3D-EDM: Tsarin Gano Matsalolin Fura da Suke Fuskanta a Na'urar Bugawa 3D - Binciken Fasaha

1. Gabatarwa

Yawaitar na'urorin bugawa 3D na FDM (Fused Deposition Modeling) masu araha sun ƙara samun dama amma sun haifar da ƙalubale masu mahimmanci na amfani, musamman a cikin daidaitawa da sarrafa matsala. Na'urorin bugawa 3D na FDM, tare da hadadden tsarin injiniyoyinsu da suka haɗa da motoci masu matakai da yawa, dogo, bel, da bututu, suna da saurin kamuwa da matsala kamar motsi na Layer, zare, karkace, da rashin fitar da filastik da ya isa. Waɗannan matsala sau da yawa ba a lura da su ba har sai an gama aikin bugawa, wanda ke haifar da ɓata lokaci da kayan aiki. Wannan takarda ta gabatar da 3D-EDM (Tsarin Gano Matsala da wuri na Na'urar Bugawa 3D), tsarin CNN (Convolutional Neural Network) mai sauƙi wanda aka ƙera don gano matsala da wuri ta amfani da bayanan hoto masu sauƙin tattarawa, da nufin sanya bugawa 3D ta zama mafi sauƙin samu da aminci ga masu amfani na gaba ɗaya.

2. Gano Matsala a Na'urar Bugawa 3D

Binciken da ya gabata ya binciko hanyoyi daban-daban don gano matsala a na'urorin bugawa 3D, waɗanda galibi sun faɗo cikin rukuni biyu.

2.1 Hanyoyin Tushen Na'urar Auna (Sensor)

Hanyoyi kamar waɗanda Banadaki [1] ya gabatar suna amfani da bayanan cikin na'urar (saurin fitarwa, zafin jiki). Wasu, kamar aikin Bing [2], suna amfani da ƙarin na'urori na auna na waje (misali, na'urori na girgiza) tare da masu rarraba kamar SVM (Support Vector Machines) don gano cikin gaggawa. Ko da yake suna da tasiri, waɗannan hanyoyin suna ƙara farashi da sarƙaƙiyar tsarin, suna iyakance amfani mai amfani ga masu sha'awar.

2.2 Hanyoyin Tushen Hoto

Wannan rukuni yana amfani da bayanan gani. Delli da sauransu [3] sun kwatanta ƙimar RGB a wuraren duba da aka ƙayyade. Kadam da sauransu [4] sun mai da hankali kan binciken Layer na farko ta amfani da tsararrun da aka riga aka horar (EfficientNet, ResNet). Jin [5] ya haɗa kyamara kusa da bututu don gano gefuna cikin gaggawa. Waɗannan hanyoyin suna nuna yuwuwar duban gani amma sau da yawa suna buƙatar takamaiman sanya kyamara ko kwatance masu sarƙaƙi.

3. Tsarin 3D-EDM da Ake Shawarar

Babban gudunmawar wannan aikin shine 3D-EDM, tsarin da aka ƙera don shawo kan iyakokin aikin da ya gabata ta kasancewa mai sauƙi kuma ya dogara da bayanan hoto masu sauƙin tattarawa, mai yiwuwa daga kyamarar yanar gizo ta al'ada da ke lura da gadon bugawa, ba tare da buƙatar haɗa na'urar auna ta musamman ba.

3.1 Tsarin Tsari & Cikakkun Bayanai na Fasaha

Duk da cewa PDF ba ta ba da cikakkun bayanai game da ainihin tsarin CNN ba, an bayyana tsarin a matsayin CNN mai sauƙi don rarraba hotuna. Hanyar al'ada don irin wannan aikin ta ƙunshi jerin Layer na haɗaɗɗun hoto (convolutional), Layer na tattarawa (pooling), da Layer masu cikakken haɗin kai (fully connected). Tsarin mai yiwuwa yana sarrafa hotunan shigar (misali, pixel 224x224) na bugu da ke ci gaba. Aikin haɗaɗɗun hoto (convolutional operation) ana iya wakilta shi kamar haka:

$(S * K)(i, j) = \sum_m \sum_n S(i-m, j-n) K(m, n)$

Inda $S$ shine hoton shigar (taswirar siffa) kuma $K$ shine kernel (tace). An horar da tsarin don rage aikin asara kamar Categorical Cross-Entropy don rarrabuwa nau'i-nau'i (multi-class classification):

$L = -\sum_{c=1}^{M} y_{o,c} \log(p_{o,c})$

inda $M$ shine adadin nau'ikan matsala, $y$ shine mai nuna alamar binary don aji $c$, kuma $p$ shine yiwuwar da aka annabta.

3.2 Sakamakon Gwaji

Tsarin da aka gabatar ya sami inganci na 96.72% don rarrabuwa biyu (matsala vs. babu matsala) da inganci na 93.38% don rarrabuwa nau'i-nau'i (gano takamaiman nau'in matsala). Wannan aikin yana da mahimmanci, yana nuna cewa tsarin gani mai sauƙi zai iya gano matsala na injiniyoyi masu sarƙaƙi cikin aminci. Sakamakon ya nuna tsarin ya koyi siffofin gani masu banbancewa da ke da alaƙa da kowane nau'in gazawa daga cikin bayanan hoto.

Bayanin Chati: Chati mai tsayi na hasashe zai nuna "Ingancin Tsarin" akan axis-y (0-100%) da "Nau'in Aiki" akan axis-x tare da sanduna biyu: "Rarrabuwa Biyu (96.72%)" da "Rarrabuwa Nau'i-nau'i (93.38%)". Zanen layi na iya nuna ingancin tabbatar da tsarin yana haɗuwa cikin sauri a kan lokutan horo, yana nuna koyo mai inganci.

4. Bincike & Fassarar Kwararru

5. Misalin Tsarin Bincike

Harka: Kimanta Lokacin Gano Matsalar "Karkace"
Manufa: Ƙayyade ko 3D-EDM zai iya gano karkace kafin ya haifar da gazawar bugu.
Tsari:

1. Rarraba Bayanai: Don aikin bugu da aka san yana karkace, ciro firam ɗin hoto a tazara na yau da kullun (misali, kowane Layer 5).
2. Ƙididdigar Tsarin: Gudanar da 3D-EDM akan kowane firam don samun maki yiwuwar matsala don "karkace."
3. Daidaituwar Gaskiya ta Ƙasa: Yi lakabin firam ɗin da karkace ya fara bayyana a fili ga ƙwararren ɗan adam.
4. Lissafin Ma'auni: Lissafa "Lokacin Jagorancin Gano da wuri" = (Lambar Layer na gano tsarin) - (Lambar Layer na gano ɗan adam). Ƙimar mara kyau tana nuna tsarin ya gano shi da wuri.
5. Binciken Kofa (Threshold): Zana makin amincewar tsarin akan lokaci. Gano kofar amincewar da ke haifar da "gargadi da wuri" yayin rage ƙididdige kuskure.

Wannan tsari ya wuce inganci mai sauƙi kuma yana kimanta amfanin tsarin a aikace don hana ɓarna.

6. Aikace-aikace na Gaba & Jagorori

• Haɗaɗɗun Tsarin OEM: Na'urorin bugawa 3D na masu amfani na gaba za su iya samun wannan tsarin da aka riga aka shigar akan microcontroller na cikin na'urar, suna ba da "Kula da Lafiyar Bugawa" a matsayin fasali na al'ada.
• Koyo na Tarayya don Keɓancewa: Na'urorin bugawa na masu amfani za su iya daidaita ainihin tsarin 3D-EDM a cikin gida akan halayen takamaiman na'urar bugawarsu da yanayin muhalli, suna inganta inganci na sirri ba tare da raba bayanan sirri ba, bin tsarin kamar na Google (Konečný da sauransu, 2016).
• Gudanar da Lafiya na Annabta: Tsawaita daga gano zuwa annabta. Ta hanyar nazarin yanayin makin amincewa don ƙananan lahani, tsarin zai iya annabta manyan gazawar da ke gabatowa (misali, annabta toshewar bututu daga ƙananan tsarin rashin fitar da filastik da ya isa).
• Koyo ta Hanyoyi Daban-daban (Cross-Modal): Duk da guje wa ƙarin na'urori na auna saboda farashi, aikin nan gaba zai iya bincika amfani da umarnin G-code na na'urar bugawa da na'urar sadarwa na al'ada a matsayin siginar kulawa mai rauni don inganta ƙarfin tsarin gani, wani nau'i na koyo mai sarrafa kansa.
• Gyara Taimakon AR: Haɗa gano tare da Gaskiyar Haɗaɗɗe (Augmented Reality). Ta amfani da wayar hannu/tabarar AR, tsarin ba zai iya gano matsala kamar zare kawai ba amma zai iya rufe kibau na gani ko umarni akan na'urar bugawa ta zahiri yana nuna wa mai amfani wace maɓallin daidaitawa zai juya.

7. Nassoshi

1. Banadaki, Y. da sauransu. (Shekara). Gano matsala a cikin ƙirar ƙari. Jarida Mai Dacewa.
2. Bing, X. da sauransu. (Shekara). Gano matsala cikin gaggawa don na'urorin bugawa 3D ta amfani da SVM. Tarihin Taro.
3. Delli, U. da sauransu. (Shekara). Kula da tsari don ƙirar ƙari ta hanyar fitarwa. Jaridar Hanyoyin Kera.
4. Kadam, V. da sauransu. (Shekara). Binciken Layer na Farko don bugawa 3D. IEEE Access.
5. Jin, Z. da sauransu. (Shekara). Gano gani cikin gaggawa don bugawa 3D. Kera da Haɗaɗɗun Kwamfuta.
6. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018). MobileNetV2: Rago Mai Juya da Makullin Layi. Tarihin Taron IEEE/CVF akan Hankali na Kwamfuta da Ganin Tsari (CVPR).
7. Konečný, J., McMahan, H. B., Yu, F. X., Richtárik, P., Suresh, A. T., & Bacon, D. (2016). Koyo na Tarayya: Dabarun Inganta Ingantacciyar Sadarwa. arXiv preprint arXiv:1610.05492.
8. Isola, P., Zhu, J. Y., Zhou, T., & Efros, A. A. (2017). Fassarar Hotuna zuwa Hotuna tare da Cibiyoyin Gaba da Gaba. Tarihin Taron IEEE/CVF akan Hankali na Kwamfuta da Ganin Tsari (CVPR). (An ambata don mahallin dabarun nazarin hoto na ci gaba).