Bing Zhang¹, Johnny Aus in Lomas², Sandeep Pu han ee il Sa heesan¹, Ma hew Be y¹, Ying Diao²
¹Na ional Cen e o Supe compu ing Applica ions; ²Depa men o Chemical and Biomolecula Enginee ing, College o Libe al A s and Sciences
Uni e si y o Illinois U bana-Champaign
A Uni e sal Edge-Cloud A chi ec u e o Au onomous 3D P in ing
ACKNOWLEDGEMENTS
● This ma e ial is based upon wo k suppo ed by he
Illinois Compu es p og am.
● The expe imen s p esen ed in his wo k we e ob ained
using he NCSA Radian cloud esou ces.
● This esea ch is in pa sponso ed by NSF unde
awa d numbe 2119172.
OUR SOLUTION
● We p opose a hyb id edge-cloud amewo k ha
uni ies he e ogeneous componen s and add esses
au onomy challenges.
○ edge compu ing ole: Edge nodes p o ide low-la ency
eedback loops, local moni o ing, and eal- ime decision
making a he machine le el.
○cloud compu ing ole: Cloud pla o ms enable scalable
compu a ion, da a s o age, and sys em-wide coo dina ion.
● This sys em a chi ec u e enables scalable, eliable,
and adap able au onomous 3D p in ing ne wo ks
ac oss domains.
○ The sys em was deployed on he UIUC Radian cloud
clus e .
○ The sys em suppo s UIUC esea che s in conduc ing
s uc u al colo 3D p in ing expe imen s.
● In eg a ed wi h p in e s unning Ma lin i mwa e.
○ Ma lin i mwa e is one o he mos popula open-sou ce
so wa e in e p e ing G-code and con olling he p in e o
c ea e 3D objec s.
○ Ma lin Fi mwa e gi hub: Fo k 19.5k, S a 17k.
INTRODUCTION
●G owing Impo ance o 3D P in ing: A co ne s one o
he u u e digi al economy, ans o ming indus ies
such as heal hca e, cons uc ion, and ae ospace, and
beyond. I s in eg a ion wi h AI, obo ics, and sma
ma e ials d i es he shi owa d in elligen , adap i e,
and decen alized manu ac u ing.
●Au onomy 3D P in ing Challenges:
○closed-loop con ol: Wi hou human in e en ion, sys em
moni o s con inuous eedback om senso s o adap /adjus
pa ame e s, co ec e o s, and ensu e p in quali y.
○coo dina ion: Mul i-de ice and mul i- ask en i onmen s
demand synch onized decision making ac oss dis ibu ed
sys ems.
○ esou ce demand: Machine lea ning, simula ion, and
op imiza ion asks exceed he compu a ional limi s o local
compu ing.
●He e ogeneous En i onmen : Di e se p in e s,
senso s, con olle s, and so wa e pla o ms and use
p in asks complica e in eg a ion and in e ope abili y.
●A Uni e sal Pla o m o accommoda e a a ie y o
use asks.
CAMPAIGN EXPERIMENTS
MANAGEMENT
Use Managemen
New Campaign Managemen
● Op ion o se ini ial pa ame e s.
● Display he con en o PCP ile.
● Dynamic G id Sizing: Loads PCP ile, calcula es bounding box
and mo emen s, and isualizes he p in bed wi h dynamically
sized g id cells.
● Colo s g id cells in eal ime as each cell is p in ed.
P e ious Campaign Expe imen Managemen
P in Image A chi al Managemen
● P in images o each campaign a e au oma ically collec ed and
sen o he a chi al se ice.
● Each campaign becomes a da ase con aining all i s images.
Use s can click a colo ed cell o access i s p in image di ec ly.
3D PRINTER WE USE
Lulzbo az p o 3D p in e in P o . Ying Diao’s lab.
CONCLUSIONS & FUTURE WORK
Sys em alida ion ia s uc u al colo 3D p in ing
applica ion, analyzing how HSV colo p ope ies ela e o
p in ing pa ame e s like p essu e, speed, and bed
empe a u e.
The sys em was deployed on he UIUC Radian cloud
clus e and es ed on mul iple PCP-de ined p in s using
Ma lin i mwa e-based 3D p in e s.
We conduc ed closed-loop campaigns wi hou human
in e en ion, guided by a Bayesian op imize o e icien ly
explo e he pa ame e space.
I s modula , adap able design enables scalable,
ep oducible expe imen s, demons a ing i s po en ial as a
uni e sal pla o m o au onomous 3D p in ing.
Fu u e wo k includes op imizing sys em pe o mance and
eliabili y, and e alua ing di e se 3D p in ing asks.
HIGH-LEVEL SYSTEM OVERVIEW
Adap e
● Running on a machine nea o 3D p in e .
○ Recei es command messages and ansla es hem in o
G-code compa ible wi h he Ma lin i mwa e.
○ Quick check o p in image quali y.
○ Au oma ically cleans he nozzle a e each cell p in .
○ S o e image iles o a chi al se ice and u he analysis.
A Cen alized, Use - iendly Web Applica ion
● P o ides an adminis a i e in e ace o managing emo e 3D
p in e s.
○ Use accoun managemen .
○ B owse his o ical campaign p in esul s.
○ S a new campaigns wi h cus omized p in pa ame e s.
Cloud-Based Scalable Mic ose ices
● Use -buil docke images a e managed by an o ches a ion
oolki , which scales mul iple con aine s ac oss a cloud-based
clus e o machines.
● High-pe o mance Rabbi MQ-based messaging sys em o
edge-cloud communica ion
● MongoDB o s uc u ed s o age o expe imen al me ada a, p in
eco ds.
● A chi al se ice (Clowde ) o managing a chi al iles and
analysis epo s.
FAILURE TOLERANCE
● Replica ed mic ose ices o eliabili y.
● Pe sis en ask commands in messaging sys em.
● Ale s on 3D p in e disconnec ions.
● Da abase main ains all p in eco ds.
