Highway Lifecycle

v1.0.1

高速公路全生命周期智能化支持,涵盖规划设计、建设施工、运营管理、养护巡检四大阶段。使用当需要:(1) 桥梁/隧道设计图纸审查与规范符合性检查,(2) 交通事件检测与视频分析,(3) 道路病害识别与损害检测,(4) 隧道地质特征分析与围岩分级,(5) 多模态模型选型与评测指导,(6) 工程文档一致性审核。触发词:高...

0· 102·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for sxy799/highway-lifecycle.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Highway Lifecycle" (sxy799/highway-lifecycle) from ClawHub.
Skill page: https://clawhub.ai/sxy799/highway-lifecycle
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install highway-lifecycle

ClawHub CLI

Package manager switcher

npx clawhub@latest install highway-lifecycle
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
The name/description (planning, design review, event detection, damage detection, tunnel analysis, model selection) matches the included documentation and three utility scripts (evaluation, norm checking, road-damage reporting). One note: the skill recommends large vision/LLM models (Qwen3-VL-235B, InternVL-241B, GLM-4.5V etc.) and shows deployment examples, but does not declare any runtime dependencies or provide a deployment/install spec — this is reasonable for a guidance-oriented skill but means the user must supply model hosting/infrastructure separately.
Instruction Scope
SKILL.md and references focus on document review, image/video analysis, model evaluation and prompts. The runtime instructions and templates are narrowly scoped to the stated tasks (RAG/LLM-as-judge evaluation, metric calculation, file-based checks). There are no instructions to read unrelated system files, access credentials, or send data to unexpected external endpoints; curl examples point to localhost model servers for local deployment.
Install Mechanism
No install spec is provided (instruction-only with included Python scripts). That is the lowest-risk pattern: nothing will be downloaded or installed automatically by the skill. The included Python scripts are plain, self-contained utilities and do not perform network downloads or execute arbitrary remote code.
Credentials
The skill declares no required environment variables, no primary credential, and no config paths. The code does not reference external credentials or secrets. Model endpoints and GPU resource recommendations are documented but not requested as environment variables — users will need to provide/host models themselves if they want to run those workflows.
Persistence & Privilege
The skill does not request always:true and has no special persistence requirements. It will not modify other skills or system-wide settings based on the provided files/instructions.
Assessment
This skill appears coherent and documentation-rich, but review before use: (1) The scripts operate on files you supply — don't feed sensitive or regulated documents to external/remote models unless you control the endpoint. (2) The skill references large models and local server examples (vLLM, curl to localhost). It does not provide or install those models — you must host them and ensure adequate GPU/compute. (3) The Python scripts are readable and do not perform network calls or secret exfiltration, but always inspect and test scripts on non-sensitive sample data before running in production. (4) Keep your norms/knowledgebase up to date (SKILL.md notes that) and verify any automated decisions with domain experts for safety-critical engineering tasks.

Like a lobster shell, security has layers — review code before you run it.

latestvk977jb01chag7na04m0n1m8w1583ypm5
102downloads
0stars
2versions
Updated 3w ago
v1.0.1
MIT-0

高速公路「设建管养」智能化支持

为高速公路全生命周期提供智能化支持,覆盖规划设计、建设施工、运营管理、养护巡检四大阶段。

核心能力矩阵

阶段核心任务技术方案
规划设计图纸审查、规范符合性检查文档分析 + 规范知识库
建设施工隧道地质分析、围岩分级多模态视觉模型
运营管理事件检测、异常识别视频流分析
养护巡检道路病害识别、损害检测图像分类模型

一、规划设计阶段:图纸审查

1.1 审查类型

规范符合性审查

检查设计文档内容是否与现行规范条款一致:

  • 规范名称、编号、版本号是否正确
  • 技术指标是否满足规范最低要求
  • 检验频率、允许偏差是否符合规范规定
  • 材料性能、施工工艺是否满足规范要求

上下文一致性审查

检查不同章节、位置之间的冲突或不一致:

  • 材料一致性(钢材牌号、混凝土等级)
  • 设计参数一致性(跨度、荷载、应力、尺寸)
  • 预应力参数一致性(摩擦系数μ、偏差系数k、张拉应力)
  • 同一结构描述一致性
  • 规范引用一致性
  • 参数范围与具体数值关系

1.2 审查输出格式

{
  "审查结果": [
    {
      "类型": "规范符合性/上下文一致性",
      "规范条款": "对应条款编号",
      "不符合内容": "具体问题描述",
      "修改建议": "如何修改",
      "严重程度": "高/中/低",
      "来源位置": "源文件位置"
    }
  ],
  "总结": "整体评估"
}

1.3 判定规则

情况判定
数值不同不一致
材料等级不同不一致
同一对象多种描述冲突
数值满足范围要求一致
数值不满足范围不一致
表达模糊但不冲突建议优化

1.4 常见问题类型

规范版本过时

  • 检查引用规范是否为最新版本
  • 例:JT/T 329-2010 → JT/T 329-2025

参数冲突

  • 主梁混凝土 C40 vs C50
  • 钢板 Q235C vs Q355D
  • 定位筋间距不一致

规范数值比对

  • 粗骨料最大粒径 vs 钢筋最小净距 3/4
  • 水胶比 vs 规范限值
  • 保护层厚度 vs 环境类别要求

二、建设施工阶段:隧道地质分析

2.1 可识别特征

视觉感知类(表现良好)

  • 节理裂隙发育程度
  • 岩石完整性
  • 地下水状态

地质推理类(需大模型)

  • 岩石坚硬程度
  • 结构面结合程度
  • 围岩等级

专业参数类(需人工/仪器)

  • 地应力状态
  • Rc值/Kv值
  • 岩体结构类型

2.2 围岩分级参考

等级特征描述支护建议
坚硬岩,完整局部锚杆
较坚硬岩,较完整锚杆+局部喷层
较软岩,完整性一般锚杆+喷层+网
软岩,破碎复合衬砌
极软岩,极破碎加强支护

三、运营管理阶段:事件检测

3.1 检测类别

宏观交通流事件

  • is_congestion 拥堵
  • is_abnormal_stop 异常停车
  • is_normal_status 正常状态

细粒度目标事件

  • is_illegal_vehicle 非法车辆
  • is_pedestrian 行人
  • is_construction_rescue 施工/救援

高危异常事件

  • is_accident 事故
  • is_fire 火灾
  • is_spillage 抛洒物

3.2 模型选型建议

任务推荐模型说明
事件检测Qwen3-VL-30B性价比高,Recall优秀
行人/非法车辆Qwen3-VL-235B精度要求高
复杂推理大模型 + Think模式需要思维链

3.3 Think模式使用原则

场景Think模式原因
高速事件检测❌ 禁用导致幻觉/禁答
隧道特征分析✅ 启用提升推理准确性
道路病害检测✅ 启用减少误判

四、养护巡检阶段:病害检测

4.1 病害分类(RDD标准)

代码类型特征难度
D10横向裂缝垂直于车道
D00纵向裂缝平行于车道
D20龟裂网状纹理
D40坑槽3D凹陷
Repair修补区域色块差异

4.2 检测要点

横向裂缝:特征明显,不易与车道线混淆,各模型表现均好

纵向裂缝:易与车道线、路肩边缘、轮胎印混淆,需语义理解

龟裂:网状密集纹理,易误判为"粗糙路面",需高分辨率

坑槽:3D特征,关注误报率(FAR),需要紧急修复优先级

4.3 模型混淆分析

常见误判:

  • 横向裂缝 → 纵向裂缝(缺乏参照系)
  • 裂缝 → 车道线(先验偏见)
  • D40坑槽 → Repair修补(需区分新修补与病害)

五、多模态模型评测指南

5.1 评测维度

维度方法指标
高速事件自动化标签评分Macro F1
隧道特征LLM-as-judgeAccuracy
道路病害自动化标签评分Macro F1

5.2 模型能力对比

模型参数量特点
Qwen3-VL-30B30B性价比高,事件检测优秀
Qwen3-VL-235B235B综合最强,细粒度检测优秀
GLM-4.5V106B地质知识好,隧道分析优秀
InternVL-241B241B修补识别好,依赖大参数

5.3 Scaling Law 效应

参数量是性能硬通货,存在明显门槛效应:

  • 30B → 235B:各项指标全面提升
  • 小模型存在能力"坍缩"风险
  • InternVL 对参数规模依赖最强

六、工程知识注入

6.1 项目上下文模板

{
  "project_type": "高速公路",
  "location": "浙江省",
  "environment": "乡村区域 → JC2",
  "design_life": "100年",
  "applicable_norms": ["公路规范", "国标"],
  "inapplicable_norms": ["铁路规范"]
}

6.2 工程常识库

除锈等级

  • 工地连接:手动/电动工具 → St3级
  • 喷砂处理:Sa3级(工地不可达)

环境腐蚀分级

  • 乡村区域:JC2
  • 工业区:JC3

规范适用映射

  • 高速公路 → 公路规范
  • 铁路项目 → 铁路规范

七、参考资源

详细内容见 references/ 目录:

  • norm-compliance.md:规范符合性审查详细指南
  • multimodal-eval.md:多模态模型评测方法
  • tunnel-analysis.md:隧道地质分析方法
  • event-detection.md:交通事件检测规范

八、注意事项

  1. RAG架构限制:无法主动扫描全文进行一致性检查,需结构化参数索引
  2. 规范更新:知识库需及时更新规范版本
  3. 样本均衡:罕见事件(火灾、抛洒物)样本少,需补充数据
  4. 标注一致性:程度描述需穷举定义,避免模糊标注

Comments

Loading comments...