Failed the lightning impulse test twice, passed after an insulation oil change
衝擊電壓試驗兩次擊穿，換油後一次就過
A high-voltage shunt reactor broke down twice at full lightning impulse voltage. The windings were fine, the design margin was fine, and the particle blamed for it was never found.
一台高壓並聯電抗器在全電壓雷擊衝擊電壓試驗連續兩次擊穿。線圈沒事，設計裕度也夠，被認定是元凶的那塊異物卻始終沒人找到。
A high-voltage shunt reactor failed its lightning impulse test at full voltage twice on the same phase, and when it was opened up there was nothing to find except a small spot where paint had come off an earthed wall.
It was a three-phase oil-immersed unit at factory acceptance, with the owner's side witnessing the dielectric tests. Two phases passed without comment. The third did not.
What happened on the test floor
After the reduced-level shot, the first full-level impulse on the third phase collapsed. The voltage trace reached its peak, started down the tail as it should, and early on the tail dropped to zero. There was an abnormal sound from the oil-filled chamber that houses that phase's bushing connection.
The test team then went back to a reduced level and stepped up again. Every step below full voltage held. At full voltage the wave collapsed again.
What the investigation ruled out
- The windings. Oil sampled from the top of the main tank after the test showed only traces of hydrogen and methane and no acetylene. An arc in oil leaves gas behind, so the main body was judged sound and attention moved to the chamber.
- The hardware. The chamber was drained and the lead connection taken apart. Bolts, nuts and lead carried no marks. The bushing shield was smooth, with no protrusion.
- The design. The field calculation between the shield and the chamber wall showed a comfortable margin at full test voltage.
What was left was the mark itself: one small spot on the inside wall of the chamber, at earth potential, where the paint had peeled.
The cause that fits, and why it is still a presumption
The manufacturer's conclusion was a floating particle. Something left inside the chamber drifted in the oil into the gap between the shield and the wall and gave the impulse a bridge. The discharge ended on the earthed wall, which is why the wall carries a mark and the live parts do not.
That reading fits everything that was seen. It is also consistent with every reduced level holding and full voltage failing twice: a contaminated gap has less margin, not none.
But nobody found the particle. The chamber was cleaned, refilled with filtered oil and retested, and the retest passed. The diagnosis rests on elimination plus that pass. When the particle is never recovered there may be no better evidence available, and the report should say "presumed", which this one did, instead of "confirmed".
What the close-out leaves open
The corrective action was cleaning, refilling and a successful retest. The preventive action for the next unit was to fill the chamber with oil, drain it completely, inspect inside once more and then refill. Both are reasonable. Three points are still worth raising.
- A flush removes particles. It does not say where they came from. The report does not identify what the particle was or which assembly step left it there, so the source is still open.
- The paint repair comes after the last test. The spot on the chamber wall was to be touched up once the acceptance test was finished. That means working inside the chamber again after the final impulse has been applied, so whatever cleanliness control applied before the test has to apply again after it.
- Nobody runs a full impulse test on site. If the chamber is opened, drained or refilled again for transport or installation, the site procedure needs the same care, because there is no impulse test at site to catch what the factory test caught here.
- Was the breakdown location physically found and opened, or only inferred from the traces?
- Where was the oil sample taken, and does that compartment share oil with the one that failed?
- Was the full impulse sequence repeated on the failed phase after the repair, and compared with the earlier records?
- Does the preventive action deal with the source of the contamination, or only with removing it?
- Will the compartment be opened again after the last dielectric test, at the factory or at site, and under what cleanliness procedure?
A retest that passes tells you the unit is clean today. Only the procedure tells you the next unit will be, and that this one still is after it has been opened again.
一台高壓並聯電抗器的雷擊衝擊電壓試驗，同一相在全電壓連續失敗兩次。拆開來看，除了接地的壁面上有一小塊掉漆，什麼都找不到。
那是一台三相油浸式設備，正在做出廠驗收的絕緣試驗，業主方在場見證。其中兩相順利通過，第三相沒有。
試驗現場發生了什麼
降低電壓那一發打完後，第三相第一發全電壓衝擊就崩潰了。電壓波形到達峰值、照正常樣子開始走波尾，才走到波尾初段就突然掉到零，同時這一相套管接線所在的油室傳出異常聲響。
試驗團隊接著退回較低的電壓重新逐級往上加，全電壓以下每一級都撐住了，一到全電壓波形再次崩潰。
調查排除了哪些可能
- 線圈本體。試驗後從主油箱上部取油做油中氣體分析，只有微量的氫和甲烷，沒有乙炔。油裡發生電弧一定會留下氣體，所以本體判定正常，目標轉向油室。
- 零件。把油室的油洩掉、拆開引線接頭逐一檢查，螺栓、螺帽和引線都沒有痕跡，套管屏蔽罩表面平滑、沒有突起。
- 設計。屏蔽罩到油室壁面之間的電場計算，在全試驗電壓下仍有充足的裕度。
剩下的只有那個痕跡：油室內壁、接地電位的位置上，一小塊油漆剝落。
說得通的原因，為什麼仍然只是推定
製造廠的結論是漂浮異物。油室裡殘留的某個東西在絕緣油中漂進屏蔽罩和壁面之間的間隙，替衝擊電壓搭了一座橋。放電的終點在接地的壁面，所以痕跡留在壁面上，帶電的零件反而乾乾淨淨。
這個解釋跟看到的現象全部對得上，也符合較低電壓都撐得住、全電壓連兩次失敗的結果：被污染的間隙是裕度變小，不是完全沒有裕度。
但那塊異物沒有人找到。油室清潔後重新注入過濾的絕緣油再試一次，通過了。整個診斷靠的是排除法加上這一次通過。異物找不回來的時候，可能也拿不出更好的證據，所以報告應該寫「推定」，這份報告也確實這樣寫，而不是「已確認」。
結案內容還沒回答的事
矯正措施是清潔、重新注油、重測通過。下一台的預防措施是先把油室注滿油、完全洩掉、再檢查一次內部，然後才正式注油。兩項都合理，但有三點值得再提出來。
- 沖洗能把異物帶走，卻沒有說明它從哪裡來。報告沒有指出那塊異物是什麼、是哪一個組裝步驟留下的，來源仍然是開放的問題。
- 補漆排在最後一次試驗之後。油室壁面那一點掉漆預定在驗收試驗全部結束後才補。也就是說最後一發衝擊打完之後，還要再進油室作業一次，試驗前做過的清潔管制，試驗後得再做一遍。
- 現場沒有人會做全電壓衝擊電壓試驗。如果油室在運輸或安裝時還會再打開、洩油或重新注油，現場程序就要有同等的管制，因為現場沒有衝擊電壓試驗可以把這次工廠試驗抓到的問題再抓一次。
- 擊穿位置是實際找到並拆開確認，還是只從波形推斷？
- 油樣取自哪裡？那個油室跟出問題的油室是不是共用同一池油？
- 修復後失敗的那一相有沒有重做完整的衝擊電壓試驗程序，並跟先前的紀錄比對？
- 預防措施處理的是污染的來源，還是只有把它清掉？
- 最後一次絕緣試驗之後，這個油室在工廠或現場還會不會再打開？用什麼清潔程序管制？
重測通過，只能說明這一台今天是乾淨的。至於下一台，或是這一台再打開之後還能不能保持乾淨，靠的是程序。