缺陷当场停线修，还是流到最后再挑出来？Stop and fix the defect at its source—or sort it out at the end?
两条线都是 3 个工位、节拍 8 秒，每 4 件出现 1 件缺陷。上线让红件继续经过 2、3 工位，最后才判废；下线由 1 工位当场检出，拉亮安灯、整线停 4 秒并修正后再放行。看清楚：停线不是目的，不把异常传给下游才是。Both lines have three stations at an 8-second cycle, with one defect in every four parts. The upper line keeps processing the red part through stations 2 and 3 before scrapping it. The lower line detects at station 1, triggers Andon, stops for four seconds, corrects the part, and only then releases it. Stopping is not the goal; containing the abnormality is.
💡 跑完看数据：停 4 秒，省下游白干 16 秒💡 Run it, then read the numbers: a 4-second stop avoids 16 seconds of downstream waste
全部来自同一 stats()One stats() source| 生产方式Production method | 完成Completed | 好件Good | 废品Scrap | 代价Penalty |
|---|---|---|---|---|
| A · 末端检验A · End inspection | 0 | 0 | 0 | 0s |
| B · 自働化B · Jidoka | 0 | 0 | 0 | 0s |
白干工时的口径：每个缺陷件已经让 2、3 工位各加工 8 秒，共浪费 16 秒。自働化把质量控制放回缺陷发生点，4 秒修正后该件仍成为好件。Wasted downstream time: every defective part consumes 8 seconds at both stations 2 and 3, for 16 seconds of work that cannot be recovered. Jidoka puts control at the source; after a 4-second correction, the same part continues as good output.
异常出现就停、就叫、就处理；不制造、不传递、不接受缺陷。Stop, call, and correct at the abnormality. Do not make, pass, or accept defects.