FMEA — failure mode, effect, severity, detection, RPN: Visual walkthrough

3.6.29 · D2 · Physics › Spacecraft Structures & Systems Engineering › FMEA — failure mode, effect, severity, detection, RPN

Hum ek sawaal ka jawab dene wale hain: ek aisi component ke liye jo shayad toot jaaye, hum apni chinta ko ek akele measurable size mein kaise badlein jise hum baaki har chinta ke saath rank kar sakein?

Step 1 — Ek failure ek event hai, aur events ke teen independent handles hote hain

KYA. Kisi bhi number se pehle, ek akeli component ki picture karo — maano ek battery cell. Woh wahaan baith ke apna kaam kar rahi hai. Phir, kisi ek moment par, woh fail ho jaati hai. Woh failure ek event hai: ek aisi cheez jo ya toh hoti hai ya nahi hoti.

KYUN. Hum "chinta" ko directly rank nahi kar sakte — chinta ek feeling hai. Rank karne ke liye humein ek size chahiye. Toh pehle humein poochna hoga: ek failure ko doosre se zyada scary kya banaata hai? Ek cheez nahi. Teen alag-alag cheezein, aur woh sach mein independent hain — ek ko jaanne se doosre ke baare mein kuch pata nahi chalta:

Agar yeh hota hai toh kitna bura hai? (yeh Severity hai, likha jaata hai $S$ )
Yeh kitni baar hoga? (yeh Occurrence hai, likha jaata hai $O$ )
Hum iske hone ke baare mein kitne andhey hain? (yeh Detection hai, likha jaata hai $D$ )

PICTURE. Figure dekho: beech mein ek failure event, teen arrows usse teen bilkul alag directions mein kheench rahe hain. Ek battery cell short severe hai (upar wala arrow lamba hai), rare hai (daayein wala arrow chota hai), aur visible hai (neeche wala arrow chota hai). Yeh teeno ek box ki length, width, aur height ki tarah hain — tum ek ko doosron se nahi nikaal sakte.

Figure — FMEA — failure mode, effect, severity, detection, RPN

Step 2 — Har handle ko 1-to-10 ruler par ek number mein badlo

KYA. Har handle ek feeling hai ("kaafi bura", "kaafi rare"). Hum ek fixed, agreed-upon ruler use karke uس par $1$ se $10$ tak ka number stamp karte hain. Random guess nahi — ek table jo team ka har member same tarike se padhta hai.

KYUN 1 to 10, aur probability kyun nahi? Ek raw probability jaise $0.0003$ padhne mein mushkil hai aur "severity" ke saath mix nahi hoti (jo ki ek probability bilkul bhi nahi hai — tum "loss of crew" par probability nahi laga sakte). Toh hum sab kuch same $1$ – $10$ ruler par map karte hain. Ab teen alag-alag cheezein ek hi bhasha bolti hain aur combine ho sakti hain. Scale jaanbooth kar coarse hai — ek insaan honestly $7$ aur $9$ mein fark bata sakta hai, lekin $71$ aur $73$ mein nahi.

PICTURE. Teen vertical rulers side by side, har ek par neeche $1$ aur upar $10$ likha hai. Har ek par ek red dot battery-cell short ko mark karta hai: $S = 9$ kaafi upar (near-catastrophe), $O = 4$ kaafi neeche (documented par rare), $D = 3$ neeche (sensors jaldi pakad lete hain). Dots padhlo — wahi failure hai, ab numbers mein.

Step 3 — Hum multiply kyun karte hain (aur kabhi add kyun nahi)

KYA. Ab hum $S$ , $O$ , $D$ ko ek number mein combine karte hain. Parent ne kaha tha "multiply karo". Yeh raha reason, drawn.

Add kyun nahi? Maano humne add kiya: $S + O + D$ . Tab $S = 10, O = 1, D = 1$ wali failure (catastrophic lekin essentially impossible aur instantly caught) $12$ score karti hai. $S = 4, O = 4, D = 4$ wali failure (moderate, common, half-visible) bhi $12$ score karti hai. Addition kehta hai woh equally worrying hain — lekin woh nahi hain. Catastrophic wali rare aur visible hone se defuse ho jaati hai. Addition ek bade term ko doosron ko chhupa dene deta hai.

Multiplication iska ulta karta hai: yeh har factor ko ek gate banaata hai. Agar koi bhi factor $1$ ke kareeb hai, toh woh poore product ko shrink kar deta hai. Ek failure tabhi sach mein khatarnaak hoti hai jab woh buri aur likely aur blind ho — teeno ek saath. Woh "teeno ek saath" exactly wahi hai jo multiplication encode karta hai.

PICTURE. Ek rectangle socho: width $S$ , height $O$ . Uska area $S \times O$ hai. Ab us rectangle ko depth $D$ se page ke andar push karo — tum ek box paate ho, aur uska volume $S \times O \times D$ hai. Poori chinta us box ka volume hai. Figure dekho: kisi bhi ek edge ko $1$ ki taraf shrink karo aur box ek sliver mein collapse ho jaata hai — chota volume, choti chinta. Wahi multiplication apna kaam kar raha hai.

Step 4 — Detection ulta score kyun hota hai

KYA. $D = 1$ matlab hum failure ko almost perfectly detect karte hain; $D = 10$ matlab woh disaster tak invisible hai. Yeh ulta lagta hai. Chalte hain isse earn karte hain.

KYUN. RPN danger measure karne ke liye hai, aur danger har factor ke saath badhta hai. Severity danger badhati hai (obvious). Occurrence danger badhata hai (obvious). Detection mein danger kya badhata hai? Acchi detection nahi — buri detection. Ek aisi failure jise tum aate nahi dekh sakte woh scary wali hai. Toh " $D$ " ruler blindness measure karna chahiye, sight nahi, taaki bada $D$ volume ko upar push kare, same direction mein jaise bade $S$ aur bade $O$ .

PICTURE. Ek hi battery failure ke do identical versions side by side. Left: sensors dekh rahe hain (good detection) — hum ise chhota $D = 2$ label karte hain, aur risk box depth mein chota hai. Right: koi sensor nahi, silent leak — bada $D = 9$ , aur same failure ab ek deep, large-volume box hai. Same $S$ , same $O$ , lekin blind wali objectively zyada khatarnaak hai, aur sirf " $D$ = blindness" hi volume ko woh reflect karwa sakta hai.

Step 5 — Volume padhna: thresholds aur danger ceiling

KYA. Ek number nikla. Ab hum ise action bands mein sort karte hain. Box ka volume sabse chhote possible se sabse bade possible tak range karta hai.

Bands kyun? $108$ jaisa raw volume akele kuch nahi kehta. Humein chahiye "kitna bada bahut bada hai?" Team mission history se cut-lines fix karti hai: $\sim 50$ se neeche bas watch karo; $\sim 150$ tak review aur test karo; $\sim 500$ tak redesign karo ya backup add karo; $500$ se upar tum fly nahi kar sakte.

PICTURE. $1$ se $1000$ tak ek akela number line, coloured bands mein split, jisme battery-short RPN $108$ ek red tick ke roop mein "test-then-decide" band mein mark hai. Tum literally dekh sakte ho woh danger ceiling se kitna door hai.

Step 6 — Corners: sabse chhota aur sabse bada possible box

KYA. Har ruler $1$ se $10$ chalta hai, toh box $1 \times 1 \times 1$ se chhota aur $10 \times 10 \times 10$ se bada kabhi nahi ho sakta. Yeh poori space ke do extreme corners hain.

Edges kyun dikhayein? Taaki tum koi aisa number na milao jise tum place na kar sako. Degenerate best case: ek harmless, impossible, obvious failure → $1 \times 1 \times 1 = 1$ . Degenerate worst case: ek catastrophic, near-certain, invisible failure → $10 \times 10 \times 10 = 1000$ . Har real failure is box-of-boxes ke andar rehti hai.

PICTURE. Poora cube chhote corner $(1, 1, 1)$ se door corner $(10, 10, 10)$ tak, jisme battery short ek red dot ke roop mein andar float kar raha hai — safe corner ke kareeb, deadly corner se door.

Step 7 — Box shrink karna: mitigation geometrically kya karta hai

KYA. Ek failure ko fix karna matlab hai uski box ki ek ya zyada edges shrink karna, volume ko tab tak kam karna jab tak woh kisi threshold se neeche na aa jaaye.

KYUN. Tum aksar $O$ nahi badal sakte (physics hai physics — cell phir bhi short ho sakta hai). Lekin tum ek fuse laga sakte ho damage cut karne ke liye ( $S$ shrink karo), aur ek redundant monitor add kar sakte ho usse jaldi pakadne ke liye ( $D$ shrink karo). Har design change ek edge ko andar kheenchta hai aur volume girta hai.

PICTURE. Left par tall battery box ( $9 \times 4 \times 3 = 108$ ) jisme do red arrows $S$ edge ko $9 \to 6$ aur $D$ edge ko $3 \to 2$ tak squeeze kar rahe hain, right par chhota box produce karte hue ( $6 \times 3 \times 2 = 36$ ). Same failure — bahut chhota box.

Ek-picture summary

Upar sab kuch, compressed: ek failure event → teen independent rulers → ek risk box jiska volume RPN hai → danger ceiling ke against sort kiya → mitigation se shrink kiya.

Recall Feynman retelling — ise apne plain words mein wapas bolo

Socho ek part jo shayad toot jaaye. Teen bilkul alag sawaal decide karte hain ki mujhe iske baare mein kitni neend khooni chahiye: agar yeh toota toh kitna bada mess hoga, yeh tootne ki kitni sambhavna hai, aur main iske tootne ke baare mein kitna andhaa hoon. Main har ek ko ek fixed ruler par ek se das tak score karta hoon jis par team mein sab agree karte hain — aur main "blindness" score karta hoon, "sight" nahi, taaki usse na dekh paana number ko bada banaye, kyunki woh genuinely zyada khatarnaak hai. Phir main teeno numbers ko add nahi karta, kyunki adding ek bade number ko do chhotey walon ko chhupa dene aur yeh pretend karne deta hai ki sab theek hai. Iske bajaaye main unhe multiply karta hoon, jaise width, height, aur depth ko multiply karke ek box ka volume milta hai. Ek failure tabhi sach mein scary hoti hai jab box teeno directions mein bada ho — aur multiplication guarantee karta hai ki agar koi bhi ek direction chhhoti hai, toh poora box chhota hai. Box ka volume Risk Priority Number hai, hamesha ek aur ek hazaar ke beech. Main apni saari failures ke boxes line up karta hoon, sabse bade walon ko pehle fix karta hoon unki edges ko fuses, backups, aur behtar sensors se squeeze karke, aur phir se measure karta hoon. Yahi FMEA hai.

Recall Quick checks

Add karne ki jagah multiply kyun karte hain? ::: Multiplication har factor ko ek gate banaata hai — $1$ ke kareeb koi bhi factor poore product ko shrink karta hai, toh failure tabhi "badi" hoti hai jab woh buri AUR likely AUR blind ho. Adding ek bade term ko chhotey walon ko mask karne deta hai. $D = 10$ , $D = 1$ se bura kyun hai? ::: $D$ blindness measure karta hai, detection quality nahi. Bada $D$ = humein kabhi aata nahi dikhta = zyada khatarnaak, toh woh RPN ko upar push kare, $S$ aur $O$ ke saath. RPN ka geometric meaning kya hai? ::: Edges $S$ , $O$ , $D$ wale box ka volume. Sabse chhota aur sabse bada possible RPN kya hai? ::: $1$ ( $1 \times 1 \times 1$ ) aur $1000$ ( $10 \times 10 \times 10$ ). Battery short mitigation se pehle/baad? ::: $9 \times 4 \times 3 = 108$ pehle; $6 \times 3 \times 2 = 36$ baad mein.

Dekho bhi: parent topic · Reliability Engineering · Risk Management in Spacecraft Design · Systems Engineering V-Model · Quality Assurance and Testing · Mission Assurance · Mars Climate Orbiter