BeamRL-EvalData-v2
收藏资源简介:
BeamRL-EvalData-v2是BeamPERL的扩展评估数据集,专门用于梁力学问题的符号计算评估。该数据集包含123个符号梁力学问题,核心任务是计算支座反力,并以载荷符号P的带符号系数作为真实答案(例如0.222222P)。数据集包含29个特征字段,涵盖配置ID、载荷信息(位置、值、类型)、支座位置、几何参数、坐标点、内力分析结果(剪力、弯矩)、变形分析结果(斜率、挠度)、各物理量的文本描述、结构离散信息(点、段)、反力、内力、变形、自然语言查询问题(每个样本4个变体)、答案、问题类别、版本号以及载荷分布参数。数据集分为7个类别:id(30个样本,单点载荷且支座位于梁端)、ood_loads(30个样本,2-4个点载荷且支座位于梁端)、ood_supports(18个样本,支座位置变化且包含1-3个点载荷)、ood_dist_left/ood_dist_middle/ood_dist_right(各5个样本,分别表示分布载荷锚定在铰支座、完全在梁内部、锚定在滚轴支座)、ood_length(15个样本,梁长度变化为7L、11L、13L)、ood_moment(15个样本,纯力偶作用无点载荷)。原始24样本评估集已完整包含为子集。所有问题均使用修改版SymBeam进行符号求解,并验证了力与力矩平衡,自然语言问题由大语言模型生成。该数据集适用于梁力学分析、符号计算、结构工程教育以及机器学习模型在工程力学问题求解能力的评估。
BeamRL-EvalData-v2 is an extended evaluation dataset derived from BeamPERL, specifically designed for symbolic computation evaluation of beam mechanics problems. This dataset contains 123 symbolic beam mechanics problems, where the core task is to calculate support reactions, using the signed coefficient of the load symbol P as the ground-truth answer (e.g., 0.222222P). It includes 29 feature fields, covering configuration ID, load information (position, value, type), support positions, geometric parameters, coordinate points, internal force analysis results (shear force, bending moment), deformation analysis results (slope, deflection), text descriptions of various physical quantities, structural discretization information (points, segments), reactions, internal forces, deformations, natural language query questions (4 variants per sample), answers, problem categories, version numbers, and load distribution parameters. The dataset is divided into 7 categories: id (30 samples, single-point load with supports located at beam ends), ood_loads (30 samples, 2-4 point loads with supports at beam ends), ood_supports (18 samples, varying support positions with 1-3 point loads), ood_dist_left / ood_dist_middle / ood_dist_right (5 samples each, indicating distributed loads anchored at pinned supports, fully within the beam, and anchored at roller supports respectively), ood_length (15 samples, with beam lengths varied to 7L, 11L, and 13L), and ood_moment (15 samples, featuring pure couple loads without point loads). The original 24-sample evaluation set is fully included as a subset. All problems were solved symbolically using a modified version of SymBeam, and force and moment equilibrium were verified. The natural language questions were generated by large language models (LLMs). This dataset is applicable to beam mechanics analysis, symbolic computation, structural engineering education, and the evaluation of machine learning models' problem-solving capabilities in engineering mechanics.
数据集概述
BeamRL-EvalData-v2 是一个用于评估梁力学问题的符号计算数据集,包含 123 个样本,每个样本都要求计算支撑处的反作用力,并以载荷符号 P 的带符号系数(例如 0.222222P)作为真实答案。
数据集特征
数据集中每个样本包含以下字段:
configuration_id: 配置标识符(字符串)load_position: 载荷位置(浮点数)load_positions: 载荷位置列表(浮点数列表)load_values: 载荷值列表(字符串列表)support_positions: 支撑位置列表(浮点数列表)parameters: 参数(字符串)x_coordinates: x 坐标列表(浮点数列表)shear_force,bending_moment,slope,deflection: 剪力、弯矩、斜率、挠度列表(浮点数列表)shear_force_info,bending_moment_info,slope_info,deflection_info: 对应的附加信息(字符串)points,segments,reactions,internal_loads,deflections: 几何与力学信息(字符串)query: 自然语言问题列表(字符串列表,每个样本 4 个变体,评估使用query[0])answer: 答案列表(字符串列表)category: 类别(字符串)version: 版本(字符串)dist_load_k,dist_load_n_end,moment_c,dist_load_n_start,length_multiplier: 整数类型参数moment_x_frac: 浮点数类型参数
数据划分
仅包含 训练集(train 划分),共 123 个样本,占用 563778 字节。下载大小为 454653 字节。
数据集类别
数据集按困难程度和问题类型分为以下类别:
| 类别 | 样本数 | 描述 |
|---|---|---|
id |
30 | 单个集中载荷,支撑在梁的两端 |
ood_loads |
30 | 2–4 个集中载荷,支撑在梁的两端 |
ood_supports |
18 | 支撑位置变化,1–3 个集中载荷 |
ood_dist_left |
5 | 分布载荷,锚定在销钉处 |
ood_dist_middle |
5 | 分布载荷,完全位于内部 |
ood_dist_right |
5 | 分布载荷,锚定在滚子处 |
ood_length |
15 | 梁长度变化(7L、11L、13L) |
ood_moment |
15 | 纯力偶作用,无集中载荷 |
原始 24 样本评估集(tphage/BeamRL-EvalData)以完全相同的位包含在 id、ood_loads 和 ood_supports 子集中。
数据生成
所有配置均使用修改后的 SymBeam 进行符号求解,并通过力和力矩平衡验证。自然语言的问题变体由 LLM 生成。该数据集由 BeamPERL 仓库中的 DataGen/generate_eval_v2.py 脚本生成。




