You are on page 1of 29

Applied Mathematical Modelling 32 (2008) 141169

www.elsevier.com/locate/apm

Mathematical modeling of molten metal dispensing: A study


of a pneumatically actuated diaphragm-driven pump
Vivek Jairazbhoy a, Randy C. Stevenson
a

b,*

Climate Control, Automotive Components Holdings, Plymouth, MI 48170, USA


Chassis CAE, Automotive Components Holdings, Dearborn, MI 48120, USA

Received 1 June 2005; received in revised form 1 August 2006; accepted 2 November 2006
Available online 19 January 2007

Abstract
This study describes a lumped-element model for a pneumatically actuated, diaphragm-driven pump. The pump was
developed to dispense molten metal for automotive electronic circuit manufacturing. In place of a CFD analysis, which
proved to involve excessively long computations because of a large range of length scales, we created a lumped-element
representation of the physics. The pump has ve main region volumes: an air lling chamber, a solder reservoir, an adjustable throttle valve, a main pumping chamber, and a dispensing nozzle or set of nozzles. The pumping process employs a
compressed gas pulse, which lls the air chamber and deforms a diaphragm, which then displaces molten solder in the main
pumping chamber. The description of the dispensing process requires the coupling of three physical domains: pneumatic,
mechanical, and hydraulic. The uid motion in the pump is modeled by creating lumped uid elements from control volumes, some with deformable surfaces. The equations for the uid elements are derived from Bernoullis equation, when
viscous eects are not important, or from the general integral momentum balance, when viscous eects are important.
A parametric study describing the relationships between variables is presented, and experimental data is used to conrm
the validity of the model.
 2006 Published by Elsevier Inc.

1. Introduction
We present a mathematical model of the operation of a pump designed for use as a molten solder dispenser
in electronic manufacturing applications. Fig. 1a shows a schematic diagram of the pump, an early concept
drawing, and Fig. 1b shows a picture of a later actual working prototype. The pump operates by displacing
molten metal with a pneumatically actuated diaphragm. The resulting dispense characteristics, such as the dispense volume, and the dispense timing, are sensitive to numerous parameters, geometrical and operational. A
rational selection of these parameters can be signicantly aided, if the pump operation can be modeled.

Corresponding author.
E-mail addresses: vjairazb@ach-llc.com (V. Jairazbhoy), rsteven7@ach-llc.com (R.C. Stevenson).

0307-904X/$ - see front matter  2006 Published by Elsevier Inc.


doi:10.1016/j.apm.2006.11.005

142

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

Nomenclature
characteristic (inverse time) for sonic ow
through solenoid
asubsonic characteristic (inverse time) for subsonic
ow through solenoid
cp
heat capacity of lling gas at constant
pressure
CS
oce discharge coecient of solenoid
DPviscous pressure drop at a port due to viscous loss
dD
diameter of diaphragm
oV
boundary of uid device
oVports boundary of all ports of uid device
oVwalls boundary of wall of uid device
E
elastic modulus for diaphragm material
g
an adjustable constant used to gauge the
jump condition in ow rates before and
after the diaphragm hit the stop
f
body force eld per unit mass
Fp
face of port p of uid device
Fwall
face of wall of uid device
u
uid velocity potential
Up
average velocity potential over port p
g
acceleration due to gravity
GN, GT frictional force (due to viscous losses) in
the nozzle and throttle volume regions,
respectively
Gret
eective G for retraction phase
Gwall
force of uid on wall through viscous and
pressure forces
GR
surface tension force in the nozzle region
hpq
vertical height between port p and port q
hN
vertical height of nozzle
hR
vertical height of reservoir
hT
vertical height of throttle
kD
the eective spring constant for the diaphragm
kin, kout adjustable dimensionless constants characterizing the degree of loss of kinetic energy at a port due to viscous surface forces
Kp
kinetic energy factor for port p
drv
K drv
;
K
eective K for the driving phase for
N
T
the nozzle stream and throttle stream
respectively
Kret
eective K for retraction phase
lpq
characteristic length scale along ow from
port p to port q
lN
ow path length of nozzle
lR
ow path length of reservoir
lT
ow path length of throttle

asonic

^l

unit vector along direction of central


stream line in lumped uid element
M
matrix of connection coecients
connection coecient between port p and
Mpq
port q
M eff
eective connection coecient from port p
pq
to port q
l
coecient of viscosity for uid
np
normal vector to face Fp
nwall
normal vector to wall face
N
number of nozzles
NP
number of ports for uid device
m
Poissons ratio for diaphragm material
p
pressure eld of uid
Pp
average value of pressure p over port p
P0
ambient pressure
PF
air pressure in air lling chamber
PG
gas pressure upstream of solenoid valve
Q
volume ow rate (used in two-port lumped
uid element)
Qp
average ow rate over port p
rthreshold pressure ratio that marks that transition
from sonic to subsonic ow
Ro
outer radius of throttle
Ri
inner radius of throttle
RN
radius of nozzles
q
density of uid
s
coordinate along mean stream line in ow
in lumped uid element
S(s)
cross-sectional area of lumped uid element at position s
Sp
area of port p of uid device
r
surface tension constant for uid
t
time
T
uid stress eld
Tvis
viscous contribution to uid stress eld
T
s
thickness of diaphragm
H()
unit step function
h
contact angle for uidairwall interface
v
uid velocity eld
vS
velocity of moving uid surface
vp
velocity of uid at port p
VD(xD) volume displaced by diaphragm with center displacement xD
VDispensed volume of uid dispensed
xD
the displacement coordinate for the center
of the diaphragm

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

v
z

gravitational potential function


Z-coordinate of gravity eld

Zp

Solder

143

average value of z-coordinate over port p

Reservoir
Throttle

Compressed
Gas

Filling
Chamber
Diaphragm
Stop

Solenoid
valve

Diaphragm

Nitrogen
Inerting
Manifold

Nozzles

Fig. 1a. Schematic of an early prototype molten solder dispenser.

Pins for
Attachment to
Robot

Thermal Couple

Inlet for Inerting


Nitrogen

Solder Reservoir
Main Chamber

Throttle
Adjustment

Optical Probe for


Diaphragm
Displacement

Heating Coil
Diaphragm Stop
Adjustment

Compressed Gas
Line

Nitrogen
Inerting Diffuser

Nozzles
(underneath
and at center)

Fig. 1b. Final prototype molten solder dispenser.

In addition, if the modeling simulations and corresponding data correspond, then one can say that one understands the operation of the pump, to an appreciable degree. The operational parameters include drive
pressure, throttle aperture, and the diaphragm displacement. Geometrical properties include the pressure head
from the reservoir, the shape of the arrangement of the interior volumes and surfaces of the pump, as well as
the length and diameter of the nozzles. Other design considerations include, the number of dispensing nozzles,
material characteristics, and very importantly, a nitrogen inerting system to limit the oxidation of the solder at
all times during the operation of the pump. Given the complexity of the system, and the desire to predict transient behavior, a comprehensive CFD model proved intractable. Thus, we were led to pursue a lumped-element type of analysis for the dispense operation. However, our lumped parameter analysis of the uid ow
excluded any detailed model of drop formation.

144

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

1.1. Previous work


The basic goal of the solder dispensing technology was to directly place solder onto substrates and device
leads without the need for printing. The technology was directly aimed at ne-pitch device applications, and
solder bumping for ip-chip applications. The trend towards smaller dimensions was one of the main drivers
for this technology. The leads (say on a microprocessor device) may have a 400-lm pitch. Conventional solder
printing techniques are problematic at these dimensions, in part, because bridging may occur between the
leads. Additionally, the promise of uxless soldering, and reduced thermal stress, were also major considerations for pursuing the technology.
There were other eorts and techniques, similar to the eort discussed in this paper, which developed working prototypes. For example, the MPM Corporation developed two metal jetting technologies [1,2]. One
technique involved an electrostatically guided stream of solder balls; a continuously dispensed stream of molten solder was broken into a stream of balls through surface-tension induced instabilities. The stream of balls
(50300 lm in diameter) was directed onto electronic substrates with sub-micron placement accuracy. MPM
also developed a piezoelectrically actuated, single nozzle, drop on demand system, more akin to the system
discussed in this paper. This technique is essentially a copy of the inkjet dispensing scheme; it employs a capillary tube, one end of which is connected to a solder reservoir and the other end is a nozzle. The nozzle end
has a surrounding piezoelectric element, which, when electrically pulsed, rapidly ejects the solder. The dynamics of the pump described in this paper is analogous to that used in the ink jet process, although at a rst
glance this may not be apparent. In both the inkjet scheme, and in our scheme, the dispensing technique relies
on a balancing of uid momentum at the two ends of a capillary tube. This balancing requires a connection of
one end of the tube to a reservoir with a free surface. The other end of the tube is dominated by capillary
eects (surface tension forces).
Another interesting technique was developed in Germany by Pac Tech-Packaging Technologies [3]. In this
technology, preformed solid solder balls (80760 lm diameters) are ejected from a capillary tube. As the balls
are ejected they are melted by a laser. What is interesting about this process is that the volume of the balls is
predetermined, and the laser not only melts the ball but can add extra heat to the balls in a controlled fashion
so that they can better form intermetallic joints with the substrate.
We mention in passing that, in addition to the mechanics of the dispensing process, the solidication
process and formation of the intermetallic joints, which the dispensed solder makes with the substrate and/
or device leads, must also be controlled. A discussion of the modeling of the joint formation can be found
in [46]. It appears to the authors of this paper that there is more in the literature regarding joint formation
than dispense modeling, probably because of the proprietary nature of the workings of the dispensers
themselves.
2. Description of the pump and its operation
The pump is constructed almost entirely of titanium, so that the pump walls are not chemically reactive
with the molten solder. The pumping action is executed with a diaphragm driven by a controlled (open loop)
actuation of pneumatic (solenoid) valves. In some applications, the pump is required to simultaneously dispense 256 drops of solder (where each drop is typically about 200 lm in diameter) onto the 4 64 = 256 leads
of a microprocessor device. Nozzle arrays (64 nozzles in one array) were fabricated with silicon micro-machining techniques. For dispensing a single drop, a single nozzle was employed. In this case, the single nozzles were
fabricated in thin titanium plates with very ne drills, as small as several thousands of an inch in diameter. The
nozzles were typically 100 lm in diameter and 1000 lm long. The ow is, of course, highly viscous within these
channels.
In preparation for dispensing, one must prime the pump; i.e., one must ll the pump entirely with liquid
solder, with virtually no air trapped inside. Because the dispense volumes are so small, small amounts of
trapped air can lead to an uncontrollable compressibility. Once the pump has been primed, it is in the resting
state, and one may wonder why the pump does not then leak. There are no valves preventing the solder from
draining out the nozzles at the bottom of the pump. Indeed, the pump does not leak because the surface tension of the solderair interface in the nozzles supports the pressure head.

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

145

One may divide that part of the pump, which contains the uid, into four sections: the solder reservoir,
main pump chamber, the adjustable throttle between the reservoir and the main pump chamber, and the dispensing nozzle or nozzles. These four sections of uid interact during the dispense process, which proceeds as
follows: the diaphragm pushes on the uid in main pump chamber; the uid then has two paths to ow. One
path goes from the diaphragm up through the adjustable throttle valve, and into the pump reservoir. The
other path from the diaphragm goes down to the nozzles. We have several control knobs to help tune the dispense process. We can control displacement and speed of the diaphragm, the pressure head in the reservoir,
and the constriction of the throttle valve. The diaphragm displacement is limited by a stop, which allows the
diaphragm to move, typically, between one to two thousandths of an inch. Once the diaphragm hits the stop,
the uid may still ow through the throttle to the reservoir, or down through the nozzles, but only in one
direction. A proper selection of the throttle position will cause the uid to ow up, creating a draw back mechanism on the nozzle ow, which is adjustable and independent of the diaphragm drawback, which occurs well
after each dispense. In the absence of the throttle-mitigated draw back, the pump may continue to dispense at
low velocities toward the end of the dispense cycle, resulting in misdirection and dribbling of the solder emanating from the nozzles. Conversely, excessive drawback can lead to high retraction velocities within the nozzles, causing air to be drawn into the pump. The optimum control of drawback is therefore essential to
successful dispensing.
To be able to perform a succeeding pumping action, one must allow the diaphragm to relax to its resting
position. We do not include this part of the process in our analytical description, since no uid is dispensed.
3. Analytical formulation
From this characterization of the dispense process, one can see that a description of the dispensing process
requires the analysis of coupled processes from at least three physical domains. There is
1. A pneumatic (air ow) process: the lling of the air chamber upstream of the diaphragm by opening and
closing a solenoid valve.
2. A mechanical process: the dynamic response of the diaphragm due the applied pressure from the air pulse
and back pressure of the uid.
3. A uid (liquid) ow process: the ow of solder within the pump volume and out from the nozzles.
To this end, we provide a representation for the air pulse that drives the diaphragm, we use a simple Hooks
Law model for lumped-element representation of the diaphragm, and we obtain an analytical representation, a
lumped-element formulation, of the physics in the four uid regions of the pump.
We treat the driving function as an applied boundary condition, and the regions are connected together
through the continuity of physical variables. A quantitative description of the pump behavior is then obtained
by simultaneous solution of the aforementioned analytical forms for a given set of design parameters.
Before we construct the various lumped elements for the volume regions of the pump, we give below a general derivation of the approximate dynamics of the uid lumped elements.
3.1. Construction of lumped uid elements general treatment
In the next two sections we develop a formalism to allow us to construct lumped-element models for the
four regions volumes of the pump in which the molten metal ows. In Section 3.2 below, we develop a general
Bernoulli lumped uid element. This element will be used for the main chamber and reservoir region volumes.
For these regions, the salient engineering approximation that we make is that the uid dynamics of the molten
solder is well approximated by potential ow. We argue that the uid motion in the main chamber and reservoir is essentially inviscid because the signicant dimensions are relatively large (the total volume in either
region is very much larger than the volume displaced in a single pulse), and the velocities are small (no more
than a few millimeters per second.) The sudden change in magnitude and, in some cases, direction of the ow
after the diaphragm hits the stop could create local vortices that might have a minor impact on the subsequent
ow, but we neglect their impact in our analysis.

146

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

In Section 3.3, we develop a uid element applicable to ow in narrow tubes, where viscous eects are
important. The treatment for these uid elements is, in one sense, more general than that for the Bernoulli
uid elements, in that we account for viscous eects. It is less general, in another sense, in that we must rely
on specic known solutions to be able to complete the dynamics of the element. The reason that we need to
invoke these specic solutions is that there is no general solution for the velocity eld within a control volume
for the NavierStokes equation, as far as the authors know, in terms of just data on the surface of the control
volume.
3.2. General formulation for a Bernoulli lumped uid element
We begin our construction of Bernoulli lumped uid elements by following Stevenson [7]. We specify an
inviscid ow of velocity v in a region V with boundary oV. The uid has density q, pressure p, and gravitational body force density f = g. The boundary of the region volume, or the control volume, is comprised of
a wall, through which no uid passes, and a series of NP ports, through which uid may pass, as illustrated
by Fig. 2. One may write for the boundary


NP
1
oV oV ports  oV walls  F p  F wall :
p1

For inviscid ow, the velocity v is derivable from a potential u.


v ru:

With conservation of mass for incompressible ow, i.e.,


r  v 0;

the potential satises Laplaces equation


r2 u 0:

We may speak of the physics within the lump volume in terms of the language of bond graphs [8]. There is a
through variable, for us this variable is volume ow rate, and a cross-variable, for us the velocity potential.
From Stevenson et al., we know that we may write an approximate integral form of Laplaces equation as
Up

NP
X

M pq Qq ; p 1; . . . ; N P :

V = FWall + F1 + F2 + ... + FN

Port N P

n1

Face FN : SN ,QN ,N ,KN

Region Volume V

Port 1

nN

Face F1 : S1 ,Q1 ,1 ,K1

f=g

Port 2
Face F2 : S2 ,Q2 , 2 ,K2

n2
Fig. 2. A uid device with Np ports. Each port has a face F, an area S, and an outward pointing normal n. There is an average ow rate Q
out of the port in the direction n. There is an average potential U (for Bernoulli ow), as well as a K constant that characterizes the kinetic
energy out of the port in terms of Q.

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

147

Here, NP is the number of ports associated with the control volume, Up is the average of the potential u over a
the face Fp of port p, and Qq is the average volume ow rate out of the port p,
Z
dS
Qq 
S p np  v:
6
F p Sp
The symmetric matrix, M = {Mpq}, is a set of connection coecients, essentially geometrical quantities, that
represents a generalized impedance relation between the through and cross variables. For example, if the potential u represented a static electric potential, then the connection coecients would represent an (inverse)
capacitance matrix elements between potential and charge.
We note that mass conservation equation yields, with the averaging process over ports, the relation
N
X

Qq 0:

We now add dynamics to our lump formalism in the form of Bernoullis equation
q

ou 1 2
qv p qgz 0:
ot 2

To convert to a lumped approximation, we average Bernoullis equation over the face Fp




Z
dS
ou 1 2
oUp 1
qv p qgz 0 ! q
qhv2 ip P p qgZ p 0:
q
S
ot
2
2
ot
p
Fp

The quantities Zp and Pp are, respectively, the average z-value, and the average pressure over the port p.
Consider the quantity hv2ip. We typically do not have detailed information on the ow elds at the ports,
but we may postulate, or impose, a boundary condition here. We will specify that the velocity at the port is
approximately proportional to the average normal ow Qp out of the port
vjp ! Qp =S p np ;

so that

1
1 Kp
qhv2 ip ! q 2 Q2p :
2
2 Sp

10

The dimensionless constant Kp is an adjustable, or empirically determined, constant that expresses our imperfect knowledge of the actual ow at the port p. We expect Kp to be less than or on the order of 1 (the average of
the square is less than the square of the average).
If we now substitute the impedance relation Eq. (5), into the averaged Bernoulli equation, Eq. (9), we
obtain the lumped-element relation that we have been seeking
q

N
X
d
1 Kp
M pq Qq q 2 Q2p qgZ p P p 0;
dt
2 Sp
q

p 1; . . . ; N :

11

Here we have a set of coupled non-linear rst order dierential equations for the ow through each port. We
must also include the mass conservation relation, Eq. (7), with this set.
3.3. General formulation for a viscous lumped element
The ow in the throttle and nozzles is dierent from that in the pump body and reservoir, since the uid
velocities are signicantly higher, and correspondingly the characteristic dimensions are smaller than that of
the main chamber of the pumb body. Viscous eects could be important during certain phases of the pulse. In
addition, in certain periods during the dispense process, the solder meniscus in the nozzles moves. Thus, our
general treatment must consider moving boundaries as well as viscous eects.
Following the procedure outlined by Slattery [9], we begin by briey deriving the generalized integral
momentum balance equations. Initially, the region volume, or control volume, has the same general geometry
as employed for the uid element illustrated in Fig. 2; we then specialize to control volumes with more tubular
shapes, more akin to the nozzle and throttle element, such as that illustrated in Fig. 3.

148

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


F1 : P1 , Q1 , S1 , Z1

n1

S(s)
V, Swall

s
v

v| |

l12 l

n2

F2 : P2 , Q2 , S2 , Z 2

vS

Fig. 3. Model for control volume V used to derive viscous lumped elements.

The momentum balance equation is


o
qv r  qvv r  TT qf;
ot

12

where we will assume that the body force density f is derivable from a potential, and that the density is constant, i.e.,
f rv;

where v g  z:

13

We shall make the conventional assumption that the stress tensor T partitions into an isotropic pressure p and
viscous stress Tvis
T pI Tvis :

14

We then integrate over our time-changing control volume V, and apply the divergence theorem.
Z
Z
Z
Z
o
dV qv 
dSn  qvv
dST  n
dSnqv:
ot
V
oV
oV
oV

15

Here, n is an outwardly pointing normal to the surface oV of V. Next, we transform the remaining volume
integral above by applying the transport theorem to the momentum density, qv, i.e.
Z
Z
Z
o
d
dV qv
dV qv 
dSn  vS qv:
16
ot
dt V
V
oV
The velocity vS is the velocity of the moving boundary. We combine the last two equations to yield the
following:
Z
Z
Z
Z
d
dV qv 
dSn  v  vS qv
dST  n
dSnqv:
17
dt V
oV
oV
oV
The reader is referred to Fig. 2 again to see that the boundary oV is partitioned into two types: ports and walls.
In principle, the surface of the port could be moving; e.g., the bottom port of a nozzle is a moving free surface.
We assume that walls do not move and, of course, no uid crosses a wall. Thus, if there are a total of NP ports,
Z
Z
NP Z
NP Z
X
X
d
dV qv 
dSnp  v  vS qv
dST  np np qv 
dSnwall  v  vS qv
dt V
Fp
Fp
F wall
p1
p1
Z

dST  nwall nwall qv:


18
F wall

We now simplify the RHS of this equation. Consider the second term on the RHS. We assume that the viscous
forces across the surface Fp of a port are negligible
Tvis  njF p ! 0;

19

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

given that the port surfaces are generally perpendicular to the ow. Hence
Z
Z
Z
Z
dST  np np qv
dSTvis  np
dSnp p qv
dSnp p qv:
Fp

Fp

Fp

149

20

Fp

For the third term on the RHS of Eq. (18), the normal component of the uid velocity nwall v vanishes at a
wall as does nwall vS. Hence
Z
dSnwall  v  vS qv ! 0:
21
F wall

For the fourth term on the RHS of Eq. (18), we substitute our expression for T, Eq. (14), and for v, Eq. (13), to
obtain
Z
dSTvis  nwall  nwall p qgz:
22
Gwall 
F wall

Here, Gwall is the force that the uid exerts on the bounding wall. Our simplied integral form then reduces to
Z
NP Z
NP Z
X
X
d
dV qv 
dSnp  v  vS qv 
dSnp p qgz Gwall :
23
dt V
Fp
Fp
p1
p1
The equation above forms the basis for our integral momentum balances in the throttle and nozzles. Below we
will apply this equation to a more specic lump geometry than that of Fig. 2. As with the Bernoulli lumped
elements, one must solve Eq. (23) in conjunction with the mass balance, Eq. (7).
3.4. Application of the viscous integral momentum form to tubular elements
We now tailor the integral momentum equation, Eq. (23), more to our application, with the simple control
volume of Fig. 3 in mind. This control volume has the basic shape of the throttle and nozzle volume regions.
First we consider the volume term of the LHS of Eq. (23), and examine Fig. 3; we dene l12^l to point from
port 1 to port 2, and positive ow is from port 1 to port 2. We wish to relate this volume integral to the average
volume ow rate Q. In general, one may split the velocity eld, v, into parallel components, vk (along the axial
direction ^l), and perpendicular components, v?. We then assume that the perpendicular components approximately average to 0, i.e.,
d
dt

dV qv
V

d
dt

0

Z
V

z
Z }|{
d
dV qvk
dV qv? :
dt V

24

We then express the remaining integral above in terms of the average volume ow rate, as follows:
vk s^l

z
}|
{
z
{
Z
Z }|
Z
Z
dS
ds
vk l12
Ssvk s ^l l12 Q^l:
dV vk
dsSs
V
l
S Ss
l l

25

From mass conservation we know that Q = Q2 = Q1, where Q2 is positive, going out of the second port. The
volume term of Eq. (24) then reduces to
Z
d
dl12 Q
:
26
dV qv ! ^lq
dt V
dt
Next we consider the rst surface term on the RHS of Eq. (23). We may write this term as
8
Z
Z
dS
dS
>
>
n1  v1 v1 qS 2
n2  v2 v2 ; F 1 and F 2 fixed;
>
2 Z
< qS 1
X
F 1 S1
F 2 S2
Z
dSnp  v  vS qv
>
dS
Fp
>
p1
> qS 1
n1  v1 v1 ; F 1 fixed; F 2 moving with v2 vS :
:
F 1 S1

27

150

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

This term is a boundary kinetic energy term, similar to that encountered in the Bernoulli lumped-element formulation. As with the Bernoulli term, we generally do not know the details of the ows at the entrances and
exits of our simple control volumes. We observe that we have an average over a quadratic dependence
on velocity, at each port. Therefore, much as we did for the Bernoulli lumped element, we postulate a
kinetic boundary condition (based on a possibly empirically determined constant.) We say that, if np  ^l
sgnQp ,


Z
dS
1
2
2
^
np  vp vp ! sgnQp l K p Qp =S p ; p 1; 2:
28
2
F p Sp
We write, with sgn(Q) = 1,


2 Z
X
1 K 2 vS 2 1 K 1 2
^
dSnp  v  vS qv l q
Q  q Q :
2
S2
2 S1
Fp
p1

29

To complete our construction of a viscous uid element, we move to the second surface term in Eq. (23) and
express this surface integral as
Z
Z
X2 Z
dS
dS
dSn
p

qgz

S
n
p

qgz

S
n2 p qgz
p
1
1
2
p1
S
1
Fp
F1
F 2 S2
! ^lS 2 P 2  S 1 P 1 ^lqgS 2 Z 2  S 1 Z 1 :

30

Finally, we consider the wall term Gwall, which is the force that the uid exerts on the wall through viscous and
pressure forces. We consider only those components along our direction of ow ^l, in which case, we pick o
primarily viscous forces, since ^l 12 tends to be orthogonal to the normal vector, the direction of the pressure
forces. So we write
^l  Gwall  GQ;

31a

where G(Q) is the force of the wall on the uid. We allow for the dependence of G on uid velocity through a
dependence of G on Q. Note, also, that a provision must be made for G to switch signs, if the ow direction
changes direction. Thus, G(Q) must have the property that
GQ sgnQjGQj:

31b

We now collect all terms together, and dot all terms in the vector equation, Eq. (23) by ^l, to remove the
vector dependence. The dynamic equation for our tubular viscous lumped element is then


d
1 K1 K2

q l12 Q q
Q2 qgS 1 Z 1  S 2 Z 2  GQ S 1 P 1  S 2 P 2 :
dt
2
S1 S2

32

As with the Bernoulli equation, the dynamical equation above is not dependent on our chosen reference port.
If one changes the sign of Q, and also interchanges the port indices, the equation remains invariant.
3.5. Lossy port condition for viscous lumped element
In our application, we have viscous lumped elements, such as the throttle, which adjoin Bernoulli lumped
elements such as the main chamber and the reservoir. However, in our formulation so far, we have not
accounted for the pressure losses that can occur at a port for ow from a constricted region to one more open.
We will do this accounting by constructing a simple viscous lumped-element model of the transition region at
the port and then ax this element to the viscous lumped element (see Fig. 4).
We will assume in the transition region, that if ow is from the viscous region to the Bernoulli region, then
we have an adjustable loss (likely, almost total loss) of the kinetic energy. This loss leads to a pressure change.
If the ow is in the reverse direction, we then assume little loss, again adjustable. We then combine the transition region with the viscous region, through the continuity of pressure, to form a single region. We shrink
this combined region back to the size of the original viscous region.

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

151

Fig. 4. Model used to develop boundary condition for a port that transitions between a viscous region and a Bernoulli region.

Referring to Fig. 4, and using Eq. (32), we write the lumped equation for the transition region




d
1 K 00 K 00
q l0 Q q 002  002 Q2 qg S 02 Z 02  S 002 Z 002  G0 Q S 02 P 02  S 002 P 002 :
dt
2
S2 S2

33

The model that we use for the port loss, as mentioned above, is to assume a pressure drop, DPviscous, that arises
from the viscous forces originating locally at the port wall. We then say,
DP viscous

G0 Q
1
Q2
! qK loss
Q
:
S2
2 2
S 22

34

We dene,
K loss
2 Q 

k out ; Q > 0;
k in ; Q < 0;

ideally

1; Q > 0;
0; Q < 0;

HQ:

Next, we write the corresponding equation for the lower viscous region


d
1 K 1 K 02
 0 Q2 qgS 1 Z 1  S 02 Z 02  GQ S 1 P 1  S 02 P 02
q lQ q
dt
2
S1 S2

35

36

We wish to eliminate the intermediate pressure P 02 from the equations for the transition region, Eq. (33), and
the lower viscous region, Eq. (36). This elimination is easily accomplished, if we just add the two equations
together; we obtain


d
1 K 1 K 002 K loss
0
2 Q
q l l Q q


37
Q2 qgS 1 Z 1  S 002 Z 002  GQ S 1 P 1  S 002 P 002 :
dt
2
S2
S 1 S 002
We then shrink the transition layer, so that
l0 ! 0; so that S 002 ! S 2 ;

Z 002 ! Z 2 ;

K 002 ! K 2 ; and P 002 ! P 2 :

38

We nd the nal form for the viscous lumped element, with a modied kinetic boundary condition to account
for port loss, is


d
1 K 1 K 2 K loss
2 Q
q l12 Q q

39
Q2 qgS 1 Z 1  S 2 Z 2  GQ S 1 P 1  S 2 P 2 :
dt
2
S2
S1

152

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

3.6. Comparison of viscous and Bernoulli uid elements


The similarity in form between the viscous uid element equation, and that of the Bernoulli, which we
reproduce below
N
X
d
1 Kp
M pq Qq  q 2 Q2p  qgZ p  P p ; p 1; . . . ; N ;
40
q
dt
2 Sp
q
is apparent from inspection. However, we can make the similarity even more apparent, if we consider applying
the Bernoulli lumped-element equation to a two-port device, with a port 1 and a port 2. We consider port 1 the
reference port; in this case, we subtract the Bernoulli lumped equation for port 1
d
1 Kp
q M 11 Q1 M 12 Q2  q 2 Q21  qgZ 1  P 1 ;
41
dt
2 Sp
from that of port 2,
d
1 Kp
q M 21 Q1 M 22 Q2  q 2 Q22  qgZ 2  P 2 ;
dt
2 Sp

42

we use the fact that Q1 + Q2 = 0, dene, Q2  Q, in addition, dene


eff
M eff
12  M 11  2M 12 M 22 trM  2M 12 M 21 :

We arrive at the Bernoulli lumped-element equation for a two-port element,


!
d
1
K1 K2
eff
q M 12 Q q 2  2 Q2 qgZ 1  Z 2 P 1  P 2 :
dt
2
S1 S2

43

44

Note that the equation is invariant, if we let Q ! Q, and switch port labels. We see that the Bernoulli formulation and the viscous formulation give rise to similar lumped-element representations, which one might
expect, but, of course, the Bernoulli element does not contain the viscous force contribution. The similarity
is closer if one identies the connection coecients, the Ms, as essentially inverse characteristic length scales,
approximately the length along the center uid stream line, divided by an area
M eff
12 $ l=S:

45

Of course, this relationship is essentially that found for the inverse of the capacitance between two charged
plates in electrostatics.
4. Derivation of lumped elements for the pump volume regions
In the next six subsections, we present a representation for the driving pneumatic pulse, and give specic
representations of the ve lumped-element volume regions for the pump.
4.1. Pneumatic driving force
In this section, we provide an analytical function that describes the time dependence of the build-up of the
pressure, PF(t), within the air lling chamber, by ow through a solenoid valve. This pressure drives the displacement of the diaphragm (see Fig. 5a); this pressure function provides a boundary condition for the lumped
element of the diaphragm.
The formation of the transient pressure pulse is well described by the lling of a hot chamber with a cold
gas through an orice. The lling of the chamber proceeds as follows: If, at the beginning of the pulse, the
pressure ratio (the pressure downstream of the orice divided by the pressure upstream of the orice, PF/
PG, is less than rthreshold = 0.526 [10]), we initially have choked (sonic) ow through the solenoid valve. As
the chamber lls the downstream pressure rises, and, if the pulse is long enough, as is generally the case in
our application, the pressure ratio, PF/PG, rises above rthreshold. The ow then becomes subsonic through
the orice. If the lling process were to continue indenitely, the downstream pressure would rise enough

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

153

Solenoid Valve
Diaphragm

PF
PG

CS SS
VF
Air Filling
Chamber

Fig. 5a. Parameters used to model the compressed air pulse pressure.

for the ow in the orice to become incompressible. In our application, the pulse is not long enough for the
incompressible state to be reached. If the initial pressure ratio is greater than rthreshold, then the choked condition never prevails; only the subsonic analysis is considered.
Since the volume displacement of the diaphragm is small compared with the volume of the pneumatic
chamber, the chamber volume may be assumed constant during the lling operation. Simultaneous mass
and energy balances for the chamber yield a relationship for the pressure transient in terms of the mass ow
rate through the orice, and the incoming gas temperature. The controlling resistance to the air ow lies in the
solenoid. Andersen [10] reports formulas for the mass ow rates through the orice in terms of the driving
pressures, and gas temperature, for both subsonic and sonic ow in the orice. These formulas are substituted
into the aforementioned pressure transient relationship. The resulting equations that dene the pressure transient in the chamber for subsonic and sonic ow in the orice are summarized below for convenience.
For sonic ow into the hot lling chamber at temperature TF and an initial pressure PG (on the upstream
side of the solenoid valve), we have the pressure PF in the chamber versus time given by
P F t P 0

asonic t;
PG
PG

46a

where
asonic

0:231cp S S C S

VF

r
T F mmol
R

and

PF
6 rthreshold :
PG

46b

The lling gas is characterized by a heat capacity at constant pressure, cp, and a molecular weight mmol; the
solenoid valve has the phenomenological parameters SS, the eective cross-sectional area, and CS, the orice
discharge coecient.
When the pressure in the lling chamber PF/PG increases above rthreshold, the ow through the solenoid
valve becomes subsonic, then the pressure in the lling chamber versus time becomes


 
P F t
P0
sin arcsin
47a
asubsonic t ;
PG
PG
where
asubsonic

0:24cp S S C S

VF

r
T F mmol
;
R

and

PF
> rthreshold :
PG

47b

Once the ow becomes incompressible, Eq. (47a) is no longer applicable. In particular, Eq. (47a) can no longer
be used when the argument of the sine function exceeds p/2. However, the relationship is accurate for the
range of pulse times used in this application.
4.2. Lumped element for the diaphragm
The diaphragm is acted upon by the pressure dierence across its two faces, as illustrated in Fig. 5b. There
is the driving air pressure from the lling chamber PF on one side, and the back pressure of the uid PD
against the diaphragm on the other. We choose to model the equation of motion for the diaphragm in terms

154

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

xD

Diaphragm Parameters
kD, SD
Diaphragm
Stop

xD
PF
PD
SD

Main
Chamber

Fig. 5b. Parameters used to model diaphragm physics.

of the static displacement xD of the center of the diaphragm, essentially Hooks Law. We will ignore the inertia
of the diaphragm in its dynamics. We know that Roark [11] for a thin, clamped diaphragm of area, SD, the
center displacement, in term of the pressure dierential across the diaphragm, is
k D xD P F  P D S D ;

48

where
kD

E
16s3
1  m2 d 2D

49

The constants E and m, are the modulus of elasticity and Poissons ratio, respectively, of the diaphragm material (in our case titanium.) The constants s and dD are the diaphragm thickness (on the order of .010 in.) and
diameter (on the order of 1.00 in.), respectively. Again, using the formulas provided by Roark [11], we can
relate the center displacement, xD, to the volume, VD(xD), displaced by the diaphragm as
V D xD 1=3S D xD :
The volume ow rate generated by the diaphragm is then
d
dxD
dxD
Q
QD V D x 1=3S D
; or
3 D:
dt
dt
dt
SD

50

51

The maximum displacement of the diaphragm is DxD.


4.3. Lumped element for the main chamber
We next apply the Bernoulli lumped element, developed in Sections 3.2 and 3.5, to the ow of solder in the
main body. There are two distinct phases of pump operation that need to be analyzed when describing the
dynamics of the ow in the pump as a whole.
A rst phase, which we will call the driving phase, begins when the solenoid valve opens to allow compressed gas into the pneumatic chamber upstream of the diaphragm. As the chamber lls, the diaphragm
deforms, only slightly, due to pressure build-up on the pneumatic side; the diaphragm displaces molten solder
in the body of the pump. During this phase, the ow in the pump body splits into two streams, one leads to the
reservoir, and the other to the dispensing nozzles, as shown in Fig. 5c. The solder meniscus in the nozzles, see
Fig. 5f, begins its displacement from its resting position at the top entrance to the nozzles. The uid motion is
therefore generally downward in the nozzles, and generally upward in the throttle and the reservoir. The rst
phase persists until the diaphragm hits the stop, which controls its maximum displacement; its motion halts
virtually instantaneously. If we ignore the possibility of cavitation, then this occurrence has the eect of instantaneously coupling the two streams, which were previously independent.
In a second phase, which we will call the retraction phase, and which begins after the diaphragm hits the
stop, the capillary forces in the nozzles work against inertial, gravitational, and viscous forces to eventually
force the solder upward, until the solder interfaces rise to the top of the channels. A single eective stream
is sucient to describe the ow in the pump body during this period. The ow associated with this stream
could be either downward or upward immediately after the diaphragm impact, depending on the throttle
and nozzle parameters.

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


Port T bot

PD , S Dtop , K Dtop , Z Dtop

155

PTbot , ST , K Tbot , Z Tbot , QT

lDT

hDT
lNT

Port D, SD QD

hNT

dD

PD , S Dbot , K Dbot , Z Dbot

l DN top

hDN top

top
top
top
Port N top PN , ( N S N ) , K N , Z N , ( N QN )

Fig. 5c. Parameters used to characterize lumped element for the main pump chamber.

In light of the preceding discussion, we will develop lumped elements for the main chamber that describe
the transient evolution of the system during the two distinct phases.
4.4. Flow in the main chamber driving phase
We apply the two-port lumped Bernoulli equation, Eq. (44), to the ow in the main chamber. During the
driving phase of the dispense process we shall ignore the mutual interactions between the ports corresponding
to the throttle entrance and the nozzle entrances. As discussed earlier, the treatment is equivalent to apportioning the pump body into two two-port devices, each beginning at the diaphragm port D, which we use
as the reference port. In our treatment, we divide the diaphragm into two eective ports, port Dtop and port
Dbot. The line that divides the diaphragm diameter dD between its top part and its bot is not necessarily in
the middle of the diaphragm, but we will represent it as so. The resulting balances, in the driving phase, are as
follows. For the path from the diaphragm to the bottom of the throttle, lDT,
Diaphragm (port Dtop) to Throttle (port T bot)
!
d
1
K top
K bot
D
eff
bot
bot
T
q M DT QT  q
 2 Q2T qgZ top
52
D  Z T P D  P T :
2
dt
2
S
S top

T
D
For the path from the diaphragm to the nozzles, lDNtop , we treat all of the nozzles as identical and independent,
so that they combine into one eective port with total ow Qtotal
NQN , and total port area S total
NS N ; thus
N
N
bot
top
Diaphragm (port D ) to Nozzles (port N ):
!
d
1
K bot
K top
top
top
N
eff
D
q M DN NQN  q

53
NQN 2 qgZ bot
D  Z N P D  P N :
2
2
dt
2
S bot

NS

N
D
We have chosen positive QT to be up, diaphragm to throttle, and positive QN to be down, diaphragm to
nozzles.
Even though the diaphragm is not necessarily divided evenly, and in fact the division dynamically shifts, the
top
bot
bot
eect on the overall dynamics is negligible. We argue that since K top
D , K T , K D , and K N are all of order 1, and
that, typically
S T  S bot
D ;

and NS N  S top
D :

54

then
K top
D
S top
D

K bot
T
;
S 2T

and

K bot
D
S bot
D

K top
N
NS N

55

156

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


2

top
bot
bot
Thus, we will ignore the terms K top
D =S D , and K D =S D . In addition, if we let
bot
Z top
D  Z T hDT  d D =4;

56a

top
Z bot
D  Z N hDNtop  d D =4;

56b

and

then we may rewrite the lump equations for the driving phase in the main chamber as
!
d
1
K bot
eff
T
q M DT QT   q
Q2T  qghDT  d D =4 P D  P bot
T ;
dt
2
S 2T
and

!
d
1
K top
N
eff
q NM DN QN  q
Q2N qghDNtop  d D =4 P D  P top
N :
dt
2
S 2N

57

58

We perform one last manipulation on the equations above, to put them in a form that will be a bit more useful
for us. We let
S T M eff
DT lDT const:

59

Similarly, we let
NS N M eff
DT lDNtop const:

60

The parameters lDT and lDNtop are characteristic length scales in the pump body associated with the streams
from the diaphragm to the throttle, and from the diaphragm to the nozzle, respectively.
Our new equations are then
!
lDT d
1
K bot
T
Q  q
q
61
Q2T  qghDT  d D =4 P D  P bot
T ;
2
S T dt T
S 2T
and

!
lDNtop d
1
K top
N
Q  q
q
Q2N qghDNtop  d D =4 P D  P top
N :
2
S N dt N
S 2N

62

One can see from these equations, with the attendant approximations used to derive them, that we have removed reference to the detailed position of dividing line across the diaphragm for the two ows in the main
chamber of the pump, and thus eectively decoupled the two streams.
Of course, the mass balance is still enforced. We must take the volume ow rate of the diaphragm QD to be
negative, for ow into main chamber. Conservation of mass for the main chamber is then
QD QT NQN 0:

63

4.5. Flow in the main chamber retraction phase


The ow in the pump body after the diaphragm is halted is signicantly dierent from the ow during forward diaphragm motion. It becomes a single two-port device, with uid moving from the nozzles to the throttle or vice versa. The change in ow pattern implies that we need a single characteristic length scale
representing the interaction between the two ports. We shall treat the nozzles as the reference port. Eq.
(44) then implies that
Nozzle (port Ntop) to Throttle (port Tbot):
!
d
1
K top
K bot
top
N
eff
bot
bot
T
q M NT QT q
 2 Q2T qgZ top
64
N  Z T P N  P T
2
dt
2
S
NS N
T

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

157

We dene
lNT  S T M eff
NT ;

top
and hNT  Z bot
T  ZN ;

65

we arrive at the desired result


!
lNT d
1
K top
K bot
bot
N
T
Q q
q
 2 Q2T  qghNT P top
N  P T :
2
S T dt T 2
ST
NS N

66

The mass balance equation for the retraction phase in the main chamber is
QT NQN 0:

67

4.6. The lumped element for the throttle


We treat the throttle as an annular conduit of a known inner radius, Ri, outer radius, Ro, and height hT (see
Fig. 5d). The amount of throttle is adjusted by moving a thin, slightly tapered, rod into the throttle neck, eectively increasing or decreasing the radius of the inner cylinder.
If we now employ our general lumped model for a viscous tubular element, but with two lossy ports, we can
write the equation of motion for the throttle as
Throttle (port Tbot) to Throttle (port Ttop):
!
!
toploss
botloss
lT d
1
K bot
QT
1
K top
QT
GT QT
2
T KT
T KT
Q q
q
QT  q
Q2T  qghT 
2
ST
S T dt T 2
S 2T
S 2T
top
P bot
T  P T :

68

We have added a second loss term to Eq. (39) for the bottom throttle port; but, because the positive ow QT is
into the bottom port, instead of out of the port, we must account for a sign change in QT in the argument of
K botloss
.
T
As a model for the frictional term GT(QT), we employ the standard steady state treatment of ow in an
annulus Slattery [9, p. 76]. In this case
 
8
lT
69a
GT QT l 2 QT ;
p Ro k
where
k 1  j4

1  j2
;
lnj

and j

Ri
:
R0

69b

Ro

Ri
top
Port T
PTtop , ST, KTtop , ZTtop

hT =lT

Viscosity
bot
Port T
bot
PT , ST, KTtop , Z Tbot

Fig. 5d. Parameters used to model the annular throttle element.

158

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


Port R PT, SR, KR, ZR,

lR
hR = lR

top
top
top
Port T top PT , ST, KT , ZT

Fig. 5e. Parameters used in the description of the lumped element for the reservoir.

4.7. The lumped element for the reservoir


We apply the lumped Bernoulli equation, Eq. (44) to the reservoir. In this case, the top of the reservoir is a
free surface.
We have (see Fig. 5e)
Throttle (port Ttop) to Reservoir (port R):
!
top
d
1
K
K
R
top
T
q M eff
 2 Q2R qgZ top
70
T  Z R P T  P R :
TR QR q
dt
2
S 2T
SR
Instead of using the volume ow rate QR of the second port, we shall this time use the ow rate QT. However,
we have chosen QT to be positive owing out of the main chamber, so we must attach a negative sign to QT in
the mass balance for the reservoir
QT QR 0:

71

In a similar manner to our previous approximations for the main chamber lumped element, we argue that
S 2R S 2T ;

2
2
or that K top
T =S T K R =S R ;

72

for our reservoir lumped element. In addition, we specify that


hR  Z R  Z top
T ;

and lTR  S T M eff


TR :

73

In addition, at the top surface of the reservoir, the total pressure PR is the sum of that due to the surface tension force, and the reference atmospheric pressure, P0, or possibly an applied bias pressure PB  P0. We
neglect the pressure arising from the surface tension force since the area of the top surface is relatively large.
Thus, we take, PR = (PB  P0) + P0. We let lR be the characteristic length scale for the reservoir. Thus, our
lumped equation for the reservoir is
q

dlR QT 1 K top
q T Q2T  qghR P top
T  P B :
dt
2 ST

74

We make one additional approximation. We observe that the free surface of the reservoir will move imperceptibly during any phase of the dispense; i.e., we will assume that dlR/dt ! 0. In this case, our nal form for the
reservoir Bernoulli lumped element is
!
lR dQT 1
K top
T
q
q
75
Q2T  qghR P top
T  P B :
2
S T dt
S 2T
This equation is valid for either phase of the dispense process.
4.8. The lumped element for a nozzle
Fig. 5f gives the basic picture of the ow of solder in the nozzles. The surface tension in the solder causes a
meniscus to form around a moving contact line. At this contact line, there is always an upward force on the
column of solder above.

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

159

top
Port N
top
top
top
QN , SN , K N , PN , Z N

RN
hN = lN
lN

Contact
Angle

i) Meniscus
Resting Position

Contact
Line
Viscosity

DN

Surface
Tension

bot
Port N
bot
bot
SN , K N = 0 , P0 , Z N

ii) Meniscus
Moving in channel in
the driving and
retraction phases

hresidual
2 DN

iii) Dispense phase

Fig. 5f. Three congurations of the nozzle meniscus and the parameters used to model the annular nozzle element.

The treatment of the ow of the solder in the nozzles, which have length, lN, and diameter DN = 2RN,
requires some nesse. We wish to describe the dispensing process accurately enough to predict the amount
of solder dispensed, but at the same time we do not wish to tackle the very dicult dynamical problem of
the instabilities that must develop to break the solder from the nozzles. (There are many papers on this topic.
Several references include Chaudhary [1315], Jeggers [16,17].)
With this philosophy in mind, we can qualitatively describe the motion of the solder column through the
channels as follows: During the driving phase, the solder meniscus in any particular nozzle moves downward
from the top entrance of the nozzle, as the diaphragm begins to move forward. One port for the nozzle is thus
a free surface and a moving boundary. The meniscus quickly reaches the nozzle exit, and solder dispensing
begins. The exiting solder stream necks, and detaches roughly two nozzle diameters beyond the exit. The diaphragm hits the stop after approximately a millisecond or two time span, and the retraction phase begins; but
the downward uid ow in the nozzle does not necessarily cease at this time. The amount of solder dispensed
is counted (with some arbitrariness) as the amount of uid that passes through a nozzle exit up to that time
that the velocity of the solder column vanishes. After the column stops, the ow in the nozzles soon reverses,
the meniscus moves upward from the nozzle exit to the top. This retraction occurs under the action of surface
tension forces in the absence of the driving diaphragm motion.
One may estimate the maximum error associated with the denition of the term, amount of solder dispensed. From high speed video footage, the quantity of residual solder that draws back into the nozzles is
at most that of a conical sliver attached to the nozzle exit (see Fig. 5f(iii)) of height hsliver  2DN. For
nozzles
D2
100 lm in diameter, an estimate of the maximum error is given by, Estimated Dispense Error
13 p 4N hsliver !
0:13 mm3 , i.e. roughly 820% of the volume dispensed for data reported in this work. This estimate is actually
quite conservative. In practice, the error is generally less than the estimate, since the conical sliver at break o
is normally based in the interior of the nozzle by about one nozzle diameter, rather than at its exit.
We again apply our equation for the tubular viscous lumped element, with a lossy boundary condition on
the top port, and moving boundary on the bottom port. Our reference port is the top of the nozzle. Hence, our
lumped element for the nozzles is (refer to Fig. 5f)


loss
d
1 K top
N K N QN
q lN QN q
Q2N S N qghN  GN QN GR S N P top
N  P 0 :
dt
2
SN

76

We take the kinetic energy boundary term to vanish for the bottom port, since it is a moving boundary. There
is the viscous force GN(QN), the sign of which depends of the direction of ow. This term will oppose the ow,
regardless of the ow direction, since it depends directly on QN. We have Slattery [9, p. 77]

160

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169



8 lN S N
QN :
GN QN l
p
R4N

77

There is also the capillary force GR, or the surface tension force Batchelor [12],
GR 2pRN r cosh:

78

The parameter r is the surface tension. The surface tension force is applied along the contact line of the solid
liquidair interface. (See Fig. 5f) Because the nozzle walls are non-wetting, the contact angle is larger than 90;
the force is always directed upwardly
One may take the term dlN/dt as the velocity of the meniscus. This moving boundary term then takes a
form that is very similar to the kinetic energy term. Since
dlN
Q2
QN ! N :
dt
SN
If we now collect all the terms together, we have for the nozzle lumped element
!
loss
lN d
1 2 1
K top
GR  GN QN
N K N QN
QN q 2 QN q
q
P top
Q2N qghN
N  P 0 :
2
2
SN
S N dt
SN
SN

79

80

To connect the nozzle volume ow rate QN to the dispensed volume, we employ the mass balance equation for
the nozzles. By denition, the volume rate QN in the nozzles is
QN

dV N
;
dt

81

We then say that the volume dispensed, VDispensed, is that volume that exits all nozzles during the time when
the variable lN is greater than the length of the nozzle length LN to the time when the volume ow rate QN
vanishes (changes from positive to negative)
Z time when QN 0
dtNQN :
82
V Dispensed
time when lN >LN

4.9. Assembly of lumped elements for the driving phase


We assemble our equations for the driving phase of the pumping action. We have two paths, one from the
diaphragm to the reservoir, D ! R, where we will thus assemble Eqs. (61), (68), and (75), and a second path,
D ! N, from the diaphragm to the end of the nozzles, where we assemble Eqs. (62) and (80). To assemble the
equations means to eliminate the unknown intermediate pressures and kinetic boundary terms between the
lumps. We aect this assembly by just adding the equations together so that the intermediate pressure and
kinetic terms pair-wise cancel. However, the port loss terms survive. We then obtain two equations, one equation for the ow QT, and another for the ow QN. The resultant equation for the ow QT is
q

lDR d
1 K drv Q
GT QT
QT q T 2 T Q2T  qghDR 
P D  P B ;
2
ST
S T dt
ST

83a

where we dene,
lDR  lDT lT lR ;

83b

hDR  hDT  d D =4 hT hR ;
botloss
K drv
QT K toploss
QT :
T
T QT  K T

83c

If assemble Eqs. (62) and (80), we nd the resultant equation for the ow QN, for the ow path D ! N,
!
lDN d
1
K drv
GR  GN QN
N QN
Q q
q
P D  P 0 ;
84a
Q2N qghND
SN
S N dt N 2
S 2N

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

161

where we dene,
lDN  lDNtop lN ;
hND  hN hDNtop  d D =4;
K drv
N QN

K loss
N QN

and

 2:

84b
84c
84d

These equations are driven by the diaphragm pressure,PD, which is subsequently driven by the air pulse gauge
pressure PF, i.e., from Eq. (48a)
P D P F  k D xD =S D :

85

To close the set of equations for the driving phase, one must include the relationship between xD, and the ow
rate QD from Eq. (51). Hence,
dxD
Q
3
Q NQN :
3 D
dt
SD SD T

86

For the driving phase, we have a system of three coupled rst order non-linear dierential equations, Eqs.
(83a), (84a), and (86) for the three unknowns, QT, QN, and xD.
We remark that the equations for the individual ows are quite transparent in their meaning. They look
very much like that for our general tubular viscous element. Under the assumption of independent ows,
one up through the throttle, and the other down through the nozzles, one may an essence write down these
equations for the ows QT, and QN, by inspection, using an eective cumulative path length, say lDR, for
example. One includes the pressure boundary conditions at the beginning and end of the path, as well as
any viscous losses along the path, and nally a pressure head that results from the vertical change in height
from the beginning port to the ending port.
4.10. Assembly of lumped elements for the retraction phase
To assemble the equations for the retraction phase, we add the dynamical equations along the path N ! R,
using Eqs. (80), (66), (68), and (75). In addition, we must employ the continuity condition for the retraction
phase, namely
QT NQN ;

QD 0:

87a

The assembled dynamic equation is then,


q

lNR d
1
Q qK ret QN Q2N qghNR  Gret QN P B :
S N dt N 2

88a

As with the dynamical equations derived for the driving phase, we see that the nal dynamical equation in the
retraction phase formally resembles a single lumped element, but with eective values for the various dynamic
quantities that depend on the entire uid path, starting from the reference port. We have,
lNR 

NS N
lNT lT lTR ;
ST

88b

and
hNR 

hNT hT hR :

88c

We will ignore the variations of lN and hN, with respect to the much larger quantities of lNR and hNR. In addition, there is an eective K constant for the retraction phase of
!
!
K toploss
QN  K botloss
QN
K loss
2
T
ret
T
N QN
 2 ;
K QN 

88d
2
S 2N
SN
S T =N
and an eective G quantity for the retraction phase,

162

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


ret

G QN 


GT QN GN QN
GR


:
S T =N
SN
SN

88e

With this last equation, we have completed the assembly of all the equations necessary to describe the dispense
process.
5. Solution methodology
The initial value problem of solder dispensing, dened by Eqs. (83), (84), (86), and (88), were solved on a
UNIX workstation with an implicit AdamsMoulton scheme available in the IMSL Library. Given material
properties and geometric information pertaining to the pump, nozzles, diaphragm, and throttle, the model
takes the drive pressure and diaphragm stop location as input and predicts the time dependence of following
quantities:
1.
2.
3.
4.
5.
6.

The
The
The
The
The
The

pressure ll of the pneumatic chamber upstream of the diaphragm.


diaphragm motion.
solder interface location, including the retraction phase.
solder interface velocity, including the retraction phase.
solder velocity in the throttle, including the retraction phase.
amount of solder dispensed.

The numerical solution proceeded in two time intervals, that of the driving phase, and that of the retraction
phase. Each phase has its own set of equations, as we have detailed above. We need to link these equations
across the ow discontinuities caused by the diaphragms abrupt stop. Appendix A shows that the volume
ow rate in the nozzles just after the diaphragm hits the stop, Q
N , is related to the volume ow rate in the
nozzles just before the diaphragm hits the stop, Q
,
by
the
relation
N


Q
N QN  gQD ;

A:5

Q
D

where
is the diaphragm volume ow rate, just before the diaphragm hits the stop. The equation above allows us to determine the starting volume ow rate for the nozzle ow in the retraction, given the nozzle and
diaphragm ows just before the diaphragm hits the stop. In the equation above, there is a constant g, which in
principle depends on ratios of intrinsic length scales. However, in the numerical simulations that we performed, we adjusted the constant, experimentally.
6. Results
6.1. Parametric study
We veried our dynamical model by comparing predictions from the model, applied to a prototype dispenser, with dispensing data gathered from the same dispenser. Typical parameters pertaining to that dispenser are given in Table 1. The results of the study are summarized in Figs. 612.
Fig. 6 shows the calculated pressure transient in the pneumatic chamber, and the calculated diaphragm displacement transient through a single dispensing pulse. The stop is placed at a diaphragm displacement of
approximately .04 mm (1.5 thousandths of an inch), and the diaphragm takes a little over 3 ms to reach
the stop.
Fig. 7 shows the solder interface position in the nozzles as the pulse progresses, and Fig. 8 shows the velocity transient of the uid in the nozzles. The interface accelerates slowly at rst (while the pressure in the pneumatic chamber builds). Rapid acceleration follows until the diaphragm hits the stop. In the featured run, the
interface takes roughly 4 ms to reach the channel exit. A rapid change in velocity is experienced when the diaphragm hits the stop. Since the throttle opening in the featured run is small, the nozzle ow is dominant over
the throttle ow just prior to impact. The opposing momenta of the two streams results in a signicant reduction in the nozzle ow, and a change in direction of the throttle ow, i.e. solder instantaneously begins to move

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

163

Table 1
Typical parametric values for model study
Length of nozzle
Diameter of nozzle
Number of nozzles
Diameter of diaphragm
Diaphragm Spring Constant
Diaphragm stop distance
Throttle length
Throttle gap width
Characteristic length (diaphragm to throttle)
Characteristic length (diaphragm to nozzle)
Characteristic length (nozzle to throttle)
Vertical height (nozzle to diaphragm)
Solder density
Solder viscosity
Solder contact angle
Reservoir height
Filling chamber temperature
Gas pulse driving pressure

lN
DN
N
dD
kD
DxD
lT
R0Ri
lDT
lDN
lNT
hDNtop
q
l
cos(h)
hR
TF
PG

1.0 mm
80 lm
256
20 mm
2.564 N/m
38.1 lm
8.0 mm
130 lm
15 mm
58 mm
66 mm
51 mm
8000 kg/m3
.002 kg/m/s
0.94
15 mm
523 K
4.1165 Pa

1.6

8.00E+05

1.2

Pressure (N /m2)

6.00E+05
1
5.00E+05
0.8
4.00E+05
0.6
3.00E+05

2.00E+05

Pressure

0.4

Diaphragm
Displacement

0.2

Displacement
(mils thousandths of an inch)

1.4

7.00E+05

1.00E+05
0

Time (ms)
Fig. 6. Pressure and diaphragm displacement transients.

from the reservoir back into the body of the pump. Hence, the solder continues to dispense after the diaphragm has stopped moving. However, the capillary forces in the non-wetting nozzles are strong enough to
overcome gravity, and the momentum of the ow, and act to retard the ow. Eventually, the ow rate through
the nozzles reaches zero, and subsequently becomes negative, i.e., the interface begins to retract. The retracting
interface achieves a maximum velocity of about 0.3 m/s before the interface reaches the top of the nozzles, at
which point the capillary forces reduce dramatically (since the diameter of the pump body is much larger than
that of the nozzles), and the interface motion ceases. The impact of throttle cross-section on the peak retraction velocity is discussed later in this section.
Fig. 9 shows the time dependence of the volume of solder that has passed through the nozzle exit. The volume dispensed through the entire pulse, represented by the value at the end of the curve in Fig. 9, is about
1.25 mm3.

164

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

Interface Position in Nozzles (mm)

Interface Position

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0

10

12

14

12

14

Time (ms)
Fig. 7. Interface position transient of uid in nozzles.

1.4

Velocity in Nozzles (m/s)

1.2

Nozzle

1
0.8
0.6
0.4
0.2
0
-0.2

10

-0.4

Time (ms)
Fig. 8. Velocity transient of uid in nozzles.

Fig. 10 shows the eect of widening the throttle opening. The upward uid momentum in the pump due to
ow through the throttle becomes signicantly larger since the throttle presents a smaller resistance to the ow
than in the previous run. Thus, after the diaphragm hits the stop, dispensing ceases, and the solder interface in
the nozzles begins to retract rapidly. The velocity in the nozzles is actually negative immediately after the diaphragm hits the stop. When the interface reaches the top of the nozzles, the uid velocity is greater than 1 m/s.
Hence, there is a greater propensity to draw gas into the pump from the nozzles if the throttle is opened wide.
Fig. 11 shows the velocity transients of the ow through the throttle for two cases in which the solder levels
in the reservoir dier, in this case by 15 mm. To exercise the predictive capability of our formalism, we studied
how the quantity of solder dispensed varies as the height of solder in the reservoir is changed. As the solder height in the reservoir increases, the upward ow through the throttle reduces (since a larger quantity of
solder must be moved), while the ow through the nozzles remains relatively unchanged. Due to the increase

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

165

1.4

Volume (mm3)

1.2
1
0.8
0.6

Solder Volume

0.4
0.2
0

6
8
Time (ms)

10

12

14

Fig. 9. Volume-dispensed transient.

1.00E+00

Velocity in Nozzles (m/s)

5.00E-01

0.00E+00
0

0.5

1.5

2.5

-5.00E-01

Nozzle Velocity (Wide Throttle)


-1.00E+00

-1.50E+00

Time (ms)
Fig. 10. Nozzle velocity for wide throttle.

in mass in the reservoir, and thus an increase in the inertia of the solder, the diaphragm takes longer to hit the
stop and, hence, the total quantity of solder dispensed increases.
Fig. 12 shows the general trend of increasing dispensed solder volume as the solder height increases.
6.2. Comparison of data and simulations
To verify the delity of our dispensing model, experimental data was collected using the prototype apparatus described in Section 2. Measurements were extracted from high speed video of the dispensing operation
linked to the recorded diaphragm motion by means of a trigger, and the operating parameters were varied to
cover a range of data. The model parameters, i.e., the characteristic length scales (the ls, particularly lDR, lDN,
and lNR), and g, were determined by tting the model predictions of
1. The dispensed solder volume.
2. The time at which the interface reached the nozzle exit.
3. The time at which the interface began receding.

166

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


3
2.5

Throttle Velocity (m/s)

Static Head = 90 mm
2

Static Head = 110 mm

1.5
1
0.5
0

-0.5
-1

Time (ms)
Fig. 11. Impact of static head on throttle velocity.

1.6

Volume Dispensed ( mm )

1.8

1.4
1.2
1
0.8
0.6
0.4

Solder Volume Dispensed

0.2
0
90

95

100

105

110

Static Head (mm)


Fig. 12. Solder volume dispensed vs. static head.

Figs. 1315 show dispensed-solder transients for three dierent throttle openings ranging from tight, or
relatively small, to wide, or relatively large.
Fig. 13 presents the results for the intermediate throttle case. The experimental displaced solder volume
(1.17 mm3) compares favorably with the computed value (1.14 mm3). The time at which the interface reaches
the nozzle exit (
3.2 ms), and the time at which the interface begins receding (
5.8 ms) also compare well with
the computed values (3.2 and 5.7 ms respectively.) The comparison of the experimental and predicted values of
these characteristic times is an excellent indicator of the ability of the model to capture the basic physics of the
pumps operation.
Fig. 14 is a similar comparison for the tight throttle case. Fluid issues from the nozzle and disappears
from the eld of view after about 4 ms, after which volume comparisons are not possible. Prior to this time,
however, the volume displaced matches well with data. Also, the time at which the interface reaches the nozzle
exit (
3 ms) compares favorably with the computed value.
Fig. 15 presents the results for the wide throttle case. The experimental data here appear somewhat
scattered. However, the computed time taken for the interface to reach the channel exit is within 10% of

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

167

Volume Dispensed (mm )

1.4
3

1.2
1

Throttle Area = 0.954 mm2


0.8

Solder exits
nozzle

0.6

Solder retracts

0.4

Model Simulation
Experimental Data

0.2
0
0

10

12

Time (ms)
Fig. 13. Experimental verication intermediate throttle position.

Volume Dispensed (mm )

3.5
3
2

Throttle Area = 0.109 mm

2.5
2
1.5

Model Simulation
Experimental

1
0.5
0
0

Time (ms)
Fig. 14. Experimental verication tight throttle position.

the experimental value, the computed time at which the interface begins receding is within 20% of the experimental value, while the computed maximum solder volume displaced is within 30% of the experimental
value.
We remark that, due to the diculties associated with the dimension measurement of tiny drops from video
photographs, the accuracy of the experimental data is somewhat limited. Hence, the aforementioned comparisons seem to indicate that the data are well represented by the model, and that the model can serve as a valuable aid while determining the best operating conditions for solder dispensing. In particular, dispensing is most
consistent when the solder ow stops abruptly immediately after the diaphragm hits the stop. If, instead, the
solder continues to dribble from the nozzles (as would happen if the throttle setting were too tight), occasional scatter of the droplets can result. Conversely, if the throttle is opened too wide, the peak retraction
velocity of the interface may be too large and air may be drawn into the pump at the end of the pulse.

168

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169


1.20E+00

Volume Dispensed (mm3)

1.00E+00

Throttle Area = 1.35 mm2


8.00E-01

6.00E-01

4.00E-01

Model Simulation

2.00E-01

Experimental Data
0.00E+00
0.00E+00

2.00E+00

4.00E+00

6.00E+00

8.00E+00

Time (ms)
Fig. 15. Experimental verication wide throttle position.

7. Conclusions
A mathematical model, employing uid lumped elements, that describes molten solder dispensing from a
pneumatically actuated, diaphragm-driven pump was developed. The model is multi-physical in that captures
the lling of the pneumatic drive chamber, the equation of motion of the diaphragm, and the ow of solder in
the pump. The simulation capability enables predictive tuning of the pump. The model indicates that the
system parameters should be chosen such that a desired amount of solder is rst dispensed, followed soon
after by the diaphragm hitting the stop. The throttle opening should be selected to ensure that the solder ow
is near zero immediately after the diaphragm hits the stop. The interface will then break cleanly, and the
retracting solder proceeds back up the nozzles with a small velocity and acceleration, reducing the propensity
to draw gas into the pump.
Appendix A. Relation of ows across the discontinuity of diaphragm stop
Here we display the continuity conditions of the ow variables at the point in time when the diaphragm hits
the diaphragm stop. At this time there is a step discontinuity in the ow driven by the diaphragm resulting in
step changes in the ow at the other two ports. Let us focus again on the main chamber, which has three ports,
the diaphragm port D, the bottom throttle port T, and the Nozzle ports N. If one integrates Eq. (11),
q

N
X
d
1 Kp
M pq Qq q 2 Q2p qgZ p P p 0; p 1; . . . ; N
dt
2 Sp
q

A:1

over a vanishing small time interval around the time of the diaphragm impact, the only terms that survive are:
Z e
Port D: M DD DQD M DT DQT M DN DQN
dtP D =q 0;
A:2a
e

Port T: M TD DQD M TT DQT M TN DQN 0; and

A:2b

Port N: M ND DQD M NT DQT M NN DQN 0:

A:3c

V. Jairazbhoy, R.C. Stevenson / Applied Mathematical Modelling 32 (2008) 141169

169

Here, DQp is the change in ow rate across any ow discontinuity at port p. Also, In principle, we must retain
the pressure term is the rst equation, since this term could be impulsive.
The change in ow DQD is a known term, an imposed change in ow from the diaphragm. One may then
take these last two equations to nd DQT and DQN in terms of DQD. One nds that jump conditions are




M TN M TD
M TD M TT
det
det
M
M ND
M
M NT
 NN
 DQD ; and DQN
 ND
 DQD :
DQT
A:4
M TT M TN
M TT M TN
det
det
M NT M NN
M NT M NN
We observe that the change in ows across the diaphragm impact time at the second and third ports is
directly proportional to the driving ow.
In the retraction phase, we need to nd the nozzle ow QN just after the diaphragm hit the stop, because we
need it for an initial condition for the ow; that ow does not start from a zero initial condition. We can call

this ow Q
N . Given Eq. (A.4), we know QN depends on the nozzle ow and the diaphragm ow in the driving

phase, just before the diaphragm hits the stop. We can call these ows, Q
N and QD , respectively. Thus, we can
write the jump for the nozzle ow as


Q
N QN  gQD :

A:5

The constant g is essentially a ratio of length scales. In our case, we used experimental data to adjust the constant to make force predictions and data to conform.
References
[1] R. Godin, S. Pearson, R. Lasky, Novel process for solder deposition, SMT Surface Mount Technol. Magazine 11 (1) (1997) 6668.
[2] R. Godin, Metal jetting solders on, Mater. World 5 (8) (1997) 451453.
[3] O. Thomas, Placement of reow solder balls for FC, BGA, wafer-level-CSP, optoelectronic components and MEMS by using solder
jetting method, Proc. SPIE Int. Society Optical Eng. 4831 (2002) 145150.
[4] R. Godin, Flip chip bumping with high speed metal jet technology, National Electron. Packag. Production Conf.Proc. Tech.
Program (West and East) 3 (1999) 14891495.
[5] J.M. Waldvogel, D. Poulikakos, Solidication phenomena in picoliter size solder droplet deposition on a composite substrate, Int.
J. Heat Mass Transfer 40 (2) (1997) 295309.
[6] A. Baggerman, D. Schwarzbach, Solder-jetted eutectic PbSn bumps for ip-chip 0 0 , IEEE Trans. Components, Packag. Manufact.
Technol. Part B: Adv. Packag. 21 (4) (1998) 371381.
[7] R. Stevenson, Z. Yang, V. Jairazbhoy, Transient Bernoulli ow in multi-port uid devices with arbitrary geometry, Appl. Math.
Modelling, in press, doi:10.1016/j.apm.2006.11.002.
[8] D.C. Karnopp, D.L. Margolis, R.C. Rosenberg, System Dynamics, Modeling and Simulation of Dynamic Systems, third ed., Wiley
Interscience, 2000.
[9] J.C. Slattery, Momentum, Energy, and Mass Transfer in Continua, Robert E. Krieger Publishing Co., Huntington, NY, 1981.
[10] B.W. Andersen, The Analysis and Design of Pneumatic Systems, John Wiley, 1967.
[11] J.R. Roark, Formulas for Stress and Strain, McGraw-Hill, 1971, p. 217.
[12] G.K. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, 1992, pp. 6068.
[13] K.C. Chaudhary, L.G. Redekopp, The nonlinear capillary instability of a liquid jet. Part I Theory, J. Fluid Mech. 96 (part 2) (1980)
257274.
[14] K.C. Chaudhary, T. Maxworthy, The nonlinear capillary instability of a liquid jet. Part II. Experiments on jet behaviour before drop
formation, J. Fluid Mech. 96 (part 2) (1980) 275286.
[15] K.C. Chaudhary, T. Maxworthy, The nonlinear capillary instability of a liquid jet. Part III. Experiments on jet behaviour, satellite
drop formation and control, J. Fluid Mech. 96 (part 2) (1980) 287297.
[16] J. Eggers, Universal pinching of 3D axisymmetric free-surface ow, Phys. Rev. Let. 71 (3) (1993) 34583460.
[17] J. Eggers, Theory of drop formation, Phys. Fluids 7 (5) (1995) 941953.

You might also like