In the third step, we discharged the EC ca-
pacitors. Their temperature thus decreases
because of the EC effect. Last, we reversed the
fluid direction of the fluid motion to trans-
port the cooled fluid to the other side (cold
side). After this cycle is repeated a few times,
a stable temperature gradient occurs along the
fluid direction in the AER. TheDTspancan be
several times theDTECof the material.
We formed our first EC regenerator, la-
beled as REG1 (fig. S2 and table S1) ( 35 ), with
15 0.9-mm-thick PST-MLCs, distributed in a
matrix of three columns by five rows. We 3D-
printed nylon and polyether ether ketone
(PEEK) pieces to form the structure and seal it
from the fluid ( 24 , 26 , 35 ). The fluid used was a
dielectric fluid to prevent short circuits. This
design generated a maximumDTspanof 1 K
(Fig. 2A) from a 2.2 KDTEC. This corresponds
to a regeneration factor of 0.45.
To improve our poor experimental perform-
ance,weusedfiniteelementnumericalmod-
eling. Our simulated geometry included some
of the structural bodies (Fig. 1C and tables S2
and S3) ( 35 ). The good match to our experi-
ments validated the model results (Fig. 2A).
This gave us confidence that we could use
modeling to explore different configurations
for prototypes in order to find ones that had
largerDTspan(Fig.2B).Wespottedthatthe
inactive structural parts in our first design
were acting as heat traps and absorbing a
non-negligible amount of heat, which com-
promised the overall performance (fig. S3) ( 35 ).
Our main finding is that extra attention has
to be paid to thermal insulation by decreas-ing as much as possible the structural parts
of the regenerator.
After we removed the problematic struc-
tural elements from our model, we obtained
aDTspanof 2 K (Fig. 2B, orange square). We
also modified the MLCs’thickness. Thinner
MLCs increases the heat exchange area. By
decreasing the MLCs’thickness from 0.9 to
0.5 mm, which is possible to implement ex-
perimentally by changing the number of PST
layers from 19 to 9, our model reaches a
DTspanof 4 K, and the period reduces to 4 s.
Increasing the length of the regenerator, by
increasing the number of MLC columns,
makesitmoredifficulttorelaxthetemper-
ature gradient. This allows for a largerDTspan
to build. We modeled a fourfold increase of
the MLCs overall length that resulted in aSCIENCEsciencemag.org 2 OCTOBER 2020•VOL 370 ISSUE 6512 127
Fig. 3. Experimental results.(AandB) Adiabatic EC temperature change (DTEC) of 0.5-mm-thick PST-MLCs (A) as a function of the starting temperatureTsfor an
applied electric fieldE= 15.8 Vmm−^1 and (B) as a function of the applied voltageVandTs= 30°C. (C) Time (t) evolutionTand (inset) temperature difference of the
REG2’s hot and cold sideDTspanfor anE= 15.8 Vmm−^1 andTs= 30°C. (D)DTspanachieved by other EC devices ( 15 , 22 , 24 , 27 , 28 ).
RESEARCH | REPORTS