sincos.cc-website/website/static/tool-physics-ai-fluent.html

<!DOCTYPE html>
<html lang="zh-CN">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Physics AI Fluent Analysis - GPU 流场仿真 | Easy Data Collection</title>
<meta name="description" content="Physics AI Fluent Analysis —— 用 GPU 加速的 3D D3Q19 格子玻尔兹曼方法求解水平卧式流化床空床气相流场：恒功率风机曲线 + 多孔布风板自洽求解，为 AI for Science 神经算子提供真值引擎与数据工厂。">
<meta name="keywords" content="Physics AI Fluent, GPU LBM, 格子玻尔兹曼, 流化床, 流场仿真, AI for Science, 神经算子, FNO, GNN">
<link rel="stylesheet" href="/static/css/style.css">
<link rel="icon" href="/static/img/logo.svg" type="image/svg+xml">
<script src="/static/js/nav.js"></script>
</head>
<body>

<header class="site-header">
  <div class="container header-inner">
    <a class="brand" href="/">
      <img src="/static/img/logo.svg" alt="Easy Data Collection">
      <span class="brand-name">Easy Data Collection</span>
    </a>
    <nav class="main-nav" id="mainNav">
      <div class="nav-item has-dropdown">
        <a class="nav-link" href="/research/">前沿技术研究</a>
        <button class="nav-caret" type="button" aria-label="展开前沿技术研究">▾</button>
        <div class="dropdown">
          <a href="http://124.220.63.9:25000">Research Digest</a>
          <a href="/research/neural/">Neural Research</a>
          <a href="/research/reslib/">ResLib</a>
        </div>
      </div>
      <div class="nav-item has-dropdown">
        <a class="nav-link active" href="/tools/">工具开发</a>
        <button class="nav-caret" type="button" aria-label="展开工具开发">▾</button>
        <div class="dropdown">
          <a href="/tools/sequence-diagram/">Sequence-Diagram</a>
          <a href="/tools/uni-viewer/">Uni-Viewer</a>
          <a href="/tools/engineer-craft/">Engineer Craft</a>
          <a href="/tools/five-phases-observer/">The Five Phases Observer</a>
          <a href="/tools/unitouch/">UniTouch Interface</a>
          <a href="/tools/edc-lite-erp/">EDC Lite ERP</a>
          <a class="active" href="/tools/physics-ai-fluent/">Physics AI Fluent</a>
        </div>
      </div>
      <a class="nav-link" href="/resources/">资源中心</a>
      <div class="nav-item has-dropdown">
        <a class="nav-link" href="/license/">License</a>
        <button class="nav-caret" type="button" aria-label="展开 License">▾</button>
        <div class="dropdown">
          <a href="/license/apply">申请 License</a>
          <a href="/license/status">查询 / 下载</a>
          <a href="/license/package">资源包获取</a>
        </div>
      </div>
      <a class="nav-link" href="/blog/">博客</a>
    </nav>
    <button class="nav-toggle" id="navToggle" aria-label="菜单">☰</button>
  </div>
</header>

<main>
  <section class="page-hero">
    <div class="container">
      <p class="hero-eyebrow">工具开发 · AI FOR SCIENCE</p>
      <h1>Physics AI Fluent Analysis<br>水平卧式流化床 · 空床流场仿真</h1>
      <p class="page-hero-sub">用 GPU 加速的 3D D3Q19 格子玻尔兹曼方法（LBM）求解流化床空床气相流场：以恒功率离心风机曲线为压力进口、布风板为多孔板，流量与压差由物理自洽涌现；为 AI for Science / 神经算子提供真值引擎与数据工厂。</p>
      <div class="hero-actions">
        <a href="http://124.220.63.9:30000/admin/Physics-Ai-Fluent-Analysis" class="btn btn-light">项目主页（Gitea）</a>
        <a href="/tools/" class="btn btn-outline-light">返回工具开发</a>
      </div>
      <div class="hero-stats">
        <div class="hero-stat"><b>~120 MLUPS</b><span>GPU 格点更新/秒</span></div>
        <div class="hero-stat"><b>827 万</b><span>网格单元（100 格/m）</span></div>
        <div class="hero-stat"><b>184 s</b><span>3900 步求解</span></div>
        <div class="hero-stat"><b>Q ≈ 1463 m³/h</b><span>自洽涌现工况点</span></div>
      </div>
    </div>
  </section>

  <section class="section">
    <div class="container">
      <h2 class="section-title">物理场景</h2>
      <p class="section-desc">水平卧式流化床空床（无颗粒）气相流场，完整还原布风板、分区风箱、渐扩段与排风结构。</p>
      <div class="grid">
        <div class="card"><div class="icon">▤</div><h3>布风板</h3><p>2400 mm × 567 mm，厚 2 mm；孔径 2 mm，开孔率 5.6%，孔轴 5° 倾斜（朝出料 -x 方向）。</p></div>
        <div class="card"><div class="icon">◫</div><h3>分区进风</h3><p>沿长度均分 4 段（每段 600 mm），每段独立风箱（高 450 mm），DN300 侧进风、风向平行床板。</p></div>
        <div class="card"><div class="icon">⇱</div><h3>沸腾 / 渐扩段</h3><p>1000 mm 高度内由 2400 × 567 渐扩至 2400 × 1200；上部稳压段 400 mm；渐缩段 500 mm。</p></div>
        <div class="card"><div class="icon">◎</div><h3>排风</h3><p>顶部居中 DN800 圆管排出，形成完整的进风—布风—沸腾—排风路径。</p></div>
        <div class="card"><div class="icon">⚙</div><h3>恒功率风机</h3><p>单段额定 2500 m³/h（合计 10000 m³/h）、额定压差 2900 Pa；恒功率 Δp · Q = P ≈ 8056 W，关死压力 4350 Pa。</p></div>
        <div class="card"><div class="icon">◈</div><h3>物理自洽的工况点</h3><p>2500 m³/h 与 2900 Pa 是风机额定点，不是独立约束。仿真只施加风机曲线压力，流量由「风机曲线 ∩ 系统阻力」唯一交点自然涌现。</p></div>
      </div>
    </div>
  </section>

  <section class="section section-alt">
    <div class="container">
      <h2 class="section-title">数值方法</h2>
      <div class="grid">
        <div class="card"><div class="icon">◍</div><h3>GPU 格子玻尔兹曼</h3><p>D3Q19 + BGK 碰撞 + 推流 + 半步反弹固体边界；NVIDIA GPU（numba.cuda）实测约 120 MLUPS。</p></div>
        <div class="card"><div class="icon">▦</div><h3>网格</h3><p>100 格/m，240 × 130 × 265 ≈ 827 万单元（8 GB 显存上限）；布风板按固体薄板 + 规则开孔离散。</p></div>
        <div class="card"><div class="icon">⇄</div><h3>边界条件</h3><p>压力进口（固定密度 = 风机曲线压力，速度零梯度自由）；压力出口（密度 = 1）；固体壁无滑移。</p></div>
        <div class="card"><div class="icon">⏱</div><h3>求解成本</h3><p>3900 步 / 184 s 完成；对比传统 CFD 小时～天级，把空床流场分析压到分钟级。</p></div>
      </div>
    </div>
  </section>

  <section class="section">
    <div class="container">
      <h2 class="section-title">仿真结果</h2>
      <p class="section-desc">压力符号、结构、量级均正确：风箱高压 → 布风板节流降压 → 沸腾段低压 → 向排风递减。</p>
      <div class="grid">
        <div class="card"><h3>工况点风量</h3><p><b>Q ≈ 1463 m³/h</b>（风机曲线与系统阻力交点）</p></div>
        <div class="card"><h3>风箱静压</h3><p><b>4061 Pa</b>（4 段侧进风）</p></div>
        <div class="card"><h3>沸腾段静压</h3><p><b>181 Pa</b>（经布风板节流后）</p></div>
        <div class="card"><h3>布风板压差</h3><p><b>+3880 Pa</b>（风箱 &gt; 沸腾段）</p></div>
        <div class="card"><h3>最大流速</h3><p><b>4.89 m/s</b>（射流区）</p></div>
        <div class="card"><h3>求解</h3><p><b>3900 步 / 184 s</b>（100 格/m，2 mm 板厚）</p></div>
      </div>
    </div>
  </section>

  <section class="section section-alt">
    <div class="container">
      <h2 class="section-title">流场图集</h2>
      <p class="section-desc">全部由 GPU 求解器输出（4× 分辨率原始图，网页为压缩版）。</p>
      <div class="gallery">
        <figure class="shot wide">
          <img src="/static/img/physics-ai-fluent/fbed-overview.jpg" alt="流化床空床流场四联总览">
          <figcaption>四联总览 —— 垂直切面（风箱 / 布风板 / 沸腾段 / 排风）、侧向渐扩、布风板近区射流与沸腾段水平切面。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-streamlines.jpg" alt="双大图流线">
          <figcaption>流线图 —— 布风板射流合并与渐扩段速度衰减。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-pressure.jpg" alt="静压场四联">
          <figcaption>静压场四联 —— 风箱高压到排风的完整压力梯度。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-cutaway.jpg" alt="三正交切面三维剖切">
          <figcaption>三正交切面三维剖切 —— 立体观察内部流场结构。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-uniformity.jpg" alt="各区段流量均匀性">
          <figcaption>各区段流量均匀性 —— 验证 4 段配气一致性。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-decay.jpg" alt="沿高度气速衰减">
          <figcaption>沿高度气速衰减 —— 沸腾段与自由空域的速度剖面。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-flowrate.jpg" alt="风量平衡">
          <figcaption>风量平衡 —— 进风 / 布风板 / 排风三段流量守恒校核。</figcaption>
        </figure>
        <figure class="shot">
          <img src="/static/img/physics-ai-fluent/fbed-tilt.jpg" alt="布风板 5° 倾斜影响">
          <figcaption>布风板 5° 倾斜对近板风向的影响分析。</figcaption>
        </figure>
      </div>
    </div>
  </section>

  <section class="section">
    <div class="container">
      <h2 class="section-title">AI for Science 闭环</h2>
      <p class="section-desc">本项目天然是 AI for Science / 神经算子的数据工厂与落地场景。</p>
      <div class="gallery">
        <figure class="shot wide">
          <img src="/static/img/physics-ai-fluent/pipeline.jpg" alt="数据与训练流水线">
          <figcaption>数据与训练流水线 —— 参数空间采样 → 批量仿真 → 数据工程 → 神经算子训练（FNO / DeepONet / GNN）→ 推理引擎部署（ONNX → TensorRT）→ 实时仿真服务（gRPC + WebSocket），并设置精度 / 稳定性 / 成本 / 可复现四道门禁。</figcaption>
        </figure>
      </div>
      <div class="grid">
        <div class="card"><div class="icon">◈</div><h3>神经算子</h3><p>FNO / DeepONet 把「解算子」参数化为网络权重：输入几何 + 边界，输出全场。</p></div>
        <div class="card"><div class="icon">▦</div><h3>图神经网络</h3><p>MeshGraphNet 类模型适配非结构 / 复杂几何，天然处理不规则边界。</p></div>
        <div class="card"><div class="icon">⚡</div><h3>代理模型加速</h3><p>用约 1% 的算力逼近求解器精度，支撑大规模参数扫描与在线优化。</p></div>
        <div class="card"><div class="icon">◉</div><h3>数字孪生</h3><p>毫秒级推理 → 参数拖拽即时刷新 → 在线优化与预测性维护。</p></div>
      </div>
    </div>
  </section>

  <section class="section section-alt">
    <div class="container">
      <h2 class="section-title">复现与技术栈</h2>
      <p class="section-desc">Python 3.11+ · numpy · numba · numba-cuda（CUDA 13.2 runtime）· matplotlib · scipy / pyamg（RANS 尝试版）</p>
      <div class="grid">
        <div class="card"><h3>主仿真</h3><p><span class="mono">python tools/fbed_simulation.py</span>（100 格/m，约 3 - 5 分钟）</p></div>
        <div class="card"><h3>批量出图</h3><p><span class="mono">python tools/fbed_figures.py</span> · <span class="mono">python tools/fbed_analysis.py</span>（秒级）</p></div>
        <div class="card"><h3>可调参数</h3><p><span class="mono">FBED_PPM</span> 网格分辨率 · <span class="mono">FBED_ULB</span> 参考格速 · <span class="mono">FBED_DP</span> 额定压差等环境变量</p></div>
        <div class="card"><h3>脚本体系</h3><p>求解器 / 出图 / 分析 / 通量诊断 / 切面探针 / RANS 尝试版（留档）</p></div>
      </div>
      <div class="pill-row mt-24">
        <span class="pill">Python</span>
        <span class="pill">D3Q19 LBM</span>
        <span class="pill">numba.cuda</span>
        <span class="pill">CUDA</span>
        <span class="pill">GPU 120 MLUPS</span>
        <span class="pill">FNO / GNN</span>
      </div>
    </div>
  </section>

  <section class="section">
    <div class="container narrow">
      <h2 class="section-title">诚实声明（局限）</h2>
      <div class="grid">
        <div class="card"><h3>板厚亚网格</h3><p>2 mm 板厚在 100 格/m 下小于一个网格，只能按 1 格最薄板近似；真正解析需 ~500 格/m（约 157 GB 显存）。</p></div>
        <div class="card"><h3>5° 倾角未几何解析</h3><p>粗网格下孔道倾斜不足 1 格，暂未施加；方向已按设计设为 -x。</p></div>
        <div class="card"><h3>不可压近似</h3><p>LBM 密度波动约 2.5%，压力场为相对值；层流求解，湍流混合未解析。</p></div>
        <div class="card"><h3>工况点偏低</h3><p>细网格把开孔阻力解析得更「阻」，工况点落在 1463 m³/h；标定到额定需放大等效开孔或减薄板厚。</p></div>
      </div>
    </div>
  </section>
</main>

<footer class="site-footer">
  <div class="container footer-inner">
    <div class="footer-brand">
      <img src="/static/img/logo.svg" alt="">
      <span>Easy Data Collection</span>
    </div>
    <span>Copyright &copy; 2026 Easy Data Collection. All rights reserved.</span>
    <span class="footer-links">
      <a href="/resources/">资源中心</a><span class="sep">·</span>
      <a href="/license/">License</a><span class="sep">·</span>
      <a href="/admin">管理页面</a><span class="sep">·</span>
      <a href="https://beian.miit.gov.cn/" target="_blank" rel="noopener">苏ICP备2026056863号</a>
    </span>
  </div>
</footer>

</body>
</html>