A multivariable optimization method and system for integrated door ring laser welding

Abstract

The application discloses a kind of integrated door ring laser welding multivariable optimization method and system, comprising: S1 establishes three-variable collaborative optimization model, defines material combination, thickness ratio and weld position;S2 establishes double-objective optimization model, sets safety objective function and lightweight objective function and constraint condition;S3 solves Pareto optimal solution set using NSGA-II algorithm;S4 selects the best variable combination from it;S5 automatically matches process parameters based on process database;S6 predicts the probability of welding defects through the quality prediction model, meets the threshold value, then outputs the result.The door ring design is improved from single-variable empirical adjustment to multi-objective intelligent optimization in the application, and based on more than 40 sets of process data, precise matching, interpolation matching and rule adjustment are realized.Combined with the 12-dimensional neural network model to evaluate the defect probability, safety and lightweight can be considered, the development cycle is greatly shortened, and 100% trial production qualification rate is realized.

Description

Technical Field

[0001] This invention belongs to the field of integrated door rings, and specifically relates to a multivariate optimization method and system for laser welding of integrated door rings. Background Technology

[0002] The integrated door ring is a core load-bearing structure for side-impact safety in automobiles. Existing technologies mainly include the following technical approaches: Traditional modular assembly technology involves stamping and spot welding multiple independent parts, which results in a large number of parts, high mold costs, poor assembly accuracy, and insufficient overall rigidity. While single-material integral hot forming technology improves rigidity, it cannot achieve variable thickness design, and the single material necessitates additional reinforcing plates for local reinforcement, leading to significant weight and hindering lightweighting. Laser-welded plate technology can weld steel plates of different thicknesses or materials and then stamp them as a whole, but currently its process parameters mainly rely on empirical settings or single-objective optimization methods such as orthogonal experiments. It lacks a collaborative optimization model for the three key variables of material combination, thickness ratio, and weld position, and also lacks an intelligent database and quality prediction system that can automatically match the optimal process parameters based on the combination of variables.

[0003] Furthermore, the design of integrated door rings faces an inherent conflict between safety (requiring high-strength thick plates) and lightweighting (requiring thin plates / lightweight materials). Existing methods are mostly single-objective optimizations or simple weighted processing, making it difficult to obtain the Pareto optimal solution set that simultaneously satisfies the following objectives: collision intrusion ≤120mm, B-pillar (the main pillar in the middle of the vehicle body that bears the side impact force) strength ≥1500MPa, door sill elongation ≥10%, and weight reduction.

[0004] Therefore, there is an urgent need for an integrated door ring design method that can balance safety and lightweight design, synergistically optimize the three variables of material, thickness and weld, automatically match process parameters and have the ability to predict welding quality. Summary of the Invention

[0005] Purpose of the invention: In order to overcome the above shortcomings, the purpose of this invention is to provide a multivariate optimization method and system for integrated door ring laser welding, which improves the door ring design from single-variable empirical adjustment to multi-objective intelligent optimization. Based on more than 40 sets of process data, it achieves accurate matching, interpolation matching and rule adjustment. Combined with a 12-dimensional neural network model to evaluate the probability of defects, it can balance safety and lightweight, significantly shorten the development cycle, and achieve a 100% trial production qualification rate.

[0006] Technical Solution: To achieve the above objectives, this invention provides a multivariate optimization method for laser welding of integrated door rings, comprising the following steps: S1: Establishing a three-variable collaborative optimization model, defining material combination variables, thickness ratio variables, and weld position variables. The material combination variable represents the material grade of each welding piece of the door ring, the thickness ratio variable represents the thickness value or thickness ratio of each piece, and the weld position variable represents the spatial position of the splicing interface between adjacent pieces; S2: Establishing a dual-objective optimization model, defining a safety objective function and a lightweight objective function, and setting constraints including collision safety, process feasibility, and variable boundaries; S3: Solving the three-variable collaborative optimization model using the NSGA-II dual-objective genetic algorithm to obtain the Pareto optimal solution set; S4: Selecting the optimal variable combination that meets the design requirements from the Pareto optimal solution set; S5: Automatically matching process parameters for the optimal variable combination based on a process database; S6: Predicting the welding defect probability of the matched process parameters using a quality prediction model. If the quality threshold is met, the optimal variable combination and the corresponding process parameters are output. By incorporating three heterogeneous variables—material, thickness, and weld location—into a unified optimization framework, the shortcomings of isolated design of these three variables in existing technologies are resolved. Through bi-objective optimization and Pareto solution, the inherent conflict between safety and lightweighting can be explicitly addressed, avoiding subjective weighting. At the same time, quality prediction is introduced as a verification step before output, forming an "optimization-matching-prediction" closed loop, which improves the first-time success rate of the final solution.

[0007] Furthermore, the parameters of the NSGA-II bi-objective genetic algorithm include: population size N=100, maximum number of iterations G. max =80, crossover probability P c =0.9, mutation probability P m =0.1; The encoding scheme is a hybrid encoding, in which the material combination variable uses integer encoding, while the thickness ratio variable and weld location variable use real number encoding. The hybrid encoding adapts to the coexistence of discrete (material grade) and continuous (thickness, coordinate) variables, avoiding the decrease in search efficiency or the generation of infeasible solutions caused by a single encoding; at the same time, the setting of population size and iteration number takes into account convergence speed and global search capability, and the empirical value of crossover / mutation probability ensures population diversity.

[0008] Furthermore, the selectable materials for the material combination variables include 2-4 of the following: 22MnB5, DP780, DP980, CP800, CP1000, MS1500, and DC04. Each material has corresponding yield strength, tensile strength, and carbon equivalent parameters. The thickness ratio variable ranges from 0.8mm to 2.5mm, and the ratio of the maximum thickness to the minimum thickness does not exceed 3. The weld position variable must meet the constraints of avoiding high-risk areas for side collisions, avoiding areas with large deformations, and maintaining a spacing of ≥30mm between adjacent welds. The materials are limited to seven typical automotive steels, with a thickness of 0.8-2.5mm covering the commonly used range for door rings. A thickness ratio ≤3 corresponds to the process limit of laser welding equipment. The weld position constraints are directly related to collision safety and formability, avoiding the problem of unmanufacturable results from optimization.

[0009] Furthermore, the security objective function f1(x) is defined as: f1(x) = 0.5·(R) min x / R ref )+0.3·(E avg x / E ref )+0.2·(C energy x / C ref ) Where R min x is the minimum bending strength of the door ring, E avg x is the average elastic modulus, C energy x represents energy absorption during a side impact, R ref E ref C ref This is the corresponding reference value; the larger the f1 value, the better the security. The lightweight objective function f2(x) is defined as follows: f2(x) = -Mass x / Mass0 Mass x Let f2 be the total mass of the door ring, and Mass0 be the baseline mass. A larger f2 indicates a lighter door ring. The safety objective function combines three sub-indicators—bending strength, elastic modulus, and impact energy absorption—with weighted coefficients reflecting the differences in importance of different failure modes of the door ring. The lightweight objective uses a relative mass ratio, which is dimensionless and easy to compare with other objectives. This formula is concise and has clear physical meaning, avoiding the poor interpretability problem of black-box proxy models.

[0010] Furthermore, the constraints include: side impact intrusion ≤ 120mm, B-pillar top strength ≥ 1500MPa, sill elongation ≥ 10%, laser welding speed ≥ 2m / min, forming limit strain ≤ 0.8, adjacent weld spacing ≥ 30mm, thickness boundary 0.8mm~2.5mm, thickness ratio 1~3, and material tensile strength 270MPa~1500MPa. Quantifiable hard constraint boundaries are given: intrusion ≤ 120mm corresponds to the CNCAP five-star crash standard; B-pillar strength ≥ 1500MPa matches the lower limit of quenched strength of 22MnB5 hot-formed steel; elongation ≥ 10% ensures the sill does not fracture; welding speed ≥ 2m / min meets production cycle time; forming strain ≤ 0.8 corresponds to the safe zone of the hot-forming limit curve. These thresholds are directly related to regulations, material physical limits, and production line capabilities, making the optimized solution inherently engineering feasible.

[0011] Furthermore, the automatic matching of process parameters in step S5 includes three methods: precise matching, which involves searching the process database for records that are exactly the same as the optimal variable combination; interpolation matching, which involves interpolating missing parameters based on carbon equivalent similarity, power-thickness relationship, or thermal accumulation model; and optimization adjustment, which involves correcting parameters such as power and decoking amount according to empirical rules based on working conditions such as the distance of the weld from the B-pillar, thickness ratio, and the presence of mild steel. A multi-level process parameter matching strategy is constructed. Precise matching ensures the rapid reuse of mature combinations; interpolation matching uses carbon equivalent similarity, power-thickness relationship, and thermal accumulation model to fill database gaps and cover untested continuous variable combinations; and optimization adjustment performs local corrections based on empirical rules. The combination of these three methods upgrades the database from a static record to a dynamic knowledge system, significantly expanding the applicable scenarios.

[0012] Furthermore, the quality prediction model is a machine learning-based neural network model. The input features include 12 dimensions: material features, geometric features, process features, and environmental features. The output is the type and probability of welding defects. The quality threshold is defined as a defect probability lower than a preset engineering allowable value. A neural network prediction framework with 12-dimensional multi-source feature input is proposed. Compared to traditional trial welding inspection, this model can predict the probability of defects such as spatter, lack of fusion, and burn-through during the parameter matching stage, achieving preventative quality control through "prediction before production." Simultaneously, the nonlinear fitting capability of the neural network can capture the interaction effects between various factors, making it more accurate than regression models.

[0013] This invention also provides a multivariate optimization system for integrated door ring laser welding, used to implement the above method, comprising: The system includes a variable combination solution module, which executes steps S1 to S4 of claim 1 and outputs the optimal variable combination; a process database, which stores multiple sets of verified laser welding process parameters and empirical rules; an automatic parameter matching module, whose input is connected to the output of the variable combination solution module and the process database, which receives the optimal variable combination and automatically matches the process parameters based on the process database, outputting the matched process parameters; and a quality prediction model module, whose input is connected to the output of the automatic parameter matching module, which predicts the welding defect probability of the matched process parameters and outputs the final optimal variable combination and corresponding process parameters when the quality threshold is met. The output of the variable combination solution module is connected to the input of the automatic parameter matching module, which in turn is connected to the input of the process database and the quality prediction model module. The inputs / outputs of each module are defined (optimal variable combination, process parameters, quality score), facilitating software implementation, unit testing, and integration with existing MES / PLM systems.

[0014] Furthermore, the process database includes at least the fields of material combination, thickness combination, thickness ratio, weld location, laser power, welding speed, defocusing amount, and shielding gas flow rate. These fields contain both categorical variables (material combination, weld location) and continuous variables (thickness, power, speed, etc.); exact matching relies on the categorical fields, while interpolation matching relies on the continuous fields. The field design of the aforementioned process database directly supports the three matching methods in weight 6, making the database not only a storage container but also the computational foundation for the matching algorithm.

[0015] As can be seen from the above technical solution, the present invention has the following beneficial effects: This invention presents a multivariate optimization method and system for integrated door ring laser welding. It establishes a three-variable collaborative optimization model for material, thickness, and weld position, elevating door ring design from single-variable empirical adjustment to multi-objective intelligent optimization. The NSGA-II algorithm is used to solve the Pareto front for safety and lightweighting, obtaining the optimal solution set under constraints such as intrusion ≤120mm and B-pillar strength ≥1500MPa. Based on a database containing over 40 sets of real process data, it achieves automatic parameter matching through precise matching, interpolation matching, and empirical rule optimization. Combined with a 12-dimensional input neural network quality prediction model, the probability of welding defects can be assessed during the matching stage, significantly shortening the development cycle and achieving a 100% trial production pass rate, balancing performance, efficiency, and quality. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the steps of a multivariate optimization method for laser welding of an integrated door ring according to the present invention. Figure 2 This is a schematic diagram of the structure of a multivariable optimization system for laser welding of an integrated door ring according to the present invention; Figure 3 This is a flowchart illustrating the NSGA-II bi-objective genetic algorithm in the multivariate optimization method for laser welding of an integrated door ring described in this invention. Detailed Implementation

[0017] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention. Example

[0018] This embodiment uses the integrated door ring of a mid-to-high-end SUV of a certain brand as an example to illustrate the technical solution of the present invention. The specific steps are as follows: Figure 1 As shown below, the details will be elaborated upon.

[0019] 1. Design Constraints Side impact intrusion ≤120mm; The top strength of the B-pillar must be ≥1500 MPa; The extension rate of the middle part of the threshold is ≥10%; Door ring weight ≤12.5kg (15% weight reduction compared to traditional solutions); Welding speed ≥ 2m / min (production cycle requirement).

[0020] 2. Step S1: Establish a three-variable collaborative optimization model 2.1 Define material combination variables The materials library is as follows: Based on the vehicle model positioning, the following material combinations were pre-selected from 7 types of materials: Based on the premise that the M2 scheme meets the 1500MPa strength requirement of the B-pillar, it has a lighter weight than M1 and a simpler structure than M4, so M2 is recommended.

[0021] 2.2 Define the thickness ratio variable Thickness T i Constraint: 0.8mm≤T i ≤2.5mm, and 1≤T max / T min ≤3.

[0022] For the M2 scheme (22MnB5+DP780), design candidate thickness combinations: T1=[1.8,1.2] (thickness ratio 1.50, estimated weight 12.8kg); T2=[1.6,1.2] (thickness ratio 1.33, estimated weight 12.2kg); T3=[1.4,1.2] (thickness ratio 1.17, estimated weight 11.8kg). T4=[1.2,1.0] (thickness ratio 1.20, estimated weight 10.8kg, excluding the strength estimation of B-pillar R) min =1500×(1.2)² / 6=360N·mm / mm (insufficient). Based on the fact that the T2 scheme meets the design requirements in terms of both safety (simulated strength of B-pillar 1520MPa≥1500MPa) and lightweight (weight 12.3kg≤12.5kg, weight reduction of 18%), and has the lowest overall cost, the T2 scheme is recommended.

[0023] 2.3 Define weld position variables Based on the structural mechanics analysis of the door ring, the candidate weld locations (distance from the root of column B) are as follows: P1=(60,150); P2=(80,150); P3=(100,150); P4=(120,150); The spacing between adjacent welds is ≥30mm.

[0024] Finite element analysis shows that the strain at 60mm is >0.8, which is a large deformation zone, so P1 is excluded; based on the fact that P2 is located in the transition zone, has moderate stiffness change, and effectively avoids the large deformation zone with strain greater than 0.8, P2 is recommended.

[0025] 2.4 Initialize the population The candidate solution space contains 18 combinations (3 materials × 3 thicknesses × 2 locations). The initial population size of NSGA-II is 100 individuals, including 18 candidate individuals and 82 random individuals. Example initial individuals: x1=[1,2,1.6,1.2,80,150]; x2=[1,2,1.8,1.2,80,150]; x3=[1,2,1.6,1.2,100,150]; x4=[1,4,1.6,0.8,80,150]; ...x 100 =[Random Individual].

[0026] 3. Step S2: Establish a dual-objective optimization model 3.1 Security objective function f1(x) Taking individual x1=[1,2,1.6,1.2,80,150] as an example, the calculation is as follows: B-pillar area (22MnB5, 1.6mm thick, accounting for 30% of the area): R min Bpillar=σ·t² / 6=1500×1.6² / 6=640N·mm / mm; Threshold area (DP780, 1.2mm thick, occupying 70% of the area): R min sill=780×1.2² / 6=187.2N·mm / mm; Take the minimum value R min =640 N·mm / mm, reference value R ref =1500 N·mm / mm, normalized R norm =640 / 1500=0.4267; Average elastic modulus E avg =0.3×210+0.7×205=206.5GPa, E ref =210GPa, E norm =206.5 / 210=0.983; Collision energy absorption (finite element estimation) C energy =52.5kJ, C ref =50kJ, C norm =52.5 / 50=1.05; In summary: f1 = 0.5 × 0.4267 + 0.3 × 0.983 + 0.2 × 1.05 = 0.7183 ≈ 0.72.

[0027] 3.2 Lightweighting the objective function f2(x) Mass per unit length in the 22MnB5 region: 7850 kg / m³ × 0.105 m × 0.0016 m = 1.319 kg / m; Mass per unit length in the DP780 region: 7850 × 0.245 × 0.0012 = 2.308 kg / m; Total mass per unit length: 3.627 kg / m; The door knocker has a total length of approximately 3.4m when unfolded, and a total mass of Mass = 3.627 × 3.4 = 12.33 kg ≈ 12.3 kg; The reference mass Mass0 = 15.0 kg, f2 = -12.3 / 15.0 = -0.82.

[0028] 3.3 Constraints Perform constraint checks on x1: Intrusion depth (simulation) = 108mm ≤ 120mm; Column B strength = 1520MPa ≥ 1500MPa; Threshold extension rate = 14% ≥ 10%; Welding speed = 2.8m / min ≥ 2m / min; The forming limit strain = 0.72 ≤ 0.8; Weld spacing = 80mm ≥ 30mm; The thickness boundary and thickness ratio meet the requirements; The tensile strength of the material is in the range of 270~1500MPa; The constraint violation degree CV(x1) = 0.

[0029] Population evaluation results: Among 100 individuals, 85 are feasible solutions (CV=0) and 15 are infeasible solutions (CV>0). Each individual has a corresponding fitness value of (f1,f2).

[0030] 4. Step S3: Solve using the NSGA-II algorithm, such as... Figure 3 As shown; 4.1 Algorithm Parameters Population size N=100; Maximum number of iterations G max =80; Crossover probability P c =0.9; Probability of mutation P m =0.1; Cross-distribution index η c =20; Distribution index η m =20; 4.2 Encoding Scheme Hybrid coding is adopted: the material combination variable [M1, M2] is encoded as an integer, and the thickness variable [t1, t2] and weld position variable [p] are encoded as integers. x ,p y Real number encoding is used. Each individual is represented as x=[M1,M2,t1,t2,p] x ,p y ].

[0031] 4.3 Non-dominated sorting Dominance definition: x1 dominates x2 if and only if f1(x1)≥f1(x2) and f2(x1)≥f2(x2), and at least one is strictly greater than the other. Since f2 is defined as a negative mass ratio, the larger its value, the lighter the door ring, i.e., the better the weight reduction effect.

[0032] First generation sorting results: F1 (First Frontier, Non-Dominant Solution): Approximately 25 individuals, including x1(0.72,-0.82), x2(0.74,-0.85), x3(0.76,-0.91), etc.; F2: Approximately 30 individuals; F3: Approximately 20 individuals; F4: Approximately 15 individuals; Infeasible solutions: Approximately 10; 4.4 Crowding Calculation Take x2 in F1 as an example: Sort by f1: x3(0.76), x2(0.74), x1(0.72)... Sort by f2: x3(-0.91), x2(-0.85), x1(-0.82)... Crowding level: CD(x2)=(0.76-0.72) / (0.76-0.58)+(-0.82-(-0.91)) / (-0.82-(-0.95))=0.222+0.692=0.914 4.5 Selection, Crossover, and Mutation Selection: Binary tournament, prioritizing individuals with small non-dominated layers and high crowding.

[0033] Cross example: Parent generation 1 [1,2,1.6,1.2,80,150] and parent generation 2 [1,3,1.8,1.4,100,180] undergo single-point crossover and arithmetic crossover to produce offspring.

[0034] Variation example: perform polynomial variation on thickness and Gaussian variation on position (sigma=5mm).

[0035] 4.6 Results after 80 iterations The Pareto front was ultimately obtained, containing 35 non-dominated solutions. Typical solutions are as follows: Solution A (safety first): [1,3,2.0,1.4,60,150], material 22MnB5+DP980, f1=0.98, f2=-0.95, weight 14.2kg; Solution B (equilibrium type): [1,2,1.6,1.2,80,150], material 22MnB5+DP780, f1=0.95, f2=-0.82, weight 12.3kg; Solution C (lightweight priority): [2,4,1.2,1.0,100,150], material DP780+CP800, f1=0.78, f2=-0.72, weight 11.2kg; Since solution B meets the safety requirements (f1=0.95, corresponding to a column strength of 1520MPa), its weight (12.3kg) is significantly lower than that of solution A (14.2kg), and its safety is significantly better than that of solution C (f1=0.78). Therefore, solution B has the lowest overall cost and is recommended.

[0036] 5. Step S4: Select the optimal combination of variables This embodiment uses a balanced solution (recommended scheme), based on the following decision criteria: Safety value f1 = 0.95, corresponding to a B-pillar strength of 1520 MPa > 1500 MPa; Lightweighting f2 = -0.82, corresponding to a weight of 12.3kg < 12.5kg, and a weight reduction of 18% > 15% target; The distance to the ideal point d = sqrt((1-0.95)² + (1-0.82)²) = 0.187 (minimum); The final optimal combination of variables is: Material: M={1,2}, i.e. 22MnB5+DP780; Thickness: T = [1.6, 1.2] mm; Weld location: P=(80,150)mm (80mm from the root of B-pillar, 150mm longitudinally).

[0037] 6. Step S5: Automatic matching of process parameters based on the process database. 6.1 Exact Match Query Searching the process database for records close to the optimal combination, three were found: Record 1: 22MnB5-DP780, [1.6, 1.2], position (85, 155), P=4.2kW, v=2.8m / min, Δz=+1.5mm, Q=22L / min, quality score 91; Record 2: 22MnB5-DP780, [1.5, 1.1], position (75, 145), P=3.9kW, v=3.0m / min, Δz=+1.2mm, Q=20L / min, quality score 87; Record 3: 22MnB5-DP780, [1.7, 1.3], position (90, 160), P = 4.5kW, v = 2.6m / min, Δz = +1.8mm, Q = 24L / min, quality score 88.

[0038] 6.2 Interpolation Matching Carbon equivalent similarity: target CE=0.348, all records have the same CE, and the weights are equal (0.333). Thickness interpolation: The target average thickness is 1.4mm. Record 1 is a perfect match. After linear interpolation, the power of records 2 and 3 both approach 4.2kW. Position interpolation: inverse distance weighting, with a weight of 0.345 for records 1 and 2, and a weight of 0.310 for record 3; The final power P is obtained by comprehensive weighted calculation.final =4.2kW; 6.3 Optimization and Adjustment Fine-tuning was performed based on a thickness ratio of 1.33 and positional distance. P opt =4.2×(1+0.05×(1.4-1.5) / 1.5)+0.085=4.27≈4.2kW; v opt =2.8 m / min; Δz opt =+1.5+0.1×(1.6-1.2)=+1.54≈+1.5mm; Q opt =22L / min; The matching results are as follows: laser power 4.2kW, welding speed 2.8m / min, defocusing amount +1.5mm, shielding gas flow rate 22L / min.

[0039] 7. Step S6: Predict the probability of welding defects using a quality prediction model. 7.1 Input Feature Preparation (12-dimensional) A 12-dimensional normalized feature vector was constructed based on material characteristics (carbon equivalent 0.348, thermal conductivity 40.5 W / (m·K), melting point 1500°C), geometric characteristics (thickness 1.6 mm, 1.2 mm, thickness ratio 1.333), process characteristics (power 4.2 kW, speed 2.8 m / min, defocusing +1.5 mm, flow rate 22 L / min), and environmental characteristics (temperature 25°C, humidity 60%). X=[0.37,0.525,0.5,0.471,0.235,0.167,0.55,0.6,0.75,0.467,0.5,0.6].

[0040] 7.2 Forward Computation of Neural Networks The network structure employs a five-layer fully connected architecture: the input layer contains 12 nodes, which are sequentially connected to the first hidden layer (64 nodes), the second hidden layer (32 nodes), and the third hidden layer (16 nodes), finally connecting to the output layer containing 5 nodes. All hidden layers use the ReLU activation function, and the output layer uses the Sigmoid activation function. The forward computation result is as follows: Y=[0.028,0.015,0.021,0.032,0.91].

[0041] 7.3 Interpretation of Results Stomatal probability = 2.8% (<5%) The probability of non-fusion is 1.5% (<5%). Crack probability = 2.1% (<5%) The probability of edge biting is 3.2% (<5%). Overall quality score = 91 points (≥85 points) Judgment: The process parameters are feasible and the quality verification is passed.

[0042] 8. Step 7: Simulation and Experimental Verification 8.1 Finite Element Simulation Welding simulation (SYSWELD): Penetration depth 1.45mm > 1.2mm, weld width 2.8mm, HAZ width 3.5mm, peak temperature 1456°C; Forming simulation (AutoForm): Maximum strain 0.72 < 0.8, springback 1.2mm < 2mm, all FLD values ​​are within the safe range; Collision simulation (LS-DYNA): Intrusion depth 108mm < 120mm, B-pillar strength 1520MPa > 1500MPa, energy absorption 52kJ, threshold elongation 14% > 10%.

[0043] 8.2 Small-batch trial production Ten samples were produced, and the test results are as follows: Weld penetration depth: 1.35-1.48mm (100% pass rate); Tensile strength: 795-820MPa (≥780MPa, 100% pass rate); Elongation rate: 12-16% (≥10%, 100% pass rate); Metallographic structure: No porosity / cracks / lack of fusion (100% pass rate); Geometric dimensions: ±0.2-0.3mm (≤±0.5mm, 100% pass rate); Surface quality: No undercut / splashing (100% pass rate); The overall pass rate was 100%, and the solution passed all verifications.

[0044] 9. Summary of Implementation Results As shown in the table below: Evaluation indicators target value actual value Status of achievement Side collision intrusion <=120mm 108mm Exceeded the target (10% better than the standard) B-pillar top strength >=1500MPa 1520MPa Achieve Threshold extension rate >=10% 14% Exceeded the target Door ring weight <=12.5kg 12.3kg Achieved (weight loss of 18%) First pass rate >=95% 100% Exceeded the target Development cycle 3-6 months 1 week 10 times more efficiency In this embodiment, as Figure 2The integrated door ring laser welding multivariate optimization system includes a variable combination solution module, a process database, an automatic parameter matching module, and a quality prediction model module. The variable combination solution module executes steps 1 to 4 (variable space definition and initialization, calculation of objective function and constraints, NSGA-II optimization solution, and user selection of optimal solution) to obtain the optimal variable combination (e.g., solution B: material 22MnB5+DP780, thickness 1.6mm+1.2mm, weld position (80,150)), and outputs it to the parameter automatic matching module. The process database stores laser welding process parameters and empirical rules (e.g., records 1, 2, and 3 queried in step 5). The parameter automatic matching module receives the optimal variable combination, connects to the process database, performs precise matching, interpolation matching, and optimization adjustment (as described in step 5), and outputs the matched process parameters (laser power 4.2kW, welding speed 2.8m / min, defocusing amount +1.5mm, shielding gas flow rate 22L / min) to the quality prediction model module. The quality prediction model module uses a neural network model to predict the welding defect probability of the process parameters, outputs a comprehensive quality score, and finally outputs the optimal variable combination and corresponding process parameters after determining that the quality threshold is met.

[0045] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A multivariate optimization method and system for laser welding of integrated door rings, characterized in that, Includes the following steps: S1: Establish a three-variable collaborative optimization model, defining material combination variable, thickness ratio variable and weld position variable. The material combination variable represents the material grade of each welded piece of the door ring, the thickness ratio variable represents the thickness value or thickness ratio of each piece, and the weld position variable represents the spatial position of the splicing interface of adjacent pieces. S2: Establish a dual-objective optimization model, define a safety objective function and a lightweight objective function, and set constraints including collision safety, process feasibility, and variable boundaries; S3: The NSGA-II bi-objective genetic algorithm is used to solve the three-variable collaborative optimization model to obtain the Pareto optimal solution set; S4: Select the optimal combination of variables from the Pareto optimal solution set that meets the design requirements; S5: Automatically match process parameters for the optimal combination of variables based on the process database; S6: Predict the probability of welding defects by matching the process parameters through the quality prediction model. If the quality threshold is met, output the optimal combination of variables and the corresponding process parameters.

2. The method according to claim 1, characterized in that, The parameters of the NSGA-II biobjective genetic algorithm include: population size N=100, maximum number of iterations G. max =80, crossover probability P c =0.9, mutation probability P m =0.1; The coding scheme is a hybrid coding scheme, in which the material combination variable uses integer coding, and the thickness ratio variable and weld position variable use real number coding.

3. The method according to claim 1, characterized in that, The optional materials for the material combination variable include 2-4 of the following: 22MnB5, DP780, DP980, CP800, CP1000, MS1500, and DC04. Each material has corresponding yield strength, tensile strength, and carbon equivalent parameters. The thickness ratio variable ranges from 0.8mm to 2.5mm, and the ratio of the maximum thickness to the minimum thickness does not exceed 3. The weld position variable must meet the constraints of avoiding high-risk areas for side collisions, avoiding areas with large deformations, and maintaining a distance of ≥30mm between adjacent welds.

4. The method according to claim 1, characterized in that, The security objective function f1(x) is defined as: f1(x)=0.5·(R min x / R ref )+0.3·(E avg x / E ref )+0.2·(C energy x / C ref ) Where R min x is the minimum bending strength of the door ring, E avg x is the average elastic modulus, C energy x represents energy absorption during a side impact, R ref E ref C ref This is the corresponding reference value; the larger the f1 value, the better the security. The lightweight objective function f2(x) is defined as follows: f2(x)=-Mass x / Mass0 Mass x f2 represents the total mass of the door ring, with Mass0 being the baseline mass. A larger f2 indicates a lighter door ring.

5. The method according to claim 1, characterized in that, The constraints include: side impact intrusion ≤ 120mm, B-pillar top strength ≥ 1500MPa, sill elongation ≥ 10%, laser welding speed ≥ 2m / min, forming limit strain ≤ 0.8, adjacent weld spacing ≥ 30mm, thickness boundary 0.8mm~2.5mm, thickness ratio 1~3, and material tensile strength 270MPa~1500MPa.

6. The method according to claim 1, characterized in that, The automatic matching of process parameters in step S5 includes three methods: precise matching, which means searching for records in the process database that are exactly the same as the optimal combination of variables; interpolation matching, which means interpolating the missing parameters based on carbon equivalent similarity, power-thickness relationship or heat accumulation model; and optimization adjustment, which means correcting parameters such as power and decoking amount according to empirical rules based on working conditions such as the distance of the weld from the B-pillar, thickness ratio, and presence of mild steel.

7. The method according to claim 1, characterized in that, The quality prediction model is a machine learning-based neural network model. The input features include 12 dimensions: material features, geometric features, process features, and environmental features. The output is the welding defect type and probability. The quality threshold is when the defect probability is lower than the preset engineering allowable value.

8. A multivariate optimization system for laser welding of an integrated door ring, used to implement the method described in any one of claims 1 to 7, characterized in that, include: The variable combination solution module is used to execute steps S1 to S4 as described in claim 1 and output the optimal variable combination. The process database stores multiple sets of verified laser welding process parameters and empirical rules; The parameter automatic matching module has its input end connected to the output end of the variable combination solution module and the process database. It is used to receive the optimal variable combination and automatically match the process parameters based on the process database, and output the matched process parameters. The quality prediction model module, whose input is connected to the output of the automatic parameter matching module, is used to predict the welding defect probability of the matched process parameters, and outputs the final optimal combination of variables and the corresponding process parameters when the quality threshold is met.

9. The system according to claim 8, characterized in that, The process database includes at least the fields of material combination, thickness combination, thickness ratio, weld location, laser power, welding speed, defocusing amount, and shielding gas flow rate.

Images