An optimized method, system, and storage medium for superplastic forming of titanium alloys.

Abstract

This invention discloses an optimization method, system, and storage medium for the superplastic forming of titanium alloys, belonging to the technical field of titanium alloy hot working. It involves collecting forming data from the titanium alloy superplastic forming process and preprocessing it to obtain a standardized forming dataset. Based on this dataset, a forming condition prediction model is used to predict the future working conditions of each region of the formed part. According to the deviation between the future working conditions and the target working conditions of each region, adaptive tuning parameters are determined for each region. These adaptive tuning parameters are then input into a multi-objective particle swarm optimization model for global process parameter optimization, resulting in the optimal parameter combination for the titanium alloy superplastic forming process. The control curves for each forming stage are dynamically adjusted based on the optimal parameter combination. This invention achieves coordinated control of local and macroscopic processes, effectively improving the quality of titanium alloy superplastic forming, significantly increasing the wall thickness uniformity and yield of the formed parts, and reducing process energy consumption, demonstrating good practicality.

Description

Technical Field

[0001] This invention belongs to the technical field of titanium alloy hot working, specifically relating to an optimized method, system and storage medium for titanium alloy superplastic forming. Background Technology

[0002] Titanium alloys are lightweight, high-strength, and corrosion-resistant metallic materials. Their density is approximately 60% that of steel, but their specific strength is far higher than that of steel and aluminum alloys. They exhibit good mechanical properties at both room and high temperatures and demonstrate high stability in various corrosive media. Due to these excellent properties, titanium alloys are widely used in aerospace, energy, shipbuilding, and chemical industries, playing a particularly important role in aerospace equipment. In the aerospace field, titanium alloys are mainly used to manufacture high-performance components such as compressor blades, engine casings, and wing skins. These components often have thin walls, complex curved surfaces, large dimensions, and high precision, placing extremely high demands on manufacturing processes. Furthermore, superplastic forming technology is an advanced manufacturing method that utilizes the superplasticity (elongation exceeding 200%) exhibited by metallic materials under specific high-temperature and strain rate conditions for forming. This technology can obtain integrated components with complex shapes, uniform wall thickness, and high precision in a single forming process, reducing subsequent processes such as welding and machining, significantly shortening the production cycle and reducing manufacturing costs. For high-performance materials such as titanium alloys, superplastic forming is particularly suitable for manufacturing thin-walled structural parts such as aero-engine casings, air intakes, and fairings. It can effectively improve material utilization and enhance the service performance of parts.

[0003] The superplastic forming process of titanium alloys is highly sensitive to forming parameters, which directly affect the material's flow behavior, strain distribution, wall thickness uniformity, and the final properties of the formed part. These forming parameters mainly include forming temperature, gas pressure loading curve, strain rate, and die heating uniformity. Currently, the setting of titanium alloy forming parameters relies primarily on engineers' experience or limited process experiments, a method with the following shortcomings:

[0004] (1) Poor parameter adaptability: It is impossible to achieve fast and accurate parameter matching for different parts shapes, sizes and wall thickness distributions, which can easily lead to local overstretching or insufficient filling, resulting in defects such as cracks and wrinkles.

[0005] (2) Lack of real-time feedback adjustment: During the forming process, the air pressure curve is generally executed according to the preset value. It cannot be dynamically adjusted according to the real-time flow state of the material, the local stress and strain distribution and the mold filling situation, resulting in low forming efficiency and unstable quality.

[0006] (3) Reliance on repeated manual trials: Traditional parameter optimization processes often require a large number of trial moldings and tests, which are costly, time-consuming, and highly dependent on the experience of operators, making it difficult to ensure consistency in mass production;

[0007] (4) Insufficient data utilization: Existing methods do not make full use of data such as temperature, air pressure, displacement, and strain collected by sensors for intelligent analysis, and lack a closed-loop optimization mechanism, resulting in low efficiency and poor accuracy of parameter optimization. Summary of the Invention

[0008] The purpose of this invention is to provide an optimized method, system, and storage medium for superplastic forming of titanium alloys, in order to solve the aforementioned problems.

[0009] This invention is mainly achieved through the following technical solutions:

[0010] An optimized method for superplastic forming of titanium alloys includes the following steps:

[0011] Step S1: Collect forming data of the titanium alloy superplastic forming process and preprocess it to obtain a standardized forming dataset; the forming data includes mold temperature, gas pressure loading curve, displacement, strain rate and wall thickness distribution;

[0012] Step S2: Based on the forming dataset, a forming condition prediction model based on a neural network model is used to predict the future working conditions of each region of the formed part; according to the deviation between the future working conditions and the target working conditions of each region, the corresponding adaptive optimization parameters for each region are determined; the future working conditions include the filling state and wall thickness distribution of the forming process, and the adaptive optimization parameters include cavity / local effective pressure, mold / sheet material local temperature / power and air pressure loading rate;

[0013] Step S3: The adaptive tuning parameters of each region are used as candidate variables and input into the multi-objective particle swarm optimization model to optimize the global process parameters and obtain the optimal parameter combination for the titanium alloy superplastic forming process. The optimization objectives include minimizing the standard deviation of wall thickness, maximizing the filling rate, and minimizing forming time and energy consumption. The constraints include upper / lower temperature limits, maximum allowable strain rate, and mold bearing limit.

[0014] Step S4: Dynamically adjust the control curves of temperature, air pressure / load and loading rate for each forming stage according to the optimal parameter combination.

[0015] Preferably, the method further includes step S5: real-time monitoring of the wall thickness distribution of the molded part and the mold filling rate; when local thinning or insufficient filling is detected, the process proceeds to step S1.

[0016] To better realize the present invention, step S2 further includes the following steps:

[0017] Step S21: Based on the standardized forming dataset, the random forest algorithm is used to extract the top N key features that affect the wall thickness distribution and forming quality;

[0018] Step S22: Input the N key features extracted after time series conversion into the forming condition prediction model, and output the future conditions of each region of the formed part;

[0019] Step S23: Dynamically output the corresponding adaptive tuning parameters according to the deviation between the future condition and the target condition.

[0020] To better implement the present invention, further, in the step S2, the deviation between the future condition and the target condition of the r-th region is:

[0021] ;

[0022] If ∣E

[0031] , ε,r ∣ < T1, then increase the air pressure loading rate by 5% or increase the local heating power of the mold by 2%;

[0023] If T1 ≤ ∣E r ∣ ≤ T2, then maintain the current parameters and enter the observation stage;

[0024] If ∣E r ∣ > T2, then reduce the air pressure loading rate by 10% or increase the local heating power of the mold by 5%;

[0025] Where: E r is the deviation between the future condition and the target condition of the r-th region;

[0026] w ε , [[ID=�7]] w t , w F are the calculated weights respectively;

[0027] e ε,r is the strain rate deviation of the r-th region;

[0028] e t,r is the wall thickness deviation of the r-th region;

[0029] e <o000010>is the filling rate deviation of the r-th region;

[0030] T1 and T2 are the preset thresholds respectively, and T2 > T1.

[0031] [[ID=¿3]]To better implement the present invention, further, in the step S2, respectively for e ε,r , e t,r and [[ID=¿0]] e F,rPerform global correction, and the correction formula is:

[0032] ;

[0033] in: E global for e ε,r , e t,r and e F,r Global deviation of any one of them;

[0034] e r for e F,r , e t,r and e ε,r Any one of them;

[0035] N is the number of regions in the formed part;

[0036] The weight coefficient for the i-th region;

[0037] w r For the first Weighting of regional operating condition deviations;

[0038] e i , e j These are the combined deviation values ​​for the i-th region and the j-th region, respectively;

[0039] Let be the coupling strength between region i and region j.

[0040] To better realize the present invention, further, if T1≤|E r If |≤T2, then the effective pressure of the cavity / local area, the local temperature / power of the mold / sheet material, and the air pressure loading rate are corrected respectively;

[0041] (1) The correction formula for cavity / local effective pressure is:

[0042] ;

[0043] ;

[0044] in: This is the correction value for the cavity / local effective pressure of the r-th region;

[0045] p rThe cavity / local effective pressure or load equivalent pressure of the r-th region;

[0046] Let the analytical target pressure for the r-th region satisfy the target strain rate.

[0047] Δp r Let be the single-cycle pressure change in the r-th region;

[0048] This represents the maximum permissible change in pressure per cycle.

[0049] α p These are the first-order filtering / distribution coefficients;

[0050] (2) The correction formula for the local temperature / power of the mold / sheet material is:

[0051] ;

[0052] ;

[0053] in: This is the local temperature / power correction value for the mold / sheet material in the r-th region;

[0054] T r The local temperature or local heating power of the mold / sheet material in the r-th region;

[0055] ΔT r For the r-th region, the temperature / power change per cycle is denoted as .

[0056] This represents the maximum permissible change in temperature / power per single cycle.

[0057] The temperature / heating power setting value for the r-th region;

[0058] α T This is a correction factor;

[0059] (3) The corrected formula for the air pressure loading rate is:

[0060] ;

[0061] ;

[0062] in: k F , k t , ks These are the correction gain coefficients;

[0063] e s,r The deformation or strain error of the r-th region;

[0064] The effective deviation for the risk of being too thin in the r-th region is defined by taking the absolute value of the negative deviation (the part that is too thin) and setting it to zero for the positive deviation (the part that is too thick).

[0065] This is the correction / change amount for the gas pressure loading rate (pressure rise rate) in region r.

[0066] To better realize the present invention, the correction formula for the air pressure loading rate is further as follows:

[0067] ;

[0068] in: k d This is the gain coefficient for the second-order correction term (the rate of change of deviation term);

[0069] t represents time.

[0070] To better realize the present invention, the correction formula for the air pressure loading rate is further as follows:

[0071] ;

[0072] in: This is a correction amount for the air pressure loading rate;

[0073] α ( t ), β ( t ), γ ( t These represent the weights for controlling the filling rate deviation, with larger initial values ​​emphasizing filling performance; and the weights for controlling the wall thickness deviation. β ( t The effect increases in the middle and later stages, with an emphasis on wall thickness uniformity; the weight of the strain rate deviation term is controlled. The strain rate is relatively large in the medium term, ensuring a stable strain rate.

[0074] e F This refers to the overall area's filling rate deviation.

[0075] e t This refers to the wall thickness deviation of the entire region.

[0076] eε This represents the strain rate deviation across the entire region.

[0077] f ( e F ) is the filling rate deviation function (reflecting the phenomenon of insufficient or overfilling of the overall area);

[0078] When e F When >0 (insufficient filling), f ( e F If positive, it will cause... Increase;

[0079] When e F When <0 (overfilled), f ( e F A negative value suppresses the loading rate.

[0080] f ( e t ): Wall thickness deviation function (reflects situations where the wall thickness is too thin or too thick);

[0081] When e t When <0 (too thin), f ( e t If ) is negative, then Reduce, thereby slowing down loading;

[0082] When e t When >0 (too thick), f ( e t A positive value increases the loading speed.

[0083] f(e ε ): Strain rate deviation function (reflects whether the deformation rate meets the target);

[0084] When e ε When <0 (rate is too low), f ( e ε If the value is positive, the speed will increase;

[0085] When e ε When >0 (rate is too high), f ( e ε If the value is negative, the rate decreases.

[0086] To better realize the present invention, further, in step S2, the forming condition prediction model includes a convolutional neural network branch, a long short-term memory network branch, and an attention mechanism layer arranged sequentially from front to back; the convolutional neural network branch is used to extract local features, the long short-term memory network branch is used to capture long-term dependencies, and the attention mechanism layer is used to dynamically assign weights to features of different modalities.

[0087] To better realize the present invention, step S3 further includes the following steps:

[0088] Step S31: Initialize the multi-objective particle swarm optimization model, wherein the forming temperature, gas pressure curve and strain rate are used as optimization objective variables, and constraints such as wall thickness uniformity, filling rate and forming time are introduced.

[0089] Step S32: Based on the input adaptive tuning parameters, the particle velocity and position are iteratively updated through the multi-objective particle swarm optimization model, the historical optimal position and the global optimal position are recorded, and the optimal parameter combination for titanium alloy superplastic forming is output.

[0090] An optimization system for superplastic forming of titanium alloys, based on the above-mentioned optimization method for superplastic forming of titanium alloys, includes a data acquisition and preprocessing module, a working condition prediction module, an adaptive tuning parameter generation module, a particle swarm optimization module, and a control and monitoring module.

[0091] The data acquisition and preprocessing module is used to acquire forming data and preprocess it to obtain a standardized forming dataset.

[0092] The working condition prediction module is used to predict the future working conditions of each region of the formed part based on the forming dataset.

[0093] The adaptive tuning parameter generation module is used to determine the corresponding adaptive tuning parameters for each region based on the deviation between the future operating conditions and the target operating conditions of the region.

[0094] The particle swarm optimization module is used to obtain the optimal parameter combination for the superplastic forming process of titanium alloys based on a multi-objective particle swarm optimization model.

[0095] The control and monitoring module is used to dynamically adjust the control curves of temperature, air pressure / load and loading rate at each forming stage based on the optimal parameter combination, and to monitor the wall thickness distribution of the formed part and the mold filling rate in real time.

[0096] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described optimized method for superplastic forming of titanium alloys.

[0097] The beneficial effects of this invention are as follows:

[0098] (1) Based on real-time monitoring data, this invention uses a forming condition prediction model to predict future working conditions, and forms local optimization parameters based on the deviation between the future working conditions and the target working conditions. Then, based on a multi-objective particle swarm optimization model, global process parameters are optimized to obtain the optimal parameter combination, realizing coordinated control of local and macroscopic processes. This effectively improves the quality of titanium alloy superplastic forming, effectively enhances the wall thickness uniformity and yield of formed parts, and reduces process energy consumption, thus having good practicality.

[0099] (2) This invention uses a multimodal fusion structure to build a forming condition prediction model based on a convolutional neural network (CNN), a long short-term memory network (LSTM), and an attention mechanism layer (Transformer layer). This model not only achieves collaborative modeling of short-term and long-term trends but also outputs the uncertainty range of the prediction results, providing a confidence basis for subsequent parameter correction and optimization. Specifically, the forming condition prediction model realizes multi-path information fusion of local detection, global memory, and dynamic weighting. The convolutional neural network branch can extract the mutation features of multi-dimensional process parameters within a local time window, which is equivalent to amplifying the sudden signal on a short time scale, avoiding the dilution of key information by long-term smoothing. The long short-term memory network branch can capture long-term dependencies across stages, maintaining the contextual dependencies of process history on a long time scale. The attention mechanism layer can dynamically weight different modal features such as temperature, air pressure, wall thickness, and strain, further establishing global correlations between different modal features, and adaptively allocating weights according to the prediction task, thereby highlighting key features and suppressing secondary interference.

[0100] (3) The present invention improves the correction formulas for cavity / local effective pressure, mold / plate local temperature / power and air pressure loading rate: by introducing a deviation change rate term into the formula, it realizes the forward adjustment of the deviation development trend and avoids correction lag. Secondly, the present invention, through the regional coupling deviation model, not only corrects the error of a single region, but also takes into account the balance between different regions, ensuring the consistency of the overall wall thickness distribution.

[0101] (4) The present invention enables the correction strategy to be adaptively adjusted with the forming stage through a phased dynamic weighting mechanism. Specifically, the initial stage focuses on filling, the middle stage focuses on strain rate, and the final stage focuses on wall thickness uniformity. Secondly, the present invention adds a second-order correction term to the correction formula of the air pressure loading rate, so that the system can sense the rate of change of deviation and adjust in advance, thereby achieving smoother and more timely control. Attached Figure Description

[0102] Figure 1 This is a flowchart of the optimized method for superplastic forming of titanium alloys according to the present invention;

[0103] Figure 2This is a schematic diagram of the optimized system for superplastic forming of titanium alloys according to the present invention;

[0104] Figure 3 This is a schematic diagram of the network structure of the forming condition prediction model. Detailed Implementation

[0105] Example 1:

[0106] An optimized method for superplastic forming of titanium alloys, such as Figure 1 As shown, it includes the following steps:

[0107] Step 1: Obtain multi-dimensional forming data of the superplastic forming process. Based on the forming condition prediction model, predict the future working conditions of different regions of the formed part. Based on the deviation between the future working conditions of the region and the target working conditions, determine the corresponding adaptive optimization parameters.

[0108] (1) Acquire multi-dimensional sensor data such as mold temperature, air pressure loading curve, displacement, strain rate, and wall thickness distribution during superplastic forming, and preprocess the data to obtain a standardized forming dataset. Preferably, local mold temperature data is collected by thermocouples or infrared temperature measurement points placed at different positions in the mold; cavity air pressure and loading rate are recorded by pressure sensor / air pressure control module; displacement / velocity of the forming plate is recorded by displacement sensor or optical measurement; wall thickness distribution information is obtained by online wall thickness measurement (springback thickness measurement or ultrasonic online measurement); and auxiliary parameters such as heating power, holding time, and cooling rate are recorded simultaneously. The acquisition frequency is 1 Hz to 10 Hz in the example. The raw data is preprocessed by outlier removal and Min-Max normalization.

[0109] (2) Based on the random forest algorithm, key features are extracted from the forming data to determine the core process parameters affecting wall thickness distribution and forming quality. Preferably, the random forest algorithm is applied to the preprocessed multidimensional time series data to evaluate the importance of each variable, and the first N (e.g., N=10) key features are retained, such as local temperature gradient, instantaneous air pressure gradient, local strain rate, real-time wall thickness decrease rate, etc. The number of trees in the random forest is 100~500.

[0110] (3) Input the key features into the preset forming condition prediction model to predict the filling state and wall thickness distribution in the future time period, and output the future working condition;

[0111] 1) The key features are converted into the input format of a temporal neural network using time series conversion technology;

[0112] 2) Adjust the structural and training parameters of the temporal neural network using parameter optimization methods;

[0113] 3) Verify the model performance based on multi-dimensional evaluation indicators, predict the material flow state and mold filling degree during the forming process in the future period, and output the future working conditions.

[0114] Preferably, such as Figure 3 As shown, the working condition prediction model is a multimodal fusion structure consisting of a convolutional neural network branch (CNN), a long short-term memory network branch (LSTM), and an attention mechanism layer (Transformer layer). Its input is multimodal time-series data such as temperature, air pressure, displacement, strain, and wall thickness.

[0115] The convolutional neural network branch employs one-dimensional convolutional layers (kernel size 364) and max-pooling layers to extract trend features within local timeframes, such as sudden temperature increases, changes in air pressure gradient, and abrupt decreases in wall thickness. The output is a local feature map with reduced dimensionality. Specifically, for example... Figure 3 As shown, a pooling layer and a feature extraction network are set up to extract changes in parameters such as temperature and air pressure, and output parameter features to capture key variables at instantaneous moments.

[0116] The Long Short-Term Memory (LSTM) branch receives local features from the CNN output and uses a two-layer LSTM (64-128 units per layer) to capture long-term dependencies throughout the forming process, such as the delayed effect of the early heating rate on the mid-to-late strain rate. The output is a context sequence representation.

[0117] Attention Mechanism Layer (Transformer Layer): Employs a multi-head self-attention structure (4-8 attention heads in the example) to dynamically assign weights to features of different modalities. For example, it automatically increases the weight of temperature features when there are temperature anomalies, and increases the weight of wall thickness features when there are drastic fluctuations in wall thickness, thereby improving the accuracy and interpretability of the prediction. Output Layer: Through fully connected layers and Softmax / regression layers, it outputs the probability distribution of wall thickness changes and the filling rate trend of each region within the next T minutes, and provides the uncertainty range of the prediction results.

[0118] The model is trained using a sliding window method (with an example window length of 120 time steps). Hyperparameters are determined through grid search or Bayesian optimization (number of hidden layers = 2, number of hidden units = 64, learning rate = 1e-4, number of iterations = 100). Compared to a single LSTM model, the fusion model in this embodiment improves the accuracy of wall thickness prediction and filling rate prediction by approximately 10% to 15%, and can dynamically adjust the correction magnitude based on the prediction confidence.

[0119] (4) Based on the predicted working conditions, calculate the deviation between the future working conditions and the target working conditions of the region, and dynamically output the corresponding adaptive tuning parameters.

[0120] Preferably, based on the predicted future operating conditions, the deviation between the current future operating conditions and the target operating conditions for each region is calculated (e.g., target wall thickness error threshold ±5%). Threshold setting logic is then established based on this deviation.

[0121] If |deviation| < the first threshold (e.g., 2%), then fine-tune the air pressure loading rate (+5%) or the local heating power (+2%).

[0122] If the first threshold ≤ |deviation| ≤ the second threshold (e.g., 2%~8%), then maintain the current parameters and proceed to the observation phase;

[0123] If the deviation exceeds the second threshold (e.g., 8%), then a larger adjustment is made, such as reducing the loading rate by 10% or increasing local heating by 5%.

[0124] Step 2: Input the adaptive tuning parameters into the multi-objective particle swarm optimization model to optimize the global process parameters and obtain the optimal parameter combination for the superplastic forming process;

[0125] (1) Initialize the multi-objective particle swarm optimization model, in which the forming temperature, gas pressure curve, strain rate and other factors are used as optimization target variables, and the preset constraints such as wall thickness uniformity, filling rate and forming time are introduced.

[0126] (2) Based on the input adaptive tuning parameters, the particle velocity and position are iteratively updated through the multi-objective particle swarm optimization model (MOPSO), the historical optimal position and the global optimal position are recorded, and the optimal parameter combination for superplastic forming is output.

[0127] Preferably, the adaptive tuning parameters for each region are used as candidate variables and input into the multi-objective particle swarm optimization model for global optimization. For example, optimization objectives include: minimizing the standard deviation of wall thickness, maximizing the filling rate, and minimizing forming time and energy consumption; constraints include upper / lower temperature limits, maximum allowable strain rate, and mold tolerance limits. Example MOPSO parameters: 40 particles, 200 maximum iterations, inertia weight decreasing linearly from 0.9 to 0.4, and learning factors c1=c2=2.0. The optimal parameter combination is output and written to the execution module.

[0128] Among them, the physical / device constraints (hard limits) are:

[0129] ;

[0130] ;

[0131] in: T min The lowest permissible temperature (lower limit) for the forming system.

[0132] This is the highest temperature (upper limit) allowed by the system.

[0133] The maximum allowable cavity pressure (MPa) of the system is determined by the mold strength, sealing performance, and safety valve settings.

[0134] The target heating temperature or equivalent heating power is sent in real time by the control system;

[0135] The target air pressure value issued by the current control system;

[0136] This represents the actual equivalent pressure in the r-th region;

[0137] The minimum equivalent cavity pressure (MPa) allowed by the system or mold, the lower limit of which is determined by equipment safety and forming stability;

[0138] The equivalent maximum cavity pressure (MPa) allowed by the system or mold;

[0139] The minimum strain rate threshold for a material in a superplastic state;

[0140] This represents the maximum allowable strain rate within the superplastic deformation range.

[0141] Let be the actual equivalent strain rate (i.e., the local deformation rate of the material) in the r-th region.

[0142] The penalty function or projection operator ΠC(·) is used to project onto the feasible region C.

[0143] This invention first defines several physical / device constraints, including temperature range, pressure range, and strain rate range. These constraints define the allowable value ranges of parameters during the forming process, collectively forming a feasible region C. Specifically, in multi-objective particle swarm optimization, the algorithm continuously generates candidate solutions (particle positions), which include variables such as temperature, pressure, and rate. However, some solutions may violate the aforementioned physical constraints (e.g., temperature exceeding the upper limit, excessive pressure, or unstable strain rate). Therefore, the aforementioned mechanism is needed to force or guide the solutions back to the physically feasible range.

[0144] Step 3: Dynamically adjust the control curves of temperature, air pressure / load, loading rate, etc. in each stage of forming according to the optimal parameter combination;

[0145] Step 4: Monitor the wall thickness distribution of the molded part and the mold filling rate in real time. When local thinning or insufficient filling is detected, proceed to step 1.

[0146] Specifically, the loading curve and displacement of the local power and air pressure of the mold heater are adjusted according to the optimal combination; at the same time, the wall thickness and filling status are monitored in real time. If local thinning or insufficient filling is detected, the process immediately returns to step 1 for re-evaluation and triggers local optimization and correction until the forming is completed.

[0147] Preferably, the deviation between the future operating conditions and the target operating conditions of the calculation area is calculated, and the corresponding adaptive tuning parameters are dynamically output; the specific details are as follows:

[0148] Define the bias (normalized form of "observation-target") for the r-th region:

[0149] ;

[0150] ;

[0151] ;

[0152]

[0153] Among them: E r The deviation between the future operating condition and the target operating condition of the r-th region (used for threshold / trigger logic or single-input control).

[0154] w ε , w t , w F Each of these is used to calculate a weight, and the sum of the three is 1.

[0155] e ε,r The strain rate deviation in the r-th region;

[0156] e t,r The wall thickness deviation of the r-th region;

[0157] e F,r The filling rate deviation of the r-th region;

[0158] Let r be the equivalent rate of change in the r-th region;

[0159] Let be the target strain rate for the r-th region;

[0160] t r Let be the wall thickness (mm) of the r-th region;

[0161] is the target wall thickness of the r-th region;

[0162] F r is the local filling rate of the r-th region ( F r is from 0 to 1);

[0163] is the target local filling rate of the r-th region.

[0164] If |E r | < T1, increase the air pressure loading rate or adjust the local heating power of the mold; specifically, increase the air pressure loading rate by 5% or increase the local heating power of the mold by 2%;

[0165] If T1 ≤ |E r | ≤ T2, maintain the current parameters and enter the observation stage;

[0166] If |E[[ID=2,7]] r | > T2, reduce the air pressure loading rate or adjust the temperature to inhibit local excessive deformation; specifically, reduce the air pressure loading rate by 10% or increase the local heating power of the mold by 5%.

[0167] Where: T1 and T2 are respectively preset thresholds, and T2 > T1; for example, T1 and T2 are 2% - 8%.

[0168] Preferably, if |E r | > T2, enable the co - ordination of temperature and pressure dual variables and reduce the upper pressure limit.

[0169] Preferably, if T1 ≤ |E r | ≤ T2, perform conventional correction according to the following formula:

[0170] (1) The correction formula for the cavity / local effective pressure is:

[0171] ;

[0172] ;

[0173] Where: is the correction value of the cavity / local effective pressure of the r-th region;

[0174] [[ID=6,0]] p r is the cavity / local effective pressure or load equivalent pressure (MPa) of the r-th region;

[0175] is the analytical target pressure of the r-th region that meets the target strain rate;

[0176] Δp r Let be the single-cycle pressure change in the r-th region;

[0177] This represents the maximum permissible pressure change per cycle in the r-th region (e.g., 0.02–0.05 MPa).

[0178] α p These are the first-order filtering / distribution coefficients. α p ∈(0,1).

[0179] Preferably, if the equipment uses the rate of pressure rise as the control variable, it can be used Achieve equivalent adjustment.

[0180] in: The amount of correction / change in the air pressure loading rate for region r;

[0181] This is the gain coefficient for the rate of change of air pressure.

[0182] The above defines a standard modified model with pressure as the control variable; the above formula points out the equivalence relationship when the equipment control quantity is different (rate control); it belongs to a kind of engineering implementation adaptation logic.

[0183] Specifically, the pressure (or equivalent load) is set based on the strain rate target. p r value:

[0184] Using the classic relationship of superplastic flow:

[0185] ;

[0186] Will σ r Linear correlation with pressure geometry (diaphragm / drum approximation):

[0187] ;

[0188] in: Let r be the equivalent rate of change in the r-th region;

[0189] σ r Let be the equivalent membrane stress in the r-th region, and m For strain rate sensitivity index, m Specifically, it is 0.3~0.6;

[0190] t r Let the wall thickness be the thickness of the r-th region.

[0191] T r Let K be the local temperature of the mold / sheet material in the r-th region (K);

[0192] Let be the local equivalent feature radius / span (mm) of the r-th region;

[0193] A is the prefactor in the material constants, comprehensively reflecting the influence of the material's microstructure characteristics and the dominant deformation mechanism (s). -1 ·MPa -m );

[0194] Q is the apparent activation energy (J·mol⁻¹). -1 );

[0195] R R is the gas constant, R = 8.314 J·mol⁻¹ -1 ·K -1 .

[0196] The analytical pressure target satisfying the target strain rate is obtained as follows:

[0197] ;

[0198] From the above formula, the target strain rate can be determined. The pressure required to reverse the process.

[0199] (2) The correction formula for the local temperature / power of the mold / sheet material is:

[0200] ;

[0201] ;

[0202] in: This is the local temperature / power correction value for the mold / sheet material in the r-th region;

[0203] T r The local temperature or local heating power of the mold / sheet material in the r-th region;

[0204] ΔT r For the r-th region, the temperature / power change per cycle is denoted as .

[0205] This represents the maximum permissible change in temperature / power per single cycle for the r-th region;

[0206] The temperature / heating power setting value for the r-th region;

[0207] αT This is a correction factor.

[0208] Specifically, temperature setpoint (Or local heating power) is used to lock the target strain rate:

[0209] Known current pressure p r The relationship with geometric parameters is as follows:

[0210] ;

[0211] make

[0212] ;

[0213] Solving

[0214] .

[0215] If heating power q r Control can be approximated using a linear thermal model:

[0216] ;

[0217] in: k T Obtained through thermal inertia identification;

[0218] σ r Let be the equivalent membrane stress in the r-th region, and m be the strain rate sensitivity index, specifically 0.3~0.6.

[0219] Selection rule: Pressure tracking is preferred. When the pressure change in a single cycle Δp r When the upper limit is reached or there is a risk of excessive thinness, temperature fine-tuning is then used. This section contains the switching logic between the main control variable and the auxiliary control variable. It is located in the section on the overall / local effective pressure and temperature correction mechanism and is used to explain the priority and switching conditions of the two control methods under different operating conditions. Main control path: pressure is the primary control, and temperature is the secondary control; when pressure adjustment is limited or wall thickness risks occur, continuous, stable, and feasible control is achieved through temperature compensation.

[0220] (3) The corrected formula for the air pressure loading rate is:

[0221] The deviation coupling correction for the pressure rise rate considers the opposing constraints of incomplete filling and excessive thinning, and gives a piecewise coupling law:

[0222] ;

[0223] ;

[0224] in: k F , k t , k s These are the correction gain coefficients;

[0225] For the effective deviation of the risk of being too thin, the absolute value of the negative deviation (the part that is too thin) is taken, and the positive deviation (the part that is too thick) is set to zero.

[0226] This is the correction amount for the air pressure loading rate in region r.

[0227] Only when it is "too thin", that is t r < It takes effect at that time. k F , k t , k s All are greater than 0. Intuitive explanation: Insufficient filling ( e F,r >0) will increase the rate; too thin risk ( e t,r <0 will inhibit the rate; the strain rate is too low ( e ε,r A value <0 indicates increased speed, while a value higher indicates decreased speed.

[0228] Preferably, in the three types of formulas—pressure correction, temperature / power correction, and air pressure loading rate correction—a deviation change rate term (derivative term) is added to each. This creates a second-order predictive correction formula with a definite deviation, building upon the original proportional correction. By adding the deviation change rate (derivative) to the original "single deviation correction," not only is the current error corrected, but deviation deterioration can also be prevented in advance, achieving predictive adjustment. This forms a similar second-order predictive correction (general formula):

[0229] ;

[0230] in: k 1 , k 2 These are the proportional gain coefficient and the differential gain coefficient, respectively.

[0231] Δu r This variable is for the r-th region;

[0232] e r This is the operating condition deviation, and it ise F,r , e t,r and e ε,r Any one of them;

[0233] To characterize trends.

[0234] Specifically, the correction formula for the air pressure loading rate is:

[0235] ;

[0236] in: k d This is the correction gain coefficient for the strain rate deviation change rate term;

[0237] t represents time.

[0238] After implementing prediction correction in the time dimension, to further improve spatial consistency, a preferred approach is to introduce regional coupling weights. This avoids focusing on a single point, resolves the conflict between local thinning and overall uniformity, and achieves global equilibrium correction. Specifically, the deviations from different regions are summarized as follows:

[0239] ;

[0240] in: E global This is a global deviation;

[0241] e r This is the operating condition deviation, and it is e F,r , e t,r and e ε,r Any one of them;

[0242] N is the number of regions divided in the formed part (such as a thin-walled titanium alloy structure);

[0243] For the first The weighting coefficient of the region;

[0244] w r For the first Weighting of regional operating condition deviations;

[0245] e i , e j These are the combined deviation values ​​(operating condition errors) for the i-th and j-th regions, respectively.

[0246] The coupling strength between region i and region j 。

[0247] Preferably, this invention proposes a phased dynamic weight correction mechanism. At different forming stages, weight functions that change with time are set respectively, thereby optimizing different deviations in stages. For example, the weight of filling rate is increased in the initial stage, the weight of strain rate is increased in the intermediate stage, and the weight of wall thickness uniformity is increased in the final stage. The correction formula is as follows:

[0248] ;

[0249] in: This is a correction amount for the air pressure loading rate;

[0250] α ( t ), β ( t ), γ ( t These represent the weights for the filling stage, wall thickness control, and strain rate, respectively.

[0251] f ( e F ) is the filling rate deviation function;

[0252] f ( e t ) is the wall thickness deviation function;

[0253] f ( e ε ) is the strain rate deviation function.

[0254] Example 2:

[0255] An optimized method for superplastic forming of titanium alloys, taking the production of complex thin-walled TC4 titanium alloy covers by an aerospace manufacturing company as an example, wherein the designed wall thickness of the TC4 titanium alloy complex thin-walled cover is 1.2 mm, and the local thickness tolerance is ±5%. The method specifically includes the following steps:

[0256] (1) In this embodiment, multiple sensors are deployed on the mold and the workpiece to collect multi-dimensional forming data of the forming process.

[0257] The mold was equipped with eight temperature measurement points, four air pressure measurement channels, two displacement measurement channels, and six online wall thickness measurement points. Specifically, eight thermocouples were deployed at key locations on the mold and workpiece, with a sampling frequency of 2 Hz; four pressure sensors were deployed in the cavity, with a sampling frequency of 5 Hz; two laser displacement gauges were deployed on the outer side of the workpiece, with a sampling frequency of 10 Hz; fiber optic grating sensors were deployed in key areas to collect strain data, with a sampling frequency of 5 Hz; and ultrasonic online wall thickness measurement was performed at six locations, with a sampling frequency of 1 Hz. After outlier removal and normalization, the collected data formed a standardized forming dataset.

[0258] (2) When analyzing the forming dataset, the Random Forest algorithm (300 trees, maximum depth 10) was used to sort the importance of features. The results showed that the local temperature gradient of the mold, the rate of increase of the cavity air pressure, the local strain rate, the heat preservation time of the previous stage, and the wall thickness reduction rate are the key features that affect the wall thickness distribution and filling rate.

[0259] (3) Based on the key features extracted above, a forming condition prediction model was constructed. The forming condition prediction model adopts a multimodal fusion structure (CNN+LSTM+Attention), with an input window length of 120 time steps (corresponding to approximately 2 minutes of historical data), two hidden layers, each with 64 units, and a learning rate of 1×10. -4 The training iterations were 100. The prediction results can predict the wall thickness change trend and filling status at each location 3 minutes in advance. Based on the validation set, the mean square error of the wall thickness prediction is less than 0.005 mm², and the accuracy of the filling rate prediction exceeds 92%.

[0260] During the initial forming process, the prediction results showed that at 30 minutes, local area A of the cover exhibited a trend of excessively rapid wall thickness reduction (estimated to be about 12% thinner), while area B showed a trend of insufficient filling. Based on the deviation between future working conditions and target working conditions, an adaptive optimization method was used to correct the parameters.

[0261] In this embodiment, the deviation correction not only adopts the basic deviation formula, but also further introduces a second-order deviation coupling and regional multi-coupling correction mechanism:

[0262] Strain rate deviation:

[0263] ;

[0264] Wall thickness deviation:

[0265] ;

[0266] Filling rate deviation:

[0267] ;

[0268] in: e ε,r The strain rate deviation in the r-th region;

[0269] e t,r The wall thickness deviation of the r-th region;

[0270] e F,r The filling rate deviation of the r-th region;

[0271] Let r be the equivalent rate of change in the r-th region;

[0272] Let be the target strain rate for the r-th region;

[0273] t r Let be the wall thickness (mm) of the r-th region;

[0274] Let the target wall thickness be the r-th region;

[0275] F r Let r be the local filling rate of the r-th region ( F r (0~1)

[0276] Let be the target local fill rate for the r-th region (usually taken as 1).

[0277] The deviation E was calculated. r Subsequently, this embodiment uses the following correction formula to dynamically adjust the process parameters:

[0278] (1) The correction formula for cavity / local effective pressure is:

[0279] ;

[0280] in: This is the correction value for the cavity / local effective pressure of the r-th region;

[0281] p r The effective pressure of the cavity / local area or the equivalent pressure of the load in the r-th region (MPa);

[0282] Let the analytical target pressure for the r-th region satisfy the target strain rate.

[0283] α p These are the first-order filtering / distribution coefficients. αp ∈(0,1];

[0284] t r Let be the wall thickness (mm) of the r-th region;

[0285] Let be the local equivalent feature radius / span (mm) of the r-th region;

[0286] Let be the target strain rate for the r-th region;

[0287] A is the prefactor in the material constants, comprehensively reflecting the influence of the material's microstructure characteristics and the dominant deformation mechanism (s). -1 ·MPa -m );

[0288] m For strain rate sensitivity index, m Specifically, it is 0.3~0.6;

[0289] T r Let K be the local temperature of the mold / sheet material in the r-th region (K);

[0290] Q is the apparent activation energy (J·mol⁻¹). -1 );

[0291] R R is the gas constant, R = 8.314 J·mol⁻¹ -1 ·K -1 .

[0292] Temperature correction formula:

[0293] ;

[0294] ;

[0295] in: This is the local temperature / power correction value for the mold / sheet material in the r-th region;

[0296] T r The local temperature or local heating power of the mold / sheet material in the r-th region;

[0297] The temperature / heating power setting value for the r-th region;

[0298] σ r The equivalent membrane stress in region r is... m For strain rate sensitivity index, m Specifically, it is 0.3~0.6;

[0299] α T This is a correction factor.

[0300] Air pressure loading rate correction formula:

[0301]

[0302] in, .

[0303] in: The amount of correction / change in the air pressure loading rate for region r;

[0304] k F , k t , k s These are the correction gain coefficients;

[0305] k d The gain coefficient of the second-order correction term;

[0306] e s,r The deformation or strain error of the r-th region;

[0307] The effective deviation for the excessive thinning risk in the r-th region;

[0308] e ε,r The strain rate deviation in the r-th region;

[0309] e t,r The wall thickness deviation of the r-th region;

[0310] e F,r The filling rate deviation of the r-th region;

[0311] t represents time.

[0312] When region A is found to have a thinner wall thickness (E) r =1.2), when it exceeds the set threshold of 1.0, according to the second-order prediction correction formula, it is calculated that the air pressure loading rate needs to be reduced by about 10%, and the local forming temperature needs to be increased by about 5K, in order to suppress the decreasing trend of strain rate and improve local filling.

[0313] When incomplete filling is detected in area B (E) rWhen the value is 0.08, which is lower than the filling rate threshold of 0.1, the modified model calculates that the air pressure loading rate needs to be increased by about 8% to promote material flow and filling.

[0314] (4) Subsequently, the adaptive tuning parameters generated in each region are input into the multi-objective particle swarm optimization model (MOPSO). In this embodiment, the number of MOPSO particles is set to 50, the maximum number of iterations is 200, and the optimization objectives include minimizing the standard deviation of wall thickness, maximizing the average filling rate, minimizing the forming time, and minimizing energy consumption. After optimization, the output global optimal parameter combination is as follows: the overall forming temperature curve is increased by about 3K, the gas pressure loading curve is adjusted to "slow rise in the early stage - rapid in the middle stage - flat in the later stage", and the strain rate is stabilized at 2.0×10 -4 s -1 .

[0315] (5) The process is executed in real time according to the optimal parameter curve, with temperature control accuracy of ±1K and air pressure control accuracy of ±0.01 MPa, and a closed-loop correction is formed during the forming process.

[0316] To verify the effectiveness of this invention, performance testing was conducted, and comparative experiments were performed on the same model of TC4 titanium alloy thin-walled cover. The cover was designed with a wall thickness of 1.2 mm, a diameter of 500 mm, and a height of 200 mm. Superplastic forming tests were performed using both the traditional empirical curve control method and the present invention, with 20 pieces formed using each method. In the traditional method, the gas pressure loading curve was preset manually based on experience, specifically a slow increase for the first 20 minutes, followed by a linear increase to the target pressure, without real-time adjustments throughout the process; the temperature was kept constant at 920℃, and the strain rate depended on the initial parameter settings. In the traditional superplastic forming process, because the process parameters depend on manual experience, problems often arise such as localized excessive wall thickness leading to cracks or insufficient filling leading to wrinkles. The wall thickness uniformity of the formed parts is poor, with a standard deviation of approximately ±12%; the average forming cycle is approximately 80 minutes, and the yield is less than 85%. Common defects include localized cracks and wrinkles; the energy consumption per piece is 100% based on the baseline value.

[0317] As shown in Table 1, the results of this embodiment are significantly better than those of the traditional method. The wall thickness uniformity of the formed parts is improved, with the standard deviation decreasing from 12% to 4%; the yield rate increases from 85% to 96%; the average forming cycle is shortened from 80 minutes to 68 minutes, a reduction of approximately 15%; and unit energy consumption is reduced by approximately 8%. All indicators of the final formed parts meet the quality requirements for complex thin-walled titanium alloy structural parts for aerospace applications, achieving stable mass production. The above comparative results demonstrate that the present invention is superior to the traditional method in terms of forming quality, production efficiency, and energy consumption control, with significant improvements.

[0318] Table 1 Processing performance parameters of the present invention and conventional methods

[0319]

[0320] Example 3:

[0321] An optimization system for superplastic forming of titanium alloys is implemented based on the aforementioned optimization method for superplastic forming of titanium alloys, such as... Figure 2 As shown, it includes a data acquisition and preprocessing module, a key feature extraction module, a working condition prediction module, an adaptive tuning parameter generation module, a particle swarm optimization module, and a control and monitoring module.

[0322] The data acquisition and preprocessing module is used to acquire forming data and preprocess it to obtain a standardized forming dataset. Specifically, it acquires forming data in multiple dimensions such as mold temperature, cavity air pressure, displacement / velocity, strain, and online wall thickness, and preprocesses the data (including cleaning, normalization, and time series slicing).

[0323] The key feature extraction module is used to evaluate the feature importance of the preprocessed forming data based on the random forest algorithm, and retain key features (such as local temperature gradient of the mold, instantaneous air pressure slope, local strain rate, wall thickness reduction rate, etc.).

[0324] The working condition prediction module is used to predict the future working conditions of each region of the formed part based on the time series data of key features and the forming working condition prediction model.

[0325] The adaptive tuning parameter generation module is used to determine the corresponding adaptive tuning parameters for each region based on the deviation between the future operating conditions and the target operating conditions of the region.

[0326] The particle swarm optimization module is used to obtain the optimal parameter combination for the superplastic forming process of titanium alloys based on a multi-objective particle swarm optimization model.

[0327] The control and monitoring module is used to dynamically adjust the control curves of temperature, air pressure / load and loading rate at each forming stage based on the optimal parameter combination, and to monitor the wall thickness distribution of the formed part and the mold filling rate in real time.

[0328] Preferably, the forming condition prediction model employs a multimodal fusion structure consisting of a convolutional neural network (CNN) branch, a long short-term memory (LSTM) branch, and an attention mechanism layer (Transformer layer) to predict the forming condition. Specifically, it includes:

[0329] Input layer: Receives time-series data containing multidimensional key features. The time steps in the input samples include multimodal features such as temperature, air pressure, displacement, strain, and wall thickness. A sliding window method is used to form a fixed-length sequence (the window length can be 120 time steps).

[0330] The convolutional neural network branch includes one-dimensional convolutional layers and pooling layers, used to extract short-term trend features within a local time window. The kernel size can be 35; the number of kernels can be 3264.

[0331] Pooling method: max pooling or average pooling. Output: Local feature map after dimensionality reduction. It can quickly capture short-term feature changes, such as sudden temperature increases, sharp changes in air pressure loading rates, and local wall thickness decreases.

[0332] Long Short-Term Memory (LSTM) branch: Models long-term dependencies on local feature maps output by CNN.

[0333] Structure: Two-layer LSTM, with 64~128 hidden units per layer, tanh activation function, and sigmoid gating mechanism.

[0334] Output: A temporal representation vector containing contextual information. It captures the temporal dependencies between different stages of the entire shaping process, avoiding the problem that simple convolution cannot identify cross-stage effects.

[0335] Attention mechanism layer (Transformer layer): Globally weights the temporal representation of the LSTM output.

[0336] Structure: Multi-Head Attention layer, with an example of 4-8 attention heads, each with a dimension of 32. Weights are dynamically assigned to different modal features (temperature, air pressure, wall thickness, etc.), focusing on the features most relevant to the current prediction. Output: A fused global representation vector. This allows for adaptive weighting of different modal features, improving prediction accuracy and interpretability. For example, when wall thickness fluctuates significantly, the weight of the wall thickness feature is automatically increased; when the rate of air pressure change is abnormal, the weight of the air pressure feature is automatically increased.

[0337] Output layer: Consists of a fully connected layer and a Softmax / regression layer. Output content includes:

[0338] Probability distribution of wall thickness change in each region within the next T minutes;

[0339] Mold filling rate trend (e.g., "insufficient filling / normal / overfilling" classification results);

[0340] The uncertainty interval of the prediction results (calculated using Monte Carlo Dropout or Bayesian methods).

[0341] It can not only provide quantitative predictions of future trends, but also provide confidence information, enabling the calculation module 16 to automatically reduce the correction magnitude when the prediction uncertainty is high, thereby improving control robustness.

[0342] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. An optimized method for superplastic forming of titanium alloys, characterized in that, Includes the following steps: Step S1: Collect forming data of the titanium alloy superplastic forming process and preprocess it to obtain a standardized forming dataset; the forming data includes mold temperature, gas pressure loading curve, displacement, strain rate and wall thickness distribution; Step S2: Based on the forming dataset, a forming condition prediction model based on a neural network model is used to predict the future working conditions of each region of the formed part; according to the deviation between the future working conditions and the target working conditions of each region, the corresponding adaptive optimization parameters for each region are determined; the future working conditions include the filling state and wall thickness distribution of the forming process, and the adaptive optimization parameters include cavity / local effective pressure, mold / sheet material local temperature / power and air pressure loading rate; Step S3: The adaptive tuning parameters of each region are used as candidate variables and input into the multi-objective particle swarm optimization model to optimize the global process parameters and obtain the optimal parameter combination for the titanium alloy superplastic forming process. The optimization objectives include minimizing the standard deviation of wall thickness, maximizing the filling rate, and minimizing forming time and energy consumption. The constraints include upper / lower temperature limits, maximum allowable strain rate, and mold bearing limit. Step S4: Dynamically adjust the control curves of temperature, air pressure / load, and loading rate for each forming stage based on the optimal parameter combination; In step S2, the deviation between the future operating condition and the target operating condition of the r-th region is: ； If |E r | < T1, then increase the air pressure loading rate by 5% or increase the local heating power of the mold by 2%; If T1≤|E r If |≤T2, then keep the current parameters and enter the observation phase; If |E r If |>T2, then reduce the air pressure loading rate by 10% or increase the local heating power of the mold by 5%; Where: E r Let be the deviation between the future operating conditions and the target operating conditions of the r-th region; w ε , w t , w F Calculate the weights respectively; e ε,r The strain rate deviation in the r-th region; e t,r The wall thickness deviation of the r-th region; e F,r The filling rate deviation of the r-th region; T1 and T2 are preset thresholds, and T2 > T1; In step S2, respectively for e ε,r , e t,r and e F,r Perform global correction, and the correction formula is: ； in: E global for e ε,r , e t,r and e F,r Global deviation of any one of them; e r for e F,r , e t,r and e ε,r Any one of them; N is the number of regions in the formed part; γ The weight coefficient for the i-th region; w r The operating condition deviation weight for the r-th region; e i , e j These are the combined deviation values ​​for the i-th region and the j-th region, respectively; Let be the coupling strength between region i and region j.

2. The optimized method for superplastic forming of titanium alloys according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Based on the standardized forming dataset, the random forest algorithm is used to extract the top N key features that affect the wall thickness distribution and forming quality; Step S22: After time series transformation, the extracted N key features are input into the forming condition prediction model, and the future working conditions of each region of the formed part are output. Step S23: Based on the deviation between the future operating conditions and the target operating conditions, dynamically output the corresponding adaptive tuning parameters.

3. An optimized method for superplastic forming of titanium alloys according to claim 1 or 2, characterized in that, If T1≤|E r If |≤T2, then the effective pressure of the cavity / local area, the local temperature / power of the mold / sheet material, and the air pressure loading rate are corrected respectively; (1) The correction formula for cavity / local effective pressure is: ； ； in: This is the correction value for the cavity / local effective pressure of the r-th region; p r The cavity / local effective pressure or load equivalent pressure of the r-th region; Let the analytical target pressure for the r-th region satisfy the target strain rate. Δp r Let be the single-cycle pressure change in the r-th region; Δp max This represents the maximum permissible change in pressure per cycle. α p These are the first-order filtering / distribution coefficients; (2) The correction formula for the local temperature / power of the mold / sheet material is: ； ； in: This is the local temperature / power correction value for the mold / sheet material in the r-th region; T r The local temperature or local heating power of the mold / sheet material in the r-th region; ΔT r For the r-th region, the temperature / power change per cycle is denoted as . ΔT max This represents the maximum permissible change in temperature / power per single cycle. The temperature / heating power setting value for the r-th region; α T This is a correction factor; (3) The corrected formula for the air pressure loading rate is: ； ； in: k F , k t , k s These are the correction gain coefficients; e s,r The deformation or strain error of the r-th region; The effective deviation for the excessive thinning risk in the r-th region; This represents the correction / change in the air pressure loading rate of region r.

4. The optimized method for superplastic forming of titanium alloys according to claim 3, characterized in that, A deviation change rate term is added to the correction formulas for cavity / local effective pressure, mold / sheet material local temperature / power, and air pressure loading rate, respectively; the corrected air pressure loading rate correction formula is as follows: ； in: k d The gain coefficient of the second-order correction term; t represents time.

5. An optimized method for superplastic forming of titanium alloys according to claim 1 or 2, characterized in that, The correction formula for the air pressure loading rate is: ； in: This is a correction amount for the air pressure loading rate; α ( t ), β ( t ), γ ( t These are the weights for the control filling rate deviation item; e F This refers to the overall area's filling rate deviation. e t This refers to the wall thickness deviation of the entire region. e ε This represents the strain rate deviation across the entire region. f ( e F ) represents the filling rate deviation function; when e F When >0, f ( e F ) is positive; when e F When <0, f ( e F ) is negative; f ( e t ) is the wall thickness deviation function; when e t When <0, f ( e t ) is negative; when e t > 0, f ( e t ) is positive; f ( e ε ) represents the strain rate deviation function; when e ε When <0, f ( e ε ) is positive; when e ε When >0, f ( e ε ) is negative.

6. The optimized method for superplastic forming of titanium alloys according to claim 1, characterized in that, In step S2, the forming condition prediction model includes a convolutional neural network branch, a long short-term memory network branch, and an attention mechanism layer arranged sequentially from front to back; the convolutional neural network branch is used to extract local features, the long short-term memory network branch is used to capture long-term dependencies, and the attention mechanism layer is used to dynamically assign weights to features of different modalities.

7. The optimized method for superplastic forming of titanium alloys according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Initialize the multi-objective particle swarm optimization model, wherein the forming temperature, gas pressure curve and strain rate are used as optimization objective variables, and constraints such as wall thickness uniformity, filling rate and forming time are introduced. Step S32: Based on the input adaptive tuning parameters, the particle velocity and position are iteratively updated through the multi-objective particle swarm optimization model, the historical optimal position and the global optimal position are recorded, and the optimal parameter combination for titanium alloy superplastic forming is output.

8. An optimization system for superplastic forming of titanium alloys, implemented based on the optimization method for superplastic forming of titanium alloys according to any one of claims 1-7, characterized in that, It includes a data acquisition and preprocessing module, a working condition prediction module, an adaptive tuning parameter generation module, a particle swarm optimization module, and a control and monitoring module; The data acquisition and preprocessing module is used to acquire forming data and preprocess it to obtain a standardized forming dataset. The working condition prediction module is used to predict the future working conditions of each region of the formed part based on the forming dataset. The adaptive tuning parameter generation module is used to determine the corresponding adaptive tuning parameters for each region based on the deviation between the future operating conditions and the target operating conditions of the region. The particle swarm optimization module is used to obtain the optimal parameter combination for the superplastic forming process of titanium alloys based on a multi-objective particle swarm optimization model. The control and monitoring module is used to dynamically adjust the control curves of temperature, air pressure / load and loading rate at each forming stage based on the optimal parameter combination, and to monitor the wall thickness distribution of the formed part and the mold filling rate in real time.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements an optimized method for superplastic forming of titanium alloys as described in any one of claims 1-7.

Images