Electronic packaging shell sintering control method and system based on tunnel type sintering furnace

Abstract

The invention provides an electronic packaging shell sintering control method and system based on a tunnel type sintering furnace. The method comprises the following steps that historical sintering data and historical finished product performance data of an electronic packaging shell by the tunnel type sintering furnace are obtained; constructing a sintering dynamic model of the electronic packaging shell in combination with the historical sintering data and the historical finished product performance data; obtaining sintering environment data of the electronic packaging shell in the sintering process through a plurality of sensors preset in the tunnel type sintering furnace; the sintering environment data are input into the sintering dynamic model, and a feedback control instruction of the tunnel type sintering furnace is generated through a model prediction control method; and sintering control parameters of the tunnel type sintering furnace are adjusted according to the feedback control instruction. The method has the effects of improving the efficiency of the sintering process and the quality of finished products and reducing the cost of the sintering process.

Description

Technical Field

[0001] The invention belongs to the technical field of electronic packaging component processing, and in particular relates to a sintering control method and system for an electronic packaging shell based on a tunnel-type sintering furnace. Background Art

[0002] As electronic products develop towards miniaturization, integration and high performance, the quality requirements for key components such as electronic packaging shells are getting higher and higher. At the same time, the global emphasis on energy efficiency and environmental protection has also driven the manufacturing industry to seek more energy-saving and environmentally friendly production methods. In this context, traditional sintering technology faces huge challenges. It is necessary to continuously improve production efficiency and reduce energy consumption and costs while ensuring product quality.

[0003] The traditional sintering process mainly relies on manual experience and fixed parameter settings, and lacks the ability to monitor and dynamically adjust the sintering process in real time. This method is difficult to cope with the impact of factors such as raw material batch differences and environmental changes, resulting in large fluctuations in product quality. The sintering process involves complex physical and chemical reactions, and traditional methods are difficult to accurately capture and control these reaction processes, which affects the stability of product performance. On this basis, existing technologies often adopt a single or simple control strategy, which makes it difficult to fully consider the synergistic effects of multiple factors in the sintering process, limiting the optimization space of the sintering process. On the other hand, traditional methods have high skill requirements for equipment operators, and the influence of human factors is relatively large, which increases the difficulty and cost of production management. Summary of the invention

[0004] The present invention provides a sintering control method and system for an electronic packaging shell based on a tunnel-type sintering furnace, so as to solve the problems of high cost and low efficiency of a traditional sintering process.

[0005] In a first aspect, the present invention provides a method for controlling sintering of an electronic packaging shell based on a tunnel sintering furnace, the method comprising the following steps:

[0006] Acquire historical sintering data and historical finished product performance data of the electronic packaging shell in the tunnel sintering furnace, wherein the historical sintering data includes historical sintering temperature data, historical sintering atmosphere data and historical sintering pressure data, and the historical finished product performance data is the performance data of the electronic packaging shell sintered based on the historical sintering data;

[0007] Combining the historical sintering data and the historical finished product performance data to construct a sintering dynamic model of the electronic packaging shell;

[0008] Acquiring sintering environment data of the electronic packaging shell during the sintering process through a plurality of sensors pre-arranged in the tunnel-type sintering furnace, wherein the sintering environment data includes sintering temperature data, sintering atmosphere data and sintering pressure data;

[0009] Inputting the sintering environment data into the sintering dynamic model, and generating feedback control instructions for the tunnel sintering furnace by a model predictive control method;

[0010] The sintering control parameters of the tunnel sintering furnace are adjusted according to the feedback control instruction.

[0011] Optionally, the combining the historical sintering data and the historical finished product performance data to construct the sintering dynamic model of the electronic packaging shell comprises the following steps:

[0012] Preprocessing the historical sintering data and the historical finished product performance data;

[0013] Combining the pre-processed historical sintering data and the historical finished product performance data, constructing a sintering physical model of the electronic packaging housing during the sintering process;

[0014] Building a sintering data association model based on a neural network model, wherein the sintering data association model is used to capture the nonlinear relationship between various physical quantities in the sintering physical model;

[0015] Synchronously performing data enhancement processing on the preprocessed historical sintering data and the historical finished product performance data to obtain a sintering data set and a corresponding performance data set;

[0016] Training the sintering data association model using the sintering data set and the performance data set until all model parameters in the sintering data association model are optimal model parameters;

[0017] The sintering physical model and the trained sintering data association model are combined into a sintering dynamic model of the electronic packaging shell.

[0018] Optionally, the expression of the sintering physical model is as follows:

[0019]

[0020] Where: ρ represents the material density of the electronic packaging shell, c p represents specific heat capacity, T represents the historical sintering temperature data, t represents the sintering time, k(T) represents thermal conductivity, Q total represents the total heat source term, ΔH r represents the heat of reaction, C i represents the atmosphere concentration of the i-th gas component in the historical sintering atmosphere data, D i (T, P) represents the diffusion coefficient of the i-th gas component as a function of temperature and pressure, P represents the historical sintering pressure data, u represents the gas flow rate in the tunnel sintering furnace, R i (T, P, Ci ) represents the reaction rate of the i-th gas component, K ρ (T, P) represents the rate constant of density change, ρ max represents the maximum theoretical material density of the electronic packaging shell, k j represents the total amount of atmosphere in the jth sintering reaction of the electronic packaging shell in the tunnel sintering furnace, ξ j represents the reaction progress of the jth sintering reaction, υ j (T, P) represents the rate constant of the jth sintering reaction, represents the stoichiometric coefficient of the i-th gas component in the j-th sintering reaction, K p (T, ρ) represents the pressure transmission coefficient, Q p (T,C i ) represents the stressor term.

[0021] Optionally, the total heat source term includes the heat density of the heating source of the tunnel sintering furnace, and also includes the radiation heat loss, convection heat loss and thermal bridge effect loss in the tunnel sintering furnace. The total heat source term is expressed as the heat density of the heating source minus the sum of the radiation heat loss, the convection heat loss and the thermal bridge effect loss.

[0022] Optionally, the sintering data association model includes a spatial feature extraction module, a temporal feature extraction module, a fully connected module, a residual connection module and an attention mechanism module, the spatial feature extraction module, the temporal feature extraction module and the fully connected module are connected in sequence, the residual connection module is configured in the fully connected module, and the attention mechanism module is configured in the spatial feature extraction module and the temporal feature extraction module.

[0023] Optionally, the step of inputting the sintering environment data into the sintering dynamic model and generating feedback control instructions for the tunnel sintering furnace by a model predictive control method comprises the following steps:

[0024] Inputting the sintering environment data into the sintering dynamic model, and predicting the future product performance of the electronic packaging housing currently in the tunnel sintering furnace through the sintering dynamic model;

[0025] Creating performance control constraint conditions of the sintering dynamic model in combination with the future product performance and the preset product sintering performance index of the electronic packaging shell;

[0026] Based on the performance control constraint conditions and with the goal of optimizing the future product performance, an optimization objective function of an optimization algorithm is generated, and optimal sintering environment data is calculated using the optimization algorithm;

[0027] Obtaining a current sintering control sequence of the tunnel sintering furnace at the current moment;

[0028] Based on a preset first mapping relationship between the sintering environment data and the sintering control parameters in the tunnel-type sintering furnace, generating an optimal sintering control sequence of the tunnel-type sintering furnace according to the optimal sintering environment data;

[0029] A feedback control instruction of the tunnel sintering furnace is generated according to the distinguishing features of the control parameters in the optimal sintering control sequence and the current sintering control sequence.

[0030] Optionally, the optimization algorithm is a particle swarm optimization algorithm or a genetic optimization algorithm.

[0031] Optionally, an image acquisition device is pre-configured at the end of the conveyor belt of the tunnel sintering furnace, and the method further comprises the following steps:

[0032] After the sintering control parameters are adjusted, the image acquisition device is used to acquire a finished product image of the finished electronic packaging shell at the end of the conveyor belt;

[0033] Identify multiple defect types in the finished product image by image recognition technology, and count the number of defect types of each different defect type;

[0034] If there is a target defect type among all the defect types whose number exceeds a preset number threshold, generating an optimization constraint condition of the optimization algorithm based on a preset second mapping relationship between the sintering environment data and the defect type;

[0035] In combination with the optimization constraint and the performance control constraint and with the goal of optimizing the future product performance, the optimization algorithm is used to update the optimal sintering environment data, and the feedback control instructions of the tunnel sintering furnace are updated through the updated optimal sintering environment data;

[0036] The sintering control parameters of the tunnel sintering furnace are adjusted according to the updated feedback control instructions.

[0037] Optionally, the defect types include temperature abnormality defects, atmosphere abnormality defects and pressure abnormality defects. The temperature abnormality defects include at least one of shell size abnormality defects, grain abnormality defects, thermal stress cracking defects, shell phase change defects, melt deformation defects and thermal shock crack defects. The atmosphere abnormality defects include at least one of shell oxidation defects, shell carbonization defects, shell abnormal porosity defects, surface impurity deposition defects and color abnormality defects. The pressure abnormality defects include at least one of shell deformation defects, shell crack defects and shell delamination defects.

[0038] In a second aspect, the present invention further provides an electronic packaging shell sintering control system based on a tunnel-type sintering furnace, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the electronic packaging shell sintering control method based on a tunnel-type sintering furnace as described in the first aspect is implemented.

[0039] The beneficial effects of the present invention are:

[0040] The present invention realizes precise control and optimization of the sintering process by establishing a sintering dynamic model and real-time monitoring of sintering environmental data, greatly improving the consistency and stability of product quality. This can not only reduce the product defect rate, but also meet increasingly stringent quality requirements and enhance the market competitiveness of enterprises. Secondly, the present invention uses the model predictive control method to generate feedback control instructions, which can realize the dynamic adjustment of sintering parameters, effectively deal with the impact of factors such as raw material batch differences and environmental changes, and improve the flexibility and adaptability of the production process. Through the analysis and utilization of historical data, the present invention can continuously optimize the sintering process and achieve continuous improvement, which is of great significance to improving production efficiency and reducing costs. In addition, the intelligent control method reduces the dependence on manual experience, reduces the influence of human factors, and improves the stability and reliability of production. At the same time, precise parameter control also helps to optimize energy utilization, reduce energy consumption, and meet the development requirements of energy conservation and environmental protection. The implementation of the present invention has accumulated a large amount of valuable production data and models for enterprises, laid a foundation for subsequent process innovation and product development, and is conducive to the improvement of long-term competitiveness of enterprises. In general, the technical solution of the present invention not only solves the problems existing in the current sintering process, but also provides a feasible solution for the intelligent transformation of the manufacturing industry, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of a sintering control method for an electronic packaging shell based on a tunnel-type sintering furnace in one embodiment of the present application.

[0042] Figure 2 This is a schematic diagram of a process for building a sintering dynamic model of an electronic packaging shell by combining historical sintering data and historical finished product performance data in one embodiment of the present application.

[0043] Figure 3 FIG. 1 is a schematic diagram of a process for adjusting sintering control parameters based on image recognition in one embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0045] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0046] Figure 1 FIG. 1 is a flow chart of a method for controlling the sintering of an electronic packaging shell based on a tunnel sintering furnace in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps in the above method may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, the electronic packaging shell sintering control method based on the tunnel sintering furnace disclosed in the present invention specifically includes the following steps:

[0047] S101. Obtain historical sintering data and historical finished product performance data of electronic packaging shells using a tunnel sintering furnace.

[0048] Among them, the historical sintering data includes historical sintering temperature data, historical sintering atmosphere data and historical sintering pressure data, and the historical finished product performance data is the performance data of the electronic packaging shell sintered based on the historical sintering data. First, various sensors, such as temperature sensors, gas composition analyzers and pressure sensors, need to be installed at key positions of the tunnel sintering furnace to record various parameters in the sintering process in real time. These sensors usually use high-precision, high-stability industrial-grade equipment to ensure the accuracy and reliability of the data. The data acquisition system records these parameters at a preset sampling frequency (for example, once per second) and transmits the data to the central control system. For historical sintering temperature data, the temperature change curves of different areas in the furnace may be recorded, including the temperature of the preheating zone, sintering zone and cooling zone; historical sintering atmosphere data may include changes in the concentration of gases such as oxygen, nitrogen, and hydrogen; historical sintering pressure data may include the pressure in the furnace and the pressure data applied to the electronic packaging shell. At the same time, it is also necessary to record the production parameters of each batch of products, such as raw material batches, sintering time, etc.

[0049] For historical finished product performance data, it is necessary to conduct product quality testing after sintering. This includes mechanical strength testing, air tightness testing, surface roughness testing, dimensional accuracy measurement, etc. of electronic packaging shells. These test data correspond one-to-one with the corresponding sintering parameters to form a complete data set. The data preprocessing stage includes data cleaning (removing outliers and noise), data standardization (unifying data of different dimensions to the same scale), and data classification (classifying by product type, batch, etc.). For example, the moving average method can be used to remove short-term fluctuations in temperature data, and the Z-score method can be used to standardize different parameters.

[0050] S102. Construct a sintering dynamic model of the electronic packaging shell by combining historical sintering data and historical finished product performance data.

[0051] Among them, the acquired historical data needs to be preprocessed first, including data cleaning, normalization and standardization to ensure data quality and consistency. Based on the preprocessed data, a sintering physical model of the electronic packaging shell during the sintering process is constructed. This physical model is usually based on physical laws such as heat conduction, mass transfer and momentum conservation, and can be described by a set of partial differential equations. For example, the heat conduction equation can describe the change of the temperature field, while the diffusion equation can describe the distribution of atmosphere components. At the same time, complex processes such as phase change and chemical reaction of materials need to be considered. Then, a sintering data association model is constructed based on the neural network model to capture the nonlinear relationship between various physical quantities in the sintering physical model. This association model may use convolutional neural network (CNN) to extract spatial features, long short-term memory network (LSTM) to extract temporal features, and enhance the expression and generalization capabilities of the model through structures such as fully connected layers, residual connections and attention mechanisms.

[0052] In order to train this association model, it is necessary to perform data enhancement processing on the preprocessed historical data, such as adding noise, rotation, scaling, etc., to generate a larger sintering data set and a corresponding performance data set. These data sets are used to train the sintering data association model, and the model parameters are continuously adjusted through the back propagation algorithm until the optimal state is reached. Finally, the sintering physical model and the trained sintering data association model are merged to form a comprehensive sintering dynamic model. This dynamic model can take into account both physical laws and capture implicit patterns in the data, thereby more accurately describing and predicting the sintering process.

[0053] S103. Acquire sintering environment data of the electronic packaging shell during the sintering process through a plurality of sensors pre-installed in the tunnel-type sintering furnace.

[0054] Among them, the sintering environment data includes sintering temperature data, sintering atmosphere data and sintering pressure data. Sensors are usually distributed in different areas of the sintering furnace, including preheating area, high temperature area and cooling area. The temperature sensor may use a thermocouple or an infrared thermometer to monitor the temperature changes at different locations in the sintering furnace in real time. For example, multiple thermocouples may be set in the high temperature area to record temperature data at a sampling frequency of 0.1 seconds to ensure that rapid temperature fluctuations can be captured. Atmosphere sensors may include oxygen concentration sensors, humidity sensors, etc., which are used to monitor the gas composition and content in the sintering furnace. These sensors may be based on electrochemical principles or spectral analysis technology and can respond quickly to atmosphere changes. For example, the oxygen concentration sensor may update data once a second with an accuracy of ±0.1%. The pressure sensor is used to monitor the pressure conditions in the sintering furnace. It may use a piezoelectric or capacitive sensor to detect small pressure changes.

[0055] The sensor data is usually collected in real time by the data acquisition system, and preliminary data processing such as filtering and calibration is performed to ensure the accuracy and reliability of the data. In addition, some auxiliary sensors may be set up, such as flow sensors for monitoring gas flow, displacement sensors for monitoring the dimensional changes of electronic packaging shells, etc. The data of these sensors together constitute the sintering environment data, including sintering temperature data, sintering atmosphere data and sintering pressure data.

[0056] S104. Input the sintering environment data into the sintering dynamic model, and generate feedback control instructions for the tunnel sintering furnace through the model predictive control method.

[0057] Among them, the sintering environment data acquired in real time is input into the sintering dynamic model constructed earlier. This model will predict the performance change trend of the electronic packaging shell in the future based on the current environmental data. For example, the model may predict the changes in performance indicators such as density and strength of the shell in the next 30 minutes. Next, compare these predicted future product performance with the preset product sintering performance indicators to create performance control constraints. These constraints may include specific indicators such as density not less than 98% of the theoretical density and strength not less than 300MPa. Based on these constraints, and with the goal of optimizing future product performance, construct the objective function of the optimization algorithm. This objective function may be a multi-objective optimization problem that requires balancing multiple performance indicators.

[0058] For example, the objective function may be in the form of: J = w1*(1-ρ / ρth)^2+w2*(σmax-σ)^2+w3*(L-L0)^2, where ρ represents density, σ represents strength, L represents size, and w1, w2, and w3 are weight coefficients. Then, an optimization algorithm (such as a particle swarm optimization algorithm or a genetic algorithm) is used to solve this optimization problem and obtain the optimal sintering environment data. The optimization algorithm may require multiple iterative calculations, and each iteration will update the particle position or gene encoding until a solution that satisfies the constraints and minimizes the objective function value is found. After obtaining the optimal sintering environment data, specific feedback control instructions can be generated through a series of mapping relationship conversions.

[0059] S105. Adjust the sintering control parameters of the tunnel sintering furnace according to the feedback control instruction.

[0060] Among them, this step first needs to parse the feedback control instructions generated in the previous step and convert them into specific control parameter adjustments. For example, if the feedback control instruction indicates that the temperature of a certain area needs to be increased by 2°C, then the corresponding increase in heating power needs to be calculated. This conversion process may involve complex thermodynamic calculations, and factors such as the heat capacity and heat loss of the sintering furnace need to be considered. At the same time, the mutual influence between the control parameters needs to be considered. For example, changes in temperature may affect the distribution of the atmosphere, so the gas flow rate may need to be adjusted at the same time. When the adjustment is actually performed, a gradual adjustment strategy is usually adopted, that is, the total adjustment amount is decomposed into multiple small steps and implemented step by step. This can avoid system instability caused by sudden and large adjustments. For example, if the temperature needs to be increased by 10°C, it may be divided into 5 times, each time by 2°C, and a certain time interval is left between each adjustment so that the system can reach a new equilibrium state.

[0061] For atmosphere control, it may be necessary to accurately control the flow ratio of multiple gases, which requires the use of a high-precision mass flow controller. Pressure control may involve adjusting the power of the exhaust system or adjusting the opening of the pressure balance valve. During the adjustment process, it is also necessary to monitor the changes in various parameters in real time to ensure that the adjustment effect meets expectations and to promptly detect and handle possible abnormalities. For example, if the temperature rise rate in a certain area is found to be abnormal during the temperature adjustment process, it may be necessary to stop the adjustment immediately and perform fault diagnosis. In addition, the timeliness of the control parameter adjustment needs to be considered, that is, according to the position and movement speed of the electronic packaging shell in the sintering furnace, the parameters are adjusted predictively to ensure that the shell is in the best sintering environment when passing through the critical area. The effect of this step is that the results of the optimization calculation can be accurately applied to the actual production process, and the sintering process can be accurately controlled, thereby improving product quality, reducing energy consumption, and improving production efficiency.

[0062] In one embodiment, referring to Figure 2 , combining historical sintering data and historical finished product performance data to build a sintering dynamic model for electronic packaging shells includes the following steps:

[0063] S201. Preprocessing historical sintering data and historical finished product performance data;

[0064] S202. Combining the historical sintering data after pretreatment and the historical finished product performance data, constructing a sintering physical model of the electronic packaging shell during the sintering process;

[0065] S203. Constructing a sintering data association model based on a neural network model, the sintering data association model is used to capture the nonlinear relationship between various physical quantities in the sintering physical model;

[0066] S204. Performing data enhancement processing on the preprocessed historical sintering data and the historical finished product performance data simultaneously to obtain a sintering data set and a corresponding performance data set;

[0067] S205. Training the sintering data association model using the sintering data set and the performance data set until all model parameters in the sintering data association model are optimal model parameters;

[0068] S206. Combining the sintering physical model and the trained sintering data association model into a sintering dynamic model of the electronic packaging shell.

[0069] In this embodiment, the process of preprocessing historical sintering data and historical finished product performance data involves multiple sub-steps, including data cleaning, normalization and standardization. Data cleaning mainly solves the problems of missing values, outliers and duplicate values ​​in the data. For missing values, a variety of methods can be used for processing, such as mean filling, median filling or predictive filling using machine learning algorithms. For example, if the temperature data at a certain time point is missing, the temperature mean of the time points before and after the location can be used to fill it. Outlier processing usually uses statistical methods such as the 3σ principle or the box plot method for identification and processing. Duplicate values ​​are directly deleted or averaged. Normalization processing unifies data of different dimensions to the same scale, and commonly used methods include minimum-maximum normalization and Z-score standardization. For example, for sintering temperature data, the possible value range is 800°C to 1200°C, which can be mapped to a range of 0 to 1 through minimum-maximum normalization. In addition, operations such as data format unification and time alignment are also required to ensure that data from different sources can be correctly associated.

[0070] Sintering physical models are usually based on thermodynamics and kinetics theory, including processes such as mass transfer, heat transfer and phase change. First, a heat conduction equation needs to be established to describe the distribution and evolution of the temperature field. In three-dimensional space, the heat conduction equation can be expressed as: Where ρ is density, c is specific heat capacity, T is temperature, t is time, k is thermal conductivity, and Q is the internal heat source term. For electronic packaging shells, it may be necessary to consider the anisotropy of the material, in which case k becomes a tensor. Secondly, a mass transfer equation needs to be established to describe the diffusion process of the substance. Fick's law can be used to describe this process: Where C is the concentration and D is the diffusion coefficient. During the sintering process, D is usually a function of temperature and can be expressed as D = D 0 exp(-Q / RT)), where D 0 is the frequency factor, Q is the diffusion activation energy, and R is the gas constant. These equations usually need to be numerically solved by the finite element method or the finite difference method. The constructed sintering physical model can describe the temperature distribution, density change, phase change process, etc. of the electronic packaging shell during the sintering process, providing a theoretical basis for subsequent optimization control.

[0071] In this embodiment, the expression of the sintering physical model is as follows:

[0072]

[0073] Where: ρ represents the material density of the electronic packaging shell, c p represents specific heat capacity, T represents historical sintering temperature data, t represents sintering time, k(T) represents thermal conductivity, Q total represents the total heat source term, ΔH rrepresents the heat of reaction, C i represents the atmosphere concentration of the i-th gas component in the historical sintering atmosphere data, D i (T, P) represents the diffusion coefficient of the i-th gas component as a function of temperature and pressure, P represents the historical sintering pressure data, u represents the gas flow rate in the tunnel sintering furnace, R i (T, P, C i ) represents the reaction rate of the i-th gas component, K ρ (T, P) represents the rate constant of density change, ρ max Indicates the maximum theoretical material density of the electronic packaging shell, k j represents the total amount of atmosphere in the jth sintering reaction of the electronic packaging shell in the tunnel sintering furnace, ξ j represents the reaction progress of the jth sintering reaction, υ j (T, P) represents the rate constant of the jth sintering reaction, represents the stoichiometric coefficient of the i-th gas component in the j-th sintering reaction, K p (T, ρ) represents the pressure transmission coefficient, Q p (T,C i ) represents the stressor term.

[0074] In the above sintering physical model, the specific derivation formula of the reaction rate constant is as follows:

[0075]

[0076] Among them: A j represents the frequency factor of reaction j, E j represents the activation energy of reaction j, R represents the gas constant, b j represents the pressure influence coefficient of reaction j.

[0077] The density change rate constant and pressure conductivity coefficient can be derived by referring to the derivation formula of the reaction rate constant. The reaction rate of gas components is affected by temperature, pressure and concentration at the same time. The specific value can be obtained through experiments or by collecting online information. The pressure source term is caused by chemical reaction or phase change. The phase change equation is as follows:

[0078]

[0079] Where: φ i is the volume fraction of phase i, v is the velocity field, Γ i is the generation / consumption rate of phase i.

[0080] In the above sintering physical model, the influencing equation of sintering pressure data is as follows:

[0081]

[0082] Where: P is the local pressure, P 0 is the reference pressure, g is the acceleration due to gravity, ρ is the current density, h is the height, μ is the dynamic viscosity, and u is the gas flow rate.

[0083] In this embodiment, the total heat source term includes the heat density of the heating source of the tunnel sintering furnace, and also includes the radiation heat loss, convection heat loss and thermal bridge effect loss in the tunnel sintering furnace. The total heat source term is expressed as the sum of the heat density of the heating source minus the radiation heat loss, convection heat loss and thermal bridge effect loss. Specifically, the total heat source term is expressed as follows:

[0084] Q total =Q heat -Q rad -Q consυ -Q bridge

[0085] Where: Q heat is the heating source term, Q rad is the radiation heat loss, Q conυ is the convective heat loss, Q bridge is the loss due to thermal bridge effect.

[0086] Heating source term:

[0087] Q heat =ηP input g(x, y, z)

[0088] Where: η is the heating efficiency, P input is the input power density and g(x, y, z) is the spatial distribution function.

[0089] Radiative heat loss:

[0090]

[0091] Where: ∈ is the emissivity, σ is the Stefan-Boltzmann constant, T surr is the ambient temperature and F is the viewing factor.

[0092] Convective heat loss:

[0093] Q conυ =h(T)(TT gas )

[0094] Where: h(T) is the temperature-dependent convective heat transfer coefficient, T gas is the gas temperature.

[0095] Thermal bridge loss:

[0096]

[0097] Where: k bridge (T) is the equivalent thermal conductivity of the thermal bridge (which may vary with temperature), is the temperature gradient, A bridge is the equivalent cross-sectional area of ​​the thermal bridge, and V is the volume of the tunnel sintering furnace that produces the thermal bridge effect. In practical applications, it may be necessary to determine k through experiments or detailed thermal analysis. bridge (T) and A bridge These parameters may vary depending on the furnace design and material. In addition, the impact of thermal bridges may vary at different stages of the sintering process, so the time dependence of these parameters may need to be considered in the model.

[0098] In this embodiment, the sintering data association model is constructed based on the neural network model to capture the nonlinear relationship between various physical quantities in the sintering physical model and supplement the complex interactions that may be ignored by the pure physical model. The sintering data association model includes a spatial feature extraction module, a temporal feature extraction module, a fully connected module, a residual connection module and an attention mechanism module. The spatial feature extraction module, the temporal feature extraction module and the fully connected module are connected in sequence, the residual connection module is configured in the fully connected module, and the attention mechanism module is configured in the spatial feature extraction module and the temporal feature extraction module. The input layer of the sintering data association model can include sintering parameters such as temperature, pressure, and atmosphere composition, and the output layer is the predicted product performance indicators, such as density, strength, etc.

[0099] Specifically, in this embodiment, the structure of the spatial feature extraction module is as follows:

[0100] 3 convolutional layers, each followed by ReLU activation function and batch normalization. The convolutional layers are used to extract spatial features, such as the spatial pattern of temperature distribution;

[0101] 2 maximum pooling layers, the pooling layer is used to reduce the dimension and extract the main features;

[0102] 1 spatial attention layer (i.e., the attention mechanism module embedded in the spatial feature extraction module) is used to implement spatial attention in the spatial feature extraction module.

[0103] The structure of the temporal feature extraction module is as follows:

[0104] 2-layer long short-term memory (LSTM) network, each followed by a dropout layer. The network structure of LSTM can be: input layer → Conv1D → MaxPooling1D → LSTM → Dense → output layer. Among them, the convolution kernel size of the Conv1D layer may be 3 or 5, which is used to capture local temporal features; the MaxPooling1D layer can reduce the data dimension; the LSTM layer can have 64 or 128 units to capture long-term dependencies; the Dense layer is used for final feature integration and output prediction. The temporal feature extraction module also includes a temporal attention layer (i.e., the attention mechanism module embedded in the temporal feature extraction module) to implement temporal attention in the temporal feature extraction module.

[0105] The fully connected module structure is as follows:

[0106] There are 3 fully connected layers, each followed by a ReLU activation function and a dropout layer, and a residual connection module from the first fully connected layer to the last fully connected layer.

[0107] Next, the preprocessed historical sintering data and historical finished product performance data need to be enhanced simultaneously. The purpose of data enhancement is to expand the scale and diversity of the training data set so that the model can learn more data changes and potential patterns. For sintering data, a variety of enhancement techniques can be used. Time series data enhancement can include methods such as time window sliding, time warping, and adding random noise. Time window sliding is to slide a fixed-length time window on the original sequence, and each slide generates a new sample. For example, for a temperature sequence of length 1000, a window of length 100 can be used, sliding 10 time steps each time, so that 90 new samples can be generated. Time warping is to simulate the time scale changes that may occur in the sintering process by performing a nonlinear transformation on the time axis. For finished product performance data, slight numerical perturbations can be used to generate new samples, such as adding or subtracting a small random value from the original value.

[0108] In addition, more complex methods can be used, such as interpolation or extrapolation based on physical models. For example, the previously constructed sintering physical model can be used to generate new virtual samples by fine-tuning certain parameters. When performing data enhancement, it is necessary to ensure that the enhanced data still conforms to the laws of physics and actual possible situations. For example, the temperature should not be negative or exceed the melting point of the material. At the same time, the enhancement of sintering data and performance data needs to be carried out simultaneously to maintain the correspondence between the two. Through these data enhancement techniques, the original data set can be expanded several times or even dozens of times, greatly increasing the number and diversity of training samples, which helps to improve the generalization ability of the model and its adaptability to abnormal situations.

[0109] The sintering data set and performance data set generated by the above process are used to train the sintering data association model. The training process usually uses the batch gradient descent (Batch Gradient Descent) or stochastic gradient descent (Stochastic Gradient Descent, SGD) algorithm. First, the data set needs to be divided into a training set, a validation set, and a test set. Before the training starts, the model parameters need to be initialized, usually using random initialization or pre-trained model parameters. During the training process, the model calculates the predicted value through forward propagation, and then calculates the gradient and updates the parameters through the back propagation algorithm. Taking the mean square error (MSE) as an example, the loss function can be expressed as: L = (1 / n) * ∑ (y p -y t ) 2 , where n is the number of samples, y p is the model prediction value, y t is the true value. The parameter update formula is: Where θ is the model parameter and η is the learning rate. The choice of learning rate is critical, and an adaptive learning rate method can be used, such as the Adam optimizer. Adam combines the advantages of momentum and RMSprop, and its parameter update formula is: Where m t and t are the first and second order moment estimates, respectively, and ε is a small constant to prevent division by zero.

[0110] To prevent overfitting, regularization techniques can be used. Early stopping techniques can also be used, that is, training is stopped when the performance on the validation set no longer improves. During training, the model performance needs to be evaluated regularly on the validation set. Various indicators can be used, such as root mean square error (RMSE), mean absolute error (MAE), etc. Training may require multiple epochs, each of which traverses the entire training set. In order to speed up training and improve generalization ability, batch training can be used, that is, a small batch of data is used each time to update parameters. The choice of batch size requires a trade-off between computational efficiency and optimization effect. Usually, you can start with 32 or 64. During training, you can draw loss curves and performance indicator curves to intuitively observe the convergence of the model. When the performance of the model on the validation set is optimal and stable, the training can be considered complete, and the model parameters at this time are the optimal model parameters. Finally, the model is finally evaluated using the test set to ensure that the model has good generalization ability.

[0111] Next, the sintering physics model and the trained sintering data association model need to be merged into a sintering dynamic model of the electronic packaging shell. This hybrid model is usually called a gray box model or a semi-physical model. There are many strategies for the merging process. One common method is the parallel structure. In the parallel structure, the sintering physics model and the sintering data association model receive input data at the same time, perform calculations separately, and then perform weighted averaging or other forms of fusion on the outputs of the two models. The weights can be fixed or dynamically adjusted. Another method is the cascade structure, that is, the output of the physical model is used as an additional input to the data model, or vice versa. The merged sintering dynamic model needs to be verified and fine-tuned. A portion of historical data that was not involved in the training can be used for model verification to compare the performance differences between the merged model and the physical model or data model used alone.

[0112] The resulting sintering dynamic model should be able to accurately predict the performance of electronic packaging shells under different sintering conditions while maintaining physical interpretability and data adaptability. The advantage of this merged model is that it can not only use known physical laws to guide predictions, but also capture complex nonlinear relationships and the influence of unknown factors through the data model. For example, when predicting the density after sintering, the physical model may be based on the traditional sintering kinetic equation, while the data model can capture the influence of factors such as material microstructure and impurity distribution that are difficult to model directly on the density. In practical applications, the merged model can dynamically adjust the weights of the physical model and the data model according to the characteristics of the input data. For example, when the input data falls within the common range of historical data, it may rely more on the data model; when the input data exceeds the historical range, it may rely more on the physical model for extrapolation. This dynamic adjustment can be achieved by designing a meta-learner that determines the final fusion strategy based on the input features and intermediate calculation results. The merged sintering dynamic model can be used not only for prediction, but also for optimization and control of the sintering process. Through back propagation or other optimization algorithms, the optimal combination of sintering parameters can be found based on the model to achieve the target performance indicators. In addition, the model can also be used for real-time monitoring and control. By comparing the difference between the actual process and the model prediction, the sintering parameters can be adjusted in time to achieve closed-loop control. In general, this method of combining physical models and data models can make full use of domain knowledge and data information to provide a sintering dynamic model that has both a theoretical basis and can adapt to actual complexity, providing strong support for the high-quality production of electronic packaging shells.

[0113] In one embodiment, the sintering environment data is input into the sintering dynamic model, and the feedback control instructions of the tunnel sintering furnace are generated by the model predictive control method, including the following steps:

[0114] Input the sintering environment data into the sintering dynamic model, and predict the future product performance of the electronic packaging shell currently in the tunnel sintering furnace through the sintering dynamic model;

[0115] Create performance control constraints for the sintering dynamic model by combining future product performance and preset product sintering performance indicators of electronic packaging shells;

[0116] Based on the performance control constraints and with the goal of optimizing future product performance, the optimization objective function of the optimization algorithm is generated, and the optimal sintering environment data is calculated using the optimization algorithm;

[0117] Get the current sintering control sequence of the tunnel sintering furnace at the current moment;

[0118] Based on a preset first mapping relationship between sintering environment data and sintering control parameters in the tunnel type sintering furnace, generating an optimal sintering control sequence of the tunnel type sintering furnace according to the optimal sintering environment data;

[0119] Feedback control instructions for the tunnel sintering furnace are generated according to the distinguishing characteristics of the control parameters in the optimal sintering control sequence and the current sintering control sequence.

[0120] In this embodiment, the real-time sintering environment data collected from the tunnel sintering furnace is first input into the sintering dynamic model, and algorithms such as Kalman filtering or particle filtering are used to combine model prediction and real-time measurement data to estimate the internal state of the current product. Based on the current state estimation, the model is used for forward prediction to obtain the evolution trajectory of the product in the future. According to the predicted product state, the key performance indicators are calculated. Taking into account the uncertainty of the prediction, the Monte Carlo method can be used to generate multiple possible evolution trajectories to obtain the probability distribution of performance indicators. Finally, the model outputs the prediction results of future product performance, including the predicted values ​​of various performance indicators and their confidence intervals, the trend of performance indicators over time, possible quality problems and their probabilities, etc. Through this process, the sintering dynamic model can predict the various performance indicators of the electronic packaging shell at future moments based on the current sintering environment data.

[0121] Next, the performance control constraints of the sintering dynamic model are created by combining the future product performance and the preset product sintering performance indicators of the electronic packaging shell. The preset product sintering performance indicators are usually determined by product design specifications and industry standards, including final density, strength, thermal conductivity, airtightness and other aspects. For example, for a specific electronic packaging shell, the final relative density may be required to be no less than 98%, the bending strength may be no less than 300MPa, and the thermal conductivity may be no less than 50W / (m·K). These indicators are compared with the future product performance predicted by the model to form a series of inequality constraints. For example, ρpred≥0.98ρtheory, σpred≥300MPa, λpred≥50W / (m·K), where the subscript pred represents the predicted value and ρtheory is the theoretical density. In addition to these final performance indicators, some intermediate constraints in the sintering process need to be considered, such as the heating rate not exceeding a certain threshold to avoid thermal shock cracks: |dT / dt|≤Rmax, where Rmax is the maximum allowable heating rate. The atmosphere concentration also needs to be kept within a certain range to prevent oxidation or carbonization: Cmin≤C≤Cmax, where C is the concentration of a specific gas component, and Cmin and Cmax are the minimum and maximum concentrations allowed, respectively. The pressure change rate also needs to be controlled to prevent deformation caused by excessive pressure gradients: |dp / dt|≤Pmax, where Pmax is the maximum allowable pressure change rate. These constraints together constitute a multi-dimensional feasible solution space, and any point in this space meets the product quality requirements. By setting these constraints, it is possible to ensure that the sintering process always develops towards the expected performance of the product while avoiding extreme process parameters that may cause product defects.

[0122] Based on the performance control constraints and with the goal of optimizing future product performance, the optimization objective function of the optimization algorithm is generated. The construction of the objective function needs to consider multiple performance indicators and adopts the form of weighted sum: J = w1f1(x) + w2f2(x) + ... + wnfn(x), where fi(x) represents the objective function of the i-th performance indicator, wi is the corresponding weight, and x is the sintering environment parameter vector. For example, f1(x) may represent the square deviation of density from the target density, f2(x) may represent the square deviation of strength from the target strength, and so on. The choice of weights reflects the relative importance of different performance indicators and can be adjusted according to specific application requirements. The optimization problem can be expressed as: minJ(x), stg(x)≤0, h(x)=0, where g(x) and h(x) represent inequality and equality constraints, respectively. This is a nonlinear constrained optimization problem that can be solved by a variety of optimization algorithms, such as gradient descent, Newton method, interior point method, etc. Considering the complexity of the problem and the possible existence of multiple local optimal solutions, a global optimization algorithm such as particle swarm optimization (PSO) or genetic algorithm (GA) is usually chosen.

[0123] Taking PSO as an example, the algorithm will initialize a group of "particles", each particle represents a set of possible sintering environment parameters. These particles move in the parameter space, and their positions and speeds are continuously updated according to the individual optimal position and the global optimal position: vi(t+1)=wvi(t)+c1r1(pi-xi(t))+c2r2(pg-xi(t)), xi(t+1)=xi(t)+vi(t+1), where vi and xi represent the speed and position of the particle respectively, pi is the individual optimal position, pg is the global optimal position, w, c1, c2 are algorithm parameters, and r1 and r2 are random numbers. Through multiple iterations, it finally converges to the optimal or approximately optimal sintering environment parameters. This method can find the global optimal or approximately optimal solution under complex nonlinear constraints, thereby obtaining the optimal sintering environment data.

[0124] Next, we need to obtain the current sintering control sequence of the tunnel sintering furnace at the current moment. The tunnel sintering furnace is usually composed of multiple temperature zones, each of which has an independent temperature control system. At the same time, it also includes an atmosphere control system and a pressure control system. The current sintering control sequence includes all the control parameters of these systems at the current moment. For example, the temperature control sequence may include the target temperature of each temperature zone, the parameters of the PID controller (proportional coefficient Kp, integral time Ti, differential time Td), etc. Specifically, for the i-th temperature zone, there may be: Ti_set(t), Kp_i(t), Ti_i(t), Td_i(t). The atmosphere control sequence may include the flow setting values ​​and mixing ratios of various gases, such as: Q_N2(t), Q_H2(t), Q_Ar(t), etc., where Q represents the volume flow rate. The pressure control sequence may include the pressure setting value and the opening of the pressure control valve, such as: P_set(t), V_open(t).

[0125] Based on the preset first mapping relationship between the sintering environment data and the sintering control parameters in the tunnel sintering furnace, the optimal sintering control sequence of the tunnel sintering furnace is generated according to the optimal sintering environment data. The first mapping relationship describes how the control parameters affect the sintering environment, which is usually a nonlinear, multi-variable complex relationship. For example, there is a dynamic response relationship between temperature control and actual temperature, which can be expressed by a transfer function: G(s) = K / (τs+1), where K is the gain, τ is the time constant, and s is the Laplace transform variable. There may be a lag effect between atmosphere control and actual atmosphere concentration, which can be described by a first-order lag plus dead zone model: C(t) = C 0 +K(1-e (-t / τ) )u(t-θ), where C_0 is the initial concentration, K is the gain, τ is the time constant, θ is the dead time, and u(t) is the unit step function.

[0126] There may be a nonlinear relationship between pressure control and actual pressure, which can be described by a polynomial model: P = a0 + a1V + a2V^2 + ... + anV^n, where P is the pressure, V is the control valve opening, and ai is the polynomial coefficient. These relationships can be obtained through system identification methods, such as least squares, maximum likelihood estimation, or neural network fitting. With these mapping relationships, the optimal sintering control sequence can be obtained by solving the inverse problem. This is usually an ill-posed problem, and there may be multiple solutions or no solution. Solutions include regularization techniques, such as Tikhonov regularization: min||Ax-b||^2+λ||x||^2, where A is the mapping matrix, x is the control parameter, b is the target environment parameter, and λ is the regularization parameter. Another method is to use an iterative optimization algorithm, such as the Levenberg-Marquardt algorithm: xk+1=xk-(JTJ+λI) ^ (-1)JTf(xk), where J is the Jacobian matrix, f(x) is the objective function, and λ is the damping factor. Through these methods, the optimal sintering environment data can be converted into the corresponding optimal sintering control sequence, realizing the reverse mapping from environmental parameters to control parameters.

[0127] Finally, the feedback control instructions of the tunnel sintering furnace are generated according to the difference characteristics of the control parameters in the optimal sintering control sequence and the current sintering control sequence. Specifically, the difference in control parameters needs to be calculated first: Δx = x_optimal-x_current, where x_optimal is the optimal control sequence and x_current is the current control sequence. This difference may include temperature difference ΔT, atmosphere concentration difference ΔC, pressure difference ΔP, etc. Then, a suitable control law needs to be designed to gradually eliminate these differences. Considering the dynamic characteristics and stability requirements of the system, a more complex control strategy is usually adopted, such as model predictive control (MPC):

[0128] min∑||y(k+i|k)-r(k+i)|| 2 +∑||Δu(k+i)|| 2

[0129] stx min ≤x≤x max ,u min ≤u≤u max

[0130] Where y is the predicted output, r is the reference trajectory, Δu is the control increment, and x and u are the constraints of the state and control variables, respectively. This method can consider multiple control objectives and constraints at the same time. In practical applications, it is also necessary to consider the rate of change limit of the control quantity to avoid drastic fluctuations in the system: |Δu|≤Δumax. At the same time, in order to improve the robustness of the control, advanced control methods such as adaptive mechanisms or sliding mode control can be introduced. The feedback control instructions finally generated should be a series of specific operation commands, such as "increase the temperature of the first temperature zone by 5℃", "increase the hydrogen flow rate by 2L / min", "reduce the pressure by 0.1MPa", etc. These instructions are sent to actuators such as heating elements, flow control valves, pressure regulating valves, etc. through the communication interface, so as to achieve precise control of the sintering process, so that the actual sintering environment gradually approaches the optimal state, and ultimately achieve the purpose of improving product quality and production efficiency.

[0131] In one embodiment, the conveyor belt end of the tunnel sintering furnace is pre-configured with an image acquisition device, referring to Figure 3 , the method further comprises the following steps:

[0132] S301. After the sintering control parameters are adjusted, the image acquisition device is used to capture the finished image of the finished electronic packaging housing at the end of the conveyor belt;

[0133] S302. Identify multiple defect types in the finished product image through image recognition technology, and count the number of defect types of each different defect type;

[0134] S303. If there is a target defect type whose number of defect types exceeds a preset number threshold among all defect types, then generating an optimization constraint condition of the optimization algorithm based on a preset second mapping relationship between the sintering environment data and the defect type;

[0135] S304. Combining the optimization constraints and the performance control constraints and aiming at optimizing the future product performance, the optimization algorithm is used to update the optimal sintering environment data, and the feedback control instructions of the tunnel sintering furnace are updated through the updated optimal sintering environment data;

[0136] S305. Adjust the sintering control parameters of the tunnel sintering furnace according to the updated feedback control instructions.

[0137] In this embodiment, a collection point is set at the end of the conveyor belt by a high-resolution camera or an industrial camera to capture images of each finished electronic packaging shell passing through. The collection device generally includes a light source system, an image sensor, an image acquisition card, and a computer system. The light source system provides stable and uniform illumination, and an LED array or a ring light source can be used to ensure image clarity and contrast. The image sensor can be a CCD or CMOS sensor, and the resolution should be at least 2048x2048 pixels to capture subtle defects. The image acquisition card is responsible for converting analog signals into digital signals and performing preliminary image processing. One embodiment is to use an array camera to trigger shooting in conjunction with a programmable logic controller (PLC). When the electronic packaging shell reaches a preset position, the PLC sends a trigger signal, and the camera takes multiple images at a shutter speed of 1 / 1000 second, and then selects the clearest one for subsequent processing. During the image acquisition process, parameters such as depth of field, focal length, and field of view need to be considered to ensure that the entire product surface can be clearly captured. For example, for a 10mm x 10mm package, a lens with a focal length of 35mm can be used and the working distance is set to 300mm to obtain a suitable field of view and resolution. The acquired images are transmitted to the image processing workstation via Gigabit Ethernet or Camera Link interface.

[0138] Image recognition technology is used to identify multiple defect types in finished product images, and the number of defect types for each different defect type is counted. Specifically, convolutional neural networks (CNNs) can be used for image classification and target detection. First, the collected high-resolution images are preprocessed, including image enhancement, noise removal, and standardization. Then, the images are analyzed using a trained deep learning model. One implementation method is to use a two-stage target detection network, such as Faster R-CNN or Mask R-CNN. The first stage uses a region proposal network (RPN) to generate candidate regions that may contain defects, and the second stage classifies and regresses these regions into bounding boxes. The training of the model requires a large number of labeled defect samples, including various types of cracks, bubbles, deformations, surface unevenness, etc. Data enhancement techniques, such as rotation, flipping, scaling, etc., are used during training to improve the generalization ability of the model.

[0139] In practical applications, the model can detect multiple defect types at the same time, such as surface cracks, edge notches, bubbles, etc. For each defect type, the model outputs a confidence score and position coordinates. By setting an appropriate confidence threshold (such as 0.7), high-confidence detection results are screened out. Then, the non-maximum suppression (NMS) algorithm is used to eliminate duplicate detections. Finally, the number of each defect type is counted and recorded in the database. Defect types include temperature abnormality defects, atmosphere abnormality defects, and pressure abnormality defects. Temperature abnormality defects include at least one of shell size abnormality defects, grain abnormality defects, thermal stress cracking defects, shell phase change defects, melting deformation defects, and thermal shock cracking defects. Atmosphere abnormality defects include at least one of shell oxidation defects, shell carbonization defects, shell abnormal pore defects, surface impurity deposition defects, and color abnormality defects. Pressure abnormality defects include at least one of shell deformation defects, shell crack defects, and shell delamination defects. By real-time statistics of defect types and quantities, abnormalities in the production process can be discovered in a timely manner, providing data support for subsequent process optimization. For example, if a large number of surface cracks are detected, it may indicate improper sintering temperature control, while frequent bubbles may indicate problems with the raw materials or the degassing process.

[0140] If there is a target defect type whose number of defect types exceeds the preset number threshold among all defect types, a correlation model between sintering environment parameters and product defects is established, and this model is used to guide the optimization process. Specifically, it is necessary to first build a comprehensive database containing historical sintering environment data and corresponding defect statistics. This database should cover a large number of samples under different sintering conditions to ensure the accuracy and robustness of the model. Then, a machine learning algorithm, such as a random forest or a gradient boosting tree, is used to establish a mapping relationship between sintering environment parameters (such as temperature curve, atmosphere composition, pressure, etc.) and the probability of occurrence of various defect types. Specifically, a multi-output regression model can be used, in which the input feature X is a sintering environment parameter vector and the output Y is a probability vector of various defect types. The model can be expressed as: Y = f(X) + ε.

[0141] Where f is a nonlinear function and ε is the error term. For the random forest model, multiple decision trees (such as 500) can be used to build it. Each tree independently predicts the probability of various defects, and the average value is finally taken as the model output. After the model training is completed, when the number of a certain defect type is detected to exceed the preset threshold, the system will automatically query the current sintering environment data and reversely deduce the key parameter range that may cause the defect through the trained model. For example, if an abnormal number of surface cracks is detected, the model may point out that the rapid temperature rise and fall rate is the main reason. Based on this information, the system generates a set of optimization constraints, such as: temperature rise rate <5℃ / min, maximum temperature holding time >30 minutes, oxygen concentration <100ppm, etc. These constraints will serve as inputs for subsequent optimization algorithms to guide the direction and range of parameter adjustment.

[0142] The effect of implementing this step is that it can quickly and accurately associate product quality problems with specific process parameters, providing a clear direction for subsequent optimization. By establishing a mapping relationship between the sintering environment and defects, accurate control and prediction of the production process can be achieved. This data-driven approach can not only solve current quality problems, but also prevent potential problems from occurring, greatly improving production efficiency and product yield. At the same time, this process is also constantly accumulating knowledge. As the amount of data increases, the accuracy of the model will continue to improve, forming a virtuous cycle of quality improvement mechanism.

[0143] Combined with the optimization constraints and performance control constraints and with the goal of optimizing future product performance, the optimization algorithm is used to update the optimal sintering environment data, and the feedback control instructions of the tunnel sintering furnace are updated through the updated optimal sintering environment data, and the sintering control parameters of the tunnel sintering furnace are adjusted according to the updated feedback control instructions. The specific process can refer to the above-mentioned specific implementation method. The final algorithm outputs a set of optimal sintering environment parameters, including temperature curve, atmosphere composition, pressure change, etc. These parameters are converted into specific control instructions, such as heating element power, gas flow, pressure valve opening, etc. The control instructions are transmitted to the programmable logic controller (PLC) or distributed control system (DCS) of the tunnel sintering furnace via industrial Ethernet or field bus (such as Profibus). Through the application of intelligent algorithms, changes in the production process can be quickly responded to, process parameters can be adjusted in time, and product quality stability can be maintained.

[0144] The present invention also discloses an electronic packaging shell sintering control system based on a tunnel-type sintering furnace, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the electronic packaging shell sintering control method based on the tunnel-type sintering furnace as described in any one of the above embodiments is implemented.

[0145] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0146] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the computer device, etc., and the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0147] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0148] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A sintering control method for an electronic packaging shell based on a tunnel sintering furnace, characterized in that: The method comprises the following steps: Acquire historical sintering data and historical finished product performance data of the electronic packaging shell in the tunnel sintering furnace, wherein the historical sintering data includes historical sintering temperature data, historical sintering atmosphere data and historical sintering pressure data, and the historical finished product performance data is the performance data of the electronic packaging shell sintered based on the historical sintering data; Combining the historical sintering data and the historical finished product performance data to construct a sintering dynamic model of the electronic packaging shell; Acquiring sintering environment data of the electronic packaging shell during the sintering process through a plurality of sensors pre-arranged in the tunnel-type sintering furnace, wherein the sintering environment data includes sintering temperature data, sintering atmosphere data and sintering pressure data; Inputting the sintering environment data into the sintering dynamic model, and generating feedback control instructions for the tunnel sintering furnace by a model predictive control method; The sintering control parameters of the tunnel sintering furnace are adjusted according to the feedback control instruction.

2. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 1, characterized in that: The step of combining the historical sintering data and the historical finished product performance data to construct the sintering dynamic model of the electronic packaging housing comprises the following steps: Preprocessing the historical sintering data and the historical finished product performance data; Combining the pre-processed historical sintering data and the historical finished product performance data, constructing a sintering physical model of the electronic packaging housing during the sintering process; Building a sintering data association model based on a neural network model, wherein the sintering data association model is used to capture the nonlinear relationship between various physical quantities in the sintering physical model; Synchronously performing data enhancement processing on the preprocessed historical sintering data and the historical finished product performance data to obtain a sintering data set and a corresponding performance data set; Training the sintering data association model using the sintering data set and the performance data set until all model parameters in the sintering data association model are optimal model parameters; The sintering physical model and the trained sintering data association model are combined into a sintering dynamic model of the electronic packaging shell.

3. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 2 is characterized in that: The expression of the sintering physical model is as follows: Where: ρ represents the material density of the electronic packaging shell, c p represents specific heat capacity, T represents the historical sintering temperature data, t represents the sintering time, k(T) represents thermal conductivity, Q total represents the total heat source term, ΔH r represents the heat of reaction, C i represents the atmosphere concentration of the i-th gas component in the historical sintering atmosphere data, D i (T, P) represents the diffusion coefficient of the i-th gas component as it changes with temperature and pressure, P represents the historical sintering pressure data, u represents the gas flow rate in the tunnel sintering furnace, R i (T, P, C i ) represents the reaction rate of the i-th gas component, K ρ (T, P) represents the rate constant of density change, ρ max represents the maximum theoretical material density of the electronic packaging shell, k j represents the total amount of atmosphere in the jth sintering reaction of the electronic packaging housing in the tunnel sintering furnace, ξ j represents the reaction progress of the jth sintering reaction, υ j (T, P) represents the rate constant of the jth sintering reaction, represents the stoichiometric coefficient of the i-th gas component in the j-th sintering reaction, K p (T, ρ) represents the pressure transmission coefficient, Q p (T,C i ) represents the stressor term.

4. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 3 is characterized in that: The total heat source term includes the heat density of the heating source of the tunnel sintering furnace, and also includes the radiation heat loss, convection heat loss and thermal bridge effect loss in the tunnel sintering furnace. The total heat source term is expressed as the heat density of the heating source minus the sum of the radiation heat loss, the convection heat loss and the thermal bridge effect loss.

5. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 2, characterized in that: The sintering data association model includes a spatial feature extraction module, a temporal feature extraction module, a fully connected module, a residual connection module and an attention mechanism module. The spatial feature extraction module, the temporal feature extraction module and the fully connected module are connected in sequence, the residual connection module is configured in the fully connected module, and the attention mechanism module is configured in the spatial feature extraction module and the temporal feature extraction module.

6. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 2, characterized in that: The step of inputting the sintering environment data into the sintering dynamic model and generating feedback control instructions for the tunnel sintering furnace by a model predictive control method comprises the following steps: Inputting the sintering environment data into the sintering dynamic model, and predicting the future product performance of the electronic packaging housing currently in the tunnel sintering furnace through the sintering dynamic model; Creating performance control constraint conditions of the sintering dynamic model in combination with the future product performance and the preset product sintering performance index of the electronic packaging shell; Based on the performance control constraint conditions and with the goal of optimizing the future product performance, an optimization objective function of an optimization algorithm is generated, and optimal sintering environment data is calculated using the optimization algorithm; Acquire the current sintering control sequence of the tunnel sintering furnace at the current moment; Based on a preset first mapping relationship between the sintering environment data and the sintering control parameters in the tunnel-type sintering furnace, generating an optimal sintering control sequence of the tunnel-type sintering furnace according to the optimal sintering environment data; A feedback control instruction of the tunnel sintering furnace is generated according to the distinguishing features of the control parameters in the optimal sintering control sequence and the current sintering control sequence.

7. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 6, characterized in that: The optimization algorithm is a particle swarm optimization algorithm or a genetic optimization algorithm.

8. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 6, characterized in that: The conveyor belt end of the tunnel sintering furnace is pre-configured with an image acquisition device, and the method further includes the following steps: After the sintering control parameters are adjusted, the image acquisition device is used to acquire a finished product image of the finished electronic packaging shell at the end of the conveyor belt; Identify multiple defect types in the finished product image by image recognition technology, and count the number of defect types of each different defect type; If there is a target defect type among all the defect types whose number exceeds a preset number threshold, generating an optimization constraint condition of the optimization algorithm based on a preset second mapping relationship between the sintering environment data and the defect type; In combination with the optimization constraint and the performance control constraint and with the goal of optimizing the future product performance, the optimization algorithm is used to update the optimal sintering environment data, and the feedback control instructions of the tunnel sintering furnace are updated through the updated optimal sintering environment data; The sintering control parameters of the tunnel sintering furnace are adjusted according to the updated feedback control instructions.

9. The electronic packaging shell sintering control method based on a tunnel sintering furnace according to claim 8, characterized in that: The defect types include temperature abnormality defects, atmosphere abnormality defects and pressure abnormality defects. The temperature abnormality defects include at least one of shell size abnormality defects, grain abnormality defects, thermal stress cracking defects, shell phase change defects, melt deformation defects and thermal shock crack defects. The atmosphere abnormality defects include at least one of shell oxidation defects, shell carbonization defects, shell abnormal pore defects, surface impurity deposition defects and color abnormality defects. The pressure abnormality defects include at least one of shell deformation defects, shell crack defects and shell delamination defects.

10. An electronic packaging shell sintering control system based on a tunnel sintering furnace, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the electronic packaging shell sintering control method based on the tunnel sintering furnace as claimed in any one of claims 1 to 9 is implemented.

Images