Tsarin Hasashen Tushen Bayanai don Matsi na Kayan Aiki a cikin Tsarin Sake Kunnawa na SMT

2.1 Tattara Bayanai & Injiniyan Fasali

An tattara bayanan gwaji don kafa alaƙa tsakanin daidaitawa da kansu da manyan abubuwan da ke tasiri. An ƙera saitin fasali da hankali don haɗawa da:

• Jometar Kayan Aiki: Ma'auni (tsayi, faɗi, tsawo).
• Jometar Pad: Girman pad, siffa, da tazara.
• Ma'auni na Tsari: Ƙarar manna guntun lantarki, matsawar sanyawa (rashin daidaito na farko).
• Maɓuɓɓukan Manufa: Matsawar ƙarshe a X ($\Delta x$), Y ($\Delta y$), da juyawa ($\Delta \theta$).

Wannan hanyar da ke dogara da bayanai ta wuce hanyoyin gargajiya masu nauyin kwaikwayo, kamar yadda aka lura a cikin bitar hakar bayanai a cikin masana'antar lantarki kamar na Lv et al., wanda ya nuna ƙarancin irin waɗannan binciken da ake amfani da su.

2.2 Tsarin Injin Koyo

An aiwatar da tsarin ƙididdiga masu ƙarfi guda uku kuma an daidaita su don yin hasashe:

• Support Vector Regression (SVR): Yana da tasiri a cikin sararin samaniya mai girma, yana neman daidaita kuskuren a cikin kofa $\epsilon$.
• Neural Network (NN): Na'urar fahimta mai yawan layuka da aka ƙera don ɗaukar hadaddun alaƙa, waɗanda ba su da layi tsakanin fasalin shigarwa da motsin kayan aiki.
• Random Forest Regression (RFR): Hanyar haɗakar da hasashe daga bishiyoyin yanke shawara da yawa, wanda aka sani da daidaitonsa da juriya ga yin wuce gona da iri.

3.1 Ma'aunin Daidaiton Hasashe

Tsarin Random Forest Regression ya nuna aiki mafi girma a duk ma'auni:

• Dacewar Tsarin (R²): ~99% don matsawar jujjuyawa (X, Y), 96% don matsawar juyawa.
• Matsakaicin Kuskuren Cikakke (MAE): 13.47 µm (X), 12.02 µm (Y), digiri 1.52 (Juyawa).

Waɗannan kurakurai sun fi ƙanƙanta sosai fiye da ma'auni na yau da kullun na kayan aiki da pad (misali, fakiti 0402 suna kusan 1000x500 µm), yana nuna babban dacewa na zahiri.

3.2 Kwatanta Aikin Tsarin

RFR ya ci gaba da fiye da SVR da NN. Wannan ya yi daidai da sanannun ƙarfin hanyoyin haɗakarwa don bayanan tebur tare da hadaddun hulɗa, kamar yadda aka haskaka a cikin wallafe-wallafen ML na asali (misali, Breiman, 2001). Ƙarancin yuwuwar NN na iya samo asali daga ƙaramin girman bayanan da aka tattara da aka saba da su a cikin gwaje-gwajen zahiri, inda ƙarfin RFR ke haskakawa.

4.1 Babban Fahimta & Tsarin Ma'ana

Babban Fahimta: "Akwatin baƙar fata" na samuwar haɗin guntun lantarki yayin sake kunnawa ba tsari ne na hargitsi ba amma tsari ne mai ƙayyadaddun ƙa'ida, wanda ke motsa shi ta hanyar kimiyyar lissafi wanda za a iya sake ginawa tare da isassun bayanai. Wannan binciken ya tabbatar da cewa hadaddun motsin ruwa da ƙarfin tashin saman, waɗanda aka saba yin samfurinsu tare da ƙididdiga masu tsada na CFD, za a iya ɗauke su da babban aminci ta hanyar koyo na haɗakar bishiyoyi. Tsarin ma'ana yana da sauƙi mai kyau: auna sakamako (matsi), a rubuta yanayin farko (fasali), kuma a bar tsarin ya koyi ɓoyayyen aikin $f$ kamar yadda $[\Delta x, \Delta y, \Delta \theta] = f(\text{jometar, manna, matsawa...})$. Wannan yana ƙetare buƙatar warware daidai daidaitattun lissafin Navier-Stokes ga kowane haɗin kayan aiki-pad.

4.2 Ƙarfafawa & Kurakurai Masu Muhimmanci

Ƙarfafawa: Hanyar zahiri, ta fara da bayanai, ita ce babban kadarta. Cimma daidaiton hasashe na matakin micron tare da RFR yana ba da ƙimar nan take don inganta tsari. Zaɓin RFR ya kasance mai hikima, saboda yana ɗaukar rashin layi da hulɗar fasali da kyau ba tare da buƙatar manyan bayanan da ake buƙata don koyo mai zurfi ba.

Kurakurai Masu Muhimmanci: Ƙafar Achilles na binciken shine rashin yuwuwar yaduwa. Kusan tabbas an horar da tsarin akan takamaiman saitin kayan aiki (mai yiwuwa guntu masu wucewa), manna guntun lantarki, da kammalawa na pad. Shin zai yi hasashe daidai ga fakiti QFN ko tare da flux mara tsaftacewa da na ruwa? Kamar yawancin tsarin ML, yana da haɗarin zama "tagwayen dijital" na takamaiman saitin dakin gwaje-gwaje. Bugu da ƙari, yayin da aka warware hasashe, sanadin bai warware ba. Tsarin bai bayyana dalilin motsin kayan aiki ba, yana iyakance amfaninsa don ƙirar ƙira ta asali. Kayan aiki ne mai kyau na alaƙa amma ba na sanadi ba.

4.3 Fahimta Mai Amfani ga Masana'antu

1. Aiwatar Yanzu: Masu samar da EMS da OEMs masu layukan SMT masu yawan haɗawa da yawa yakamata su gwada wannan hanyar. Fara ta hanyar gina bayanai daga tsarin ku—ROI daga rage lahani na ginin kabari da gada kadai ya ba da hujjar ƙoƙarin.
2. Inganta Sanyawa: Haɗa tsarin hasashe cikin software na injin Zaɓi & Sanya. Maimakon nufin tsakiyar pad na yau da kullun, yakamata injin ya nufi wurin "da aka riga aka rama" $P_{comp} = P_{nominal} - \text{matsawar da aka hasashe}$, yana amfani da tsarin sake kunnawa a matsayin matakin ƙarshe na daidaitawa ta atomatik.
3. Gada Gibin Kimiyyar Lissafi-ML: Gaba gaba shine Hybrid AI. Yi amfani da ƙayyadaddun tsarin tushen kimiyyar lissafi (misali, lissafin lokutan tashin saman) don samar da bayanan horo na roba ko a matsayin fasali da kansa, sannan a inganta tare da bayanan zahiri. Wannan, kamar yadda tsarin jijiyoyi masu sanin kimiyyar lissafi (PINNs) ke aiki, zai magance kurakuran yaduwa.

4.4 Misalin Tsarin Nazari (Babu Lamba)

Yanayi: Injiniyan tsari yana buƙatar rage lahani don sabon haɗin capacitor 0201. Aiwatar Tsarin: 1. Layer na Bayanai: Ga allunan 50, da gangan canza matsawar sanyawa a cikin iyakar da aka sarrafa (misali, ±50 µm). Rubuta matsawar farko X, Y, $\theta$, ma'aunin pad, da girman buɗaɗɗen stencil. 2. Layer na Ma'auni: Bayan sake kunnawa, yi amfani da Binciken Gani ta Atomatik (AOI) ko na'urar gani mai daidaito don auna ƙarshen $\Delta x, \Delta y, \Delta \theta$. 3. Layer na Samfurin: Shigar da bayanan da aka tattara cikin tsarin RFR (ta amfani da ɗakunan karatu kamar scikit-learn). Horar da tsarin don hasashen matsawa. 4. Layer na Aiki: Tsarin yana fitar da taswirar ramawa. Ciyar da wannan cikin injin P&P don amfani da sanyawar da aka riga aka rama ga allunan 500 na gaba. 5. Tabbatarwa: Lura da ƙimar lahani (ginin kabari, matsawa) daga rukunin na gaba don ƙididdige ingantawa.