Hz or the fluctuation is >±0.015mm, the fusion data of the displacement sensor and torque sensor is automatically activated. The detection results are corrected through the displacement, torque and gap mapping relationship to ensure that the data update interval after fusion is ≤10ms and the detection error is ≤±0.01mm.

[0035] The Kalman filter algorithm is used to filter out detection noise caused by tooling vibration, temperature change, and material parameter fluctuation. It is adapted to riveting scenarios and links the interference factors in the model construction stage to the associated sub-model. It identifies the main interference factors affecting the gap in real time, such as material hardness fluctuation, tooling wear, and sudden changes in ambient temperature. It outputs the interference signal strength, interference type, and impact on the gap in real time. The detected interference information is synchronously fed back to the parameter adjustment module.

[0036] The scenario-based adaptation settings for the filtering algorithm are as follows: the noise figure associated with the rotational speed of the riveting head is introduced into the state equation. When the rotational speed is >1200rpm, the noise figure increases to 1.2 times, and when the rotational speed is <1000rpm, the noise figure decreases to 0.8 times. The size of the filtering window in the observation equation is dynamically adjusted to 5-10 sampling points, and automatically switched according to the fluctuation amplitude of the detection data. When the fluctuation is >±0.01mm, the window is expanded to 10 sampling points, and when the fluctuation is ≤±0.005mm, the window is shrunk to 5 sampling points, ensuring that noise is filtered while retaining the true trend of gap change.

[0037] In the parameter adjustment stage, the real-time gap value and gap change detected in the detection and identification stage are compared with the reference gap calibrated in the acquisition and calibration stage to calculate the gap deviation value. The dynamic prediction model, gap rebound prediction sub-model, interference factor correlation sub-model, and interference identification results in the linkage model construction stage, as well as the interference identification results in the detection and identification stage, adopt a PID adaptive adjustment algorithm to optimize the adjustment coefficient, adapt to the riveting characteristics of rotary fasteners, and adjust the riveting parameters in real time to achieve closed-loop control of gap detection, deviation calculation, interference identification, and parameter adjustment. When the real-time gap is greater than the reference gap, the riveting pressure is automatically increased, the feed speed is reduced, the riveting head speed is finely adjusted, the initial pressure holding time is extended, the riveting deformation is accelerated, and the gap is reduced. When the real-time gap is less than the reference gap, the riveting pressure is reduced, the feed speed is increased, and the riveting angle is reduced to avoid excessive deformation and cracking of the fastener, while protecting the riveting equipment. The specific parameters and adaptation rules of the PID algorithm are as follows: ① Initial coefficient settings: proportional coefficient Kp=5.0, integral coefficient Ki=0.1, differential coefficient Kd=0.5; ② Dynamic adjustment rules for coefficients: When the gap deviation is >0.01mm, Kp increases to 6.0, Ki remains at 0.1, and Kd decreases to 0.3; when the gap deviation is <0.005mm, Kp decreases to 3.0, Ki increases to 0.2, and Kd remains at 0.5. ③Scene adaptation logic: For fasteners with a load > 20MPa, the Kp value range is 5.5-6.5, and for fasteners with a load < 10MPa, the Kp value range is 3.5-4.5; when the material hardness HV > 300, Ki increases to 0.15, and when HV < 150, Ki decreases to 0.08. ④ The algorithm inputs are the real-time gap deviation value and the influence of interference factors, and the output is the riveting parameter adjustment amount. The input and output response is completed within an adjustment interval of 10ms.

[0038] For interference factors identified during the detection and identification phase, targeted adjustments are made. For example, if the material hardness is too high and the gap shrinks slowly, the riveting angle and pressure are increased; if the tooling wear is aggravated and the gap deviation is caused, the displacement of the riveting head is dynamically compensated; if the temperature fluctuation causes the gap to be abnormal, the holding time is finely adjusted. The entire adjustment process does not require manual intervention, and the adjustment interval is 10ms. This achieves real-time linkage between riveting parameters and gap changes and interference factors. The adjustment logic is adapted to the structural characteristics and load requirements of the rotating fastener, ensuring that the gap is always within a reasonable range during the riveting process.

[0039] The parameter adjustment range is limited as follows: the riveting pressure should not be adjusted by more than 10% of the current value in a single adjustment, and the cumulative adjustment should not exceed 30% of the initial value; the riveting head speed should not be adjusted by more than ±100 rpm in a single adjustment, and the feed speed should not be adjusted by more than ±0.1 mm / s in a single adjustment; the riveting angle should not be adjusted by more than ±2° in a single adjustment, and the holding time should not be adjusted by more than ±0.2 s in a single adjustment; to avoid sudden changes in parameters that may cause damage to fasteners or overload of equipment.

[0040] The minimum riveting pressure is 3MPa to prevent weak riveting, and the maximum is 30MPa to prevent equipment overload. The minimum riveting head speed is 600rpm to prevent material from hardening due to cold, and the maximum is 1800rpm to prevent excessive centrifugal force. The minimum feed speed is 0.1mm / s, and the maximum is 1.0mm / s. The holding time is no more than 2.0s to prevent excessive material deformation, and no less than 0.3s to ensure effective springback compensation. If the limits are exceeded, the system will automatically lock the adjustment function and trigger an alarm.

[0041] In the springback verification stage, when the real-time gap value detected in the detection and identification stage is close to the reference gap, i.e., the deviation is ≤0.015mm, the riveting equipment is controlled to reduce the riveting pressure and enter the unloading stage. By predicting the springback amount in the gap springback prediction sub-model during the model construction stage, and associating it with the elastic recovery characteristics of the material of the fastener to be riveted, the holding time of the riveting head (high pressure holding in the early stage and low pressure holding in the later stage) and the unloading speed (gradient unloading to avoid abnormal springback amount) are automatically adjusted to compensate for the springback of the gap. That is, the real-time gap is adjusted in advance to a preset value that is slightly smaller than the reference gap. The preset value = reference gap - predicted springback amount, ensuring that the gap springs back to the reference gap range after unloading. The gradient pressure holding parameters are set as follows: in the initial high-pressure stage, the pressure is 80%-90% of the current riveting pressure, and the holding time is 0.3-0.5s; in the later low-pressure stage, the pressure is 30%-50% of the current riveting pressure, and the holding time is 0.5-0.8s; the gradient unloading speed decreases in 3 levels, with an initial unloading speed of 0.2mm / s, decreasing by 0.05mm / s every 0.2s until complete unloading, ensuring smooth elastic recovery of the material.

[0042] After unloading, the final gap after riveting is checked a second time using a laser gap sensor. Combined with the final torque value detected by the torque sensor, a dual judgment of gap qualification is completed to determine whether the gap value is within the design allowable error range (±0.02mm). If it is not qualified, the process returns to the parameter adjustment stage. The riveting parameters are readjusted in conjunction with the springback amount and interference factors for a second fine-tuning of the riveting until the gap is qualified. If it is qualified, the entire process of riveting data of the fastener is recorded and entered into the associated database. The springback compensation mechanism of gradient pressure holding and gradient unloading realizes the prediction and compensation of springback amount, ensuring the final accuracy and consistency of the gap after riveting.

[0043] The acceptable torque range for low load level (≤10MPa) is 0.8-1.5N・m, for medium load level (10-20MPa) it is 1.5-3.0N・m, and for high load level (>20MPa) it is 3.0-5.0N・m. It is considered qualified only when the final clearance value is within ±0.02mm and the torque value is within the corresponding grade threshold range. If any indicator is not met, a secondary fine adjustment is triggered.

[0044] The learning and optimization phase is based on the full-process riveting data of the aforementioned six phases. It performs self-learning optimization of the model and parameters, and records the full-process riveting data of each rotary fastener, including fastener multi-dimensional parameters, load level, initial riveting parameters in the initial matching phase, real-time adjustment parameters in the parameter adjustment phase, gap deviation value, springback amount, type and influence of interference factors, final gap detection results, real-time data of tooling wear, and ambient temperature fluctuation data. Reinforcement learning algorithms are employed to analyze the correlation between fastener multi-dimensional parameters, riveting parameters, interference factors, clearance control effect, and load adaptability. The parameters of the dynamic prediction model, clearance springback prediction sub-model, and interference factor correlation sub-model in the model building phase are optimized. The parameter matching relationships in the fastener multi-dimensional parameters, reference clearance, and load level correlation database are corrected in the data acquisition and calibration phase. A riveting parameter optimization library is constructed, and based on riveting data of different models and load levels of rotary fasteners, a dedicated parameter adaptation scheme is formed to achieve adaptive parameter optimization for subsequent riveting operations, gradually improving clearance control accuracy and production efficiency. The optimization library's calling mechanism is as follows: priority is given to searching by dual keywords of fastener model and load level. When a perfectly matching solution is found, it is directly called. If no perfectly matching solution is found, the solution with the closest structural parameters (hole diameter, wall thickness) under the same load level is matched according to the load level priority principle. Then, based on the parameter difference between the current fastener and the matching solution, the riveting parameters are finely adjusted proportionally. For every 10% increase in the difference, the parameter fine-tuning range does not exceed 8%. When there is no solution with the same load level in the optimization library, a solution with an adjacent load level is called and corrected by combining dynamic prediction model to ensure the continuity and accuracy of parameter adaptation.

[0045] A tooling wear early warning module is added. Based on the recorded tooling wear data and the corresponding clearance interference, the timing of tooling replacement is predicted to avoid clearance deviation caused by excessive tooling wear. The adaptive control system of riveting assembly is improved to achieve upgraded and stable and reliable riveting clearance control.

[0046] The logic for determining tooling wear warning is as follows: ① Set the tooling wear baseline value. The initial wear amount is 0. When the cumulative wear amount reaches 0.1mm, a first-level warning is triggered to prompt attention; when it reaches 0.2mm, a second-level warning is triggered, and spot checks are recommended; when it reaches 0.3mm, a third-level warning is triggered, and the machine is forced to stop and the tooling is replaced. ② Correlation rule between wear amount and clearance interference amount: If the clearance interference amount increases by ≥0.008mm for every 0.05mm increase in wear amount, the warning level will be automatically upgraded by one level. ③ The early warning signal is output simultaneously through the system interface pop-up window and audible and visual alarm. At the same time, the tooling usage time and the number of processed fasteners are recorded at the time of the early warning, providing data support for the tooling maintenance cycle.

[0047] This invention also provides an adaptive control system for the riveting clearance of rotary fasteners, based on the above method, including: The data acquisition and calibration module is equipped with a laser rangefinder, hardness tester, vision recognition module and elastic modulus tester. It completes the acquisition of multi-dimensional parameters of the fastener to be riveted, the positioning and calibration of the riveting reference surface, the calibration of the reference riveting gap, the establishment of a database linking fastener multi-dimensional parameters, reference gap and load level, and the input of tooling initial wear data and ambient temperature reference value. The model prediction and initial matching module embeds an LSTM neural network and Monte Carlo method, runs a dynamic prediction model for riveting gap, a gap springback prediction sub-model and an interference factor correlation sub-model to complete gap and interference prediction; it links the database matching results to automatically adapt the initial riveting parameters, and uses a six-degree-of-freedom positioning mechanism to align the riveting head and complete sensor zero-point calibration and initial parameter pre-verification. The real-time detection and interference identification module relies on a collaborative detection system composed of displacement sensors, torque sensors, and laser gap sensors to achieve dual detection of gaps, both indirect and direct. It uses a Kalman filter algorithm to filter detection noise, identifies interference factors such as material hardness fluctuations and increased tooling wear in real time, outputs the type and magnitude of interference, and feeds it back to the parameter adjustment and springback verification module. The parameter adjustment and springback verification module adopts a PID adaptive adjustment algorithm, links detection data and prediction model, and fine-tunes riveting parameters in real time to achieve closed-loop control; it controls the riveting equipment to maintain pressure and unload gradient to complete springback compensation, and performs gap qualification judgment through laser gap sensor and torque sensor. If it fails to meet the qualification, it returns to readjustment. The self-learning optimization module uses reinforcement learning algorithms to record riveting data throughout the entire process, optimizes the matching relationship between the parameters of each prediction model and the database, builds a riveting parameter optimization library, and adds a tooling wear early warning function.

[0048] The interaction timing requirements for each module of the system are as follows: ① After the data acquisition and calibration module completes parameter acquisition, it needs to transmit the data to the model prediction and initial matching module within 100ms; ② The real-time detection and interference identification module feeds back detection data and interference information to the parameter adjustment and rebound verification module every 10ms. The parameter adjustment module needs to complete parameter calculation and output adjustment instructions within 5ms. ③ After the springback verification module completes the pass / fail determination, it will transmit the entire process data to the self-learning optimization module within 20ms. The self-learning optimization module will start model optimization once every 20 sets of data accumulated, and the single optimization time will not exceed 500ms to avoid affecting the continuity of subsequent riveting operations.

[0049] The system's fault self-diagnosis and handling mechanism is as follows: ① Sensor failure: When any sensor fails to provide data for 3 consecutive times or the data fluctuation is greater than ±0.02mm, it is determined to be a fault and automatically switches to the remaining sensor fusion detection mode. If the laser sensor fails, displacement and torque are combined for correction. If the displacement / torque sensor fails, the laser data is used as the primary data. ② Data transmission failure: If the data transmission interruption exceeds 20ms, the system will pause the riveting operation, save the current processing status, and restart from the parameter adjustment stage after the fault is recovered; ③ Model prediction failure: When the model prediction deviation is greater than 0.02mm for 5 consecutive times, the historical best parameters of the same type of fastener in the riveting parameter optimization library will be automatically called, and emergency iterative optimization will be triggered at the same time.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive control of the riveting clearance of a rotary fastener, characterized in that, The specific steps include the following: Data collection and calibration stage: The rotating fastener to be riveted is sent to the pre-processing station, and the multi-dimensional key parameters of the fastener are collected simultaneously by the detection equipment and the riveting positioning benchmark is calibrated; based on the load-bearing requirements of the fastener and the riveting process specifications, the benchmark riveting gap is calibrated, a multi-parameter correlation database is established, and the basic data related to tooling and environment are collected and entered into the database. Model building phase: Based on the collected basic data and combined with the cold rolling physical characteristics of rotary fastener riveting, a dynamic prediction model for riveting gap is built to predict gap changes in advance; a gap springback prediction sub-model is built to predict springback and optimize accuracy. Construct a sub-model for the correlation of interference factors to clarify the correspondence between various interference factors and the amount of gap interference. Initial matching phase: Combining database matching results, the initial parameters of the riveting are automatically matched with the actual parameters and load levels of the fasteners through various models, and the triple adaptation principle is followed; the initialization of the riveting equipment and sensor calibration are completed. Detection and identification phase: Start the riveting operation and simultaneously start the multi-sensor collaborative detection system to achieve multiple verification and filtering of gaps and interferences, and link the interference factor correlation sub-model to identify the main interference factors in real time and output relevant information; Parameter adjustment stage: Based on real-time gap data and interference information, combined with various prediction models, real-time adaptive adjustment of riveting parameters is performed; By comparing the deviation between the real-time clearance and the reference clearance, an adaptive adjustment algorithm is used to adjust the riveting parameters; corresponding adjustments are performed for various interference factors to achieve real-time linkage between parameters, clearance, and interference factors; Springback verification stage: When the real-time gap is close to the reference value, the springback amount predicted by the gap springback prediction sub-model is combined with the elastic recovery characteristics of the fastener material to perform springback compensation on the gap. After unloading, the gap qualification is double-checked by the testing equipment. If it fails, it returns to the parameter adjustment stage for readjustment. If it passes, the entire process data is recorded and entered into the database. Learning and optimization phase: Reinforcement learning algorithms are used to analyze the correlation between fastener multi-dimensional parameters, riveting parameters, interference factors, clearance control effect and load adaptability, optimize various model and database parameters, build a riveting parameter optimization library, and add tooling wear early warning function.

2. The adaptive control method for the riveting clearance of rotary fasteners according to claim 1, characterized in that, During the data acquisition and calibration phase, a multi-dimensional parameter synchronous acquisition system is used to acquire key parameters specific to the fasteners and the three-dimensional coordinates of the riveting reference surface. These key parameters include the fastener hole diameter form and position tolerance, wall thickness deviation, riveting surface roughness, material hardness, and elastic modulus fluctuation. The associated database links the fastener's multi-dimensional parameters, reference gap, and load level, and incorporates an automatic fastener model recognition function to achieve integrated adaptation of pre-detection of fasteners to be riveted, automatic calibration of reference gap, and load level matching.

3. The adaptive control method for the riveting clearance of rotary fasteners according to claim 2, characterized in that, In the model construction phase, the dynamic prediction model for riveting gap uses the hardness, elastic modulus fluctuation value, and wall thickness deviation of the fastener material as the core input variables, and the reference gap as the target variable. It introduces the correlation coefficient of riveting process parameters and the weight of load level, and trains and optimizes it through an adaptation algorithm to improve the convergence speed and prediction accuracy. The gap springback prediction sub-model combines the elastic recovery characteristics of the material and historical springback data to predict the springback amount, and incorporates the model correction term to optimize the accuracy. The interference factor correlation sub-model clarifies the correspondence between tooling wear, temperature fluctuation, material parameter deviation and clearance interference.

4. The adaptive control method for the riveting clearance of rotary fasteners according to claim 3, characterized in that, In the initial matching stage, the initial riveting parameters include the riveting head rotation speed, feed speed, riveting pressure, riveting angle, and initial pressure holding time. The triple adaptation principle is differential adaptation, load adaptation, and interference prediction adaptation. During equipment initialization, the riveting head is aligned with the riveting reference surface and the sensor zero point is calibrated. An initial parameter pre-verification step is added to verify the parameter adaptability through small-scale trial riveting. If the deviation exceeds the allowable range, the system returns to the initial matching stage to re-match the parameters.

5. The adaptive control method for the riveting clearance of a rotary fastener according to claim 4, characterized in that, In the detection and identification stage, multi-sensor collaborative detection achieves dual verification of indirect and direct detection, and the linkage interference factor correlation sub-model achieves interference detection, forming a triple verification; among them, the displacement sensor indirectly reflects the gap change, the torque sensor supplements the detection of the gap state and assists in judging interference, and the laser gap sensor directly detects the real-time gap value. The system employs a filtering algorithm to filter out detected noise, uses a linkage interference factor correlation sub-model to identify the main interference factors, outputs the interference signal strength, interference type and impact, and feeds it back to the parameter adjustment stage.

6. The adaptive control method for the riveting clearance of a rotary fastener according to claim 5, characterized in that, The parameter adjustment stage employs a PID adaptive adjustment algorithm to achieve closed-loop control of gap detection, deviation calculation, interference identification, and parameter adjustment. When the real-time gap is greater than the reference gap, the riveting pressure is increased, the feed speed is reduced, and relevant process parameters are fine-tuned to narrow the gap. When the real-time gap is less than the reference gap, the riveting pressure is reduced, the feed speed is increased, and the riveting angle is decreased. Targeted adjustments are performed for different interference factors, and the adjustment interval is adapted to the fastener structure characteristics and load requirements.

7. The adaptive control method for the riveting clearance of a rotary fastener according to claim 6, characterized in that, During the springback verification stage, when the real-time gap approaches the reference value, the riveting equipment reduces the pressure and enters the unloading stage. Springback compensation is achieved by adjusting the holding time and unloading speed, and the real-time gap is adjusted to the preset range in advance. The pass / fail judgment combines the final gap detection value and the torque detection value. If it fails, the springback amount and interference factors are adjusted for a second time. If it passes, the entire riveting process data is entered.

8. The adaptive control method for the riveting clearance of a rotary fastener according to claim 7, characterized in that, During the learning and optimization phase, the full-process riveting data includes fastener multi-dimensional parameters, load level, initial riveting parameters, real-time adjustment parameters, gap deviation value, springback amount, interference factor type and influence, detection results, tooling wear data, and ambient temperature fluctuation data. The matching relationship between each model parameter and the database is optimized through reinforcement learning algorithms. The riveting parameter optimization library contains dedicated parameter adaptation schemes for fasteners of different models and load levels.

9. A rotary fastener riveting clearance adaptive control system, based on the method of claim 8, characterized in that, include: The data acquisition and calibration module is equipped with various testing equipment and recognition modules to complete the acquisition of multi-dimensional parameters of fasteners, the positioning and calibration of riveting reference surfaces, the calibration of reference riveting gaps, the establishment of a related database and the input of tooling and environmental basic data. The model prediction and initial matching module runs the dynamic prediction model of riveting gap, the gap springback prediction sub-model and the interference factor correlation sub-model, and links the database to complete the initial parameter adaptation of riveting, realizing the riveting head alignment, equipment and sensor initialization and initial parameter pre-verification. The real-time detection and interference identification module relies on a multi-sensor collaborative detection system to achieve dual gap detection, uses a filtering algorithm to filter noise, identifies interference factors in real time, and outputs relevant information to feed back to the parameter adjustment stage. The parameter adjustment and springback verification module uses a PID adaptive adjustment algorithm to link detection data and prediction models to achieve real-time fine-tuning of riveting parameters; it controls the equipment to complete springback compensation through gradient pressure holding and gradient unloading, performs gap qualification judgment, and links related modules to complete the readjustment of unqualified parameters. The self-learning optimization module uses reinforcement learning algorithms to analyze the data throughout the entire process, optimize the matching relationship between the parameters of each prediction model and the database, build a riveting parameter optimization library, and realize tooling wear early warning.

Images