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

1. Gabatarwa

Fasahar Sanyawa ta Surface (SMT) ita ce babbar hanyar haɗa na'urorin lantarki inda ake sanya kayan aiki kai tsaye a kan allunan da'ira da aka buga (PCBs). Wani muhimmin mataki shine tsarin gudanar da sake kunnawa, inda narkakken man gini ke nuna halayen motsin ruwa, wanda ke sa kayan aiki suyi motsi—wannan al'amari ana kiransa "daidaita kai." Duk da yake wannan na iya gyara ƙananan kurakuran sanyawa, daidaita kai mara daidai yana haifar da lahani kamar ginin kabari da gado. Wannan binciken yana magance gibi a cikin aiki, hasashen motsin da ya dogara da bayanai ta hanyar haɓaka tsarin injin koyo don hasashen matsawar kayan aiki a cikin hanyoyin x, y, da juyawa ($\theta$) tare da ingantaccen daidaito, da nufin inganta sigogin na'urar ɗauka da sanyawa.

2. Hanyoyi & Tsarin Gwaji

Binciken ya bi hanyar matakai biyu: na farko, bincika bayanan gwaji don fahimtar alaƙa tsakanin daidaita kai da abubuwa kamar siffar kayan aiki/pad; na biyu, amfani da ingantattun tsarin ML don hasashe.

2.1 Tattara Bayanai & Injiniyan Fasali

An tattara bayanan gwaji da suka haɗa da nau'ikan kayan aiki masu wucewa na SMT (misali, resistors, capacitors). Manyan fasali sun haɗa da:

• Siffar Kayan Aiki: Tsayi, faɗi, tsayi.
• Siffar Pad: Tsayin pad, faɗi, tazara.
• Sigogin Tsari: Girman man gini, ƙirar buɗaɗɗen stencil, matsawar sanyawa ta farko.
• Maƙasudai Masu Manufa: Matsawar ƙarshe a cikin X ($\Delta x$), Y ($\Delta y$), da juyawa ($\Delta \theta$).

2.2 Tsarin Injin Koyo

An aiwatar da tsarin koma baya guda uku kuma an kwatanta su:

• Support Vector Regression (SVR): Yana da tasiri a cikin sararin samaniya mai girma, ta amfani da kernel na radial basis function (RBF).
• Neural Network (NN): Multi-layer perceptron (MLP) tare da ɓoye yadudduka don ɗaukar alaƙar da ba ta layi ba.
• Random Forest Regression (RFR): Haɗin bishiyoyin yanke shawara, mai ƙarfi ga wuce gona da iri kuma yana iya jerin muhimmancin fasali.

3. Sakamako & Bincike

3.1 Kwatancen Aikin Tsari

Random Forest Regression (RFR) ya fi duka SVR da Neural Networks a cikin dukkan ayyukan hasashe guda uku (X, Y, juyawa). Ya cimma matsakaicin ma'aunin ƙaddara (R²) na 99% don matsawar matsayi da 96% don matsawar juyawa, tare da ƙananan kurakuran ma'anar cikakke (misali, ~13 µm). Wannan yana nuna babban ikon RFR na sarrafa rikitattun, alaƙar da ba ta layi ba, da yuwuwar hulɗa a cikin bayanan tsarin sake kunnawa na SMT.

3.2 Mahimman Abubuwan Hasashe

Binciken muhimmancin fasali na tsarin RFR ya bayyana:

• Matsawar Sanyawa ta Farko: Babban abu guda ɗaya mafi mahimmanci don hasashen matsawar ƙarshe.
• Siffar Pad & Tazara: Muhimmi a cikin tantance ƙarfin dawowa da matsayin daidaito.
• Girman Man Gini: Yana shafar girman ƙarfin tashin hankali na saman kai tsaye.
• Siffar Kayan Aiki: Yana shafar lokacin rashin motsi na kayan aiki da martani ga ƙarfin gini.

4. Tsarin Fasaha & Bincike

Ra'ayin mai binciken masana'antu game da ƙimar dabarun binciken da iyakoki.

4.1 Cikakken Fahimta

Wannan takarda ba kawai game da hasashen matsawar micron-level ba ne; yana da ma'ana mai mahimmanci daga simintin da ya dogara da kimiyyar lissafi zuwa ƙwaƙƙwaran bayanai a cikin ƙera daidaito. Marubutan sun gano daidai cewa tsarin ka'idodin samuwar haɗin gini, duk da kyawunsa, sau da yawa yana kasawa a cikin gaskiyar rikitarwar samarwa mai haɗawa. Ta hanyar ɗaukar tanda sake kunnawa a matsayin "akwatin baƙar fata" da amfani da RFR don taswirar shigarwa (fayilolin ƙira, bayanan sanyawa) zuwa fitarwa (matsayi na ƙarshe), suna ba da mafita mai amfani wanda ke ketare buƙatar warware rikitattun lissafi, lissafin yanayi da yawa a cikin ainihin lokaci. Wannan yayi kama da falsafar da ke bayan nasarorin AI a wasu fagage, kamar amfani da CNNs don gane hoto maimakon yin lambar na'urori na fasali a sarari.

4.2 Kwararar Ma'ana

Ma'anar binciken tana da inganci kuma tana da alaƙa da samarwa: 1) Amincewa da Matsala: Daidaita kai takobi ne mai kaifi biyu. 2) Gano Gibin: Rashin kayan aiki masu amfani, masu hasashe. 3) Amfani da Bayanan da Ake da su: Yi amfani da sakamakon gwaji azaman mai horarwa. 4) Aiwatar da Kayan Aiki na Zamani: Gwada tsarin ML da yawa. 5) Tabbatar da Gano Zakarun: RFR ya ci nasara. 6) Ba da Shawara Aikace-aikace: Ciyar da hasashe zuwa na'urorin sanyawa. Wannan yayi daidai da daidaitaccen tsarin CRISP-DM (Cross-Industry Standard Process for Data Mining), yana mai da shi zane mai maimaitawa ga sauran ƙalubalen inganta tsari a cikin haɗa na'urorin lantarki.

4.3 Ƙarfi & Kurakurai

Ƙarfi: Zaɓin RFR yana da kyau sosai—yana da fassara (ta hanyar muhimmancin fasali), yana sarrafa rashin layi da kyau, kuma yana da ƙarancin saukarwa ga wuce gona da iri akan ƙayyadaddun bayanai idan aka kwatanta da koyo mai zurfi. An ba da rahoton daidaito (~13µm kuskure) yana da ban sha'awa kuma yana iya yin aiki ga yawancin layukan SMT. Mai da hankali kan kayan aiki masu wucewa da farko shine farkon farko mai hankali, mai iya aiki.

Kurakurai & Makafin Ido: Giwa a cikin ɗaki shine iyakar bayanai da gama gari. An horar da tsarin akan takamaiman saitin kayan aiki, man gini, da kammala allon. Ta yaya yake aiki tare da sabbin nau'ikan kayan aiki da ba a gani ba (misali, manyan QFPs, BGAs) ko gami da gini mara gubar tare da kaddarorin jika daban-daban? Binciken ya nuna amma bai magance cikakken ƙalubalen ci gaba da koyo da daidaita tsari a cikin yanayin masana'anta mai ƙarfi ba. Bugu da ƙari, duk da yake ma'aunin kuskure yana da ƙasa a matsakaici, muna buƙatar ganin rarraba kuskuren—wasu ƙananan abubuwan da ba a saba gani ba na iya haifar da asarar yawan amfanin ƙasa.

4.4 Fahimta Mai Aiki

Ga injiniyoyin tsarin SMT da masu kera kayan aiki:

1. Matukin Jirgi Nan take: Maimaita wannan binciken akan layin samarwa naku don samfur mai yawan amfani. Fara tattara bayanai masu tsari akan matsawar sanyawa da ma'aunin bayan sake kunnawa (ta amfani da SPI da AOI). Gina tsarin RFR na ku na musamman.
2. Mai da hankali kan Haɗawa: Ƙimar gaske ita ce sarrafa madauki rufe. Yi aiki tare da masu siyar da na'urorin sanyawa (kamar Fuji, ASM SIPLACE) don haɓaka API wanda ke ciyar da gyaran da tsarin ya hasasce ($-\Delta x, -\Delta y, -\Delta \theta$) zuwa cikin daidaitattun sanyawa don allon na gaba.
3. Faɗaɗa Saitin Fasali: Haɗa masu canji na tsari na ainihin lokacin da takarda ta rasa: yanayin zafi na yankin tanda sake kunnawa, saurin mai ɗauka, yawan nitrogen, da danshin yanayi. Wannan yana haifar da tsarin daidaitawa na gaske.
4. Benchmark Dangane da Kimiyyar Lissafi: Kar a watsar da simintin gwaji. Yi amfani da hanyar haɗaka: bari tsarin ML ya yi saurin hasashe, kan layi, amma yi amfani da simintin gwaji na kimiyyar lissafi (misali, ta amfani da kayan aiki kamar ANSYS) a kashe layi don tabbatarwa da fahimtar yanayin gefe, ƙirƙirar zagayowar ingantawa mai kyau.

5. Bincike na Asali & Ra'ayi na Masana'antu

Wannan binciken yana wakiltar muhimmin aikace-aikacen injin koyo na lokaci-lokaci ga ƙalubalen masana'antu na daɗewa. Canji daga tsarin ka'idodin motsin ruwa zuwa hasashen da ya dogara da bayanai yana kwatanta wani babban yanayi a cikin Masana'antu 4.0, inda bayanan ƙwaƙƙwaran sau da yawa suka wuce tsarin ƙa'idodin farko a cikin rikitattun mahalli, masu hayaniya. Nasarar marubutan tare da Random Forest ba abin mamaki bane; yanayin haɗakar sa yana sa shi mai ƙarfi ga wuce gona da iri akan ƙayyadaddun bayanai—wata matsala ta gama gari a cikin masana'antu inda tattara miliyoyin samfuran da aka yiwa alama ba zai yiwu ba. Wannan ya yi daidai da binciken a wasu yankuna, kamar amfani da tsarin bishiyoyi don hasashen kulawa akan kayan aikin semiconductor, inda sau da yawa suka fi ƙwararrun hanyoyin sadarwar jijiyoyi a kan bayanai na tebur masu tsari.

Duk da haka, iyakar binciken shine babban iyakansa. An nuna tsarin akan kayan aiki masu wucewa, inda ƙarfin daidaita kai yana da ɗabi'a sosai. Gwaji na gaske zai zama kayan aiki masu aiki kamar fakitin lebur huɗu (QFPs) ko tsararrun ƙwallon ƙwallon ƙafa (BGAs), inda samuwar haɗin gini ya fi rikitarwa kuma ya ƙunshi adadi mafi girma na haɗin gwiwa masu dogaro da juna. Bugu da ƙari, tsarin ya bayyana yana tsaye. A cikin layin SMT na gaske, tsarin man gini yana canzawa, stencils suna lalacewa, da kuma bayanan tanda suna karkata. Tsarin gaske mai ƙarfi zai buƙaci ɓangaren koyo akan layi, kama da tsarin sarrafawa masu daidaitawa da ake amfani da su a cikin injinan mutum-mutumi, don ci gaba da sabunta tsarin. Bincike daga cibiyoyi kamar Cibiyar Fraunhofer don Injiniyanci da Kera Kayan Aiki IPA akan tsarin samarwa masu inganta kai suna jaddada wannan buƙatar daidaitawa.

Yiwuwar tasiri yana da girma. Ta hanyar hasashen matsawa daidai, wannan fasaha na iya ba da damar "sanyawa mai hasashe," inda kayan aiki da gangan suka sanya su ba daidai ba ta hanyar matsawar da algorithm ta ƙididdige don haka su daidaita kansu zuwa cikakkiyar matsayi. Wannan zai iya sassauta buƙatun daidaito (da farashi) na na'urorin sanyawa masu daidaito, rage buƙatar sake aikin bayan sake kunnawa, da ƙara yawan amfanin ƙasa, musamman ga ƙananan kayan aiki kamar fakitin 0201 ko 01005. Yana haɗa gibin tsakanin ƙira na dijital (bayanin CAD) da sakamakon zahiri, yana ba da gudummawa ga hangen nesa na "tagwaye na dijital" don tsarin haɗa SMT.

6. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Babban aikin hasashe shine matsalar koma baya mai yawa. Don wani ɓangaren kayan aiki $i$, tsarin yana koyon aikin taswira $f$ daga vector fasali $\mathbf{X_i}$ zuwa vector manufa $\mathbf{Y_i}$: $$\mathbf{Y_i} = f(\mathbf{X_i}) + \epsilon_i$$ inda $\mathbf{Y_i} = [\Delta x_i, \Delta y_i, \Delta \theta_i]^T$ kuma $\mathbf{X_i}$ ya haɗa da fasali kamar girman kayan aiki $(L_c, W_c)$, girman pad $(L_p, W_p, S)$, girman man gini $V_s$, da matsawar farko $(x_{0,i}, y_{0,i})$.

Algorithm na Random Forest yana aiki ta hanyar gina bishiyoyin yanke shawara da yawa yayin horo. Hasashen ƙarshe shine matsakaicin hasashen bishiyoyin ɗaiɗaiku don koma baya. Muhimmancin fasali don wani fasali $j$ sau da yawa ana ƙididdige shi azaman raguwar gaba ɗaya a cikin ƙazantar kumburi (wanda aka auna ta hanyar Matsakaicin Kuskuren Square, MSE) wanda aka matsaka akan duk bishiyoyin inda ake amfani da fasalin don raba: $$\text{Muhimmanci}(j) = \frac{1}{N_{trees}} \sum_{T} \sum_{t \in T: \text{raba akan } j} \Delta \text{MSE}_t$$ inda $\Delta \text{MSE}_t$ shine raguwar MSE a kumburi $t$.

7. Sakamakon Gwaji & Bayanin Ginshiƙi

Bayanin Ginshiƙi (Hasashen bisa rubutu): Ginshiƙi zai yi tasiri sosai don kwatanta tsarin injin koyo guda uku. X-axis zai jera ayyukan hasashe guda uku: "X-Matsi," "Y-Matsi," da "Matsawar Juyawa." Ga kowane aiki, ginshiƙi guda uku masu rukuni za su wakilci aikin SVR, Neural Network (NN), da Random Forest (RFR). Babban y-axis (hagu) zai nuna Ma'aunin Ƙaddara (R²) daga 90% zuwa 100%, tare da sandunan RFR suna kaiwa kusa da saman (99%, 99%, 96%). Na biyu y-axis (dama) zai iya nuna Matsakaicin Kuskuren Cikakke (MAE) a cikin micrometers (don X, Y) da digiri (don juyawa), tare da sandunan RFR suna zama mafi gajarta, yana nuna mafi ƙarancin kuskure (13.47 µm, 12.02 µm, 1.52°). Wannan na gani zai bayyana sosai mafi girman daidaito da daidaito na RFR a cikin dukkan ma'auni.

Mahimman Sakamako na Lamba: Tsarin Random Forest ya cimma matsakaicin kuskuren hasashe na 13.47 micrometers don matsawar gefe, wanda bai kai faɗin gashin ɗan adam ba (~70 µm), yana nuna ingantaccen daidaito na aiki don haɗa SMT.

8. Tsarin Bincike: Misalin Lamari Ba tare da Lamba ba

Yanayi: Mai ba da sabis na EMS yana fuskantar asarar amfanin ƙasa na 2% akan allon saboda ginin kabari na resistors 0402.

Aiwatar da Tsarin:

1. Tattara Bayanai: Don alluna 10,000 na gaba, yi rikodin ga kowane resistor 0402: ƙirar pad daga fayil ɗin Gerber, girman buɗaɗɗen stencil, binciken man gini (SPI) girma, daidaitattun $(x_0, y_0)$ na'urar sanyawa, da bayan sake kunnawa $(x_f, y_f, \theta_f)$ daidaitattun daga Binciken Gani ta Atomatik (AOI).
2. Horo na Tsari: Gina tsarin RFR ta amfani da wannan bayanan, tare da fasali (girman pad, girman man gini, matsawar farko) da manufa (matsawar ƙarshe).
3. Samar da Fahimta: Muhimmancin fasali na tsarin ya nuna cewa rashin daidaituwa a cikin girman man gini tsakanin pad biyu shine mafi ƙarfin mai hasashen matsawar juyawa ($\Delta \theta$) wanda ke haifar da ginin kabari, har ma fiye da kuskuren sanyawa na farko.
4. Aiki: Maimakon ƙoƙarin inganta daidaiton sanyawa (mai tsada), hankali ya koma inganta ƙirar stencil da tsarin bugawa don tabbatar da daidaiton girman man gini. Tsarin kuma na iya ba da "maki haɗari" ga kowane sanyawar kayan aiki a ainihin lokaci, yana alamar sanyawa masu haɗari don gyara nan take kafin sake kunnawa.

9. Aikace-aikacen Gaba & Hanyoyin Ci Gaba

• Sanyawa mai Daidaitawa Madauki Rufe: Haɗa tsarin hasashe kai tsaye cikin software ɗin sarrafa na'urar ɗauka da sanyawa don daidaita daidaitattun sanyawa a ainihin lokaci, ƙirƙirar layin haɗawa mai gyara kai.
• Faɗaɗa zuwa Kayan Aiki Masu Aiki: Aiwatar da tsarin don hasashen daidaitawar rikitattun kayan aiki kamar BGAs, QFNs, da masu haɗawa, inda daidaita kai ya fi takurawa amma har yanzu yana da mahimmanci.
• Haɗin Tagwaye na Dijital: Yin amfani da tsarin a matsayin babban ɓangaren tagwaye na dijital na tsarin SMT, yana ba da damar inganta tsarin kama-da-wane da gwajin yanayin "menene idan" kafin samarwa na zahiri.
• Tsarin Kimiyyar Lissafi-AI Haɗaka: Haɗa tsarin RFR mai dogaro da bayanai tare da sauƙaƙan lissafin kimiyyar lissafi (misali, don ƙarfin tashin hankali na saman) don inganta daidaiton ƙididdiga zuwa sabbin nau'ikan kayan aiki ko kayan da ba a gani ba.
• Koyo Ba tare da Harbi/Ƙananan Harbi: Haɓaka dabarun hasashen matsawa don sabbin fakitin kayan aiki tare da ƙananan sabbin bayanan horo, yin amfani da canja wurin koyo daga babban tushen tsarin kayan aiki na yanzu.

10. Nassoshi

1. Parviziomran, I., Cao, S., Srihari, K., & Won, D. (Shekara). Tsarin Hasashen Dogaro da Bayanai na Matsawar Kayan Aiki yayin Tsarin Sake Kunnawa a cikin Fasahar Sanyawa ta Surface. Sunan Jarida, Volume(Issue), shafuka. (Tushen PDF)
2. Böhme, B., et al. (2022). Tsarin inganta kai a cikin samar da lantarki. Fraunhofer IPA. [https://www.ipa.fraunhofer.de/]
3. Lv, C., et al. (2020). Cikakken bita na hakar bayanai a cikin masana'antar lantarki. Jaridar Ƙera Mai Hankali, 31(2), 239-256.
4. Breiman, L. (2001). Gandun daji bazuwar. Injin Koyo, 45(1), 5-32. (Takarda mai mahimmanci akan algorithm da aka yi amfani da shi)
5. ANSI/IPC J-STD-001. (2020). Bukatun Haɗaɗɗun Wutar Lantarki da Lantarki. IPC. (Ma'auni na masana'antu don hanyoyin SMT)