A vapor phase welding process parameter inversion method based on a vacuum process stage dynamic weighting

Abstract

The application provides a vapor phase welding process parameter inversion method based on a vacuum process stage dynamic weighting, and belongs to the technical field of vapor phase welding process parameter identification and numerical simulation. The method acquires measured temperature data of a vapor phase welding process, divides the process into multiple vacuum process stages, calculates root mean square errors of temperature residuals of the stages, generates dynamic stage weights through dimensionless, amplitude limiting and smoothing processing, and forms time point by time point weights in combination with stage sample number compensation; based on the weights, a multi-measurement-point weighted residual, a weighted Jacobian matrix, a gradient and an approximate Hessian matrix are constructed to generate a Levenberg-Marquardt candidate direction and a conjugate gradient candidate direction respectively, process parameters are selected and updated according to the ratio of an actual descent amount to a predicted descent amount, and vapor phase welding process parameter inversion results are obtained through iteration. The application can be used to improve the balance and stability of parameter inversion in different process stages.

Description

Technical Field

[0001] This invention relates to the fields of vapor phase welding process parameter identification, computational fluid dynamics numerical simulation, and nonlinear parameter inversion, and particularly to a method for inverting vapor phase welding process parameters based on dynamic weighting of the vacuum process stage. Background Technology

[0002] Vapor phase welding utilizes the exothermic condensation of saturated vapor to heat the components to be welded. Its temperature, pressure, and heat transfer state change with each stage of the process, including heating, holding, vacuuming, and cooling. To ensure that the numerical simulation results match the actual process, it is usually necessary to invert the process parameters in the simulation model based on measured temperature data.

[0003] Existing inversion methods typically apply uniform or fixed weights to all temperature sampling points, making it difficult to dynamically adjust the inversion focus based on the actual errors at different process stages. Furthermore, the varying number of sampling points at different stages can easily lead to stages with longer sampling times occupying a large proportion of the objective function; using a single parameter update direction can also result in convergence instability or poor parameter update performance. Therefore, it is necessary to propose an inversion method for vapor phase welding process parameters based on dynamic weighting of vacuum process stages. Summary of the Invention

[0004] The technical problem to be solved:

[0005] This invention aims to address the problem in existing vapor phase welding process parameter inversion methods where temperature sampling data from different vacuum process stages are processed with uniform or fixed weights, making it difficult to dynamically adjust the inversion focus based on the current temperature error at each process stage. Furthermore, this invention also addresses the issues of inconsistent sampling point numbers across different process stages leading to an imbalance in the objective function's contribution, and the instability of parameter update effects when using a single parameter update direction.

[0006] To this end, the present invention dynamically calculates the stage weights based on the temperature residuals of each vacuum process stage, and forms time-point weights by combining the number of sampling points in each stage. The weights are then applied to the residuals, Jacobian matrix, parameter update direction, and candidate parameter evaluation process to achieve iterative inversion of vapor phase welding process parameters.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] This invention provides a method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages, comprising the following steps.

[0009] S1. Obtain the measured temperature time series of at least one temperature measurement point in the vapor phase welding process, establish a numerical simulation model of heat transfer in vapor phase welding, and determine the vector of process parameters to be inverted, the initial values ​​of process parameters, and the range of values ​​for each process parameter.

[0010] The vector of process parameters to be inverted is represented as follows:

[0011]

[0012] Where P represents the number of process parameters to be inverted. Let be the i-th process parameter to be inverted.

[0013] The process parameter vector satisfies:

[0014]

[0015] S2. Based on the preset stage boundary times, the vapor phase welding process is divided into multiple vacuum process stages, and corresponding process stages are assigned to each temperature sampling time.

[0016] Let M be the number of vacuum process stages, and let the end times of each stage be represented as follows:

[0017]

[0018] The end time of the last stage Not less than the last sampling time of the temperature time series.

[0019] The set of sampling times corresponding to the s-th vacuum process stage is represented as follows: The number of sampling points in the set of sampling times is expressed as .

[0020] The vacuum process stages may include a heating stage, a heat preservation and equalization stage, a rapid heating stage, a vacuuming stage, and a cooling stage. The boundaries of each stage can be determined according to preset process timings, or according to equipment control signals, changes in vacuum pressure, or changes in temperature.

[0021] S3. Input the current process parameter vector into the vapor phase welding heat transfer numerical simulation model to obtain the simulation temperature of each temperature measurement point, and calculate the temperature residual between the simulation temperature and the measured temperature.

[0022] In the k-th iteration, the root mean square error of the s-th vacuum process stage is expressed as:

[0023]

[0024] Where L is the number of temperature measurement points. Let be the simulated temperature of the i-th temperature measurement point at the j-th sampling time during the k-th iteration. This corresponds to the measured temperature.

[0025] The average stage error of all vacuum process stages is expressed as:

[0026]

[0027] Calculate the dimensionless original weight of the s-th vacuum process stage based on the ratio of the root mean square error of the stage to the average stage error:

[0028]

[0029] in, To prevent zero parameters, The error amplification index is... As the lower limit of the weight, This represents the upper limit of the weight.

[0030] The original weights are iteratively smoothed to obtain the dynamic stage weights:

[0031]

[0032] in, This is the weighted smoothing coefficient.

[0033] The dynamic stage weights are further compensated based on the number of sampling points in each stage. For the jjj-th sampling time belonging to the sss-th vacuum process stage, its time-point weight is expressed as:

[0034]

[0035] Where N represents the total number of temperature sampling times. This sample number compensation ensures that different vacuum process stages have a relatively balanced overall weight contribution when the stage error is the same.

[0036] S4. Apply finite difference perturbation to each process parameter in the current process parameter vector, obtain the corresponding temperature response change through the vapor phase welding heat transfer numerical simulation model, and construct the Jacobian matrix.

[0037] In the k-th iteration, the finite difference perturbation of the i-th process parameter is expressed as:

[0038]

[0039] in, This represents the relative perturbation ratio. This is the preset minimum disturbance amount.

[0040] When the positive disturbance exceeds the upper bound of the corresponding parameter, the negative disturbance is used, and the actual disturbance after parameter boundary constraints is used as the denominator of the finite difference.

[0041] The Jacobian matrix column corresponding to the i-th process parameter is represented as follows:

[0042]

[0043] in, This represents the actual disturbance after boundary constraints. Let be the i-th unit direction vector.

[0044] The temperature residuals and Jacobian matrices of all temperature measurement points are stacked in the same order to obtain the residual vector. And Jacobi matrix .

[0045] Construct a diagonal weight matrix based on the weights at each time point. The weighted residual vector and the weighted Jacobian matrix are expressed as follows:

[0046]

[0047] The weighted gradient and the approximate Hessian matrix are expressed as follows:

[0048]

[0049] S5. Generate Levenberg-Marquardt candidate directions based on the weighted gradient and the approximate Hessian matrix:

[0050]

[0051] in, Let be the damping coefficient for the k-th iteration. This is the diagonal damping matrix.

[0052] When calculating the first conjugate gradient candidate direction or when no valid previous conjugate gradient direction exists, the conjugate gradient candidate direction is set to the negative gradient direction; when a valid previous conjugate gradient direction exists, the conjugate gradient candidate direction is generated according to the PRP+ method.

[0053]

[0054] When the conjugate gradient candidate direction does not satisfy the descent direction condition, it is reset as the negative gradient direction.

[0055] The Levenberg-Marquardt candidate directions and conjugate gradient candidate directions are superimposed onto the current process parameter vector, and constraints are applied based on the parameter value range to obtain two candidate parameter vectors.

[0056] S6. Input the two candidate parameter vectors into the vapor phase welding heat transfer numerical simulation model respectively, and obtain the corresponding candidate simulation temperatures.

[0057] During the evaluation of candidate parameters, the time-step weights calculated in this iteration remain unchanged. The weighted objective function is expressed as:

[0058]

[0059] Determine the effective step size based on the actual parameter changes after the candidate parameters have undergone boundary constraints. .

[0060] The actual decrease corresponding to the candidate parameter is expressed as follows:

[0061]

[0062] The predicted decrease is expressed as:

[0063]

[0064] When both the actual decrease and the predicted decrease are greater than zero, the decrease ratio of the candidate parameter is expressed as:

[0065]

[0066] Calculate the decrease ratio for each of the two candidate parameters, and select the effective candidate parameter with the larger decrease ratio as the parameter to be updated in this iteration.

[0067] When the descent ratio corresponding to the selected candidate parameter is greater than zero, the candidate parameter is accepted, and the current process parameters, simulation temperature, dynamic stage weight, and damping coefficient are updated; when the descent ratio is not greater than zero, the current process parameters are kept unchanged, and the damping coefficient is increased.

[0068] S7. Repeat steps S3 to S6 until the preset stop condition is met.

[0069] The preset stopping conditions include at least one of the following: the maximum absolute error between the simulated temperature and the measured temperature is less than a preset error threshold; the damping coefficient exceeds a preset upper limit; the relative decrease rate of the weighted objective function for a consecutive preset number of times is less than a preset decrease threshold; the infinite norm of the weighted gradient is less than a preset gradient threshold; the relative norm of the effective step size corresponding to the accepted candidate parameter is less than a preset step size threshold; or the number of iterations reaches a preset maximum number of iterations. After the preset stopping conditions are met, the parameter inversion is terminated, the final process parameter vector and its corresponding simulated temperature are output, and the root mean square error, dynamic stage weight, objective function change and candidate direction selection record of each vacuum process stage can be output simultaneously.

[0070] The beneficial technical effects of this invention are as follows:

[0071] First, the present invention dynamically adjusts the stage weights based on the current temperature error of each vacuum process stage, so that the parameter inversion can take into account the fitting state of different process stages.

[0072] Second, this invention reduces the impact of differences in the absolute magnitude of error and the number of sampling points on the inversion results by dimensionless error reduction, weight smoothing, and compensation for the number of stage samples.

[0073] Third, this invention uses dynamic weights in the evaluation process of multi-point residuals, Jacobian matrix and candidate parameters to improve the comprehensive utilization of data from different temperature measurement locations and process stages.

[0074] Fourth, the present invention employs a competitive selection between the Levenberg-Marquardt direction and the conjugate gradient direction, combined with parameter boundary and damping adjustment, which is beneficial to improving the stability of the parameter inversion process. Attached Figure Description

[0075] Figure 1 This is a schematic diagram of the overall process flow of the vapor phase welding process parameter inversion method based on dynamic weighting of vacuum process stages as described in this invention.

[0076] Figure 2 This is a schematic diagram illustrating the stage division and dynamic stage weight calculation of the vapor phase welding process in an embodiment of the present invention.

[0077] Figure 3 This is a schematic diagram illustrating the construction of the multi-measurement point weighted residual and the weighted Jacobian matrix in an embodiment of the present invention;

[0078] Figure 4 This is a schematic diagram illustrating the competition between the Levonburg-Marquardt direction and the conjugate gradient direction, as well as the closed-loop numerical simulation, in an embodiment of the present invention. Detailed Implementation

[0079] To enable those skilled in the art to more clearly understand the technical solution of the present invention, the present invention will be further described below with reference to specific embodiments. These embodiments are only used to illustrate the implementation of the present invention and are not intended to limit the scope of protection of the present invention.

[0080] Example 1

[0081] This embodiment uses the measured temperature data of the vacuum vapor phase welding process and the corresponding heat transfer numerical simulation model as a basis to invert the process parameters in the numerical simulation model.

[0082] First, obtain the temperature time series of at least one temperature measurement point during the vacuum vapor phase welding process. All temperature measurement points use the same sampling time, and the temperature sampling time vector is represented as:

[0083]

[0084] The measured temperature vector of the i-th temperature measurement point is represented as:

[0085]

[0086] A numerical simulation model of heat transfer in vacuum vapor phase welding was established, and temperature monitoring points corresponding to the measured temperature locations were set in the numerical simulation model. The parameter inversion program wrote the process parameters into a parameter file, and the numerical simulation program read the parameter file and updated the corresponding process control conditions or boundary conditions. After completing the solution, the simulated temperature of each temperature monitoring point was output.

[0087] The process parameters to be inverted in this embodiment include thermal conductivity adjustment parameters. Specific heat capacity adjustment parameter and the temperature control parameters corresponding to different process time intervals. , , and The vector of process parameters to be inverted is represented as follows:

[0088]

[0089] The process parameters correspond to the pre-set process control quantities or boundary condition parameters in the numerical simulation model of heat transfer in vacuum vapor phase welding. Initial values, lower bounds, and upper bounds are set for each process parameter based on the actual equipment and process requirements.

[0090] According to the vacuum vapor phase welding process, the temperature-time series is divided into a heating stage, a holding stage, a rapid heating stage, a vacuuming stage, and a cooling stage. The set of sampling times corresponding to the s-th vacuum process stage is represented as follows: The number of sampling points in this stage is expressed as .

[0091] Input the current process parameters into the heat transfer numerical simulation model to obtain the simulated temperature at each temperature measurement point. In the k-th iteration, the root mean square error of the s-th vacuum process stage is expressed as:

[0092]

[0093] Where L is the number of temperature measurement points. To simulate temperature, This is the measured temperature.

[0094] Calculate the average stage error for all vacuum process stages, and generate dimensionless raw weights based on the ratio of each stage error to the average stage error. Then, perform amplitude limiting and smoothing on these raw weights to obtain dynamic stage weights.

[0095]

[0096] in, Let be the dimensionless original weight of the s-th vacuum process stage. This is the weighted smoothing coefficient.

[0097] Further compensation is made to the dynamic stage weights based on the number of sampling points in each stage. The time-point weight of the j-th sampling moment belonging to the s-th vacuum process stage is expressed as:

[0098]

[0099] Where M represents the number of vacuum process stages, and N represents the total number of temperature sampling moments.

[0100] Subsequently, finite difference perturbations are applied to the current process parameters, and the temperature response after perturbation is obtained by calling the heat transfer numerical simulation model. The perturbation amount is determined based on the absolute value of the current parameters and the preset minimum perturbation amount; when the positive perturbation exceeds the upper limit of the parameters, a reverse perturbation is adopted, and the actual perturbation amount after boundary constraints is used as the denominator of the difference, thereby constructing the Jacobian matrix.

[0101] Stack the temperature residuals and Jacobian matrices of all temperature measurement points in the same order. Construct a diagonal weight matrix based on the time-point weights. The weighted residual vector and the weighted Jacobian matrix are expressed as follows:

[0102]

[0103] Calculate the weighted gradient and approximate Hessian matrix based on the weighted residual vector and the weighted Jacobian matrix:

[0104]

[0105] Based on the weighted gradient and the approximate Hessian matrix, Levonburg-Marquardt candidate directions are generated, and PRP+ conjugate gradient candidate directions are generated based on the current weighted gradient and the previous effective conjugate gradient direction. The two candidate directions are then superimposed onto the current process parameters and constrained according to the upper and lower bounds of the parameters to obtain two candidate parameter vectors.

[0106] The two candidate parameter vectors are input into the heat transfer numerical simulation model to obtain the corresponding candidate simulation temperatures. During the candidate parameter evaluation process, the time-point weights of this iteration remain unchanged, and the weighted objective function for each candidate parameter is calculated.

[0107] For any candidate parameter, its actual decrease and predicted decrease are expressed as follows:

[0108]

[0109] When both the actual decrease and the predicted decrease are greater than zero, the decrease ratio is expressed as:

[0110]

[0111] The descent ratios of the Levenberg-Marquardt candidate parameters and the conjugate gradient candidate parameters are compared, and the effective candidate parameter with the larger descent ratio is selected. When the descent ratio of the selected candidate parameter is greater than zero, the candidate parameter is accepted, and the process parameters, simulation temperature, dynamic stage weights, and damping coefficients are updated. When the descent ratio is not greater than zero, the current process parameters and simulation temperature are kept unchanged, and the damping coefficient is increased. The process of stage error calculation, dynamic weight update, Jacobian matrix construction, candidate parameter generation, and candidate parameter evaluation is repeated until the aforementioned preset stopping conditions are met. After the parameter inversion terminates, the final process parameter vector and its corresponding simulation temperature are output, and the root mean square error, dynamic stage weights, objective function changes, and candidate direction selection results of each vacuum process stage can be recorded, thereby completing the vapor phase welding process parameter inversion based on measured temperature data.

Claims

1. A method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages, characterized in that, Includes the following steps: S1. Obtain the measured temperature time series of at least one temperature measurement point in the vapor phase welding process, establish a numerical simulation model of heat transfer in vapor phase welding, and determine the vector of process parameters to be inverted, the initial values ​​of process parameters, and the range of values ​​for each process parameter. S2. Based on the preset stage boundary times, the vapor phase welding process is divided into multiple vacuum process stages, and corresponding process stages are assigned to each temperature sampling time. S3. Input the current process parameter vector into the vapor phase welding heat transfer numerical simulation model to obtain the simulated temperature of each temperature measurement point, calculate the temperature residual between the simulated temperature and the measured temperature, and calculate the stage error based on the temperature residual within each vacuum process stage; perform dimensionless, amplitude limiting and smoothing processing on the stage error to obtain the dynamic stage weight, and expand the dynamic stage weight into time-point weights according to the number of sampling points in each vacuum process stage. S4. Apply finite difference perturbation to each process parameter in the current process parameter vector, obtain the corresponding temperature response change through the vapor phase welding heat transfer numerical simulation model and construct the Jacobian matrix; construct the weighted residual vector and weighted Jacobian matrix according to the time-point weights, and calculate the weighted gradient and approximate Hessian matrix. S5. Generate Levenberg-Marquardt candidate directions based on the weighted gradient and the approximate Hessian matrix, and generate conjugate gradient candidate directions based on the current weighted gradient and the previous effective conjugate gradient direction. The two candidate directions are superimposed onto the current process parameter vector and their value ranges are constrained to obtain two candidate parameter vectors. S6. Input the two candidate parameter vectors into the vapor phase welding heat transfer numerical simulation model respectively. While keeping the weights of each time point of this iteration unchanged, calculate the ratio of the actual decrease to the predicted decrease corresponding to the two candidate parameter vectors, and select the effective candidate parameter vector with the larger ratio to update the current process parameter vector. S7. Repeat steps S3 to S6 until the preset stopping condition is met, output the final process parameter vector and its corresponding simulation temperature, and obtain the vapor phase welding process parameter inversion result.

2. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 1, characterized in that: In step S3, the stage error of the s-th vacuum process stage in the k-th iteration is expressed as the root mean square error: in, Let be the set of sampling times corresponding to the s-th vacuum process stage. The number of sampling points in the set of sampling times. The number of temperature measurement points, Let be the simulated temperature of the i-th temperature measurement point at the j-th sampling time during the k-th iteration. This corresponds to the measured temperature.

3. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 2, characterized in that: In step S3, the average stage error of all vacuum process stages is expressed as: The dimensionless original weight of the s-th vacuum process stage is expressed as: The dynamic stage weight of the s-th vacuum process stage is represented as: The time-by-time weight of the j-th sampling moment belonging to the s-th vacuum process stage is expressed as: Where M represents the number of vacuum process stages, and N represents the total number of sampling times. To prevent zero parameters, The error amplification index is... and These are the lower and upper weight limits, respectively. This is the weighted smoothing coefficient.

4. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 1, characterized in that: In step S4, the temperature residuals of all temperature measurement points are stacked according to the temperature measurement points to obtain a residual vector. The Jacobian matrices corresponding to each temperature measurement point are stacked in the same order to obtain the Jacobian matrix. ; Construct a diagonal weight matrix based on the weights at each time point. The weighted residual vector and the weighted Jacobian matrix are expressed as follows: The weighted gradient and the approximate Hessian matrix are respectively expressed as:

5. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 1, characterized in that: In step S4, the preset finite difference perturbation of the i-th process parameter is expressed as: in, This represents the relative perturbation ratio. The minimum disturbance amount; Forward differencing is used when the process parameters after a positive disturbance do not exceed the corresponding upper bound; reverse differencing is used when the process parameters after a positive disturbance exceed the corresponding upper bound, and the actual disturbance amount is constrained by the value range. As the denominator of the difference, the Jacobian matrix column corresponding to the i-th process parameter is represented as: in, Let be the i-th unit direction vector; when a disturbance direction cannot produce an effective disturbance, the opposite direction is used for disturbance.

6. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 4, characterized in that: In step S5, the Lewenberg-Marquardt candidate directions are represented as follows: in, The damping coefficient is... The diagonal damping matrix; When calculating the first conjugate gradient candidate direction or when no valid previous conjugate gradient direction exists, the conjugate gradient candidate direction is the negative gradient direction; when a valid previous conjugate gradient direction exists, the PRP+ coefficient is expressed as: The candidate directions of the conjugate gradient are represented as follows: When the conjugate gradient candidate direction does not satisfy the descent direction condition, it is reset as a negative gradient direction.

7. The method for inverting vapor phase welding process parameters based on dynamic weighting of vacuum process stages according to claim 1, characterized in that: In step S6, under the condition that the weights remain unchanged at each time point in this iteration, the weighted objective function is expressed as: For any candidate direction, the effective step size is determined based on the difference between the candidate parameter vector constrained by the value range and the current process parameter vector. The actual decrease is expressed as: The predicted decrease is expressed as: When both the actual decrease and the predicted decrease are greater than zero, the decrease ratio is expressed as: Select a candidate parameter vector with a large decrease ratio; when the decrease ratio corresponding to the candidate parameter vector is greater than zero, accept and update the current process parameter vector; when the decrease ratio is not greater than zero, keep the current process parameter vector unchanged and increase the damping coefficient. The preset stopping conditions include at least one of the following: the maximum absolute error between the simulated temperature and the measured temperature is less than a preset error threshold; the damping coefficient exceeds a preset upper limit; the relative decrease rate of the weighted objective function for a consecutive preset number of times is less than a preset decrease threshold; the infinite norm of the weighted gradient is less than a preset gradient threshold; the relative norm of the effective step size corresponding to the accepted candidate parameter is less than a preset step size threshold; or the number of iterations reaches a preset maximum number of iterations.

Images