Titanium steel composite plate manufacturing optimization method and system based on neural network model

Abstract

The invention provides a titanium-steel composite plate manufacturing optimization method and system based on a neural network model, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting interface data in a titanium-steel composite plate manufacturing process, and generating interface combination features; inputting the interface bonding features into an adaptive memory network, and predicting interface bonding strength; generating an optimized process parameter combination through a double-delay depth Q network structure with a priority experience playback mechanism; and taking the optimized process parameter combination as an initial population, and generating an optimal process parameter scheme based on performance indexes and process stability by adopting a phase change temperature point staged control and theory-differential gradient fusion mode.

Description

Technical Field

[0001] The present invention relates to artificial intelligence technology, and particularly to an optimization method and system for manufacturing titanium-steel composite plates based on a neural network model. Background Art

[0002] As a new type of composite material, the control of the interfacial bonding strength during the manufacturing process of titanium-steel composite plates is a key technical problem. Existing neural network optimization methods mainly use a single deep learning model to predict the interfacial bonding strength, lacking consideration of the material constitutive relationship and physical constraints of the manufacturing process, resulting in insufficient prediction accuracy and unreliable optimization results.

[0003] Traditional process parameter optimization methods for titanium-steel composite plates usually rely on expert experience or simple deep reinforcement learning algorithms, and cannot accurately capture the dynamic changes of interfacial bonding characteristics in the spatio-temporal dimension. At the same time, when dealing with parameter optimization near the phase change temperature point, due to the lack of an adaptive control strategy and theoretical model support, performance fluctuations and local convergence problems are likely to occur.

[0004] Currently, the application of neural networks in the optimization of titanium-steel composite plate manufacturing fails to effectively integrate the theoretical model with experimental data, and also lacks systematic consideration of the stability of process parameters. Especially during the optimization process, existing methods are difficult to balance the relationship between exploration and exploitation, and lack a dynamic feedback mechanism, which affects the engineering practicability of the optimization effect. Summary of the Invention

[0005] Embodiments of the present invention provide an optimization method and system for manufacturing titanium-steel composite plates based on a neural network model, which can solve the problems in the prior art.

[0006] In the first aspect of the embodiments of the present invention, an optimization method for manufacturing titanium-steel composite plates based on a neural network model is provided, including: collecting interfacial data during the manufacturing process of titanium-steel composite plates, and generating interfacial bonding characteristics through spatio-temporal dimension feature extraction; inputting the interfacial bonding characteristics into an adaptive memory network, and predicting the interfacial bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; generating an optimized process parameter combination based on the interfacial bonding strength prediction result and process parameters through a double-delayed deep Q-network structure with a prioritized experience replay mechanism; using the optimized process parameter combination as an initial population, and generating an optimal process parameter scheme based on performance indicators and process stability by means of staged control at the phase change temperature point and integration of theory-difference gradient; feeding back the deviation between the actually measured performance indicator and the prediction result to the deep Q-network and genetic algorithm to dynamically adjust the process parameter optimization strategy.

[0007] In an alternative embodiment, inputting the interface bonding feature into an adaptive memory network and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit includes: inputting the interface bonding feature into the material constitutive equation library, which introduces an interface layer strengthening relation on the basis of the Johnson-Cook model, and calculating the stress-strain relation of the interface bonding layer; based on the stress-strain relation, using the coupling calculation unit of the temperature field-stress field to perform a separated solution to obtain the stress distribution of the interface layer; converting the stress distribution of the interface layer into a multi-resolution tensor field representation, extracting stress gradient features, stress field singularity distribution features, and temperature-stress covariance features and mapping them into a deep learning feature space; Inputting the deep learning feature into a long short-term memory network with a physical constraint embedding layer and an interface heterogeneity adaptive attention mechanism, and outputting the prediction result of the interface bonding strength.

[0008] In an alternative embodiment, the method further includes: the interface layer strengthening relation is: ; where is the interface layer stress, is the matrix stress, is the grain size, is the dislocation density, and are material constants used to calculate the stress-strain relation of the interface bonding layer; the coupling calculation unit includes: the heat conduction equation: ; where, is the material density, is the specific heat capacity; is the partial derivative of temperature with respect to time, representing the change rate of temperature with time; is the temperature, is the time, is the thermal conductivity, is the plastic work heat conversion coefficient, is the stress, is the strain rate, is the friction coefficient, is the interface pressure, is the relative slip velocity, is the gradient operator, is the divergence operator; the thermal stress mapping: ; where, is the thermal strain, is the thermal expansion coefficient, is the temperature, is the initial temperature; the stress field equilibrium equation: ; is the stress tensor, is the body force, is the divergence operator; the physical constraint embedding layer embeds the first law of thermodynamics and the continuity equation into the network structure as loss function terms. The interface heterogeneity adaptive attention mechanism calculates the spatial attention score based on the query vector and key vector at the interface position, and at the same time captures the temporal dependence of the interface features by combining the cell state and the hidden state, and outputs the prediction result of the interface bonding strength.

[0009] In an alternative embodiment, based on the interface bonding strength prediction result and process parameters, through a double-delayed deep Q-network structure with a prioritized experience replay mechanism, the optimized process parameter combination is generated, including: constructing a reinforcement learning state vector from the interface bonding strength prediction value, the process parameter set of the titanium-steel composite material, the trend of process parameter change, the target deviation of the interface bonding strength, and the process stability index; constructing an action space from the adjustment amount of the process parameter set, and constraining the maximum adjustment amplitude of the action space based on the process parameter safety window; constructing a basic reward function from the strength reward term, the stability reward term, and the efficiency reward term, and performing proportional adjustment through the physical constraint correction term to obtain a multi-objective reward function; mapping the state vector and the action space to the feature space through an encoder respectively and then fusing them to input into a double-delayed deep Q-network with a main network and two target networks with different update periods, and the two target networks update the network parameters with different time delays respectively; setting a prioritized experience replay mechanism based on the temporal difference error, where the temporal difference error is used to calculate the priority, and the priority is used to determine the sampling probability; based on the output value of the double-delayed deep Q-network, using an ε-greedy strategy to select the process parameter adjustment action, and performing constraint processing on the parameters that exceed the process parameter safety window, and outputting the optimized process parameter combination.

[0010] In an alternative embodiment, the multi-objective reward function includes: the strength reward term is calculated based on the deviation between the interface bonding strength target value and the predicted value, the stability reward term is calculated based on the ratio of the process parameter adjustment amount to the current parameter value, and the efficiency reward term is calculated based on the ratio of the process parameter adjustment amount to the maximum adjustment amplitude; calculating the influence coefficient of the intergranular reaction layer thickness based on the deviation relationship between the intergranular reaction layer thickness and the critical thickness and the optimal thickness; calculating the temperature gradient influence coefficient based on the deviation relationship between the interface temperature gradient and the optimal temperature gradient; calculating the thermal stress influence coefficient based on the ratio relationship between the thermal stress and the allowable stress; determining the constraint weight according to the sensitivity of the influence coefficient to the process parameters, and constructing a physical constraint correction term from the weighted sum of each influence coefficient and the corresponding constraint weight; performing proportional adjustment on the basic reward function through the physical constraint correction term to obtain a multi-objective reward function; according to the process stage characteristics, increasing the temperature gradient constraint weight in the heating stage, increasing the intergranular reaction layer thickness constraint weight in the insulation stage, and increasing the thermal stress constraint weight in the cooling stage.

[0011] In an alternative embodiment, taking the optimized process parameter combination as the initial population, adopting the method of phased control of the phase change temperature point and the fusion of theory-difference gradient, and generating the optimal process parameter scheme based on the performance index and process stability includes: taking the optimized process parameter combination as the initial population, obtaining the phase change temperature point of the titanium-steel composite material, calculating the current optimization stage according to the population distribution, where the optimization stage includes an exploration stage, a transition stage, and a refinement stage, setting corresponding gradient step size parameters and mutation probability parameters for different optimization stages; when the process parameter value is close to the phase change temperature point, reducing the perturbation amplitude of the crossover operation and improving the local search accuracy; in the non-dominated sorting process, calculating the theoretical gradient value of the process parameter on the performance index according to the phase change temperature point and the optimization stage, and obtaining the fusion gradient value by confidence weighted with the difference gradient value; the performance index includes interface bonding strength, tensile strength, and elongation; based on the population evolution process, calculating the diffusion sufficiency of the titanium-steel interface, the rationality of the interface layer thickness, and the brittle phase content respectively, determining the constraint violation degree, and calculating the influence degree of the process parameter fluctuation on the performance index; in each generation of population evaluation, obtaining the actual fluctuation value of the process parameter, calculating the process stability score in combination with the influence degree, and calculating the process cost score based on the energy consumption value, processing time, and equipment operation parameters; performing weighted calculation on the process stability score, process cost score, fusion gradient value, and constraint violation degree to obtain a comprehensive score, and selecting the process parameter combination with the highest comprehensive score as the optimization result.

[0012] In an alternative embodiment, obtaining the fusion gradient value by confidence weighted with the theoretical gradient value and the difference gradient value includes: calculating the change rate of element concentration with temperature and time during the interface diffusion process, the temperature sensitivity coefficient of the intermetallic compound phase fraction near the phase change temperature point, and the contribution coefficient of the interface microstructure characteristic parameters to the bonding strength, and obtaining the corresponding theoretical gradient; selecting the top N parameter combinations with the highest scores in the current population, applying a perturbation amount, and calculating the ratio of the change value of the performance index to the perturbation amount to obtain the difference gradient value; determining the empirical support degree based on the similarity degree between the process parameters and historical data, reducing the theoretical prediction credibility near the phase change temperature point, and determining the confidence interval in combination with the historical prediction error; increasing the difference gradient weight in the exploration stage to expand the search range, increasing the theoretical gradient weight in the refinement stage, implementing differential fusion on the performance index near the phase change temperature point, and obtaining the fusion gradient by confidence weighted with the theoretical gradient and the difference gradient.

[0013] In the second aspect of the embodiments of the present invention, there is provided an optimization system for manufacturing a titanium-steel composite plate based on a neural network model, including: A first unit, configured to collect interface data during the manufacturing process of the titanium-steel composite plate, and generate interface bonding characteristics through spatio-temporal dimension feature extraction; A second unit, configured to input the interface bonding features into an adaptive memory network, and predict the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; A third unit, configured to generate an optimized process parameter combination based on the interface bonding strength prediction result and process parameters through a double-delayed deep Q-network structure with a prioritized experience replay mechanism; A fourth unit, configured to use the optimized process parameter combination as an initial population, and generate an optimal process parameter solution based on performance indicators and process stability by means of phased control at the phase change temperature point and the integration of theory-difference gradient; A fifth unit, configured to feedback the deviation between the actually measured performance indicators and the prediction result to the deep Q-network and genetic algorithm, and dynamically adjust the process parameter optimization strategy.

[0014] In a third aspect of the embodiments of the present invention, there is provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] By introducing a material constitutive equation library with an interface layer strengthening relational expression and a temperature field-stress field coupling calculation unit, and combining an adaptive memory network with a physical constraint embedding layer, the present invention improves the accuracy and interpretability of interface bonding strength prediction. The double-delayed deep Q-network with a prioritized experience replay mechanism effectively balances exploration and exploitation in the process parameter optimization process, and improves the stability of parameter adjustment.

[0017] By means of the phased control strategy at the phase change temperature point and the theory-difference gradient fusion mechanism, the present invention realizes the precise regulation of process parameters near the phase change temperature point; based on the comprehensive evaluation system of process stability score and process cost score, the engineering practicability of the optimization result is ensured; combined with the dynamic feedback mechanism, the optimization strategies of the deep Q-network and genetic algorithm are continuously optimized, and the adaptive ability of the system is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a schematic flowchart of a titanium-steel composite plate manufacturing optimization method based on a neural network model according to an embodiment of the present invention; Figure 2 is a flowchart of fusion gradient value calculation. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0021] Figure 1 The following is a schematic flowchart of an optimization method for manufacturing titanium-steel composite plates based on a neural network model in an embodiment of the present invention, as Figure 1 shown. The method includes: collecting interface data during the manufacturing process of titanium-steel composite plates, and generating interface bonding features through spatio-temporal dimension feature extraction; inputting the interface bonding features into an adaptive memory network, and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; generating an optimized process parameter combination based on the interface bonding strength prediction result and process parameters through a double-delayed deep Q-network structure with a prioritized experience replay mechanism; using the optimized process parameter combination as an initial population, and generating an optimal process parameter solution based on performance indicators and process stability by means of phase transformation temperature point staged control and theory-difference gradient fusion; feeding back the deviation between the actually measured performance indicators and the prediction result to the deep Q-network and genetic algorithm to dynamically adjust the process parameter optimization strategy.

[0022] In an alternative embodiment, inputting the interface bonding features into an adaptive memory network and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit includes: inputting the interface bonding features into the material constitutive equation library, and the material constitutive equation library introduces an interface layer strengthening relational expression on the basis of the Johnson-Cook model to calculate the stress-strain relationship of the interface bonding layer; based on the stress-strain relationship, using a coupling calculation unit of the temperature field-stress field for separated solution to obtain the interface layer stress distribution; converting the interface layer stress distribution into a multi-resolution tensor field representation, extracting stress gradient features, stress field singular point distribution features, and temperature-stress covariance features and mapping them to a deep learning feature space; inputting the deep learning features into a long short-term memory network with a physical constraint embedding layer and an interface heterogeneity adaptive attention mechanism, and outputting the interface bonding strength prediction result; where the interface layer strengthening relational expression is: ; where is the interface layer stress, is the matrix stress, is the grain size, is the dislocation density, and are material constants used to calculate the stress-strain relationship of the interface bonding layer; The coupling calculation unit includes: Heat conduction equation: ; where is the material density, is the specific heat capacity; is the partial derivative of temperature with respect to time, representing the rate of change of temperature with time; is the temperature, is the time, is the thermal conductivity, is the plastic work heat conversion coefficient, is the stress, is the strain rate, is the friction coefficient, is the interface pressure, is the relative slip velocity, is the gradient operator, is the divergence operator; Thermal stress mapping: ; where is the thermal strain, is the coefficient of thermal expansion, is the temperature, is the initial temperature; Stress field equilibrium equation: ; is the stress tensor, is the body force, is the divergence operator; The physical constraint embedding layer is embedded in the network structure by constructing the first law of thermodynamics and the continuity equation as loss function terms. The interface heterogeneity adaptive attention mechanism calculates the spatial attention score based on the query vector and key vector at the interface position, and at the same time captures the temporal dependence relationship of the interface features by combining the cell state and hidden state, and outputs the prediction result of the interface bonding strength.

[0023] Exemplarily, interface data is collected during the manufacturing process of the titanium-steel composite plate, and dynamic features such as the temperature change rate, pressure fluctuation period, and diffusion rate are extracted in the time dimension; In the spatial dimension, static features such as the interface thickness distribution, grain size change, and element concentration gradient are extracted. These features together constitute the interface bonding feature set.

[0024] The material constitutive equation library uses an improved Johnson-Cook model and introduces an interface layer strengthening relationship on the basis of the original model. This relationship takes into account the effects of matrix stress, grain size, and dislocation density. The matrix stress is obtained through tensile tests, the grain size is obtained through image analysis, and the dislocation density is measured by the X-ray diffraction peak width method. Material constants and are calibrated through experiments. The specific operation is to conduct tensile tests under different process parameters and obtain the optimal constant values through curve fitting. For typical titanium-steel composite materials, ranges from 110 - 130 MPa·μm 0.5 , ranges from 0.15 - 0.25 MPa·m. During calculation, first input the basic parameters of the matrix material, including the yield strength (850 MPa), hardening coefficient (0.15), strain rate sensitivity index (0.025), and thermal softening index (1.1) of the titanium alloy, as well as the corresponding parameters of the steel (yield strength 520 MPa, hardening coefficient 0.18, strain rate sensitivity index 0.018, thermal softening index 1.05). Then superimpose the calculation results of the interface layer strengthening relationship with the results of the basic Johnson-Cook model to obtain the complete stress-strain relationship of the interface layer.

[0025] After obtaining the stress-strain relationship, a decoupled solution is carried out using a temperature field-stress field coupling calculation element to obtain the stress distribution of the interface layer. This calculation element mainly consists of three parts: the heat conduction equation, the thermal stress mapping, and the stress field equilibrium equation. During calculation, first set the initial temperature field (room temperature 25 °C) and boundary conditions (such as convective heat transfer boundary), and then calculate the temperature distribution at each time step according to the heat conduction equation. For typical titanium-steel composite plates, the material densities are taken as ρ_titanium = 4500 kg / m³ and ρ_steel = 7800 kg / m³, the specific heat capacities are taken as c_titanium = 544 J / (kg·K) and c_steel = 460 J / (kg·K), the thermal conductivities are taken as k_titanium = 21.9 W / (m·K) and k_steel = 50.2 W / (m·K), the plastic work heat conversion coefficient η is taken as 0.9, and the interface friction coefficient μ is taken as 0.3. The thermal stress mapping calculates the thermal strain generated according to the calculated temperature field in combination with the thermal expansion coefficients of the materials (8.6×10 -6 / K for the titanium alloy and 12×10 -6 / K for the steel). Finally, through the stress field equilibrium equation, combined with the boundary conditions (such as fixed constraints and load conditions), the stress distribution of the interface layer is solved. The calculation domain is divided into a grid of 100×50×20, the time step is 0.01 s, and the total calculation time is the entire manufacturing process.

[0026] The obtained interfacial layer stress distribution is transformed into a multi-resolution tensor field representation. The three-dimensional stress distribution data is reconstructed into a fourth-order tensor, where the first three dimensions of the tensor correspond to spatial coordinates, and the fourth dimension contains six independent stress components. Then, the tensor is decomposed into different resolution levels through wavelet transform, usually with 4-level decomposition, which respectively captures the stress characteristics at the macro, meso, and micro scales. Based on this, three types of key features are extracted: stress gradient features, stress field singular point distribution features, and temperature-stress covariance features. The stress gradient features are obtained by calculating the stress difference between adjacent grid points, with particular attention paid to the stress gradient changes near the interface. The stress field singular point distribution features are identified by detecting the mutation points and discontinuity points in the stress field, using a local extreme value detection algorithm, and setting the threshold to 20% of the average stress value. The temperature-stress covariance features are obtained by calculating the covariance matrix between the temperature field and each component of the stress field, reflecting the correlation between temperature changes and stress responses. The extracted features are processed by normalization and dimensionality reduction and mapped into a 128-dimensional deep learning feature space.

[0027] The deep learning features are input into a long short-term memory network with a physically constrained embedding layer and an interfacial heterogeneity adaptive attention mechanism to output the prediction result of the interfacial bonding strength. The network structure includes an input layer (128 nodes), a physically constrained embedding layer (64 nodes), an LSTM layer (two layers, each with 128 units), an interfacial heterogeneity adaptive attention layer, and an output layer (1 node). The physically constrained embedding layer constructs the first law of thermodynamics and the continuity equation as loss function terms. The specific implementation is to set the violation metrics of energy conservation and mass conservation, and when the network prediction result violates the physical laws, a corresponding penalty term is added. The penalty coefficient is set to 0.2 and tuned through the validation set. The interfacial heterogeneity adaptive attention mechanism calculates the spatial attention scores based on the query vector and key vector of the interface position. The number of attention heads is set to 8, and the feature dimension is 16. At the same time, it combines the cell state and hidden state to capture the temporal dependence relationship of the interface features. In the LSTM network, the cell state stores long-term memory, while the hidden state represents the output information at the current moment. The cell state retains the key information of the entire process of interface formation, while the hidden state reflects the interface state at the current moment. The information of the two states is weighted and fused, enabling the network to simultaneously consider the historical accumulation effect and the current instantaneous characteristics, so as to accurately identify the temporal evolution law of the interfacial bonding strength. The network training uses a batch size of 32, a learning rate of 0.001, the Adam optimizer, and the number of training epochs is 200. The early stopping method is used to prevent overfitting, and the training stops when the validation set loss does not decrease for 10 consecutive epochs.

[0028] Through the combination of multi-source data collection and deep learning methods, this technical solution has successfully achieved high-precision prediction of the interfacial bonding strength of titanium-steel composite plates. It not only considers the basic principles of materials science and mechanics, but also introduces advanced data processing and deep learning technologies, overcoming the limitations of traditional methods that rely on a large number of experiments and empirical knowledge. The high accuracy of the prediction results can effectively guide the optimization of process parameters, significantly improve production efficiency and product quality, reduce R & D costs, provide reliable technical support for the design and manufacturing of composite materials, and has important theoretical and practical values.

[0029] In an alternative embodiment, based on the interfacial bonding strength prediction results and process parameters, an optimized process parameter combination is generated through a double-delayed deep Q-network structure with a prioritized experience replay mechanism, including: constructing a reinforcement learning state vector from the interfacial bonding strength prediction value, the process parameter set of the titanium-steel composite material, the trend of process parameter changes, the target deviation of interfacial bonding strength, and the process stability index; constructing an action space from the adjustment amount of the process parameter set, and constraining the maximum adjustment amplitude of the action space based on the process parameter safety window; constructing a basic reward function from the strength reward term, the stability reward term, and the efficiency reward term, and obtaining a multi-objective reward function through proportional adjustment with a physical constraint correction term; respectively mapping the state vector and the action space to the feature space through an encoder and then fusing and inputting them into a double-delayed deep Q-network with a main network and two target networks with different update periods, and the two target networks update network parameters with different time delays; setting a prioritized experience replay mechanism based on the temporal difference error, where the temporal difference error is used to calculate the priority, and the priority is used to determine the sampling probability; based on the output value of the double-delayed deep Q-network, selecting a process parameter adjustment action using an ε-greedy strategy, and constraining the parameters that exceed the process parameter safety window, and outputting an optimized process parameter combination.

[0030] Exemplarily, the reinforcement learning state vector consists of five parts: the predicted value of the interface bonding strength, the set of process parameters of the titanium-steel composite material, the trend of process parameter changes, the target deviation of the interface bonding strength, and the process stability index. The predicted value of the interface bonding strength is provided by the aforementioned prediction. The set of process parameters includes parameters such as pressing temperature, pressing pressure, holding time, cooling rate, and interface cleanliness, and each parameter is normalized to the range of 0 - 1. For example, the pressing temperature range is 700 - 1000 °C, and it is normalized according to (actual temperature - 700) / (1000 - 700). The trend of process parameter changes is obtained by calculating the change rate of continuous 5 parameter adjustments, reflecting the dynamic adjustment direction and amplitude of the parameters. The target deviation of the interface bonding strength is calculated as the difference ratio between the predicted strength and the target strength (usually set to 350 MPa). For example, if the predicted strength is 320 MPa, the deviation is (320 - 350) / 350 = -8.57%. The process stability index includes three sub-indices: temperature fluctuation degree, pressure consistency, and interface diffusion uniformity. Among them, the temperature fluctuation degree is calculated as the ratio of the standard deviation of temperature to the average temperature during the manufacturing process, the pressure consistency is the coefficient of variation of the pressure in different regions of the interface, and the interface diffusion uniformity is characterized by the variance of the element distribution gradient. The dimension of the entire state vector is 50, and normalization is performed to ensure that the numerical ranges of all dimensions are the same.

[0031] The action space is composed of the adjustment amounts of the set of process parameters, including the pressing temperature adjustment amount (±50 °C), the pressing pressure adjustment amount (±5 MPa), the holding time adjustment amount (±2 minutes), the cooling rate adjustment amount (±2 °C / s), and the interface cleanliness adjustment amount (±10%). Each adjustment amount is discretized into 9 levels, from -4 to +4, corresponding to different adjustment amplitude percentages. For example, the pressing temperature adjustment amount level 2 corresponds to an increase of 20 °C. To ensure process safety, a process parameter safety window is set to restrict the maximum adjustment amplitude of the action space. The safety window is determined according to material characteristics and equipment limitations. For example, the pressing temperature is not lower than 750 °C and not higher than 950 °C, the pressing pressure is not lower than 10 MPa and not higher than 25 MPa, and the holding time is not less than 5 minutes and not more than 20 minutes. When the parameter adjustment amount output by the optimization algorithm causes the parameter to exceed the safety window, the system automatically limits the parameter value to the window boundary.

[0032] Construct a multi-objective reward function. After mapping the state vector and the action space to the feature space through the encoder respectively, they are fused and input into the double-delayed deep Q-network. The state encoder adopts a three-layer fully-connected neural network structure. The number of nodes in the input layer is 50 (corresponding to the dimension of the state vector), the number of nodes in the hidden layers are 128 and 64 respectively, and the number of nodes in the output layer is 32. The action encoder is a two-layer fully-connected network. The input dimension is the dimension of the process parameter adjustment amount (usually 5), the number of nodes in the hidden layer is 16, and the output dimension is 8. The encoded features are fused into a 40-dimensional vector through a concatenation operation and input into the double-delayed deep Q-network. This network consists of a main network and two target networks. The main network is used for current decision-making, and the two target networks are used to calculate the target Q-values. The network structure is a four-layer fully-connected network. The number of nodes in the hidden layers are 128, 256 and 128 respectively, and the number of nodes in the output layer is equal to the size of the action space. The two target networks update the network parameters with different time delays. The first target network is updated every 10 training iterations, and the second target network is updated every 20 iterations. This double-delayed structure effectively reduces the bias of Q-value estimation.

[0033] Based on the temporal difference error, a prioritized experience replay mechanism is set up. An experience replay pool is constructed to store the experience tuples generated during the reinforcement learning interaction process. Each tuple contains the current state, the executed action, the obtained reward, the next state, and the termination flag. The temporal difference error is calculated as the actually obtained reward plus the discount factor multiplied by the maximum Q-value of the next state, and then subtracted by the Q-value estimation corresponding to the current state action. This error reflects the gap between prediction and reality. The absolute value of the error plus a small constant (such as 0.01) is used as the priority of the experience. The higher the priority, the greater the sampling probability of the experience. The sampling probability is calculated as the α-th power of the priority (α is set to 0.6). At the same time, importance sampling weights are introduced to prevent overlearning of high-priority experiences. Each time during training, a batch of 64 experiences are sampled from the experience replay pool for network update.

[0034] Based on the output value of the double-delayed deep Q-network, an ε-greedy strategy is adopted to select the process parameter adjustment action. The initial ε value of the strategy is set to 0.9, indicating a 90% probability of random exploration and a 10% probability of selecting the action with the maximum Q-value. As the training progresses, the ε value decreases at a decay rate of 0.995, and the minimum is not less than 0.05 to ensure that the system maintains a certain exploration ability. After selecting the action, the parameters that exceed the safety window of the process parameters are constrained. When a certain parameter exceeds the lower limit of the safety window after adjustment, it is set to the lower limit value; when it exceeds the upper limit, it is set to the upper limit value. Finally, the optimized process parameter combination is output.

[0035] This technical solution realizes the intelligent optimization of the process parameters of titanium steel composite materials by integrating deep reinforcement learning with knowledge in the field of material manufacturing. Compared with the traditional parameter adjustment method, this solution can simultaneously consider multiple objectives such as interface bonding strength, process stability and production efficiency, greatly reducing the cost of manual trial and error and the optimization cycle. In particular, the introduction of the dual-delay Q network structure and the priority experience replay mechanism effectively improves the learning efficiency and decision-making accuracy, enabling the system to quickly adapt to different production conditions and material properties. This solution provides an efficient and intelligent solution for the optimization of process parameters of titanium steel composite materials and other composite materials, with significant practical value and promotion prospects.

[0036] In an optional embodiment, the multi-objective reward function includes: the strength reward item is calculated based on the deviation between the target value and the predicted value of the interface bonding strength, the stability reward item is calculated based on the ratio of the process parameter adjustment amount to the current parameter value, and the efficiency reward item is calculated based on the ratio of the process parameter adjustment amount to the maximum adjustment amplitude; the intergranular reaction layer thickness influence coefficient is calculated based on the deviation relationship between the intergranular reaction layer thickness and the critical thickness and the optimal thickness; the temperature gradient influence coefficient is calculated based on the deviation relationship between the interface temperature gradient and the optimal temperature gradient; the thermal stress influence coefficient is calculated based on the ratio relationship between thermal stress and allowable stress; the constraint weight is determined according to the sensitivity of the influence coefficient to the process parameters, and the physical constraint correction term is constructed by the weighted sum of each influence coefficient and the corresponding constraint weight; the basic reward function is proportionally adjusted by the physical constraint correction term to obtain a multi-objective reward function; according to the characteristics of the process stage, the temperature gradient constraint weight is increased in the heating stage, the intergranular reaction layer thickness constraint weight is increased in the insulation stage, and the thermal stress constraint weight is increased in the cooling stage.

[0037] Exemplarily, the strength bonus item is calculated based on the deviation between the target value and the predicted value of the interface bonding strength. In the specific implementation, the target interface bonding strength is set to 350MPa, and the predicted value is provided by the aforementioned interface bonding strength prediction model. When the predicted strength is lower than the target value, the bonus value decreases in proportion to the predicted value and the target value; when the predicted strength is equal to the target value, the bonus value reaches the maximum (set to 1.0); when the predicted strength exceeds the target value, the bonus will also increase appropriately, but the increase is limited to avoid excessive optimization and material performance redundancy. For example, when the predicted strength is 320MPa, the strength bonus is calculated as 1.0-(350-320) / 350=0.914; when the predicted strength is 370MPa, the strength bonus is calculated as 1.0+0.2×(370-350) / 350=1.114, where 0.2 is the gain coefficient after exceeding the target, to prevent excessive pursuit of high strength and ignoring other goals.

[0038] The stability reward term is calculated based on the ratio of the process parameter adjustment amount to the current parameter value. This term aims to encourage smooth parameter adjustment and avoid process instability caused by drastic fluctuations in parameters. When calculating, first calculate the percentage of the adjustment amplitude for each process parameter, that is, the ratio of the adjustment amount to the current parameter value. For example, if the current pressing temperature is 830 °C and the adjustment amount is +15 °C, then the percentage of the adjustment amplitude is 15 / 830 = 1.81%. Then calculate the weighted average of the percentage of the adjustment amplitude of all parameters, and the weights are determined according to the sensitivity of the parameters to the interface bonding strength. Usually, the weights of temperature and pressure are relatively high (0.3 each), the holding time and cooling rate are the second (0.15 each), and the interface cleanliness is relatively low (0.1). Finally, substitute the weighted average into a decreasing function. When the average adjustment amplitude is 0, the stability reward is the largest (1.0); when the average adjustment amplitude increases, the stability reward decreases. Specifically, an exponential decreasing function is used. When the average adjustment amplitude is 5%, the stability reward drops to 0.6; when it reaches 10%, it drops to 0.25.

[0039] The efficiency reward term is calculated based on the ratio of the process parameter adjustment amount to the maximum adjustment amplitude. This term mainly considers the impact of process parameter adjustment on production efficiency and encourages improving production efficiency on the premise of ensuring quality. For key parameters that affect production efficiency, such as the holding time and cooling rate, calculate the ratio of their adjustment amount to the maximum allowable adjustment amplitude. For example, if the current value of the holding time is 12 minutes, the adjustment amount is -2 minutes, and the maximum adjustment amplitude is ±5 minutes, then the ratio is (-2) / 5 = -0.4. For adjustments that shorten the holding time or increase the cooling rate (in line with the direction of improving efficiency), the efficiency reward increases linearly according to the ratio; for adjustments that extend the holding time or slow down the cooling rate, the efficiency reward decreases linearly. For other parameters such as temperature and pressure, a small reward is given when the adjustment is beneficial to efficiency improvement, and a small penalty is given otherwise. Finally, the efficiency rewards of each parameter are weighted and summarized to form the efficiency reward term.

[0040] After the basic reward function is constructed, it is adjusted by the physical constraint correction term. The physical constraint correction term takes into account three key physical factors: the thickness of the intergranular reaction layer, the interfacial temperature gradient, and the thermal stress. Based on the deviation relationship between the thickness of the intergranular reaction layer and the critical thickness and the optimal thickness, the influence coefficient of the intergranular reaction layer thickness is calculated. The thickness of the intergranular reaction layer of the titanium-steel composite material has a significant impact on the interfacial strength, and there is an optimal thickness range. Through metallurgical analysis, it is determined that when the reaction layer is too thin (less than 2 μm), the interfacial bonding is insufficient; when the reaction layer is too thick (more than 15 μm), the interface becomes brittle and the strength decreases. The optimal thickness range is 5 - 8 μm. The influence coefficient is calculated using a piecewise function: when the thickness is within the optimal range, the coefficient is 1.0; when the thickness deviates from the optimal range but does not reach the critical value, the coefficient decreases linearly; when the thickness is below the critical lower limit or exceeds the critical upper limit, the coefficient rapidly drops below 0.3. For example, when the reaction layer thickness is 10 μm, the influence coefficient is 1.0 - (10 - 8) / (15 - 8)×0.7 = 0.84.

[0041] Based on the deviation relationship between the interfacial temperature gradient and the optimal temperature gradient, the influence coefficient of the temperature gradient is calculated. An excessive temperature gradient at the titanium-steel composite interface will lead to thermal stress concentration, while too small a temperature gradient is not conducive to interfacial diffusion and bonding. Through thermodynamic analysis, it is determined that the optimal temperature gradient range is 20 - 40 °C / mm. The calculation method of the influence coefficient is similar to that of the intergranular reaction layer, using a piecewise function: when the temperature gradient is within the optimal range, the coefficient is 1.0; when it deviates from the optimal range, the coefficient gradually decreases; when the temperature gradient exceeds 80 °C / mm or is lower than 10 °C / mm, the coefficient drops below 0.2. For example, when the temperature gradient is 50 °C / mm, the influence coefficient is 1.0 - (50 - 40) / (80 - 40)×0.8 = 0.8.

[0042] Based on the ratio relationship between the thermal stress and the allowable stress, the influence coefficient of the thermal stress is calculated. Excessive thermal stress will cause interface cracking or deformation, affecting the bonding quality. According to the analysis of material mechanics, the upper limit of the allowable thermal stress in the titanium-steel interface area is about 200 MPa. The influence coefficient is calculated using an exponentially decreasing function: when the ratio of the thermal stress to the allowable stress is 0, the coefficient is 1.0; when the ratio is 0.5, the coefficient is 0.9; when the ratio is 0.8, the coefficient is 0.7; when the ratio is 1.0, the coefficient is 0.5; when the ratio exceeds 1.0, the coefficient rapidly drops below 0.2. For example, when the thermal stress is 160 MPa, the ratio is 160 / 200 = 0.8, and the influence coefficient is 0.7.

[0043] Determine the constraint weights according to the sensitivity of process parameters to influence coefficients, and construct a physical constraint correction term by the weighted sum of each influence coefficient and the corresponding constraint weight. Different process parameters have different influence degrees on the three physical factors, so differential constraint weights are set. For example, the pressing temperature has a greater influence on the thickness of the intergranular reaction layer and the temperature gradient, and their constraint weights are 0.4 and 0.4 respectively, and has a smaller influence on the thermal stress, with a weight of 0.2; the pressing pressure has a medium influence on the thickness of the intergranular reaction layer, with a weight of 0.3, has a smaller influence on the temperature gradient, with a weight of 0.2, and has a greater influence on the thermal stress, with a weight of 0.5. Calculate the weighted physical constraint influence of each process parameter to obtain the physical constraint correction term.

[0044] Perform proportional adjustment on the basic reward function through the physical constraint correction term to obtain the final multi-objective reward function. The basic reward function is the weighted sum of the strength reward term, the stability reward term and the efficiency reward term, with weights of 0.5, 0.3 and 0.2 respectively. The physical constraint correction term is used as a multiplicative factor to adjust the basic reward to ensure that the reward function conforms to physical laws. For example, when the basic reward value is 0.85 and the physical constraint correction term is 0.9, the final reward value is 0.85×0.9 = 0.765. To adapt to the characteristics of different process stages, dynamically adjust the physical constraint weights. In the heating stage, the temperature gradient has a significant influence on the interface bonding, so the temperature gradient constraint weight is increased to 1.5 times the standard value; in the heat preservation stage, the formation of the intergranular reaction layer is the key process, and the constraint weight of the intergranular reaction layer thickness is increased to 1.5 times the standard value; in the cooling stage, the thermal stress concentration is the main risk, and the thermal stress constraint weight is increased to 1.5 times the standard value. This dynamic weight adjustment mechanism enables the reward function to accurately reflect the key objectives of different process stages and improve the optimization effect.

[0045] For example, in the heat preservation stage of the manufacturing process of a certain titanium-steel composite material, the current thickness of the intergranular reaction layer is 4μm, the temperature gradient is 45℃ / mm, and the thermal stress is 120MPa. The calculated influence coefficient of the intergranular reaction layer thickness is 0.95, the influence coefficient of the temperature gradient is 0.9, and the influence coefficient of the thermal stress is 0.8. Considering that it is in the heat preservation stage, the constraint weight of the intergranular reaction layer thickness is increased from the standard value of 0.4 to 0.6, and the constraint weights of the temperature gradient and the thermal stress remain at the standard values. Finally, the calculated physical constraint correction term is 0.6×0.95 + 0.25×0.9 + 0.15×0.8 = 0.915. If the basic reward value is 0.82 at this time, the final reward value is 0.82×0.915 = 0.75.

[0046] This multi-objective reward function construction method scientifically evaluates the process parameter optimization process of titanium-steel composite materials by comprehensively considering three objectives: interface bonding strength, process stability, and production efficiency, and dynamically adjusts them in combination with key physical constraint factors. By deeply integrating materials science and reinforcement learning, it can not only guide the system to find the optimal process parameter combination but also ensure that the optimization results comply with physical laws and process requirements. Compared with traditional single-objective optimization methods, this method takes into account production efficiency and process stability while improving material properties, significantly reducing the trial-and-error costs during the optimization process.

[0047] In an alternative embodiment, taking the optimized process parameter combination as the initial population, a method of phase transformation temperature point staged control and theory-difference gradient fusion is adopted to generate the optimal process parameter scheme based on performance indicators and process stability, including: taking the optimized process parameter combination as the initial population, obtaining the phase transformation temperature point of the titanium-steel composite material, calculating the current optimization stage according to the population distribution, where the optimization stage includes an exploration stage, a transition stage, and a refinement stage, and setting corresponding gradient step size parameters and mutation probability parameters for different optimization stages; when the process parameter value is close to the phase transformation temperature point, reducing the perturbation amplitude of the crossover operation and increasing the local search accuracy; during the non-dominated sorting process, calculating the theoretical gradient value of the process parameter with respect to the performance indicator according to the phase transformation temperature point and the optimization stage, and obtaining the fusion gradient value by confidence weighting with the difference gradient value; the performance indicators include interface bonding strength, tensile strength, and elongation; based on the population evolution process, calculating the titanium-steel interface diffusion sufficiency, interface layer thickness rationality, and brittle phase content respectively to determine the constraint violation degree, and calculating the influence degree of process parameter fluctuations on the performance indicators; in each generation of population evaluation, obtaining the actual fluctuation value of the process parameter, calculating the process stability score in combination with the influence degree, and calculating the process cost score based on the energy consumption value, processing time, and equipment operation parameters; weighting and calculating the process stability score, process cost score, fusion gradient value, and constraint violation degree to obtain a comprehensive score, and selecting the process parameter combination with the highest comprehensive score as the optimization result.

[0048] Exemplarily, the process parameter combinations optimized by reinforcement learning are used as the initial population. This initial population contains multiple groups of process parameter combinations, and each group includes key parameters such as pressing temperature, pressing pressure, holding time, cooling rate, and interface cleanliness. To ensure population diversity, 20 sets of parameter variants are generated around the benchmark combination. The variant generation rule is to add perturbations according to a Gaussian distribution based on the benchmark parameters, and the perturbation amplitude is ±5% of the parameter value. For example, if the benchmark pressing temperature is 865 °C, the temperature range of the variants is approximately distributed between 822 - 908 °C. After the initial population is constructed, the phase transformation temperature points of the titanium-steel composite material are obtained through the material database, including key temperature points such as the α→β phase transformation temperature of the titanium alloy (about 882 °C), the austenite transformation temperature of the steel (about 723 °C), and the activation temperature of the interface diffusion reaction (about 750 °C). These temperature points serve as important reference bases during the optimization process.

[0049] The optimization process is divided into three stages: the exploration stage, the transition stage, and the refinement stage. The current stage is judged by calculating the dispersion degree of the population parameters. The specific method is to calculate the coefficient of variation (the ratio of the standard deviation to the mean) of each process parameter in the population. When the coefficient of variation is greater than 0.15, it is judged as the exploration stage; when the coefficient of variation is between 0.08 - 0.15, it is judged as the transition stage; when the coefficient of variation is less than 0.08, it is judged as the refinement stage. In the exploration stage, a relatively large gradient step size parameter (0.1) and a relatively high mutation probability (0.3) are set to encourage the algorithm to search in a large range; in the transition stage, the gradient step size parameter is reduced to 0.05 and the mutation probability is reduced to 0.15; in the refinement stage, the gradient step size parameter (0.02) and the mutation probability (0.05) are further reduced to concentrate resources for fine search near the optimal solution.

[0050] Particularly, when the process parameter values are close to the phase transformation temperature points, more cautious parameter adjustment is required. For example, when the pressing temperature is close to 882 °C (within the range of ±10 °C), the α→β phase transformation of the titanium alloy occurs, and at this time, the interface reaction mechanism changes significantly. To prevent the algorithm from crossing this sensitive area and missing the optimal solution, the perturbation amplitude of the crossover operation is reduced to half of the standard value, and the local search accuracy is improved. When the parameter value enters the range of ±1.5% of the phase transformation temperature point, the parameter perturbation coefficient of the crossover operator is reduced from the standard value of 0.2 to 0.1, and at the same time, 10 additional sampling points are added in this area for fine evaluation.

[0051] During the non-dominated sorting process, according to the phase transition temperature point and the optimization stage, the theoretical gradient values of process parameters on performance indicators are calculated, and the confidence-weighted sum of the differential gradient values is obtained to get the fusion gradient value. First, based on material science theory, the theoretical gradient values of each process parameter on performance indicators are calculated, such as the special change law of the theoretical gradient near the phase transition temperature point. At the same time, the individuals with higher scores in the population are selected, and the differential gradient values are calculated by applying small perturbations. The weight configuration is adjusted according to the optimization stage. For the area near the phase transition temperature point, the credibility of the theoretical gradient is reduced, and the weight of the local differential gradient is increased to improve adaptability. The performance indicator evaluation includes three aspects: interface bonding strength, tensile strength, and elongation. The target value of the interface bonding strength is set to 350 MPa and obtained through the aforementioned interface bonding strength prediction model; the target value of the tensile strength is set to 620 MPa; the target value of the elongation is set to 15%. The weights of the three indicators are set according to application requirements. Usually, the weight of the interface bonding strength is the highest (0.5), followed by the tensile strength (0.3), and the elongation is the lowest (0.2).

[0052] Based on the population evolution process, process constraint indicators are calculated. The diffusion sufficiency of the titanium-steel interface is calculated, which reflects the integrity of element diffusion at the interface. The ratio of the estimated penetration depth of elements to the ideal depth is obtained through a diffusion model, and the complete sufficiency is 1.0. For example, when kept at a pressure of 15 minutes at 865 °C, the diffusion depth of titanium elements on the steel side of the interface is about 5 μm, and the ideal diffusion depth is 6 μm, then the sufficiency is 0.83. Secondly, the rationality of the interface layer thickness is calculated. According to the aforementioned optimal thickness range of the interface layer (5 - 8 μm), the compliance of the actual thickness with the optimal range is evaluated. Thirdly, the brittle phase content is calculated, mainly focusing on the formation amount of brittle intermetallic compounds such as TiFe2, and the lower the content, the better. Finally, the constraint violation degree is determined according to the deviation degree of the three indicators, and the penalty function method is used to convert the constraint violation into a fitness penalty.

[0053] Calculate the influence degree of process parameter fluctuations on performance indicators. In actual production, it is difficult to precisely control process parameters, and there are certain fluctuations. Through sensitivity analysis, the influence degree of each parameter fluctuation on performance is evaluated. For example, the influence of the pressing temperature fluctuation of ±5 °C on the interface bonding strength is about ±3 MPa, while the influence of the pressing pressure fluctuation of ±1 MPa is about ±2 MPa. Based on the sensitivity analysis results, a sensitivity matrix of process parameters - performance indicators is established to quantify the influence weights of different parameter fluctuations.

[0054] In each generation of population evaluation, the actual fluctuation values of process parameters are obtained, and the process stability score is calculated by combining the influence degree. The actual fluctuation values can be obtained through industrial data collection or simulation estimation. For example, the temperature control fluctuation of a typical production line is about ±3°C, and the pressure control fluctuation is about ±0.5 MPa. The process stability score is calculated as the negative exponential function of the weighted sum of the fluctuations of each parameter, and the score range is 0 - 1, where a higher score indicates greater stability. At the same time, the process cost score is calculated based on the energy consumption value, processing time, and equipment operation parameters. The energy consumption mainly considers the heating energy consumption (related to temperature and time) and the pressing energy consumption (related to pressure and time); the processing time includes the heating-up time, holding pressure time, and cooling time; the equipment operation parameters include the equipment load rate and the maintenance cost factor. The process cost score is also normalized to the range of 0 - 1, where a higher score indicates lower cost.

[0055] Finally, the process stability score, process cost score are weighted with the fusion gradient value and the degree of constraint violation to obtain the comprehensive score. The weighting coefficients are set according to actual production requirements. Usually, the weight of performance indicators is 0.5, the weight of process stability is 0.3, and the weight of process cost is 0.2. The degree of constraint violation is deducted from the total score as a penalty term. The process parameter combination with the highest comprehensive score is selected as the final optimization result.

[0056] Regularly collect the actual performance indicators of the titanium-steel composite plate and calculate the deviation from the model prediction value. The deviation data is used to update the experience replay pool of the deep Q-network, adjust the reward function weight, and correct the state transition model. At the same time, the deviation data is used to adjust the fitness function of the genetic algorithm, update the theoretical-difference weight ratio in the gradient fusion, and calibrate the prediction model near the phase transition temperature point. For example, a global parameter update is performed every 50 batches of materials produced, and a local fine-tuning is performed every 10 batches to continuously align the prediction model with the actual process.

[0057] For example, the initial process parameters of a titanium-steel composite plate production line are: pressing temperature 865°C, pressing pressure 20 MPa, holding pressure time 15 minutes, cooling rate 4°C / s, interface cleanliness 90%, and interface bonding strength 348 MPa. After optimization by this method, the optimal process parameters are adjusted to: pressing temperature 878°C (close to but not exceeding the phase transition temperature of the titanium alloy), pressing pressure 21 MPa, holding pressure time 16 minutes, cooling rate 4.5°C / s, and interface cleanliness 92%. The interface bonding strength after optimization is increased to 356 MPa, the tensile strength reaches 628 MPa, and the elongation is 16.2%. All performance indicators meet or exceed the target values. At the same time, the process stability score is 0.91 (showing a significant improvement compared to the initial 0.85), and the process cost score is 0.88 (slightly lower than the initial 0.9, indicating a slight increase in cost but within an acceptable range). The comprehensive score is 0.925, which is significantly higher than the initial plan's 0.87.

[0058] This technical solution realizes the fine optimization of the process parameters of titanium-steel composite materials by combining the evolutionary algorithm and material science theory. The innovation of this method lies in introducing the phased control of the phase transition temperature point and the theory-difference gradient fusion mechanism, enabling the optimization process to fully consider material characteristics and phase transition laws. By comprehensively evaluating performance indicators, process stability, and process costs, multi-objective balanced optimization is achieved. Compared with traditional methods that rely solely on experience or a single algorithm, this solution significantly improves the optimization accuracy and efficiency, reduces the trial-and-error cost, and at the same time ensures the engineering feasibility of the optimization results, providing a systematic and intelligent solution for the process optimization of titanium-steel composite materials and other similar composite materials.

[0059] In an optional implementation manner, obtaining the fusion gradient value by performing confidence weighting on the theoretical gradient value and the difference gradient value includes: calculating the change rate of element concentration with temperature and time during the interface diffusion process, the temperature sensitivity coefficient of the intermetallic compound phase fraction near the phase transition temperature point, and the contribution coefficient of the interface microstructure characteristic parameters to the bonding strength, and obtaining the corresponding theoretical gradient; selecting the top N parameter combinations with the highest scores in the current population, applying a perturbation amount, and calculating the ratio of the change value of the performance indicator to the perturbation amount to obtain the difference gradient value; determining the empirical support degree based on the similarity between the process parameters and historical data, reducing the credibility of theoretical predictions near the phase transition temperature point, and determining the confidence interval in combination with historical prediction errors; increasing the difference gradient weight in the exploration stage to expand the search range, increasing the theoretical gradient weight in the refinement stage, implementing differential fusion of the performance indicators near the phase transition temperature point, and obtaining the fusion gradient by performing confidence weighting on the theoretical gradient and the difference gradient.

[0060] Exemplarily, in combination with Figure 2 the flowchart of the fusion gradient value calculation is described as follows: For the interface diffusion process, the modified Fick's second law is used to describe the concentration distribution of elements at the interface. The relationship between the diffusion coefficient of titanium element in the steel matrix and temperature follows the Arrhenius formula. In the typical process temperature range (800 - 900 °C), the diffusion coefficient of titanium element in steel is about 1×10-12 to 5×10-11 m² / s. Based on this, the change rate of titanium element concentration with temperature is calculated. At 850 °C, for every 10 °C increase in temperature, the increase rate of the titanium element concentration gradient at the interface is about 0.033 μm-1, and the corresponding increase in diffusion depth is about 0.5 μm / 15 min. Combining the quantitative relationship between the diffusion depth and the bonding strength (for every 1 μm increase in diffusion depth, the bonding strength increases by 10 - 15 MPa until saturation), the theoretical gradient of temperature on strength is obtained.

[0061] Regarding the temperature sensitivity coefficient of the intermetallic compound phase fraction near the phase transition temperature point, the changes near the α→β phase transition temperature (about 882 °C) of the titanium alloy are focused on. Through thermodynamic calculations and phase diagram analysis, the temperature sensitivity coefficients of each phase are quantitatively determined: in the range of 870 - 880 °C, the temperature sensitivity coefficient of the TiFe phase fraction is 0.2% / °C, and in the range of 880 - 890 °C, the temperature sensitivity coefficient of the TiFe2 phase fraction is 0.3% / °C.

[0062] The contribution coefficients of the interface microstructure characteristic parameters to the bonding strength are quantitatively determined through microstructure analysis and mechanical tests. In the titanium-steel composite material, the interface layer is composed of a solid solution layer, a diffusion layer, and an intermetallic compound layer. Through scanning electron microscopy and transmission electron microscopy analysis, combined with nanoindentation tests, the contribution coefficients of each parameter are quantified: when the interface layer thickness is in the range of 5 - 8 μm, the contribution coefficient is 25 MPa / μm; when the TiFe phase fraction is in the range of 30 - 40%, the contribution coefficient is 3 MPa / %; when the TiFe2 phase fraction exceeds 15%, the contribution coefficient is -5 MPa / %. Based on these quantitative relationships, the theoretical gradient values of each process parameter for the performance index are calculated.

[0063] For the different characteristics of the process parameters, the theoretical gradients are calculated separately. The theoretical gradient of the pressing temperature on the interface strength is about 0.8 MPa / °C in the range of 800 - 870 °C, increases to 1.2 MPa / °C in the range of 870 - 882 °C, and then decreases to 0.5 MPa / °C or even becomes negative after exceeding 882 °C. The theoretical gradient of the pressing pressure on the interface strength is about 1.5 MPa / MPa in the range of 15 - 25 MPa, and the gradient decreases after exceeding 25 MPa. The theoretical gradient of the holding pressure time on the interface strength shows a non-linear relationship. The gradient is relatively large (about 3 MPa / minute) in the initial stage (5 - 10 minutes), decreases in the middle stage (10 - 20 minutes) (about 1 MPa / minute), and approaches zero in the later stage (>20 minutes). The theoretical gradient of the cooling rate on the interface strength is about -0.5 MPa / (°C / s) in the range of 1 - 5 °C / s, indicating that appropriately reducing the cooling rate is beneficial to reducing thermal stress and improving the bonding strength.

[0064] Select the top N parameter combinations with the highest scores in the current population, apply a perturbation amount, and calculate the ratio of the change value of the performance index to the perturbation amount to obtain the differential gradient value. The value of N is usually set to 20% of the population size (for example, if the population size is 50, then N = 10). For each selected parameter combination, apply a small perturbation to a single process parameter while keeping other parameters unchanged, and then calculate the performance change through the aforementioned interface strength prediction model. The perturbation amount is set to ±1% of the standard range of the parameter. For example, for the pressing temperature of 865 °C, the perturbation amount is ±8.65 °C. To reduce the influence of random errors, each parameter is subjected to 3 positive perturbations and 3 negative perturbations, and the average effect is taken. The differential gradient value is calculated as the ratio of the change value of the performance index to the perturbation amount. For example, if the interface strength increases by 7.2 MPa when the temperature increases by 8.65 °C, the differential gradient value is 7.2 / 8.65 = 0.83 MPa / °C.

[0065] For the multi-dimensional parameter space, the orthogonal design method is used to reduce the number of experiments. Each of the N high-scoring combinations is regarded as a center point, and a fractional factorial orthogonal experiment is constructed around it. Only two levels (the center point value ± the perturbation amount) are considered for each process parameter. Calculate the main effects and interaction effects of each parameter through the results of the orthogonal experiment, so as to obtain a more accurate differential gradient value matrix. In addition, to improve the calculation efficiency, cluster analysis is performed on the gradient values in similar parameter regions, and regions with a similarity exceeding 95% share the same gradient value.

[0066] Determine the empirical support degree based on the similarity between the process parameters and historical data, reduce the credibility of the theoretical prediction near the phase change temperature point, and combine the historical prediction error to determine the confidence interval. First, construct a historical database of process parameter - performance index, including valid records from past experiments and production. For the current process parameter combination to be evaluated, calculate its Euclidean distance from the historical data points, select the nearest K historical data points (K is usually set to 5), and calculate the mean and standard deviation of their prediction errors. The empirical support degree is calculated as the negative exponential function of the distance. The smaller the distance, the higher the support degree. For example, when the Euclidean distance is 0.1 in the standardized parameter space, the empirical support degree is 0.9; when the distance is 0.3, the support degree drops to 0.7. In particular, for the parameter region near the phase change temperature point, the credibility of the theoretical prediction needs to be adjusted. By analyzing the prediction accuracy near the phase change temperature point in the historical data, a credibility adjustment function is established. For example, for the α→β phase change temperature (882 °C) of titanium alloy, within the range of ±5 °C, the average error of the theoretical prediction increases by about 50%, so the credibility of the theoretical gradient is reduced to 0.6 times that of the standard region. Similarly, for the austenite transformation temperature (723 °C) and the interface diffusion activation temperature (750 °C) of steel, similar credibility adjustments are also applied. Combining the historical prediction error and the characteristics of the phase change region, determine the confidence interval of the theoretical gradient value, usually expressed as the range of ±2 times the standard deviation.

[0067] At different stages of optimization, a differential gradient fusion strategy with different weights is adopted. The differential gradient weight is set to 0.7 and the theoretical gradient weight is 0.3 during the exploration stage; both weights are 0.5 during the transition stage; the differential gradient weight is reduced to 0.3 and the theoretical gradient weight is increased to 0.7 during the refinement stage. This dynamic weight allocation strategy takes advantage of the differential gradient in exploring new regions and the accuracy of the theoretical gradient in fine optimization.

[0068] Differential fusion is implemented for performance indicators near the phase transition temperature point. Within the range of ±3% from the phase transition temperature point, a more conservative fusion strategy is adopted, increasing the detection point density and reducing the step size. For example, near the α→β phase transition temperature (882°C) of titanium alloy, the temperature step size is reduced from the standard 5°C to 2°C, and at the same time, the weight of the local differential gradient is increased in the gradient fusion in this region to improve adaptability. When the differential gradient and the theoretical gradient are in the same direction, a larger learning step size is adopted; when the two gradient directions are inconsistent, a smaller step size is adopted and the detection point density is increased to avoid missing the optimal solution.

[0069] The final calculation formula for the fused gradient value is the weighted sum of the theoretical gradient and the differential gradient: Fused Gradient = W_theoretical × Theoretical Gradient × Credibility + W_differential × Differential Gradient, where W_theoretical and W_differential are the weights of the theoretical gradient and the differential gradient respectively, and W_theoretical + W_differential = 1. The credibility is a theoretical gradient credibility factor adjusted based on empirical support and the phase transition region.

[0070] This technical solution realizes the intelligent fusion of gradient information by combining theoretical models with actual data, and provides precise optimization guidance especially for the complex material behavior near the phase transition temperature point; effectively overcomes the limitations of single-gradient methods, and improves the optimization efficiency and accuracy; the theoretical gradient provides global guidance based on material science theory, while the differential gradient captures local behaviors that may be ignored by the theory in the actual system. The combination of the two achieves a balance between exploration and refinement of the optimization algorithm; especially the handling of special regions such as near the phase transition temperature point reflects the in-depth consideration of material characteristics by this method, making the optimization results more in line with actual process requirements.

[0071] In the second aspect of the embodiments of the present invention, an optimization system for manufacturing titanium-steel composite plates based on a neural network model is provided, including: A first unit for collecting interface data during the manufacturing process of titanium-steel composite plates and generating interface bonding characteristics through spatio-temporal dimension feature extraction; A second unit for inputting the interface bonding characteristics into an adaptive memory network and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; The third unit is configured to generate an optimized process parameter combination based on the interface bonding strength prediction result and process parameters through a double-delayed deep Q-network structure with a prioritized experience replay mechanism; The fourth unit is configured to use the optimized process parameter combination as an initial population and generate an optimal process parameter scheme based on performance indicators and process stability by means of phased control of the phase transition temperature point and the integration of theory and differential gradient; The fifth unit is configured to feedback the deviation between the actually measured performance indicator and the prediction result to the deep Q-network and genetic algorithm to dynamically adjust the process parameter optimization strategy.

[0072] In a third aspect of the embodiments of the present invention, there is provided an electronic device, including: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the foregoing method.

[0073] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the foregoing method is implemented.

[0074] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are uploaded.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A neural network optimization method for the manufacture of titanium-steel composite plates, characterized in that, Including: Collecting interface data during the manufacturing process of titanium-steel composite plates, and generating interface bonding characteristics through spatio-temporal dimension feature extraction; Inputting the interface bonding characteristics into an adaptive memory network, and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; Based on the predicted result of the interface bonding strength and process parameters, generating an optimized process parameter combination through a double-delayed deep Q-network structure with a prioritized experience replay mechanism; Taking the optimized process parameter combination as the initial population, and generating an optimal process parameter solution based on performance indicators and process stability by means of phase change temperature point staged control and theory-difference gradient fusion; Feeding back the deviation between the actually measured performance indicators and the predicted results to the deep Q-network and genetic algorithm, and dynamically adjusting the process parameter optimization strategy.

2. The method according to claim 1, wherein Inputting the interface bonding characteristics into an adaptive memory network, and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit includes: inputting the interface bonding characteristics into the material constitutive equation library, the material constitutive equation library introducing an interface layer strengthening relation formula on the basis of the Johnson-Cook model to calculate the stress-strain relation of the interface bonding layer; based on the stress-strain relation, using a coupling calculation unit of the temperature field-stress field to perform a separated solution to obtain the interface layer stress distribution; converting the interface layer stress distribution into a multi-resolution tensor field representation, extracting stress gradient features, stress field singular point distribution features and temperature-stress covariance features and mapping them to a deep learning feature space; inputting the deep learning features into a long short-term memory network with a physical constraint embedding layer and an interface heterogeneity adaptive attention mechanism, and outputting the predicted result of the interface bonding strength.

3. The method according to claim 2, wherein The method further includes: The interface layer strengthening relation formula is: ; where is the interfacial layer stress, is the matrix stress, is the grain size, is the dislocation density, and are material constants used to calculate the stress-strain relationship of the interface bonding layer; The coupling calculation unit includes: Heat conduction equation: ; where, is the material density, is the specific heat capacity; is the partial derivative of temperature with respect to time, representing the rate of change of temperature with time; is the temperature, is the time, is the thermal conductivity, is the plastic work heat conversion coefficient, is the stress, is the strain rate, is the friction coefficient, is the interface pressure, is the relative slip velocity, is the gradient operator, is the divergence operator; Thermal stress mapping: ; wherein, is the thermal strain, is the coefficient of thermal expansion, is the temperature, is the initial temperature; the stress field equilibrium equation: ; is the stress tensor, is the body force, is the divergence operator; the physical constraint embedding layer is embedded into the network structure by constructing the first law of thermodynamics and the continuity equation as loss function terms, and the interface heterogeneity adaptive attention mechanism calculates the spatial attention score based on the query vector and key vector at the interface position, and at the same time captures the temporal dependence relationship of the interface features by combining the cell state and the hidden state, and outputs the prediction result of the interface bonding strength.

4. The method according to claim 1, wherein Based on the predicted result of the interface bonding strength and process parameters, generating an optimized process parameter combination through a double-delayed deep Q-network structure with a prioritized experience replay mechanism includes: constructing a reinforcement learning state vector from the predicted value of the interface bonding strength, the process parameter set of the titanium-steel composite material, the process parameter change trend, the interface bonding strength target deviation and the process stability index, constructing an action space from the adjustment amount of the process parameter set, and constraining the maximum adjustment amplitude of the action space based on the process parameter safety window; constructing a basic reward function from the strength reward term, the stability reward term and the efficiency reward term, and performing proportional adjustment through a physical constraint correction term to obtain a multi-objective reward function; respectively mapping the state vector and the action space to the feature space through an encoder and then fusing and inputting them into a double-delayed deep Q-network with a main network and two target networks with different update periods, and the two target networks respectively updating the network parameters with different time delays; setting a prioritized experience replay mechanism based on the temporal difference error, the temporal difference error being used to calculate the priority, and the priority being used to determine the sampling probability; based on the output value of the double-delayed deep Q-network, selecting a process parameter adjustment action by using an ε-greedy strategy, and performing constraint processing on the parameters exceeding the process parameter safety window, and outputting an optimized process parameter combination.

5. The method according to claim 4, wherein The multi-objective reward function includes: the intensity reward term is calculated based on the deviation between the interface bonding strength target value and the predicted value, the stability reward term is calculated based on the ratio of the process parameter adjustment amount to the current parameter value, and the efficiency reward term is calculated based on the ratio of the process parameter adjustment amount to the maximum adjustment range; based on the deviation relationship between the intergranular reaction layer thickness and the critical thickness and the optimal thickness, the influence coefficient of the intergranular reaction layer thickness is calculated; based on the deviation relationship between the interface temperature gradient and the optimal temperature gradient, the temperature gradient influence coefficient is calculated; based on the ratio relationship between the thermal stress and the allowable stress, the thermal stress influence coefficient is calculated; the constraint weights are determined according to the sensitivity of the influence coefficients to the process parameters, and the weighted sum of each influence coefficient and the corresponding constraint weight is used to construct a physical constraint correction term; the physical constraint correction term is used to perform proportional adjustment on the basic reward function to obtain the multi-objective reward function; according to the characteristics of the process stage, the temperature gradient constraint weight is increased in the heating stage, the intergranular reaction layer thickness constraint weight is increased in the holding stage, and the thermal stress constraint weight is increased in the cooling stage.

6. The method according to claim 1, characterized in that Taking the optimized process parameter combination as the initial population, adopting the method of phase transformation temperature point staged control and theory-difference gradient fusion, and generating the optimal process parameter scheme based on the performance index and process stability includes: taking the optimized process parameter combination as the initial population, obtaining the phase transformation temperature point of the titanium-steel composite material, calculating the current optimization stage according to the population distribution, and the optimization stage includes an exploration stage, a transition stage and a refinement stage, and corresponding gradient step size parameters and mutation probability parameters are set for different optimization stages; when the process parameter value is close to the phase transformation temperature point, the perturbation amplitude of the crossover operation is reduced and the local search accuracy is improved; in the non-dominated sorting process, according to the phase transformation temperature point and the optimization stage, the theoretical gradient value of the process parameter to the performance index is calculated, and the confidence weighted sum with the difference gradient value is used to obtain the fusion gradient value; the performance index includes interface bonding strength, tensile strength and elongation; based on the population evolution process, the diffusion sufficiency of the titanium-steel interface, the rationality of the interface layer thickness and the brittle phase content are calculated respectively, the constraint violation degree is determined, and the influence degree of the process parameter fluctuation on the performance index is calculated; in each generation of population evaluation, the actual fluctuation value of the process parameter is obtained, and the process stability score is calculated by combining the influence degree, and the process cost score is calculated based on the energy consumption value, processing time and equipment operation parameters; the process stability score, the process cost score, the fusion gradient value and the constraint violation degree are weighted and calculated to obtain the comprehensive score, and the process parameter combination with the highest comprehensive score is selected as the optimization result.

7. The method according to claim 6, wherein Obtaining a fused gradient value by performing confidence weighting on the theoretical gradient value and the differential gradient value includes: calculating the change rates of element concentrations with respect to temperature and time during the interface diffusion process, the temperature sensitivity coefficient of the intermetallic compound phase fraction near the phase transformation temperature point, and the contribution coefficient of the interface microstructure characteristic parameters to the bonding strength to obtain the corresponding theoretical gradient; selecting the top N parameter combinations with the highest scores in the current population, applying a perturbation amount, and calculating the ratio of the change value of the performance index to the perturbation amount to obtain the differential gradient value; determining the empirical support degree based on the similarity between the process parameters and historical data, reducing the credibility of theoretical predictions near the phase transformation temperature point, and determining the confidence interval in combination with historical prediction errors; increasing the differential gradient weight in the exploration stage to expand the search range, increasing the theoretical gradient weight in the refinement stage, implementing differential fusion of the performance indexes near the phase transformation temperature point, and weighting the theoretical gradient and the differential gradient according to the confidence degree to obtain the fused gradient.

8. An optimization system for manufacturing titanium-steel composite plates based on a neural network model, which is used to implement the method described in any one of the foregoing claims 1-7, characterized in that, Including: A first unit for collecting interface data during the manufacturing process of the titanium-steel composite plate, and generating interface bonding characteristics through spatio-temporal dimension feature extraction; A second unit for inputting the interface bonding characteristics into an adaptive memory network, and predicting the interface bonding strength based on a material constitutive equation library and a temperature field-stress field coupling calculation unit; A third unit for generating an optimized process parameter combination through a double-delay deep Q-network structure with a priority experience replay mechanism based on the interface bonding strength prediction result and the process parameters; A fourth unit for using the optimized process parameter combination as an initial population, and generating an optimal process parameter solution based on the performance index and process stability by adopting a method of controlling in stages at the phase transformation temperature point and fusing the theory and the differential gradient; A fifth unit for feeding back the deviation between the actually measured performance index and the prediction result to the deep Q-network and the genetic algorithm, and dynamically adjusting the process parameter optimization strategy.

9. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images