Taimakon Na'ura Mai Kwakwalwa (ML) don Gano Tsarin Kima na Ƙarfin Juzu'i (UTS) a cikin Samfuran PLA da aka Yi ta Hanyar FDM

1. Gabatarwa

Hankali na Wucin Gadi (AI) da Koyon Na'ura (ML) suna kawo sauyi a masana'antu, suna ba da damar da ba a taɓa gani ba don inganta tsari da hasashen bincike. A fagen Ƙara Kera (AM), musamman Tsarin Zubarwa da aka Haɗa (FDM), hasashen kaddarorin injiniya kamar Ƙarfin Juzu'i na Ƙarshe (UTS) yana da mahimmanci don tabbatar da amincin sassa da faɗaɗa aikace-aikacen masana'antu. Wannan binciken ya fara aiwatar da algorithms na rarrabuwa masu kulawa—Rarrabawar Logistic, Haɓaka Gradient, Bishiyar Shawara, da K-Nearest Neighbor (KNN)—don kimanta UTS na samfuran Polylactic Acid (PLA). Ta hanyar danganta mahimman sigogi na tsari (Kashi na Ciki, Tsayin Layer, Gudun Buga, Zafin Fitarwa) da sakamakon ƙarfin juzu'i, wannan binciken yana nufin kafa tsarin da ke dogara da bayanai don hasashen inganci a cikin FDM, yana rage dogaro ga gwajin jiki mai tsada da ɗaukar lokaci.

2. Hanyoyi & Tsarin Gwaji

Hanyar binciken an tsara ta ne a kusa da gwaji mai sarrafawa sannan kuma binciken lissafi.

31

An Kera Samfuran PLA

4

Mahimman Sigogi na Shigarwa

4

An Kimanta Algorithms na ML

2.1. Kera Samfura & Sigogi

An yi amfani da na'urar buga 3D ta FDM don kera samfuran PLA guda 31. Ƙirar gwaji ta bambanta sigogi huɗu masu mahimmanci na tsari, waɗanda suka zama saitin siffofi don samfuran ML:

Kashi na Ciki: Yawan girma na tsarin ciki.
Tsayin Layer: Kauri na kowane layer da aka ajiye.
Gudun Buga: Gudun kan fitarwa.
Zafin Fitarwa: Zafin jikin filament ɗin da ya narke.

An auna UTS na kowane samfur ta hanyar gwajin juzu'i na yau da kullun, wanda ya haifar da saitin bayanai mai lakabi don koyo mai kulawa.

2.2. Algorithms na Koyon Na'ura (ML)

An aiwatar da algorithms daban-daban guda huɗu na rarrabuwa masu kulawa don hasashen ajin UTS (misali, ƙarfi mai girma da ƙarancin ƙarfi). Mai yiwuwa an raba ma'aunin manufa (UTS) zuwa azuzuwan don rarrabuwa.

Rarrabawar Logistic: Samfurin layi don rarrabuwa binary.
Rarrabawar Haɓaka Gradient: Dabarar haɗin gwiwa da ke gina bishiyoyi a jere don gyara kurakurai.
Bishiyar Shawara: Samfurin kamar bishiya na yanke shawara dangane da ƙimar siffofi.
K-Nearest Neighbor (KNN): Algorithm na koyo wanda ba na sigogi ba, wanda ya dogara da misali.

An kimanta aikin samfurin ta amfani da ma'auni kamar Makin F1 da Yankin Ƙarƙashin Lankwasa (AUC).

3. Sakamako & Bincike

3.1. Kwatancen Aikin Algorithm

Binciken ya haifar da tsari bayyananne a cikin aikin algorithm don wannan aiki na musamman. Dukansu algorithms na Bishiyar Shawara da K-Nearest Neighbor sun sami makin F1 daidai na 0.71, wanda ke nuna daidaiton daidaito tsakanin daidaito da tunawa. Duk da haka, algorithm ɗin KNN ya nuna ƙarfin rarrabuwa mafi girma tare da Makin Yankin Ƙarƙashin Lankwasa (AUC) mafi girma na 0.79, wanda ya fi Bishiyar Shawara da sauran algorithms guda biyu (Logistic da Gradient Boosting).

3.2. Fitarwar K-Nearest Neighbor

Makin AUC mafi girma na KNN yana nuna ƙarfin sa na ƙara rarrabuwa tsakanin azuzuwan biyu na ƙarfin juzu'i na ƙarshe a duk matakan rarrabuwa. Wannan yana nuna cewa ga saitin bayanai da aka bayar—wanda ke da sigogi huɗu na masana'antu da kuma alaƙa mai rikitarwa, wanda ba na layi ba tare da UTS—tunani na gida, wanda ya dogara da nisa na KNN ya fi tasiri fiye da dokokin duniya da Bishiyar Shawara ta koya ko iyakokin layi/logistic. Sakamakon ya jaddada mahimmancin zaɓin algorithm da ya dace da tsarin da ke cikin bayanan.

Fassarar Chati (Ra'ayi): Zane mai hasashe na Lankwasa Aiki Mai Karɓa (ROC) zai nuna lankwasa KNN yana kusanci kusanci da kusurwar sama-hagu (AUC=0.79) idan aka kwatanta da sauran algorithms, yana tabbatar da fitaccen aikin rarrabuwa. Lankwasa Bishiyar Shawara zai kasance a ƙasa kaɗan, yana raba maki makinta F1 iri ɗaya amma tare da ƙarancin yanki gabaɗaya a ƙarƙashin lankwasa.

4. Tsarin Fasaha & Tsarin Lissafi

Tushen yanke shawarar algorithm na KNN don sabon ma'auni na bayanai $\mathbf{x}_{\text{sabo}}$ (wanda aka ayyana ta sigoginsa huɗu na FDM) ya dogara ne akan ma'aunin nisa (yawanci Euclidean) da tsarin zaɓe a tsakanin $k$ mafi kusancin maƙwabta a sararin siffofi.

Nisa na Euclidean: Ana ƙididdige nisa tsakanin sabon ma'auni da ma'aunin horo $\mathbf{x}_i$ kamar haka: $$d(\mathbf{x}_{\text{sabo}}, \mathbf{x}_i) = \sqrt{\sum_{j=1}^{4} (x_{\text{sabo},j} - x_{i,j})^2}$$ inda $j$ ke nuna siffofi huɗu na shigarwa (Kashi na Ciki %, Tsayin Layer, da sauransu).

Dokar Rarrabuwa: Bayan gano samfuran horo $k$ tare da mafi ƙanƙanta nisa zuwa $\mathbf{x}_{\text{sabo}}$, ana ba da ajin UTS (misali, 'High') ta hanyar zaɓen rinjaye: $$\text{Aji}(\mathbf{x}_{\text{sabo}}) = \arg\max_{c \in \{\text{High, Low}\}} \sum_{i \in \mathcal{N}_k} I(y_i = c)$$ inda $\mathcal{N}_k$ shine saitin fihirisa don $k$ mafi kusancin maƙwabta, $y_i$ shine ainihin ajin maƙwabcinsa na $i$, kuma $I$ shine aikin nuna alama.

Mafi kyawun ƙimar $k$ yawanci ana ƙaddara ta hanyar giciye-tabbatarwa don guje wa yin wuce gona da iri (ƙananan $k$) ko sassauƙa mai yawa (babban $k$).

5. Tsarin Bincike: Nazarin Lamari Ba tare da Lambar Ba

Ka yi la'akari da masana'anta da ke nufin buga ƙirar PLA mai aiki wanda ke buƙatar mafi ƙarancin UTS na 45 MPa. Maimakon buga takardun gwaji da yawa, za su iya amfani da samfurin KNN da aka horar da shi a matsayin tagwaye na dijital.

Tambayar Shigarwa: Injiniya ya ba da shawarar saitin sigogi: {Ciki: 80%, Tsayin Layer: 0.2 mm, Gudu: 60 mm/s, Zafi: 210°C}.
Ƙididdigar Samfuri: Samfurin KNN ($k=5$) yana ƙididdige nisan Euclidean tsakanin wannan tambaya da duk samfuran 31 a cikin bayanan horo.
Daukar Maƙwabta: Yana gano bugu 5 na tarihi tare da mafi kusancin saitin sigogi.
Hasashe & Shawara: Idan 4 daga cikin waɗannan maƙwabta 5 suna da UTS da aka rarraba a matsayin 'High' (>45 MPa), samfurin yana hasashen 'High' don sababbin saituna. Injiniya ya sami babban kwarin gwiwa don ci gaba. Idan zaɓen ya kasance 3-2 don 'Low', an faɗakar da injiniya don daidaita sigogi (misali, ƙara ciki ko zafin jiki) kafin a yi kowane bugu na jiki.

Wannan tsarin yana canza inganta tsari daga ƙoƙarin jiki na gwaji-da-kuskue zuwa saurin siminti na lissafi.

6. Ayyukan Gaba & Hanyoyin Bincike

Nasarar wannan binciken ta buɗe hanyoyi da yawa don ci gaba:

Hasashen Abubuwa Da Yawa & Kaddarori Da Yawa: Faɗaɗa tsarin zuwa wasu kayan AM na gama gari (ABS, PETG, haɗaɗɗun) da hasashen jerin kaddarori (ƙarfin lanƙwasa, juriyar tasiri, watsa zafi) lokaci guda.
Haɗawa tare da Kulawar Tsari na Ainihi: Haɗa samfurin ML tare da na'urori masu auna firikwensin a cikin wuri (misali, kyamarori infrared, fitarwar sauti) don sarrafa madauki, kamar yadda aka bincika a cikin ayyuka kamar America Makes da MIT Self-Assembling Systems Lab. Wannan yana motsawa daga hasashen bayan haka zuwa gyara na ainihi.
Gine-ginen ML Na Ci Gaba: Yin amfani da samfuran koyo mai zurfi kamar Cibiyoyin Jijiyoyi na Convolutional (CNNs) don nazarin hotunan micro-CT na bugu don alaƙar lahani-kaddara kai tsaye, kamar hanyoyin da ake amfani da su a cikin nazarin hoton likita.
Matsalar Ƙirar Ƙirar Halitta: Juya samfurin don yin aiki azaman kayan aikin ƙirƙira: shigar da kaddarorin injiniya da ake so don fitar da mafi kyawun saitin sigogi na bugu, yana haɓaka tsarin ƙira-don-AM.

7. Ra'ayi na Mai Nazarin Masana'antu

Fahimta ta Asali: Wannan takarda ba game da KNN ya doke Bishiyar Shawara kawai ba ne; shaida ce cewa ko da samfuran ML masu sauƙi, masu fassara za su iya ɗaukar ilimin kimiyyar lissafi mai rikitarwa, wanda ba na layi ba na FDM da kyau don yin hasashe masu amfani. Ainihin ƙimar shawarar ita ce democratization na siminti mai zurfi—kawo hasashen bincike ga SMEs da benayen aiki ba tare da buƙatar PhD a cikin injiniyan lissafi ba.

Kwararar Hankali & Ƙarfafawa: Hanyar marubutan ta kasance mai aiki kuma bayyananne: ayyana gwaji mai sarrafawa, fitar da siffofi, gwada masu rarraba na yau da kullun. Ƙarfin yana cikin sake yiwuwa da kuma bayyanannen ƙarshe, wanda ya dogara da ma'auni (AUC > Makin F1 don zaɓin samfur). Yana haɗa gibin tsakanin kimiyyar kayan aiki da kimiyyar bayanai yadda ya kamata.

Kurakurai & Gibin Mai Muhimmanci: Giwa a cikin ɗaki shine ƙaramin saitin bayanai (n=31). A duniyar ML, wannan binciken matukin jirgi ne, ba samfurin da aka shirya don samarwa ba. Yana da haɗarin yin wuce gona da iri kuma ba shi da ƙarfi a cikin masu buga daban-daban, rukunin filament, ko yanayin muhalli. Bugu da ƙari, rarraba UTS zuwa azuzuwan yana rasa bayanai masu ci gaba masu ƙima; hanyar koma baya (misali, Regression na Gaussian Process, Regression na Daji bazuwar) mai yiwuwa ta kasance mafi bayani don ƙirar injiniya.

Fahimta Mai Aiki: Ga masu amfani da masana'antu: Fara nan, amma kar a tsaya nan. Yi amfani da wannan hanyar don gina nasa saitin bayanai na mallakar mallaka. Ga masu bincike: Mataki na gaba dole ne ya zama sikelin samun bayanai ta hanyar sarrafa kansa da bincika haɗaɗɗun cibiyoyin jijiyoyi masu sanin ilimin kimiyyar lissafi (PINNs)—kamar yadda aka haskaka a cikin aikin farko na Raissi et al. (2019) akan Journal of Computational Physics—wanda ke sanya sanannun dokokin kimiyyar lissafi (misali, daidaito na damuwa na zafi) cikin samfurin ML. Wannan haɗakar hanyar, haɗa koyo na tushen bayanai tare da ilimin yanki, shine mabuɗin haɓaka tagwaye na dijital masu ƙarfi, masu yaduwa, da aminci don ƙara kera waɗanda za su iya motsawa daga dakin gwaje-gwaje zuwa benen masana'anta.

8. Nassoshi

Du, B., et al. (Shekara). Nazari kan samuwar ramuka a cikin haɗin gwiwar walda mai gogayya ta amfani da bishiyar yanke shawara da cibiyar jijiyoyi ta Bayesian. Take na Jarida.
Hartl, R., et al. (Shekara). Aiwatar da Cibiyoyin Jijiyoyi na Wucin Gadi a cikin nazarin bayanan tsarin FSW. Take na Jarida.
Du, Y., et al. (Shekara). Hanyar haɗin gwiwa mai haɗa ilimin kimiyyar lissafi da koyon na'ura don rage lahani a cikin AM. Nature Communications.
Maleki, E., et al. (Shekara). Hanyar tushen ML don hasashen rayuwar gajiya a cikin samfuran AM da aka yi magani da su. International Journal of Fatigue.
Raissi, M., Perdikaris, P., & Karniadakis, G.E. (2019). Cibiyoyin jijiyoyi masu sanin ilimin kimiyyar lissafi: Tsarin koyo mai zurfi don warware matsalolin gaba da juzu'i waɗanda suka haɗa da daidaito na ɓangarori marasa layi. Journal of Computational Physics, 378, 686-707.
America Makes. (b.t.k.). Kundin Binciken Ƙara Kera. An samo daga https://www.americamakes.us
MIT Self-Assembling Systems Lab. (b.t.k.). Bincike kan Kera Mai Cin Gashin Kansa. An samo daga http://selfassemblylab.mit.edu
Zhu, J.Y., Park, T., Isola, P., & Efros, A.A. (2017). Fassarar Hoton zuwa Hoton mara Haɗin gwiwa ta amfani da Cibiyoyin Adawa na Haɗin kai. Proceedings of the IEEE International Conference on Computer Vision (ICCV). (An ambata a matsayin misalin gine-ginen ƙirar ML na ci gaba).