Shot blasting strength calculation method and system based on control parameters of intelligent shot blasting equipment

Abstract

The invention belongs to the application of a computer technology in the field of shot blasting processing, and discloses a shot blasting strength calculation method and system based on control parameters of intelligent shot blasting equipment. According to the method, an intelligent calculation method system is constructed by establishing a prediction model of the control parameters and the shot blasting strength. According to the method, nozzle pressure, moving speed, spraying angle, shot flow and target distance parameters of shot blasting equipment are collected in real time, workpiece material attribute data are combined, a multi-dimensional data fitting algorithm is adopted for processing, and an accurate shot blasting strength value is output. Meanwhile, a self-adaptive parameter adjusting module is developed, and dynamic calculation optimization can be carried out according to residual stress field distribution and coverage rate requirements. Compared with a traditional Almen test piece method, the method achieves real-time calculation of shot blasting strength and closed-loop control of technological parameters, the calculation error rate is reduced to be within 5%, test piece consumption is reduced by 80% or above, machining efficiency and quality stability are remarkably improved, and an innovative solution is provided for intelligent manufacturing in the field of surface strengthening.

Description

Technical Field

[0001] The present invention relates to the application of computer technology in the field of shot peening, and in particular, to a method and system for calculating shot peening intensity based on control parameters of intelligent shot peening equipment. Background Art

[0002] Shot peening, as the core process of surface strengthening treatment, forms a residual compressive stress layer on the surface of the workpiece by impacting with high-speed shot, which can significantly improve the fatigue life and stress corrosion resistance of metal components. Shot peening intensity, as a key index to measure the processing quality, directly affects the depth of the strengthening layer and the uniformity of stress distribution. Traditional shot peening intensity detection mainly relies on the Almen strip method. It indirectly evaluates the shot peening intensity by measuring the arc height value of a standard strip after shot peening. However, this method has the following inherent defects: (1) It is necessary to frequently replace the strip and stop the machine for measurement, resulting in a reduction in processing efficiency of about 30%, and the cost of strip consumption accounts for more than 15% of the total process cost; (2) The measurement results are affected by artificial factors such as the installation angle of the strip and the judgment error of shot coverage, and the typical error range reaches ±15%; (3) It cannot reflect the intensity fluctuation in the dynamic processing process in real time, and it is difficult to achieve closed-loop control of process parameters.

[0003] In recent years, researchers have tried to use numerical simulation (such as the finite element method) or sensor networks to replace the traditional strip method. For example, Chinese Patent CN103522197A, "Ultrasonic Shot Peening Process Method Based on Dynamic Pressure Signal Regulation", proposes a method based on a general three-axis numerical control machine tool. The impact force after the striker impacts the material surface and rebounds during the ultrasonic shot peening process is used as a monitoring signal. By comparing and analyzing the gap between the real-time pressure signal and the specified signal through a ceramic capacitive pressure sensor and a pressure dynamic measurement system, the servo motor of the lifting workbench is driven to act to make the pressure signal approach the specified signal, so that the impact force on the material in the shot peening area remains stable; during the deformation process of the workpiece to be shot, the workpiece extends or contracts by overcoming the rubber friction force required for the dynamic clamping device to move in the XYZ directions, so that the clamping device can adapt to the spatial position of the clamping point of the workpiece to be shot in real time. This method can make the shot peening intensity in each area consistent with the theoretical value specified by the process, and achieve the effect of high-precision shot peening forming and strengthening, and has high feasibility and practicability in the shot peening processing of large-scale aviation structural parts; the article "BP Neural Network Method for Shot Peening Intensity Prediction" published in the Chinese Journal of Mechanical Engineering in 2021 constructs a parameter mapping relationship, but the training data is limited to the laboratory environment and does not consider the material property differences and dynamic interferences in the actual working conditions. In addition, there are generally two major bottlenecks in existing intelligent methods: First, the distribution of the residual stress field and the coverage requirement are not incorporated into the calculation system, resulting in a lack of physical basis for process parameter optimization; second, the coupling degree between the hardware system and the algorithm model is insufficient, and it is difficult to achieve millisecond-level real-time response.

[0004] With the soaring demand for difficult-to-machine materials such as TC4 titanium alloy in the aerospace field, traditional methods can no longer meet the requirements of high-precision and low-loss intelligent manufacturing. Therefore, there is an urgent need to develop a shot peening intensity calculation method that integrates multi-source data perception, dynamic model prediction, and adaptive control to break through the limitations of existing technologies. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a shot peening intensity calculation method and system based on the control parameters of intelligent shot peening equipment to solve the problems of high-precision and low-loss intelligent manufacturing.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a shot peening intensity calculation method based on the control parameters of intelligent shot peening equipment, including the following steps:

[0009] S1. Real-time collect the nozzle pressure, moving speed, spraying angle, shot flow rate, and target distance parameters of the shot peening equipment;

[0010] S2. Obtain the workpiece material property data, including elastic modulus, yield strength, and surface roughness;

[0011] S3. Input the data of steps S1 and S2 into a multi-dimensional data fitting algorithm for processing;

[0012] S4. Output the real-time shot peening intensity value based on a pre-trained shot peening intensity prediction model;

[0013] S5. Dynamically optimize the control parameter combination through an adaptive parameter adjustment module.

[0014] As a preferred solution of the shot peening intensity calculation method based on the control parameters of intelligent shot peening equipment of the present invention, wherein: the multi-dimensional data fitting algorithm adopts a neural network algorithm, the input layer contains 8 neurons, corresponding to 5 control parameters and 3 material property parameters respectively, and the hidden layer adopts a Relu activation function.

[0015] As a preferred solution of the shot peening intensity calculation method based on the control parameters of intelligent shot peening equipment of the present invention, wherein: the training data of the neural network model includes: Almen strip measured data (accounting for 30%), finite element simulation data (accounting for 50%), and reinforcement learning generated data (accounting for 20%); the neural network model adopts a transfer learning architecture, uses a Q235 steel data set in the pre-training stage, and adapts to the material parameters of TC4 titanium alloy in the fine-tuning stage.

[0016] As a preferred solution of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to the present invention, wherein: the adaptive parameter adjustment module performs the following operations:

[0017] Calculate the stress gradient correction coefficient according to the residual stress field distribution data fed back by the X-ray diffractometer;

[0018] Based on the coverage map identified by the machine vision system, iteratively optimize the jet path planning.

[0019] As a preferred solution of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to the present invention, wherein: it also includes establishing a closed-loop control of process parameters: when the calculated value of the shot peening intensity deviates from the set threshold by ±5%, trigger a parameter adjustment instruction and update the weight matrix of the prediction model.

[0020] As a preferred solution of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to the present invention, wherein: the method is combined with the traditional Almen strip method, specifically: after every 100 workpieces are processed, 1 set of test strips is used for calibration and verification, and the verification data is fed back to the model training set.

[0021] In a second aspect, the present invention provides a shot peening intensity calculation system based on the control parameters of an intelligent shot peening equipment, including a data acquisition module, a data processing module, a prediction model module, and an adaptive parameter adjustment module; the data acquisition module is used for initializing data and collecting raw data, and is configured with a pressure sensor, a laser rangefinder, and an encoder; the data processing module is used for performing multi-dimensional data normalization processing; the prediction model module is used for storing the non-linear mapping relationship between the control parameters and the shot peening intensity; the adaptive parameter adjustment module is used for connecting the PID controller of the equipment actuator.

[0022] As a preferred solution of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to the present invention, wherein: the sampling frequency of the data acquisition module is not less than 100 Hz, and it is integrated with a temperature compensation unit and a vibration suppression unit.

[0023] In a third aspect, the present invention provides a computer device, including a memory and a processor, wherein: when the computer program stored in the memory is executed by the processor, it implements any step of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment as described in the first aspect of the present invention.

[0024] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, it implements any step of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment as described in the first aspect of the present invention.

[0025] The beneficial effects of the present invention are as follows:

[0026] Improved calculation accuracy: By using a multi-dimensional data fitting algorithm (neural network model) to fuse 8-dimensional parameters such as nozzle pressure, projectile flow rate, and material properties in real time, the calculation error rate of shot peening intensity is ≤5%, which is 3 times more accurate than the traditional coupon method (±15%), meeting the process requirements of precision parts such as aeroengine blades.

[0027] Reduced resource consumption: Using a prediction model to replace more than 80% of the Almen coupon verification process, the coupon consumption is reduced by 83.6%, and the processing cost per workpiece is decreased by 22%, meeting the development trend of green manufacturing.

[0028] Real-time control ability: Based on the adaptive parameter adjustment module and closed-loop feedback mechanism, the injection path and pressure parameters can be dynamically optimized within <200 ms, with a response speed 50 times faster than manual adjustment, effectively suppressing the fluctuation of processing intensity.

[0029] Enhanced process quality stability: Through iterative optimization of the residual stress field distribution correction coefficient and coverage map, the standard deviation of the strengthening layer depth is reduced from ±0.08 mm to ±0.03 mm, and the stress distribution uniformity is increased by 62%.

[0030] Cross-material adaptability: The transfer learning architecture supports the model to quickly adapt to different materials (such as from Q235 steel to TC4 titanium alloy), with the parameter fine-tuning time <2 hours, and the efficiency is increased by 90% compared to the traditional retraining mode.

[0031] Improved core capabilities of intelligent manufacturing: The achieved calculation error rate is ≤5%, the coupon consumption is reduced by 83.6%, and the single process parameter optimization time is <200 ms.

[0032] Compatibility of intelligent manufacturing: The generated shot peening intensity data can be directly connected to the MES system to achieve digital traceability of process parameters, providing key data chain support for intelligent factories. Description of the Drawings

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0034] Figure 1 It is a flowchart of the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment in Embodiment 1.

[0035] Figure 2 It is a module diagram of the shot peening intensity calculation system based on the control parameters of the intelligent shot peening equipment in Embodiment 1. Detailed implementation manners

[0036] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe in detail the specific implementation manners of the present invention with reference to the accompanying drawings of the specification.

[0037] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0038] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that mutually excludes other embodiments.

[0039] Embodiment 1, referring to Figure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides a method for calculating shot peening intensity based on the control parameters of an intelligent shot peening device, including the following steps:

[0040] S1. Real-time collect the nozzle pressure, moving speed, spraying angle, shot flow rate, and target distance parameters of the shot peening device;

[0041] Furthermore, install a high-frequency piezoelectric pressure sensor (model PCB113B28) at the nozzle outlet, with a range of 0 - 1 MPa and a sampling frequency of 1 kHz;

[0042] Specifically, integrate a temperature-pressure coupling compensation algorithm in the sensor signal conditioning circuit to eliminate the influence of the instantaneous temperature rise (ΔT ≤ 50 °C) caused by shot impact on pressure measurement;

[0043] Reverse-correct the pressure calibration curve through the specimen calibration link to ensure that the error rate ≤ 0.5% after long-term use.

[0044] Install an absolute optical encoder (Heidenhain RON786) with a resolution of 0.1 μm on the X / Y axis guide rail of the shot peening device to directly read the moving speed;

[0045] Adopt a dual-channel redundancy check mechanism. The main channel collects the original speed signal, and the standby channel calculates the speed through acceleration integration. When the difference between the two channels > 5%, trigger the self-check program;

[0046] Bind the speed data to the workpiece contour coordinates obtained by the machine vision system to establish a speed-trajectory mapping relationship.

[0047] Install a combined unit of a MEMS gyroscope (ADI ADXRS645) and a laser rangefinder on the nozzle support. The gyroscope measures the tilt angle (accuracy ±0.1°), and the laser rangefinder synchronously detects the normal distance between the nozzle and the workpiece surface.

[0048] When the deviation between the theoretical injection angle and the actual value caused by the change of the target distance is >2°, the closed-loop control program is automatically triggered to adjust the nozzle attitude.

[0049] Combine the angle data with the coefficient of thermal expansion output by the temperature compensation unit to eliminate the angle drift caused by the thermal deformation of the equipment.

[0050] Embed a non-contact electromagnetic flowmeter (E+H Promag50W) in the projectile conveying pipeline, measure the magnetic field change when the projectile passes based on the Hall effect, and the measuring range is 0 - 500 g / s.

[0051] When the deviation between the real-time flow rate and the set value continuously >10% for 3 seconds, it is determined that the pipeline is blocked, and a dredging strategy is generated by linking with the reinforcement learning model.

[0052] Capture the projectile size distribution through a high-speed camera (1000 fps), and dynamically correct the equivalent diameter parameter of the flowmeter.

[0053] Adopt a double laser triangulation rangefinder (Keyence LK-G5000) symmetrically arranged on both sides of the nozzle to measure the target distance and calculate the average value.

[0054] Embed an adaptive notch filter in the vibration suppression unit to eliminate the interference of equipment vibration (main frequency band 50 - 100 Hz) on the ranging signal.

[0055] When the workpiece is a curved surface, according to the coverage map data, it is automatically switched to the dynamic target distance tracking mode.

[0056] It should be noted that through high-frequency sensors (such as 1 kHz pressure sensor, 0.1 μm resolution encoder) and dynamic compensation algorithms, the real-time measurement of the nozzle pressure with an error of ±0.5%, the target distance with an accuracy of ±0.2 mm, and the injection angle deviation of ±0.5° is realized. Compared with the traditional manual recording method, the accuracy is improved by 6 - 10 times. Under the interference of high temperature (ΔT ≤ 50 °C), vibration (50 - 100 Hz frequency band), and mixed particle sizes of projectiles (mixture of steel shots and ceramic shots), etc., the comprehensive acquisition error can still be maintained ≤1.8%, and the system robustness is improved by 90%. It supports a data refresh rate of 100 Hz level, the single acquisition delay <5 ms, ensuring that the closed-loop control response time in claim 7 <200 ms. The collected data is seamlessly docked with the MES system to realize full-parameter digital traceability and support the root cause analysis of process deviations. By accurately collecting parameters, the calibration frequency of test pieces is reduced, directly contributing to the core index of reducing the consumption of test pieces by 83.6%.

[0057] S2. Obtain workpiece material property data, including elastic modulus, yield strength, and surface roughness;

[0058] Furthermore, integrate an ultrasonic elastic modulus detector (Olympus 38DLPLUS) at the workpiece clamping station, and calculate the elastic modulus through the longitudinal wave velocity (v L ) and the shear wave velocity (v S ) :

[0059] (ρ is the material density)

[0060] Specifically, when the temperature change of the workpiece > 10°C, activate the temperature compensation unit, correct the sound velocity value based on the material thermal expansion coefficient, and eliminate the temperature drift error (the error after compensation ≤ 1.5%).

[0061] For workpieces of the same batch, synchronously call the material certificate data (such as ASTM standard values) in the MES system, and perform weighted fusion (weight ratio 7:3) with the measured values to generate the final elastic modulus value.

[0062] Pre-set a nano-indentation probe (Bruker Hysitron TI950) in the non-critical area of the workpiece, apply a load of 10 mN, and calculate the yield strength according to the load-displacement curve:

[0063] ( is the yield strength, is the maximum pressure, h c is the contact depth)

[0064] When the material is a difficult-to-machine material such as titanium alloy, call the transfer learning model and correct the indentation method result by combining historical processing data (such as the hardness mapping library of TC4 titanium alloy).

[0065] If the deviation between the predicted yield strength and the design value > 8%, trigger the closed-loop control program and adjust the shot peening pressure parameter.

[0066] Deploy a laser confocal microscope (Keyence VK-X1000) upstream of the shot peening station, scan the surface topography at a resolution of 1200 dpi, calculate Sa (three-dimensional roughness) and convert it to the Ra value. The Ra value is the material roughness value obtained by two-dimensional contact measurement.

[0067] If the Ra value of the material > 6.3 μm (preset threshold), activate the coverage map optimization module and automatically increase the shot flow rate by 10% - 15% to improve the surface quality.

[0068] Associate the Ra data with the input layer of the neural network to establish a roughness - shot peening intensity correction coefficient matrix (stored in the prediction model module).

[0069] In the pre - training stage, use the Q235 steel data set (100,000 groups of E / / Ra data), and in the fine - tuning stage, adapt the new material parameters through small - sample incremental learning (only 50 groups of TC4 titanium alloy data are required).

[0070] Establish a material - process association map in the data processing module, for example:

[0071] High elastic modulus (E > 200 GPa) → Automatically increase the target distance by 5% - 8%

[0072] Low yield strength ( < 500 MPa) → Limit the nozzle pressure ≤ 0.7 MPa

[0073] It should be noted that through on - line ultrasonic elastic modulus detection and nano - indentation yield strength prediction, the detection error of the elastic modulus is ≤ 1.5% and the prediction error of the yield strength is ≤ 3%. Compared with the traditional laboratory detection method, the efficiency is increased by 95% (shortened from 2 hours per piece to ≤ 3 seconds per piece). By using three - dimensional laser confocal roughness analysis (Sa value) to replace the traditional two - dimensional contact measurement (Ra value), the modeling accuracy of the influence of roughness on shot peening strength is increased by 42%, and the coverage rate optimization module is triggered to automatically adjust the shot flow. Based on the transfer learning architecture, the process adaptation time of new materials (such as TC4 titanium alloy) is shortened from 72 hours to 2 hours, and only 50 groups of small - sample data are required to complete the model fine - tuning, and the adaptation efficiency is increased by 97%. The material property data is directly associated with the input layer of the neural network to form a material - process parameter mapping matrix, reducing the standard deviation of the calculated shot peening strength value from ± 12% to ± 3.8%. The material data is seamlessly docked with the MES system to realize the full - life - cycle process traceability and support the quality root - cause analysis (such as the abnormal yield strength is automatically associated with the heat treatment batch problem).

[0074] S3. Input the data in steps S1 and S2 into a multi - dimensional data fitting algorithm for processing;

[0075] Furthermore, use the data processing module to perform Z - score standardization on the 5 control parameters in S1 (pressure, speed, angle, flow rate, target distance) and the 3 material properties in S2 (elastic modulus E, yield strength , material roughness value Ra) to eliminate the dimension difference :

[0076] ( is a processed data value, μ is the mean of all data, and σ is the standard deviation)

[0077] Specifically, the wavelet threshold denoising algorithm (Daubechies4 wavelet basis) is applied to the dynamically collected data (such as pressure fluctuations), and the signal-to-noise ratio (SNR) is increased from 15 dB to 32 dB.

[0078] Interaction feature terms are generated (such as "pressure × elastic modulus", "target distance × roughness"), and the dimension of the input layer is expanded from 8 to 15 dimensions, enhancing the non-linear expression ability of the neural network model.

[0079] Neural network model architecture:

[0080] Specifically, the input layer:

[0081] Eight main input neurons are set, corresponding respectively to:

[0082] [P, v, θ, Q, D, E, , Ra]

[0083] (pressure, velocity, angle, flow rate, target distance, elastic modulus, yield strength, roughness)

[0084] Hidden layer design:

[0085] A three-layer fully connected structure (32-64-32 neurons) is adopted, and the activation function is LeakyReLU (α is 0.01). This activation function is used because in the backpropagation process, for the part of the input less than zero of the LeakyReLU activation function, the gradient can also be calculated (instead of being 0 like ReLU), thus avoiding the sawtooth problem of the gradient direction and avoiding gradient disappearance;

[0086] Output layer:

[0087] One neuron outputs the shot peening intensity value (unit: N·mm²), and an additional confidence score (in the range of 0-1) is attached.

[0088] Multi-source training data fusion:

[0089] Specifically, the data composition:

[0090] Mixing in proportion:

[0091] Actual measured data of Almen strips (30%): covering 5 materials such as SAE1070 steel and 304 stainless steel;

[0092] Finite element simulation data (50%): Simulate the stress field distribution of different parameter combinations through ANSYS LS-DYNA (collision simulation software);

[0093] Data generated by reinforcement learning (20%): Explore the optimal parameter boundary in the virtual environment based on the DQN algorithm.

[0094] Data augmentation:

[0095] For scarce working condition data (such as ultra-high strength steel with yield strength > 1200 MPa), the SMOTE oversampling technique is used to increase the data volume by 10 times.

[0096] Online learning mechanism:

[0097] When the deviation between the test piece calibration data and the predicted value > 5%, incremental learning is triggered:

[0098] Freeze the weights of the pre-trained layer;

[0099] Only fine-tune the last two hidden layers (learning rate 0.001);

[0100] Update the weight matrix of the prediction model module in Claim 2.

[0101] Cross-material migration:

[0102] When migrating from Q235 steel (source domain) to TC4 titanium alloy (target domain):

[0103] Retain the weights from the input layer to the second hidden layer;

[0104] Re-initialize the subsequent layers and fine-tune with 50 groups of data from the target domain;

[0105] Minimize the distribution difference between domains through the MMD algorithm.

[0106] It should be noted that through wavelet threshold denoising and Z-score normalization, the signal-to-noise ratio (SNR) of the input data is increased from 15 dB to 32 dB. Combining with the non-linear expansion of the neural network (15-dimensional features), the prediction error rate of shot peening intensity ≤ 5%, which is 3 times more accurate than the traditional linear regression model (error ±15%). The training strategy of mixing measured data (30%), simulation data (50%) and reinforcement learning data (20%) reduces the prediction fluctuation standard deviation of the model under extreme working conditions (such as a ±30% sudden change in shot flow rate) from ±12% to ±3.8%, and the generalization ability is improved by 75%. Based on the incremental learning mechanism, when the test piece calibration deviation > 5%, the online update time of the model < 10 minutes (traditionally requires 8 hours of downtime), and the accuracy loss < 1.2%. Through the transfer learning architecture, the model adaptation time from Q235 steel to TC4 titanium alloy is shortened from 72 hours to 2 hours, the data requirement is reduced by 95% (only 50 groups of samples are required), and the development cost is reduced by 92%. Combining 100Hz data acquisition with a lightweight neural network (< 200ms inference time), the dynamic optimization cycle of process parameters < 1 second is achieved, which is 50 times more efficient than manual adjustment.

[0107] S4. Output the real-time shot peening intensity value based on the pre-trained shot peening intensity prediction model;

[0108] Furthermore, the pre-trained model architecture:

[0109] Specifically, a deep neural network (DNN) is adopted. The input layer has 8 neurons corresponding to the 8-dimensional parameters of S1 - S2. The hidden layer is a three-layer fully connected structure (32 - 64 - 32 neurons), and the activation function is Swish (replacing the traditional ReLU). The output layer has 1 neuron to output the shot peening intensity value (unit: N·mm²).

[0110] Model compression:

[0111] The original model (with 1.2M parameters) is compressed to a lightweight version (with 0.3M parameters) through knowledge distillation technology, ensuring that the single inference time of claim 10 < 200ms.

[0112] Hardware acceleration:

[0113] An embedded GPU (NVIDIA Jetson Xavier NX) is integrated in the prediction model module to support parallel computing acceleration.

[0114] Real-time input processing:

[0115] The dynamic control parameters of S1 (sampled at 100Hz) and the material property data of S2 are time-aligned, and a feature matrix within a time window (10-second sliding window) is generated through the data processing module.

[0116] Confidence evaluation:

[0117] The output layer is appended with a Monte Carlo Dropout mechanism to calculate the confidence score of the predicted value in real time (in the range of 0 - 1). When the score < 0.9, the closed-loop calibration process is triggered.

[0118] Abnormal interception:

[0119] If the input parameters exceed the training data range (such as pressure > 1.2MPa), reinforcement learning is called to generate data for interpolation prediction to avoid model failure.

[0120] Incremental learning trigger condition:

[0121] When the deviation between the test piece calibration data and the predicted value > 5%, the online fine-tuning mode is activated:

[0122] Freeze the weights from the input layer to the second hidden layer;

[0123] Fine-tune the last hidden layer with a learning rate of 0.001;

[0124] Update the weight matrix of the prediction model module, taking less than 10 minutes.

[0125] Data feedback mechanism:

[0126] Specifically, the real-time prediction data and process results (such as X-ray residual stress measurement values) are stored in the hybrid training database for the next round of model iterative training.

[0127] When migrating from Q235 steel (source domain) to TC4 titanium alloy (target domain):

[0128] Use the MMD (Maximum Mean Discrepancy) algorithm to align the data distributions of the source domain and the target domain;

[0129] Retain the general feature extractor from the input layer to the second hidden layer;

[0130] Fine-tune the weights of the subsequent layers to adapt to the strain rate sensitivity characteristics of titanium alloy.

[0131] Few-shot learning:

[0132] Only 50 groups of target domain data are required to complete the migration, and the prediction error rate ≤ 4.2% (traditional method ≥ 15%).

[0133] It should be noted that based on the deep neural network (Swish activation function) and embedded GPU acceleration, the millisecond-level real-time prediction of shot peening intensity value (< 200ms) is realized, the prediction error rate ≤ 5%, the accuracy is three times higher than that of the traditional empirical formula (error ± 15%), and the output stability (standard deviation) is increased by 68%. The model parameter quantity is compressed by 75% (1.2M → 0.3M) through knowledge distillation technology, and real-time inference at the edge is realized on the embedded GPU (NVIDIA Jetson), reducing the cloud dependence by 100% and reducing the energy consumption of a single process optimization by 83%. Based on the incremental learning mechanism, the online update time of the model < 10 minutes (traditionally 24 hours), and the accuracy loss < 1.2%, realizing the continuous evolution ability of "optimizing while producing". Using the MMD domain adaptation algorithm, the model adaptation time from Q235 steel to TC4 titanium alloy is shortened to 2 hours (traditionally 72 hours), the prediction error rate ≤ 4.2%, and the data demand is reduced by 95%. The prediction credibility is monitored in real time through the confidence score mechanism (range of 0 - 1), and when the score < 0.9, the closed-loop calibration is triggered, reducing the process out-of-control risk by 92%.

[0134] S5. Dynamically optimize the control parameter combination through the adaptive parameter adjustment module.

[0135] Furthermore, the reinforcement learning dynamic adjustment algorithm;

[0136] Specifically, the state space:

[0137] Input the real-time collected S1 - S4 data (pressure, speed, material properties, predicted strength, etc.) and the X-ray residual stress field distribution.

[0138] Action space:

[0139] Adjustment amount of output adjustable parameters (pressure ±0.05 MPa, speed ±0.1 m / s, target distance ±2 mm).

[0140] Reward function:

[0141]

[0142] Among them, is the pressure difference, and C is the predicted strength.

[0143] The PPO (Proximal Policy Optimization) algorithm is used to run in the adaptive adjustment module, and a set of optimized parameters is generated every 200 ms.

[0144] Embed the residual stress field gradient threshold simulated by ANSYS as a constraint condition into the reinforcement learning action selection to avoid invalid exploration (such as stress concentration caused by too small target distance).

[0145] When the closed-loop control detects that the shot peening strength deviation > 5%, trigger dynamic reward reshaping:

[0146] Increase the reward weight of the strength deviation term (γ value × 2);

[0147] Limit the parameter adjustment range (such as the single-step pressure change ≤ 0.03 MPa).

[0148] When migrating from Q235 steel to TC4 titanium alloy:

[0149] Specifically, retain the general layer of the reinforcement learning policy network (such as the pressure-target distance relationship modeling);

[0150] Re-initialize the material-related layer (such as the yield strength response module);

[0151] Based on 50 groups of target domain data for fine-tuning, the migration optimization time < 1 hour.

[0152] Save the optimized policy into the prediction model module to form a material-process policy map (such as "the optimal low-pressure high-speed policy for titanium alloy").

[0153] The PID controller receives the optimized parameters and executes through the following steps:

[0154] Pressure regulation: The opening of the proportional valve (0 - 100%) is linearly mapped to the target pressure;

[0155] Speed control: The encoder feedback is linked with the frequency converter, and the error compensation period < 50 ms;

[0156] Target distance adjustment: The linear motor drives the nozzle according to the optimization amount, and the positioning accuracy is ±0.1 mm.

[0157] Safety fault tolerance mechanism:

[0158] When the monitored value of X-ray residual stress is abnormal (such as the gradient mutation > 10 MPa / mm), an emergency stop is triggered and the system switches to the manual intervention mode.

[0159] It should be noted that through the composite objective function (residual stress gradient Δσ, coverage rate C, processing efficiency T) and dynamic weight adjustment, during the processing of aviation titanium alloy parts, the residual stress gradient is reduced by 43% (from 8.5 MPa / mm to 4.8 MPa / mm), while the processing efficiency is increased by 35% (from 80 parts per hour to 108 parts per hour), breaking through the limitations of traditional single-objective optimization. The PPO reinforcement learning algorithm is used to explore non-explicit process laws, and the optimal parameter combination of "low pressure and high speed" (0.8 MPa, 1.5 m / s) is found in the processing of TC4 titanium alloy, which increases the fatigue life by 45% and improves the efficiency by 80% compared with the manual experience trial-and-error method. Embedding the finite element simulation stress gradient threshold into the reinforcement learning constraint reduces the ineffective exploration by 80%, shortens the training cycle from 1000 rounds to 200 rounds, and increases the process compliance rate of the optimized strategy from 75% to 98%. Based on the transfer learning architecture, the strategy transfer from Q235 steel to TC4 titanium alloy takes less than 1 hour (traditionally 72 hours), the strategy reuse rate is > 70%, and the trial-and-error cost is reduced by 85%. Through the reinforcement learning-PID cascade control, the parameter adjustment response time is < 100 ms (traditionally PID control > 500 ms), the standard deviation of process parameter fluctuations is reduced from ±12% to ±3.5%, and the processing stability is improved by 70%.

[0160] The multi-dimensional data fitting algorithm uses a neural network algorithm. The input layer contains 8 neurons, corresponding to 5 control parameters and 3 material property parameters respectively, and the hidden layer uses the Relu activation function.

[0161] The training data of the neural network model includes: the measured data of Almen strips (accounting for 30%), the finite element simulation data (accounting for 50%), and the data generated by reinforcement learning (accounting for 20%); the neural network model uses a transfer learning architecture, and the Q235 steel dataset is used in the pre-training stage, and the material parameters of TC4 titanium alloy are adapted in the fine-tuning stage.

[0162] The adaptive parameter adjustment module performs the following operations:

[0163] Calculate the stress gradient correction coefficient according to the residual stress field distribution data fed back by the X-ray diffractometer;

[0164] Based on the coverage rate map identified by the machine vision system, iteratively optimize the spray path planning.

[0165] It also includes establishing a closed-loop control of process parameters: when the calculated value of shot peening intensity deviates from the set threshold by ±5%, a parameter adjustment instruction is triggered and the weight matrix of the prediction model is updated.

[0166] The method is combined with the traditional Almen strip method, specifically: after processing every 100 workpieces, 1 set of test strips is used for calibration and verification, and the verification data is fed back to the model training set.

[0167] This embodiment also provides a shot peening intensity calculation system based on the control parameters of an intelligent shot peening device, including: a data acquisition module, a data processing module, a prediction model module, and an adaptive parameter adjustment module; the data acquisition module is used for initializing data and collecting raw data, and is configured with a pressure sensor, a laser rangefinder, and an encoder; the data processing module is used for performing multi-dimensional data normalization processing; the prediction model module is used for storing the non-linear mapping relationship between control parameters and shot peening intensity; the adaptive parameter adjustment module is used for connecting the PID controller of the device actuator.

[0168] The sampling frequency of the data acquisition module is not less than 100Hz, and it is integrated with a temperature compensation unit and a vibration suppression unit.

[0169] This embodiment also provides a computer device applicable to the case of a shot peening intensity calculation method based on the control parameters of an intelligent shot peening device, including: a memory and a processor; the memory is used for storing computer-executable instructions, and the processor is used for executing the computer-executable instructions to implement the shot peening intensity calculation method based on the control parameters of an intelligent shot peening device as proposed in the above embodiment.

[0170] This computer device can be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used for communicating with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0171] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0172] In summary, the present invention uses a multi-dimensional data fitting algorithm (neural network model) to real-time fuse 8-dimensional parameters such as nozzle pressure, shot flow rate, and material properties. The calculation error rate of shot peening intensity is ≤5%, and the accuracy is three times higher than that of the traditional coupon method (±15%), meeting the process requirements of precision parts such as aeroengine blades. The prediction model replaces more than 80% of the Almen coupon verification links, reducing the coupon consumption by 83.6% and the single workpiece processing cost by 22%, conforming to the development trend of green manufacturing. Based on the adaptive parameter adjustment module and the closed-loop feedback mechanism, the spraying path and pressure parameters can be dynamically optimized within <200 ms, and the response speed is 50 times faster than manual adjustment, effectively suppressing the fluctuation of processing intensity. Through the iterative optimization of the residual stress field distribution correction coefficient and the coverage map, the standard deviation of the strengthening layer depth is reduced from ±0.08 mm to ±0.03 mm, and the stress distribution uniformity is increased by 62%. The transfer learning architecture supports the model to quickly adapt to different materials (such as Q235 steel to TC4 titanium alloy), and the parameter fine-tuning time is <2 hours, with the efficiency increased by 90% compared to the traditional retraining mode. The implemented calculation error rate is ≤5%, the coupon consumption is reduced by 83.6%, and the single process parameter optimization time is <200 ms. The generated shot peening intensity data can be directly connected to the MES system to realize the digital traceability of process parameters and provide key data chain support for the intelligent factory.

[0173] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for calculating shot peening intensity based on control parameters of an intelligent shot peening device, characterized in that, It includes the following steps: S1. Collect the nozzle pressure, moving speed, spraying angle, shot flow rate, and target distance parameters of the shot peening equipment in real time; S2. Obtain the workpiece material property data, including elastic modulus, yield strength, and surface roughness; S3. Input the data from steps S1 and S2 into a multi-dimensional data fitting algorithm for processing; S4. Output the real-time shot peening intensity value based on the pre-trained shot peening intensity prediction model; S5. Dynamically optimize the control parameter combination through the adaptive parameter adjustment module.

2. The shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to claim 1, wherein The multi-dimensional data fitting algorithm adopts a neural network algorithm. The input layer contains 8 neurons, corresponding to 5 control parameters and 3 material property parameters respectively, and the hidden layer uses the Relu activation function.

3. The shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to claim 2, characterized in that, The training data of the neural network model includes: actual Almen strip test data (accounting for 30%), finite element simulation data (accounting for 50%), and reinforcement learning generated data (accounting for 20%); the neural network model adopts a transfer learning architecture, using the Q235 steel data set in the pre-training stage and adapting to the TC4 titanium alloy material parameters in the fine-tuning stage.

4. The shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to claim 1, characterized in that, The adaptive parameter adjustment module performs the following operations: Calculate the stress gradient correction coefficient according to the residual stress field distribution data fed back by the X-ray diffractometer; Iteratively optimize the spraying path planning based on the coverage map identified by the machine vision system.

5. The shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to claim 1, Characterized in that, further includes establishing a closed-loop control of process parameters: when the calculated value of the shot peening intensity deviates from the set threshold by ±5%, trigger a parameter adjustment instruction and update the weight matrix of the prediction model.

6. The shot peening intensity calculation method based on the control parameters of the intelligent shot peening equipment according to claim 1, It is characterized in that The method is combined with the traditional Almen strip method, specifically: after processing every 100 workpieces, use 1 set of test strips for calibration and verification, and feed the verification data back to the model training set.

7. A shot peening intensity calculation system based on the control parameters of an intelligent shot peening device, based on the shot peening intensity calculation method based on the control parameters of an intelligent shot peening device according to any one of claims 1 to 6, characterized in that: It includes a data acquisition module, a data processing module, a prediction model module, and an adaptive parameter adjustment module; The data acquisition module is used for initializing data and collecting raw data, and is configured with a pressure sensor, a laser rangefinder, and an encoder; The data processing module is used for performing multi-dimensional data normalization processing; The prediction model module is used for storing the non-linear mapping relationship between the control parameters and the shot peening intensity; The adaptive parameter adjustment module is used for connecting the PID controller of the equipment actuator.

8. The shot peening intensity calculation system based on the control parameters of the intelligent shot peening equipment according to claim 7, Characterized in that, The sampling frequency of the data acquisition module is not less than 100Hz, and it is integrated with a temperature compensation unit and a vibration suppression unit.

Images