Welding method and system for conical sealed cold storage box cover

Abstract

The invention provides a conical sealing cold storage box cover welding method and system, and relates to the field of intelligent control, and the method comprises the steps: collecting temperature field data of a conical contact surface, building a stress distribution prediction model, and outputting a stress concentration coefficient matrix; based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed, and a welding power density distribution function and a welding seam track correction curve are generated; a layered progressive heat input compensation strategy is adopted for conducting dynamic compensation on welding heat input; and a welding parameter optimization model is established in combination with the welding seam track correction curve and a heat input compensation result, an optimal parameter combination is solved, and welding path planning data is generated. In this way, the problems of uneven heat input and stress concentration in the welding process can be solved, and the welding stability and precision are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control, and more specifically, to a conical sealing cold storage box cover welding method and system. BACKGROUND

[0002] With the continuous development of industrial technology, the application of sealed cold storage devices in cold chain transportation, medical storage, aerospace and other temperature control fields is becoming increasingly widespread. As a special structure designed cold storage device, the conical sealing cold storage box gradually becomes a research hotspot in the industry because it can provide excellent sealing performance and thermal efficiency under high pressure and low temperature conditions. The welding between the conical sealing cold storage box body and the box cover, as one of the key processes in device manufacturing, directly affects its sealing performance, durability and service life. Existing welding technologies mainly include laser welding, electron beam welding and electric arc welding, etc. These technologies have made significant progress in the welding of planar structures or regular geometric shapes, but still face many challenges in the welding of complex contact surfaces of conical sealing boxes and covers. The conical contact surface has a complex shape, and the temperature field distribution of the weld is extremely uneven, which can easily lead to stress concentration, welding defects (such as cracks and pores) and welding deformation. In addition, due to the instability of heat input during the welding process, the quality and consistency of the weld are difficult to guarantee, which further affects the sealing performance and mechanical strength of the overall device.

[0003] In the prior art, some research has proposed improvement methods for welding temperature field distribution and stress concentration. For example, by optimizing the welding path or adjusting the heat input parameters, the quality of the weld can be improved to some extent. However, these methods usually rely on manual experience and lack precise temperature field and stress distribution data support, and the optimization effect is not ideal. Especially for conical sealing structures, due to the complex geometric characteristics of the contact surface such as arc surface and inclined surface, it is difficult for traditional algorithms to achieve precise optimization of welding power density distribution and welding trajectory. In addition, existing heat input compensation strategies mostly use single fixed parameter adjustment, which cannot dynamically adapt to the real-time changes of the temperature field during the welding process, thus leading to the continuous existence of welding stress concentration and heat input deficiency or overload problems. Overall, the existing technology has significant shortcomings in welding quality consistency, stress concentration relief and welding path optimization, which limits the further development of conical sealing cold storage box welding technology. SUMMARY

[0004] In order to solve the above technical problems, the present application is proposed. The present application provides a conical sealing cold storage box cover welding method and system, which can to some extent solve the problems of uneven weld quality, decreased sealing performance and increased welding defects caused by the complex temperature field distribution of the conical contact surface, the difficulty in predicting stress concentration, the insufficient optimization of welding power density and weld trajectory, and the inability to dynamically compensate heat input.

[0005] According to one aspect of the present application, a conical sealing cold storage box cover welding method is provided, comprising:

[0006] Collecting temperature field data of the conical contact surface of the conical sealing cold storage box body and the conical sealing cold storage box cover, establishing a stress distribution prediction model based on the temperature field data, and outputting a stress concentration coefficient matrix;

[0007] Based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed, and a welding power density distribution function and a weld trajectory correction curve are output;

[0008] According to the welding power density distribution function, a hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input;

[0009] Based on the output results of the weld trajectory correction curve and the hierarchical progressive heat input compensation strategy, a welding parameter optimization model is established, and welding path planning data is generated by solving the optimal parameter combination.

[0010] Further, an infrared sensor array is used to collect temperature field data of the conical contact surface, which is denoised through triple filtering processing and reconstructed through hyperboloid interpolation algorithm to obtain complete temperature field data.

[0011] Further, based on the temperature field data, a stress distribution prediction model is constructed to output a stress concentration coefficient matrix;

[0012] The stress distribution prediction model is realized by using an improved finite element neural network structure, which includes a finite element calculation path and a neural network prediction path;

[0013] The finite element path calculates the initial thermal stress, and the neural network prediction path extracts temperature features and dynamically adjusts the weight to cope with temperature field changes.

[0014] Further, based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed;

[0015] The multi-parameter collaborative optimization algorithm uses an improved particle swarm optimization algorithm, with welding power density and trajectory offset as key optimization variables;

[0016] When the improved particle swarm optimization algorithm obtains the optimal solution, the welding power density is reconstructed, and the welding power density distribution function and the trajectory correction curve are output.

[0017] Further, based on the welding power density distribution function, a hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input;

[0018] The hierarchical progressive heat input compensation strategy adopts a double-layer control structure, including a fast response layer and a fine adjustment layer.

[0019] The fast response layer uses a proportional-differential controller to quickly correct short-term deviations;

[0020] The fine adjustment layer accurately compensates for long-term errors through a fuzzy adaptive controller.

[0021] Further, the fast response layer adjusts the proportional coefficient and the differential coefficient according to the deviation characteristics to improve the response capability and damping characteristics;

[0022] The fine adjustment layer optimizes the compensation effect by adaptively adjusting the fuzzy rule weight, and combines the membership function offset and weight recovery strategy to cope with complex deviation characteristics.

[0023] Further, a welding parameter optimization model is established based on the trajectory correction curve and the layered progressive heat input compensation strategy;

[0024] The welding parameter optimization model adopts a three-layer nested structure: an outer main optimization loop, a middle constraint processing loop, and an inner parameter refinement loop.

[0025] Further, the outer main optimization loop is responsible for global search, and convergence control is achieved by dynamically adjusting the search range. When the search range is insufficient, the current direction is marked as invalid;

[0026] The middle constraint processing loop ensures the feasibility of the solution, and gradually adjusts the parameters through constraint checking and coordination mechanism;

[0027] The inner parameter refinement loop performs local optimization near the constraint solution, and combines micro-grid and adaptive encryption mechanism to improve optimization precision.

[0028] Further, based on the optimal solution of the welding parameter optimization model, a control instruction sequence containing position, angle, depth, speed and time is generated by projecting it along the weld trajectory, and converted into a segmented path planning data packet to control the welding path.

[0029] According to another aspect of the present application, a conical sealing cold storage box cover welding system is provided, which comprises:

[0030] The acquisition module is used for acquiring the temperature field data of the conical contact surface of the conical sealing cold storage box body and the conical sealing cold storage box cover, establishing a stress distribution prediction model, and outputting a stress concentration coefficient matrix;

[0031] The correction module is used for constructing a multi-parameter collaborative optimization algorithm according to the stress concentration coefficient matrix, and outputting a welding power density distribution function and a weld trajectory correction curve;

[0032] A compensation module is configured to dynamically compensate welding heat input by using a hierarchical progressive heat input compensation strategy according to the welding power density distribution function.

[0033] A planning module is configured to establish a welding parameter optimization model based on the welding seam trajectory correction curve and the hierarchical progressive heat input compensation strategy, solve an optimal parameter combination, and generate welding path planning data.

[0034] Compared with the prior art, the present application can optimize the welding parameter distribution from a global perspective by calculating a stress concentration coefficient matrix, constructing a multi-parameter collaborative optimization algorithm, generating a welding power density distribution function and a welding seam trajectory correction curve, and guaranteeing welding quality and trajectory accuracy. The hierarchical progressive heat input compensation strategy can be used to dynamically compensate welding heat input, effectively solving the welding deformation and quality defect problems caused by uneven heat input. The welding parameter optimization model can be established by combining the welding seam trajectory correction curve and the heat input compensation result, and the welding path planning data can be generated by solving the optimal parameter combination, thereby realizing accurate control and stable operation of the welding process. The welding precision and stability are significantly improved, the stress concentration and welding defects are reduced, the product quality and service life are improved, the production efficiency is optimized, and the application value is high. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0036] Figure 1 A flowchart of the conical sealing cold storage box cover welding method according to the embodiment of the present application;

[0037] Figure 2 A flowchart of the hierarchical progressive heat input compensation strategy according to the embodiment of the present application. DETAILED DESCRIPTION

[0038] In the following, the example embodiments according to the present application will be described in detail with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.

[0039] As described in the above background, the prior art mainly has the following prominent problems in the welding process of the conical sealing cold storage box cover: first, due to the difficulty in accurately predicting the stress distribution of the welding area, stress concentration phenomenon is easy to occur in the welding process, which leads to the decrease of the strength of the welded joint and the shortening of the service life; second, the welding heat input control is not accurate enough, and uneven heat input will cause welding deformation, cracks and other quality defects, affecting the overall performance and reliability of the product.

[0040] Figure 1 The system flowchart of the conical sealing cold storage box cover welding method according to the embodiment of the present application. As shown in the figure, in the conical sealing cold storage box cover welding method, it comprises: Figure 1

[0041] S1: Collecting the temperature field data of the conical contact surface of the conical sealing cold storage box body and the conical sealing cold storage box cover, establishing a stress distribution prediction model according to the temperature field data, and the stress distribution prediction model outputs a stress concentration coefficient matrix.

[0042] A high-precision infrared sensor array is used to arrange a temperature collection network, and every 20 mm in the radial direction of the conical contact surface is provided with a temperature measurement ring, and a total of 8 temperature measurement rings are provided. Each temperature measurement ring is uniformly arranged with 16 infrared sensor temperature measurement points in the circumferential direction, forming a 16x8 temperature collection matrix.

[0043] The determination of the arrangement density of the temperature measurement points in the radial and circumferential directions is based on the temperature field test of the standard test piece. The conical contact surface is divided into three characteristic regions of top, middle and bottom, and an initial temperature measurement point array is arranged in each region for temperature collection. The temperature evolution data of each region is obtained through welding pre-experiment, and the relationship between the temperature rising rate and the distance of the heat source is analyzed to ensure that the change characteristics of the temperature field are captured.

[0044] The temperature collection process is divided into a preheating stage and a welding stage. The sampling frequency of the preheating stage is 1 Hz, and the duration is 300 seconds, so as to obtain the initial temperature distribution of the conical contact surface. When entering the welding stage, the sampling frequency is increased to 10 Hz to ensure that the transient temperature change in the welding process can be accurately captured. If a temperature mutation is detected by a certain temperature measurement point, i.e. the temperature change rate exceeds 50℃ / s, the sampling frequency of the region is increased to 20 Hz.

[0045] The original temperature data collected by the sensor is subjected to triple filtering processing:

[0046] The median filter is used to remove outliers, and the filter window size is 5 data points. The data in the window is sorted according to the temperature value, and the middle value is taken as the effective data.

[0047] ​The Kalman filter is used to eliminate random noise, and the state transition matrix reflecting the evolution of the temperature field is constructed, which has a dimension of twice the number of temperature measurement points, including two state quantities of temperature value and temperature change rate. When the temperature change rate is less than 5℃ / s, linear prediction is used, and when it exceeds, nonlinear prediction is switched.

[0048] Low-pass filtering is used to remove high-frequency interference, and a Butterworth fourth-order filter is used with a cutoff frequency of 1 / 4 of the sampling frequency.

[0049] If the deviation between the filtered temperature data and the original data exceeds 5℃, it is marked as an abnormal data point for review.

[0050] In actual measurement, due to the limitations of sensor installation space, cost and interference, etc., it is not possible to arrange an infinite number of temperature measurement points on the conical contact surface. Even if a 16x8 temperature measurement matrix is used, there are still blank areas between the measurement points. In order to obtain the complete temperature distribution state of the contact surface, a hyperboloid interpolation algorithm is used to reconstruct the continuous temperature field. The interpolation process combines the thermal conductivity characteristics of the material and the geometric characteristics of the structure. When the temperature difference between adjacent temperature measurement points is greater than 100℃, the interpolation grid is automatically encrypted, and the grid size can be refined to 1 / 4 of the original.

[0051] At the same time, a real-time data verification mechanism is set up, including sensor fault detection and temperature anomaly alarm. When the output value of a certain sensor is abnormal, such as exceeding the range or fluctuating too much, switch to the adjacent sensor for compensation measurement. If the local temperature exceeds the maximum allowable use temperature of the material, an alarm signal is immediately sent out, and the temperature data of the region is recorded at a frequency of 100Hz for subsequent analysis.

[0052] All collected temperature field data is stored in the local database in real time, and automatically backed up every 300 seconds. The data recording format includes temperature measurement point coordinates, time stamp, temperature value and sensor status code. When the database capacity reaches 90% of the preset value, the early data is compressed and archived, and only the complete record of the key moment is retained.

[0053] Based on the obtained temperature field data, a stress distribution prediction model is constructed to output the stress concentration coefficient matrix.

[0054] The stress distribution prediction model is implemented using an improved finite element neural network structure, which includes a finite element calculation path and a neural network prediction path.

[0055] The finite element path divides the conical contact surface based on tetrahedral elements, each element containing 20 nodes, and adjacent elements sharing boundary nodes. The neural network prediction path uses a four-layer structure, the input layer receives temperature field data and boundary conditions, the two hidden layers contain 128 and 64 neurons respectively, and the output layer generates stress distribution prediction values.

[0056] When new temperature field data is input, the finite element path first calculates the initial thermal stress. The thermal stress calculation uses the linear elastic constitutive equation, considering the volume deformation and elastic modulus temperature effect caused by temperature. The stress tensor is decomposed into hydrostatic stress and deviatoric stress, which are calculated separately and then superimposed to obtain the total stress. If singular points are found during calculation, the grid is refined until the convergence condition is met.

[0057] The neural network prediction path synchronously processes input data and extracts features using an improved residual connection structure. The first hidden layer uses the GELU activation function to extract local features of the temperature field. The second hidden layer uses an attention mechanism to assign different weights to features in different regions. When the temperature gradient in a certain region is greater than 1.5 times the average of the surrounding regions, the corresponding attention weight is automatically increased. The output layer uses a linear activation function to generate stress distribution prediction values.

[0058] The result fusion of the dual-path uses an adaptive weight mechanism. When the temperature field changes smoothly, the weight of the finite element path is larger, with a typical value of 0.7; when the temperature field changes dramatically, the weight of the neural network prediction path increases, with a maximum of 0.6. The weight adjustment is based on the time derivative of the temperature field, and the sigmoid function is used to achieve smooth transition.

[0059] The online learning mechanism of the stress distribution prediction model consists of two stages:

[0060] The first stage compares the prediction results of the dual-path, and when the difference exceeds 10%, the sample is marked as a key sample.

[0061] In the second stage, after collecting 100 key samples, the neural network parameters are updated using these samples, while the finite element path remains unchanged. The update process uses a small batch training method, with each batch containing 16 samples and a learning rate of 0.0001.

[0062] When constructing the stress concentration coefficient matrix, normal elements far from the stress concentration area are selected as the reference benchmark. For each node, the ratio of local stress to nominal stress is calculated to obtain the initial stress concentration coefficient. Spatial smoothing is performed on the stress concentration coefficient to ensure the continuity of the numerical distribution.

[0063] The output basic matrix structure adopts a 80x80 two-dimensional array form, covering the entire conical contact surface. Each matrix element contains three key information: the stress concentration coefficient value of the point, the spatial position coordinates, and the stress state identifier.

[0064] S2: Based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed, which uses an improved particle swarm structure to output the welding power density distribution function and the weld trajectory correction curve.

[0065] Based on the constructed stress concentration coefficient matrix, an improved particle swarm optimization algorithm is constructed. The particle swarm consists of 200 particles, each particle containing two key optimization variables: welding power density and trajectory offset. Among them, the welding power density describes the energy input distribution per unit area, and the trajectory offset represents the position offset relative to the theoretical weld centerline.

[0066] When the regional stress concentration coefficient exceeds 1.5 times the average value, the initial value of the welding power density of the corresponding region is set to 80% of the standard power density; when the stress concentration coefficient exceeds 2 times the average value, the initial value of the power density is reduced to 65% of the standard power density; when the stress concentration coefficient exceeds 3 times the average value, the initial value of the power density is further reduced to 50% of the standard power density. The reduction of power density adopts a stepwise decrease, and a 10% transition interval is set between adjacent levels to ensure smooth transition of power distribution.

[0067] When the regional stress concentration coefficient is less than 0.8 times the average value, the initial value of the welding power density of the corresponding region is increased to 120% of the standard power density; when the stress concentration coefficient is less than 0.6 times the average value, the initial value of the power density is increased to 135% of the standard power density; when the stress concentration coefficient is less than 0.4 times the average value, the initial value of the power density is increased to 150% of the standard power density. The power increase also adopts a stepwise increase, and a 5% transition interval is set near the critical value of the stress concentration coefficient to ensure the smoothness of power adjustment.

[0068] When the calculated power density is lower than the minimum power density required to ensure the penetration, the power density of the region is forced to be limited to the minimum power density value. To prevent local overheating, set the upper limit of the power density to 160% of the standard power density, and perform truncation processing when it exceeds this value.

[0069] In the particle swarm initialization stage, the Latin hypercube sampling method is used to generate the initial particle distribution, ensuring that the search space is uniformly covered. The initial speed of each particle is randomly generated by Gaussian distribution, and the speed range is limited to [-10%, 10%] of the variable definition domain. To prevent the particle swarm from falling into local optimum too early, set the dynamic inertia weight, with an initial value of 0.9, linearly decreasing to 0.4 with the iteration number.

[0070] The fitness function of the algorithm is composed of stress uniformity index and temperature field stability index. The stress uniformity index is obtained by calculating the standard deviation of the stress field, and the smaller the standard deviation, the more uniform the stress distribution; the temperature field stability index is obtained by integrating the absolute value of the temperature field time derivative, and the smaller the integral value, the more stable the temperature field. These two indicators are combined to form the final fitness value according to the weight ratio of 3:2.

[0071] In each iteration process, stress field simulation calculation is performed for each particle in the population. The simulation adopts adaptive mesh technology, which densifies the mesh in areas with large stress gradients to improve calculation accuracy. When the maximum stress concentration coefficient is detected to decrease by more than 15% compared with the initial value, the current particle corresponding to the welding power density and the trajectory offset parameter combination is recorded.

[0072] To improve the convergence efficiency of the algorithm, a local search mechanism is introduced. Every 50 iterations, a fine search is performed on the neighborhood of the current optimal solution, and the search step is dynamically adjusted with the iteration number. At the same time, a population diversity maintenance strategy is set, and when the population aggregation degree exceeds the threshold, some particles are reinitialized to avoid falling into a local optimal solution.

[0073] After 1000 iterations of optimization, the solution with the optimal fitness value is selected from all solutions that meet the stress reduction requirement as the final output. The welding power density corresponding to this solution is converted into a continuous distribution function, and the trajectory offset is converted into a smooth correction curve.

[0074] Specifically, after the particle swarm optimization algorithm obtains the optimal solution, the welding power density is reconstructed. The reconstruction process is based on the partition fitting strategy, which divides the conical contact surface into three characteristic regions along the radial direction: the top transition zone, the middle stable zone, and the bottom end zone. In the top transition zone, a quadratic function is used to describe the rapid rise of the power density; in the middle stable zone, a piecewise linear function is used to maintain stable power input; and in the bottom end zone, an exponential function is used to achieve a gentle decay of the power.

[0075] When the power density difference between adjacent regions exceeds 20%, a transition section is added at the boundary of the region, and the length of the transition section is 15% of the length of the two characteristic regions. In the transition section, a cubic spline function is used to achieve smooth transition, ensuring the continuity of the power density and the continuity of the first derivative.

[0076] For the weld trajectory correction curve, first determine the key control points based on the stress field distribution, which are located at positions where the stress gradient changes significantly. The initial correction curve is fitted by the least squares method to ensure the position constraints of the starting point and the ending point. When the curvature of the curve exceeds the maximum value allowed by the welding process, a new control point is inserted in the region to re-fit. To avoid excessive trajectory deviation, a transition circular arc is inserted at positions where the tangent angle changes by more than 30 degrees, and the radius of the circular arc is dynamically calculated according to the welding speed. Finally, the correction curve is discretized into a sequence of identifiable path points.

[0077] S3: According to the welding power density distribution function, a hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input;

[0078] A hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input. The compensation strategy adopts a double-layer control structure, including a fast response layer and a fine adjustment layer. The fast response layer is responsible for processing short-term fluctuations, and the fine adjustment layer is responsible for eliminating long-term cumulative errors. The two layers work together to achieve accurate heat input control.

[0079] The fast response layer uses a proportional-differential controller to process short-term power deviations. The proportional coefficient Kp is initially set to 0.8, and the differential coefficient Kd is initially set to 0.15. The controller samples the welding power density value every 0.1 seconds. When the power density deviation is detected to be within ±5% of the set value, the proportional term generates an immediate correction based on the current deviation, and the differential term provides lead compensation based on the deviation rate to suppress fluctuation trends.

[0080] When the power density deviation is within ±2% for 3 consecutive sampling periods, control is switched to the fine adjustment layer.

[0081] The fine adjustment layer uses a fuzzy adaptive controller with a control period of 0.2 seconds. The controller's input variables include the cumulative power density deviation and the deviation rate, and the output variable is the power compensation amount. The fuzzy rule base contains 49 rules covering 7 power deviation levels and 7 change rate levels.

[0082] In the fine adjustment layer, the cumulative deviation is calculated through a sliding time window with a window length of 2 seconds. When the deviation consistently deviates in the same direction, the control effect is enhanced by adjusting the center of gravity of the membership function; when the deviation shows a fluctuation trend, the control gain is reduced by widening the membership function.

[0083] The switching between the two control layers uses a dynamic threshold mechanism. When the fast response layer detects a power deviation exceeding ±5% or the fine adjustment layer finds a cumulative deviation exceeding ±2%, the control layer switching is triggered. During the switching process, a gradual transition strategy is adopted, and the new control layer gradually takes over control within the first 0.3 seconds, ensuring smooth transition.

[0084] If a power surge occurs during the transition period, the transition process is immediately suspended, and the original control layer resumes full control.

[0085] When a control layer cannot control the deviation within the target range within a specified time, the control parameters are checked for adjustment. For the fast response layer, the proportional and differential coefficients are mainly adjusted; for the fine adjustment layer, the weight distribution of the fuzzy rules is optimized.

[0086] Specifically, for the parameter adjustment of the fast response layer, the proportional coefficient Kp and the differential coefficient Kd are dynamically updated based on the deviation characteristics. When it is detected that the deviation is positive and increasing for 5 consecutive sampling periods, the Kp value is increased by 20%, and the Kd value is increased by 15% to enhance the system response capability. When the deviation shows oscillation characteristics and the oscillation period is less than 0.3 seconds, the Kp value is reduced by 25%, and the Kd value is increased by 30% to enhance the system damping characteristics. If the deviation randomly fluctuates within ±3%, the Kp value is reduced by 10%, and the Kd value is kept unchanged. When the adjusted Kp value exceeds 1.5 or is less than 0.4, an alarm is triggered and the default parameters are restored.

[0087] For the fuzzy rule weight optimization of the fine adjustment layer, a performance index driven adaptive adjustment strategy is adopted. When the cumulative deviation exceeds ±2% for 1 second, the weight of the corresponding rule is adjusted. If the cumulative deviation is positive, the weight coefficient of the corresponding negative output rule is increased by 15% of the original weight; if the cumulative deviation is negative, the weight coefficient of the positive output rule is increased by 15% of the original weight. When reverse overshoot occurs, the weight of the triggered rule is reduced by 20%, and the weight of the adjacent rule is increased by 10%.

[0088] In the fuzzy rule weight optimization process, the upper limit of the weight of a single rule is set to 1.8, and the lower limit is set to 0.4. When the weight of a rule reaches the upper limit value and the system performance has not improved, the membership function center position related to the rule is shifted 5% in the desired direction. If the system performance does not improve significantly after 5 consecutive weight adjustments, reset all rule weights to the initial value and reduce the coverage of the membership function.

[0089] In the fuzzy rule library, the rule weight adjustment for the deviation rate of change is more sensitive. When it is detected that the deviation rate of change exceeds 2 times the preset threshold, the weight of the corresponding rule is immediately increased by 30%. If the rate of change always exceeds the threshold within 3 control periods, the weights of the adjacent rules are gradually increased, with a step size of 10%. When the rate of change returns to the normal range, the weights are gradually restored to the standard value within 5 control periods, with a weight difference of 20% restored each period.

[0090] S4: Based on the weld seam trajectory correction curve and the output result of the layered progressive heat input compensation strategy, a welding parameter optimization model is established, and welding path planning data is generated by solving the optimal parameter combination.

[0091] Based on the output results of the weld trajectory correction curve and the hierarchical progressive heat input compensation strategy, a welding parameter optimization model is established. The model adopts a three-layer nested structure: the outer main optimization loop is a global search, the middle constraint processing loop processes the constraint conditions, and the inner parameter refinement loop performs parameter refinement. The decision variables include the welding angle, the welding depth, and the welding speed. The objective function represents the joint strength, which is constructed by considering the tensile strength, the bending strength, and the impact toughness.

[0092] Specifically, the outer main optimization loop is responsible for the global search strategy. At the beginning of each iteration, a search space is constructed based on the current optimal solution. The initial range of the search space is ±3 degrees for the welding angle, ±2% for the welding depth, and ±3% for the welding speed. When the improvement of the objective function in three consecutive iterations is less than 0.5%, the search range is reduced to 70% of the original. When the search range is less than the minimum allowed value, i.e., the welding angle is less than 0.5 degrees, the welding depth is less than 0.3%, and the welding speed is less than 0.5%, the outer optimization is considered to be converged.

[0093] The middle constraint processing loop is responsible for the feasible solution search. When a new candidate solution is generated by the outer loop, a constraint check is performed first. If the welding angle change rate exceeds 5° / s, the angle adjustment amount is proportionally reduced until the constraint is met. When the welding depth exceeds the range of 75%-85%, the bisection method is used to adjust the depth value gradually, with a step size of half the excess. If the welding speed fluctuation exceeds ±10%, the speed is immediately controlled within the constraint range, and the relevant parameters are recalculated.

[0094] When a constraint adjustment causes other constraints to be violated, a constraint coordination mechanism is triggered. First, the constraint violation degrees are sorted, and then the constraints with the largest violation degrees are processed one by one. If a feasible solution cannot be obtained after three coordinations, the search step size is enlarged to 1.5 times the original, and the constraint processing is restarted. When the search step size exceeds twice the original range, the last feasible solution is returned, and the current search direction is marked as invalid.

[0095] The inner parameter refinement loop is responsible for local optimization. After a solution that satisfies the constraints is obtained, a fine search is performed around the solution. First, the parameter space is divided into fine grids, with a grid size of 1 / 10 of the outer search step size. Then, the grid points are evaluated, and the point with the optimal objective function value is selected as the refinement center. If the objective function value of the refinement center is better than the current solution, the point is accepted as the new current solution.

[0096] In the inner layer optimization process, an adaptive grid refinement mechanism is set. When the objective function value of a certain grid point is more than 5% higher than that of the surrounding points, the grid near the point is refined, and the new grid size is 1 / 4 of the original. If the difference between the optimal solution after refinement and the optimal solution of the original grid is less than 0.1%, the refinement process is stopped. When the number of grid refinements exceeds 3, the inner layer optimization is stopped, and the current optimal solution is returned.

[0097] The data transmission between the three layers adopts a bidirectional mutual feedback mechanism. When the inner layer finds that the fine search in a certain direction is continuously effective, it feeds back to the middle layer to expand the constraint margin in that direction. When the middle layer finds that a certain type of constraint is often violated, it feeds back to the outer layer to adjust the search strategy of the corresponding parameter. When the outer layer search falls into local optimum, it requires the inner layer to expand the search range and find possible escape directions.

[0098] The optimal solution parameters are projected along the weld trajectory with a projection interval of 5mm, and the complete parameter combination is recorded at each projection point. When the parameter difference between adjacent projection points exceeds the preset threshold, i.e., the welding angle difference exceeds 2 degrees, the depth difference exceeds 1%, and the speed difference exceeds 3%, a transition point is automatically inserted to make the parameter change more gentle.

[0099] Then, a time-discrete control instruction sequence is generated based on the projection point sequence, each control instruction containing the following information: spatial position coordinates, welding angle, welding depth control parameter, welding speed, and target time to reach the point. The time interval between adjacent instructions is dynamically calculated according to the welding speed to ensure motion continuity.

[0100] The control instruction sequence is converted into path planning data packets recognizable by the welding equipment. The data packets adopt a segmented structure, each segment containing start point parameters, end point parameters, transition characteristic description, and control compensation. The transition characteristic description defines how the parameters gradually change from the start point to the end point, using a cubic spline curve to achieve smooth transition. The control compensation comes from the real-time output of the multi-layer progressive heat input compensation strategy.

[0101] In summary, a conical sealing cold storage box cover welding method based on the embodiment of the present application is illustrated, which solves the problems of complex temperature field distribution, stress concentration, insufficient welding power and trajectory control precision, and difficulty in dynamic adjustment of heat input during the welding process through high-precision temperature field data acquisition, stress distribution prediction, multi-parameter collaborative optimization, and dynamic heat input compensation. At the same time, based on the stress concentration coefficient matrix, the improved particle swarm optimization algorithm realizes the dynamic optimization of welding power density and trajectory, alleviates the overheating phenomenon in the stress concentration area, and improves the uniformity and stability of welding. Moreover, the welding parameter optimization model generates accurate welding path planning data, significantly improves the welding quality and reliability, reduces the welding defect rate, improves the strength and consistency of the welded joint, and has wide industrial application value and economic benefits.

Claims

1. A method of welding a cover of a conical sealing regenerator, characterized in that, The method comprises the following steps: Collecting temperature field data of the conical contact surface of the conical sealing cold storage box body and the conical sealing cold storage box cover, establishing a stress distribution prediction model based on the temperature field data, and outputting a stress concentration coefficient matrix; Based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed, and a welding power density distribution function and a welding seam trajectory correction curve are outputted; According to the welding power density distribution function, a hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input; Based on the output results of the welding seam trajectory correction curve and the hierarchical progressive heat input compensation strategy, a welding parameter optimization model is established, and welding path planning data is generated by solving the optimal parameter combination.

2. The method of claim 1, wherein The temperature field data of the conical contact surface is collected by an infrared sensor array, denoised by triple filtering, and reconstructed by a hyperboloid interpolation algorithm to obtain complete temperature field data.

3. The method of claim 2, wherein the welding is performed by a laser beam. Based on the temperature field data, a stress distribution prediction model is constructed to output a stress concentration coefficient matrix; The stress distribution prediction model is realized by using an improved finite element neural network structure, which includes a finite element calculation path and a neural network prediction path; The finite element path calculates the initial thermal stress, and the neural network prediction path extracts temperature features and dynamically adjusts the weight to cope with temperature field changes.

4. The method of claim 3, wherein Based on the stress concentration coefficient matrix, a multi-parameter collaborative optimization algorithm is constructed; The multi-parameter collaborative optimization algorithm adopts an improved particle swarm optimization algorithm, taking welding power density and trajectory offset as key optimization variables; When the improved particle swarm optimization algorithm obtains the optimal solution, the welding power density is reconstructed, and the welding power density distribution function and the trajectory correction curve are outputted.

5. The method of claim 4, wherein the welding is performed by a laser beam. Based on the welding power density distribution function, a hierarchical progressive heat input compensation strategy is adopted to dynamically compensate the welding heat input; The hierarchical progressive heat input compensation strategy adopts a double-layer control structure, including a fast response layer and a fine adjustment layer; The fast response layer uses a proportional-differential controller to quickly correct short-term deviations; The fine adjustment layer accurately compensates for long-term errors through a fuzzy adaptive controller.

6. The method of claim 5, wherein the welding is performed by a laser beam. The fast response layer adjusts the proportional coefficient and the differential coefficient according to the deviation characteristics to improve the response ability and damping characteristics; The fine adjustment layer optimizes the compensation effect by adaptively adjusting the fuzzy rule weight, and combines the membership function offset and weight recovery strategy to cope with complex deviation characteristics.

7. The method of claim 6, wherein the welding is performed by a laser beam. Based on the trajectory correction curve and the hierarchical progressive heat input compensation strategy, a welding parameter optimization model is established; The welding parameter optimization model adopts a three-layer nested structure: an outer main optimization cycle, a middle constraint processing cycle, and an inner parameter refinement cycle.

8. The method of claim 7, wherein the welding is performed by a laser beam. The outer main optimization cycle is responsible for global search, and realizes convergence control by dynamically adjusting the search range. When the search range is insufficient, the current direction is marked as invalid; The middle constraint processing cycle ensures the feasibility of the solution, and gradually adjusts the parameters through constraint inspection and coordination mechanism; The inner parameter refinement cycle performs local optimization near the constraint solution, and improves the optimization precision by combining fine mesh and adaptive encryption mechanism.

9. The method of claim 8, wherein the welding is performed by a laser beam. Based on the optimal solution of the welding parameter optimization model, a control instruction sequence containing position, angle, depth, speed and time is generated by projecting it along the weld trajectory, and is converted into a segmented path planning data packet to control the welding path.

10. A welding system for a conical sealed cold storage box cover, comprising welding a conical sealed cold storage box cover using the welding method for a conical sealed cold storage box cover according to any one of claims 1-9, characterized in that, The method comprises the following steps: The acquisition module is used to acquire the temperature field data of the conical contact surface of the conical sealing cold storage box body and the conical sealing cold storage box cover, establish a stress distribution prediction model, and output a stress concentration coefficient matrix; The correction module is used to construct a multi-parameter collaborative optimization algorithm according to the stress concentration coefficient matrix, and output a welding power density distribution function and a weld trajectory correction curve; The compensation module is used to dynamically compensate the welding heat input by using a hierarchical progressive heat input compensation strategy according to the welding power density distribution function; The planning module is used to establish a welding parameter optimization model based on the weld trajectory correction curve and the hierarchical progressive heat input compensation strategy, solve the optimal parameter combination, and generate a welding path planning data.

Images