AI推动工业智能化转型~系列文章07：分类与诊断算法体系：故障识别的完整工具箱

第 7 期 | 分类与诊断算法体系：故障识别的完整工具箱

工业分类任务的终极形态是"诊断"：判断设备是否故障、炉况是否异常、产品是否合格。与互联网分类（猫狗图片）不同，工业诊断面临类别极端不平衡、标签模糊、误报代价不对称三重挑战。本期给出工业分类的算法工具箱与不平衡应对体系。

一、工业分类任务的四大类型 🗂️

#mermaid-svg-gta1xiHgwXaRc6Yi{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-gta1xiHgwXaRc6Yi .error-icon{fill:#552222;}#mermaid-svg-gta1xiHgwXaRc6Yi .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-gta1xiHgwXaRc6Yi .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-gta1xiHgwXaRc6Yi .marker{fill:#333333;stroke:#333333;}#mermaid-svg-gta1xiHgwXaRc6Yi .marker.cross{stroke:#333333;}#mermaid-svg-gta1xiHgwXaRc6Yi svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-gta1xiHgwXaRc6Yi p{margin:0;}#mermaid-svg-gta1xiHgwXaRc6Yi .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster-label text{fill:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster-label span{color:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster-label span p{background-color:transparent;}#mermaid-svg-gta1xiHgwXaRc6Yi .label text,#mermaid-svg-gta1xiHgwXaRc6Yi span{fill:#333;color:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi .node rect,#mermaid-svg-gta1xiHgwXaRc6Yi .node circle,#mermaid-svg-gta1xiHgwXaRc6Yi .node ellipse,#mermaid-svg-gta1xiHgwXaRc6Yi .node polygon,#mermaid-svg-gta1xiHgwXaRc6Yi .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-gta1xiHgwXaRc6Yi .rough-node .label text,#mermaid-svg-gta1xiHgwXaRc6Yi .node .label text,#mermaid-svg-gta1xiHgwXaRc6Yi .image-shape .label,#mermaid-svg-gta1xiHgwXaRc6Yi .icon-shape .label{text-anchor:middle;}#mermaid-svg-gta1xiHgwXaRc6Yi .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-gta1xiHgwXaRc6Yi .rough-node .label,#mermaid-svg-gta1xiHgwXaRc6Yi .node .label,#mermaid-svg-gta1xiHgwXaRc6Yi .image-shape .label,#mermaid-svg-gta1xiHgwXaRc6Yi .icon-shape .label{text-align:center;}#mermaid-svg-gta1xiHgwXaRc6Yi .node.clickable{cursor:pointer;}#mermaid-svg-gta1xiHgwXaRc6Yi .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-gta1xiHgwXaRc6Yi .arrowheadPath{fill:#333333;}#mermaid-svg-gta1xiHgwXaRc6Yi .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-gta1xiHgwXaRc6Yi .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-gta1xiHgwXaRc6Yi .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-gta1xiHgwXaRc6Yi .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-gta1xiHgwXaRc6Yi .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-gta1xiHgwXaRc6Yi .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster text{fill:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi .cluster span{color:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-gta1xiHgwXaRc6Yi .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-gta1xiHgwXaRc6Yi rect.text{fill:none;stroke-width:0;}#mermaid-svg-gta1xiHgwXaRc6Yi .icon-shape,#mermaid-svg-gta1xiHgwXaRc6Yi .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-gta1xiHgwXaRc6Yi .icon-shape p,#mermaid-svg-gta1xiHgwXaRc6Yi .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-gta1xiHgwXaRc6Yi .icon-shape .label rect,#mermaid-svg-gta1xiHgwXaRc6Yi .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-gta1xiHgwXaRc6Yi .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-gta1xiHgwXaRc6Yi .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-gta1xiHgwXaRc6Yi :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

工业分类任务类型

设备故障诊断输入: 振动/温度/电流输出: 故障模式

过程状态识别输入: 工艺参数特征输出: 顺行/异常模式

产品质量判定输入: 图像/工艺曲线输出: 合格/缺陷等级

操作事件检测输入: 参数变化模式输出: 事件类型

二、主流分类算法的工业画像 🧰

逻辑回归与 SVM：小样本时代的遗产

• 逻辑回归：输出概率、可解释，适合作为诊断基线；
• SVM：小样本高维场景的经典（如数百样本的故障分类），核技巧处理非线性；短板是大数据慢、概率输出需校准。

树模型家族：诊断主力

随机森林与 XGBoost 在工业诊断中的统治力与回归场景一致，另有两个独特优势：

• 特征重要性=诊断依据：输出"振动 3 倍频能量贡献最大"直接对应机械故障机理（不对中、松动）；
• 类权重机制：class_weight 参数一行代码应对不平衡。

CNN：图像类诊断的标准解

表面缺陷检测、仪表读数识别、料面状态判定的默认选择（第 10 期专述）。

一维 CNN / LSTM：波形类诊断

振动信号、电流波形的故障诊断，直接以原始波形或频谱为输入，端到端学习——替代传统"人工提特征+分类器"路线，但需更多故障样本支撑。

三、类别不平衡：工业诊断的核心难题 ⚖️

工业数据的典型失衡比：正常样本 : 故障样本 = 10000 : 1 甚至更高。此时"全部判正常"的模型准确率 99.99% 却毫无价值。

应对体系的五个层次：

#mermaid-svg-cSvgGPQIF4JMYScv{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-cSvgGPQIF4JMYScv .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-cSvgGPQIF4JMYScv .error-icon{fill:#552222;}#mermaid-svg-cSvgGPQIF4JMYScv .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-cSvgGPQIF4JMYScv .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-cSvgGPQIF4JMYScv .marker{fill:#333333;stroke:#333333;}#mermaid-svg-cSvgGPQIF4JMYScv .marker.cross{stroke:#333333;}#mermaid-svg-cSvgGPQIF4JMYScv svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-cSvgGPQIF4JMYScv p{margin:0;}#mermaid-svg-cSvgGPQIF4JMYScv .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-cSvgGPQIF4JMYScv .cluster-label text{fill:#333;}#mermaid-svg-cSvgGPQIF4JMYScv .cluster-label span{color:#333;}#mermaid-svg-cSvgGPQIF4JMYScv .cluster-label span p{background-color:transparent;}#mermaid-svg-cSvgGPQIF4JMYScv .label text,#mermaid-svg-cSvgGPQIF4JMYScv span{fill:#333;color:#333;}#mermaid-svg-cSvgGPQIF4JMYScv .node rect,#mermaid-svg-cSvgGPQIF4JMYScv .node circle,#mermaid-svg-cSvgGPQIF4JMYScv .node ellipse,#mermaid-svg-cSvgGPQIF4JMYScv .node polygon,#mermaid-svg-cSvgGPQIF4JMYScv .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-cSvgGPQIF4JMYScv .rough-node .label text,#mermaid-svg-cSvgGPQIF4JMYScv .node .label text,#mermaid-svg-cSvgGPQIF4JMYScv .image-shape .label,#mermaid-svg-cSvgGPQIF4JMYScv .icon-shape .label{text-anchor:middle;}#mermaid-svg-cSvgGPQIF4JMYScv .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-cSvgGPQIF4JMYScv .rough-node .label,#mermaid-svg-cSvgGPQIF4JMYScv .node .label,#mermaid-svg-cSvgGPQIF4JMYScv .image-shape .label,#mermaid-svg-cSvgGPQIF4JMYScv .icon-shape .label{text-align:center;}#mermaid-svg-cSvgGPQIF4JMYScv .node.clickable{cursor:pointer;}#mermaid-svg-cSvgGPQIF4JMYScv .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-cSvgGPQIF4JMYScv .arrowheadPath{fill:#333333;}#mermaid-svg-cSvgGPQIF4JMYScv .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-cSvgGPQIF4JMYScv .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-cSvgGPQIF4JMYScv .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-cSvgGPQIF4JMYScv .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-cSvgGPQIF4JMYScv .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-cSvgGPQIF4JMYScv .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-cSvgGPQIF4JMYScv .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-cSvgGPQIF4JMYScv .cluster text{fill:#333;}#mermaid-svg-cSvgGPQIF4JMYScv .cluster span{color:#333;}#mermaid-svg-cSvgGPQIF4JMYScv div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-cSvgGPQIF4JMYScv .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-cSvgGPQIF4JMYScv rect.text{fill:none;stroke-width:0;}#mermaid-svg-cSvgGPQIF4JMYScv .icon-shape,#mermaid-svg-cSvgGPQIF4JMYScv .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-cSvgGPQIF4JMYScv .icon-shape p,#mermaid-svg-cSvgGPQIF4JMYScv .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-cSvgGPQIF4JMYScv .icon-shape .label rect,#mermaid-svg-cSvgGPQIF4JMYScv .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-cSvgGPQIF4JMYScv .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-cSvgGPQIF4JMYScv .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-cSvgGPQIF4JMYScv :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

不平衡应对五层次

1 数据层: 重采样SMOTE过采样/随机欠采样注意: SMOTE在时序数据慎用

2 算法层: 代价敏感class_weight调故障类权重阈值移动替代0.5默认

3 评估层: 换指标弃用准确率用召回率/精确率/F1/AUC

4 范式层: 换问题定义分类→异常检测只用正常样本建模

5 知识层: 规则融合稀有故障用专家规则兜底模型+规则双通道

评估指标的正确选择：

实战权衡：安全相关诊断（如炉缸烧穿预警）优先保召回率（≥99%），容忍较高误报，由人工复核消化；效率相关诊断（如质量分级）追求精确率与召回率平衡。

四、报警阈值：被忽视的决策变量 🎚️

分类模型输出概率，阈值决定最终判定。默认 0.5 几乎总是错的：

#mermaid-svg-XngsGn8pdaPgkji9{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-XngsGn8pdaPgkji9 .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-XngsGn8pdaPgkji9 .error-icon{fill:#552222;}#mermaid-svg-XngsGn8pdaPgkji9 .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-XngsGn8pdaPgkji9 .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-XngsGn8pdaPgkji9 .marker{fill:#333333;stroke:#333333;}#mermaid-svg-XngsGn8pdaPgkji9 .marker.cross{stroke:#333333;}#mermaid-svg-XngsGn8pdaPgkji9 svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-XngsGn8pdaPgkji9 p{margin:0;}#mermaid-svg-XngsGn8pdaPgkji9 .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-XngsGn8pdaPgkji9 .cluster-label text{fill:#333;}#mermaid-svg-XngsGn8pdaPgkji9 .cluster-label span{color:#333;}#mermaid-svg-XngsGn8pdaPgkji9 .cluster-label span p{background-color:transparent;}#mermaid-svg-XngsGn8pdaPgkji9 .label text,#mermaid-svg-XngsGn8pdaPgkji9 span{fill:#333;color:#333;}#mermaid-svg-XngsGn8pdaPgkji9 .node rect,#mermaid-svg-XngsGn8pdaPgkji9 .node circle,#mermaid-svg-XngsGn8pdaPgkji9 .node ellipse,#mermaid-svg-XngsGn8pdaPgkji9 .node polygon,#mermaid-svg-XngsGn8pdaPgkji9 .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-XngsGn8pdaPgkji9 .rough-node .label text,#mermaid-svg-XngsGn8pdaPgkji9 .node .label text,#mermaid-svg-XngsGn8pdaPgkji9 .image-shape .label,#mermaid-svg-XngsGn8pdaPgkji9 .icon-shape .label{text-anchor:middle;}#mermaid-svg-XngsGn8pdaPgkji9 .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-XngsGn8pdaPgkji9 .rough-node .label,#mermaid-svg-XngsGn8pdaPgkji9 .node .label,#mermaid-svg-XngsGn8pdaPgkji9 .image-shape .label,#mermaid-svg-XngsGn8pdaPgkji9 .icon-shape .label{text-align:center;}#mermaid-svg-XngsGn8pdaPgkji9 .node.clickable{cursor:pointer;}#mermaid-svg-XngsGn8pdaPgkji9 .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-XngsGn8pdaPgkji9 .arrowheadPath{fill:#333333;}#mermaid-svg-XngsGn8pdaPgkji9 .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-XngsGn8pdaPgkji9 .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-XngsGn8pdaPgkji9 .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-XngsGn8pdaPgkji9 .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-XngsGn8pdaPgkji9 .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-XngsGn8pdaPgkji9 .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-XngsGn8pdaPgkji9 .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-XngsGn8pdaPgkji9 .cluster text{fill:#333;}#mermaid-svg-XngsGn8pdaPgkji9 .cluster span{color:#333;}#mermaid-svg-XngsGn8pdaPgkji9 div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-XngsGn8pdaPgkji9 .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-XngsGn8pdaPgkji9 rect.text{fill:none;stroke-width:0;}#mermaid-svg-XngsGn8pdaPgkji9 .icon-shape,#mermaid-svg-XngsGn8pdaPgkji9 .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-XngsGn8pdaPgkji9 .icon-shape p,#mermaid-svg-XngsGn8pdaPgkji9 .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-XngsGn8pdaPgkji9 .icon-shape .label rect,#mermaid-svg-XngsGn8pdaPgkji9 .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-XngsGn8pdaPgkji9 .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-XngsGn8pdaPgkji9 .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-XngsGn8pdaPgkji9 :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

模型输出概率分布

阈值左移: 高召回高误报安全攸关场景

阈值右移: 高精确低召回成本敏感场景

阈值优化方法:按验证集F1最优点或按误报/漏报成本比求解期望成本最小阈值

成本敏感阈值公式：设漏报成本 C_miss、误报成本 C_false，最优阈值 t* ≈ C_false / (C_false + C_miss)。漏报成本是误报的 100 倍时，阈值应降至约 0.01——这就是工业报警阈值必须业务化定制的原因。

五、诊断模型的工业验收标准 ✅

诊断模型上线前必须通过四项验收：

六、从分类到诊断系统：工程闭环 🔧

#mermaid-svg-iyXmacfZzZn47gWB{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-iyXmacfZzZn47gWB .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-iyXmacfZzZn47gWB .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-iyXmacfZzZn47gWB .error-icon{fill:#552222;}#mermaid-svg-iyXmacfZzZn47gWB .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-iyXmacfZzZn47gWB .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-iyXmacfZzZn47gWB .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-iyXmacfZzZn47gWB .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-iyXmacfZzZn47gWB .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-iyXmacfZzZn47gWB .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-iyXmacfZzZn47gWB .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-iyXmacfZzZn47gWB .marker{fill:#333333;stroke:#333333;}#mermaid-svg-iyXmacfZzZn47gWB .marker.cross{stroke:#333333;}#mermaid-svg-iyXmacfZzZn47gWB svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-iyXmacfZzZn47gWB p{margin:0;}#mermaid-svg-iyXmacfZzZn47gWB .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-iyXmacfZzZn47gWB .cluster-label text{fill:#333;}#mermaid-svg-iyXmacfZzZn47gWB .cluster-label span{color:#333;}#mermaid-svg-iyXmacfZzZn47gWB .cluster-label span p{background-color:transparent;}#mermaid-svg-iyXmacfZzZn47gWB .label text,#mermaid-svg-iyXmacfZzZn47gWB span{fill:#333;color:#333;}#mermaid-svg-iyXmacfZzZn47gWB .node rect,#mermaid-svg-iyXmacfZzZn47gWB .node circle,#mermaid-svg-iyXmacfZzZn47gWB .node ellipse,#mermaid-svg-iyXmacfZzZn47gWB .node polygon,#mermaid-svg-iyXmacfZzZn47gWB .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-iyXmacfZzZn47gWB .rough-node .label text,#mermaid-svg-iyXmacfZzZn47gWB .node .label text,#mermaid-svg-iyXmacfZzZn47gWB .image-shape .label,#mermaid-svg-iyXmacfZzZn47gWB .icon-shape .label{text-anchor:middle;}#mermaid-svg-iyXmacfZzZn47gWB .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-iyXmacfZzZn47gWB .rough-node .label,#mermaid-svg-iyXmacfZzZn47gWB .node .label,#mermaid-svg-iyXmacfZzZn47gWB .image-shape .label,#mermaid-svg-iyXmacfZzZn47gWB .icon-shape .label{text-align:center;}#mermaid-svg-iyXmacfZzZn47gWB .node.clickable{cursor:pointer;}#mermaid-svg-iyXmacfZzZn47gWB .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-iyXmacfZzZn47gWB .arrowheadPath{fill:#333333;}#mermaid-svg-iyXmacfZzZn47gWB .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-iyXmacfZzZn47gWB .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-iyXmacfZzZn47gWB .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-iyXmacfZzZn47gWB .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-iyXmacfZzZn47gWB .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-iyXmacfZzZn47gWB .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-iyXmacfZzZn47gWB .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-iyXmacfZzZn47gWB .cluster text{fill:#333;}#mermaid-svg-iyXmacfZzZn47gWB .cluster span{color:#333;}#mermaid-svg-iyXmacfZzZn47gWB div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-iyXmacfZzZn47gWB .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-iyXmacfZzZn47gWB rect.text{fill:none;stroke-width:0;}#mermaid-svg-iyXmacfZzZn47gWB .icon-shape,#mermaid-svg-iyXmacfZzZn47gWB .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-iyXmacfZzZn47gWB .icon-shape p,#mermaid-svg-iyXmacfZzZn47gWB .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-iyXmacfZzZn47gWB .icon-shape .label rect,#mermaid-svg-iyXmacfZzZn47gWB .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-iyXmacfZzZn47gWB .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-iyXmacfZzZn47gWB .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-iyXmacfZzZn47gWB :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

单点分类模型

诊断系统四层架构

信号层: 数据质量门传感器失效先行排除

检测层: 异常检测无监督初筛

诊断层: 故障分类有监督定性

决策层: 处置建议规则引擎+案例检索

报警输出: 分级报警预警/报警/紧急

分级报警制是工业诊断系统的标准设计：低置信度异常→提示级（记录不打扰）、中置信度→预警级（看板提示）、高置信度+高严重度→报警级（强制确认）——避免"狼来了"式的报警疲劳。

🎯 本期小结

• 工业分类四大任务：设备诊断、状态识别、质量判定、事件检测；
• 类别不平衡五层应对：重采样、代价敏感、换指标、换范式（异常检测）、规则兜底；
• 报警阈值是业务决策变量，按误报/漏报成本比优化，默认 0.5 几乎必错；
• 验收四关：分工况混淆矩阵、提前量测试、误报消化评估、失效演练；
• 工程落地采用"检测→诊断→决策"分层架构与分级报警制。

下期预告：第 8 期攻克工业数据的时间本质——时序建模方法谱系：从平稳性检验到时序交叉验证，为深度学习时序篇（第 11 期）奠基。⏳

作者：高炉炼铁智能化技术研究者，专注钢铁冶金与人工智能 交叉领域。

👍 如果觉得有帮助，请点赞、收藏、转发！
版权归作者所有，未经许可请勿抄袭，套用，商用(或其它具有利益性行为)。
🔔 关注专栏，不错过后续精彩内容