2026-10-14T22:54:16Z https://www.scidb.cn/oai

10.57760/sciencedb.0138w 2026-09-28T09:40:00Z

2026-09-28 Design of a Torque-Balancing Mechanism and Optimization of Characteristic Parameters for a Multi-Terrain Mobile Robot doi:10.57760/sciencedb.0138w en This dataset is the "IANSGA-II Algorithm Stage Threshold Sensitivity Simulation and Statistical Analysis Dataset," sourced from IANSGAII_algorithm.zip. The data was generated through the SKM4.slx in MATLAB/Simulink, employing the dynamic model and the improved Non-dominated Sorting Genetic Algorithm IANSGA-II, and is used to compare the effects of stage thresholds qs=0.70, 0.75, 0.80 on multi-objective optimization performance and convergence process. The dataset contains a total of 174 files, including 38 MATLAB program files, 126 MAT data files, 2 Excel summary files, 3 PNG figures, 1 MATLAB FIG figure, 1 PDF figure, and the Simulink model. The algorithm population size N=100, maximum iteration Gmax=200, parent pool size=50, and tournament size=2. The four decision variables are suspension elastic element stiffness k2/k3, suspension damping coefficient C2/C3, torque balance mechanism stiffness k3/kh, and torque balance mechanism damping coefficient C3/Ch, with ranges of 4~40 kN/m, 0.1~1.0 kN·s/m, 20~200 kN/m, and 1~4 kN·s/m respectively. The crossover probability and mutation probability for the fixed operator stage are 0.90 and 0.10 respectively; for the adaptive stage, the preset crossover and mutation probabilities are 0.60 and 0.05 respectively, with crossover and mutation distribution indices both set to 20. The algorithm uses simulated binary crossover, polynomial mutation, non-dominated sorting, and crowding distance maintenance strategy. When q=i/Gmax ≤ q, the fixed probability operator is used; when it exceeds the threshold, the adaptive probability operator based on normalized fitness is applied. Therefore, the phase boundaries corresponding to the three thresholds are the 140th, 150th, and 160th generations, and the adaptive stage begins from the 141st, 151st, and 161st generation, respectively.Each candidate solution injects ks, cs, kh, and ch into the SKM4 model through Simulink.SimulationInput for a single simulation. The simulation time of the model is from 0 to 10 seconds, with a fixed solver step size of 0.001 seconds, so the physical time resolution is 1 ms. The data for each generation is represented by the average of the two objectives in the first non-dominated front. The relative rate of change is calculated at 10-generation intervals, i.e., the rate of change of the two objectives relative to the generation i−10 is computed separately, and the maximum R(i) among the two objectives is taken; a small term of 10^-6 is added to the denominator to avoid division by zero. Then, a backward moving average Rsmooth of length 10 generations is calculated. The stable generation i_c is defined as the first generation in the adaptive phase that satisfies the criterion "all Rsmooth < 0.01 from this generation to generation 200"; if the condition is not met within 200 generations, 201, which is Gmax+1, is used as a mark of non-convergence. The final Pareto front is first filtered by rank 1, then rows with non-finite objective values, duplicate rows, and dominated points are removed. All 60 sets of final fronts are normalized using the same minimum and maximum values, and then combined into an empirical reference front containing 1,084 points; the two-dimensional hypervolume (HV) uses a normalized reference point [1.10,1.10][1.10,1.10], with larger HV being better, and the inverted generational distance (IGD) uses the average Euclidean distance to the empirical reference front, with smaller IGD being better. Statistical analysis includes mean, sample standard deviation, paired Wilcoxon signed-rank test, Friedman test, Holm multiple comparison correction, and 95% bootstrap confidence intervals of the mean based on 5,000 resamples. The final comprehensive ranking assigns double weight to HV and IGD, and single weight to convergence rate and ic.The 60 qs_070/075/080_run_01–20.mat files in the qs_sensitivity_results_paper_consistent folder are the core data files for the formal experiments. Each file contains thresholds, random seeds, algorithm parameters, phase boundaries, stable generations, and seven generational vectors of length 200; among them, 'generation' represents the generation number, 'history_f1' and 'history_f2' represent the root mean square of vehicle vertical acceleration and drive torque difference, respectively, and 'history_R1', 'history_R2', 'history_R', and 'R_smooth' represent the corresponding rates of change. In each file, 'final_chromosome' consists of 100 rows by 8 columns, with each row representing a final population individual. The 8 columns correspond sequentially to k2, c2, k3, c3, original objective 1 (drive torque difference RMS), original objective 2 (vehicle vertical acceleration RMS), Pareto rank, and crowding distance. The 60 files in total contain 12,000 generational records and 6,000 final individual records. The checkpoint_run_01–20_gen140/150/160.mat files, totaling 60, each save a 100×8 population from generation 140, 150, or 160, generational structures of length 200, and the state of the Mersenne Twister random number generator, which are used to share fixed-phase computation results or restore computations. The trailing NaN values in the checkpoints corresponding to generations not yet run are reserved positions and do not indicate data file corruption.SKM4.slx is a simulation model, while SKM4.slxc and the slprj directory are cache or verification files generated by Simulink and can be regenerated in a compatible environment. The remaining .m files include the official test entry, parameter configuration, objective functions, initialization, non-dominated sorting, tournament selection, fixed and adaptive genetic operators, population replacement, convergence recording, statistical analysis, plotting, and other programs.MAT, M, FIG, SLX, and SLXC are proprietary or semi-proprietary formats specific to MATLAB/Simulink. It is recommended to open them using MathWorks MATLAB and Simulink, with the installation program available for download from the official MathWorks website. The signrank, friedman, and boxplot analyses additionally require the Statistics and Machine Learning Toolbox. The precise software version was not recorded, and model metadata only indicates that it contains information on 64-bit Windows and earlier Simulink components. The SLXC and slprj caches may be related to the software version and operating system and should not be considered essential data for cross-platform reproduction. Numeric MAT v5 files can also be read by third-party software that supports this format; however, analyzing MATLAB table objects in MAT files as well as FIG and SLX files remains best achieved with MATLAB/Simulink compatibility. moment-balancing mechanism; multi-objective optimisation; IANSGA-II xu lei Wang Buyun PUBLIC https://creativecommons.org/licenses/by-nc-sa/4.0/ dataset Science Data Bank