Modeling 6.1 Development of Non-Linear Model: Tuning of Injection Molding Machine
Modeling 6.1 Development of Non-Linear Model: Tuning of Injection Molding Machine
CHAPTER 6
include such difficulties as a time delay "e ts#, saturation and valve dead $one. %ecause of the difficulty involved in solving general non linear e!uations, the e!uations were put into block diagram form and then modeled using Matlab&s 'imulink program. (sing 'imulink, the simulated response of the injection molding model could be compared with the actual response of the injection molding machine.
Q= Cq
../ 0here - is a coefficient which includes the fluid mass density and the area of the restrictor as such: -1/a2. P is the pressure differential between the inlet and the outlet of the valve. *n the case of the pump flow valve, P1"3sup 3line#. *t is assumed that the supply pressure from the pump, 3 sup, is constant, The final term, -! is a flow coefficient which accounts for valve specific losses, and must be found e,perimentally for each valve design. +ere in this work e,perimental data from the valve manufacturer was used to find a
DEPARTMENT OF MACHANICAL ENGINEERING
piece wise continuous function which described the response of the valve to input voltage. The term -! is taken as 4nom for simulink modeling where 4nom accounts for valve losses and should be found for each valve design from manufacturer data. 5s per manufacture manual the valve has been designed so that when it is operated for 67879, the response of the valve is relatively linear, when operation outside of this range, the values become very non linear. )or input voltages less than /.:/ volts there is no flow through the valve, causing a dead $one effect to occur, and for voltages greater than ;./ volts, the valve is wide open, and there is no change in the output of the valve. 5 plot of for which, is in )igure ../.
The final e!uation for the flow valve "not including the transmission lines# becomes:
Q1= Q nom P sup P1 C
..2
DEPARTMENT OF MACHANICAL ENGINEERING
P P sup 1 C
..6 *n addition to the time delay due to the hydraulic lines, the appro,imate volume of the lines was also found inductively by measuring the length and outside diameter of the lines, an appro,imation of the *.=. was
DEPARTMENT OF MACHANICAL ENGINEERING
made from the measured >.=. )or fi,tures other than straight pipe, the appro,imate total length of the piece was used as the length. (sing these appro,imated length and diameter values, the estimated volume of the hydraulic lines was calculated. Table / 'hows all of these calculations.
dV1 V dP Q1 Q2 = + l dt dt
..4 0here is %ulk Modulus Term specified by the volume of the fluid divide by its bulk modulus,
P= Vl
(Q Q )
1 2
... There is a pressure relief valve installed in the pump which is discussed in section <.2.2. This valve is modeled in flow model as a saturation of the pressure in the line. The saturation is designed so that pressures below $ero are e!ual to $ero, and any pressure above the ma,imum pressure set by the relief valve remains at the ma,imum pressure.
*ncluding the pressure relief valve, the final e!uation used to model the pressure in the lines is:
P= (Q1 Q2 ) saturation Vl
..:
6.'.. Orifi$e
The hydraulic fluid flows from the hydraulic lines into the cylinders, but what is physically happening in that connection is unknown. %etween the hydraulic line and the actuator there is a small connector restriction that causes a pressure drop between the line pressure and pressure in the actuator. This unknown pressure drop or restrictor is modeled as an orifice. The e!uation for an orifice comes from %ernoulli's e!uation for steady, incompressible pipe flow:
P
..9
v2 ++ gz = const 2
0hen the energy on both sides of the orifice are e!uated, the e!uation for the orifice becomes:
2 2 P P v v2 2 1 +1 + g z + + g 2 z2 1 1 2 2
= headloss
...
DEPARTMENT OF MACHANICAL ENGINEERING
P P 1 2
v2 v 21 = 2
../@
(sing the minor head loss due to a change in area, with flow substituted for velocity to give:
P P 1 2 Q2 2 A2
= K
..// 0here, A is the loss coefficient, 4 is the flow through the orifice, and 5 is the area of the orifice. 'olving this for 4, the e!uation for the flow through the orifice becomes:
A 2 Q= P P 1 2 K
6.'./ A$t#ator
5 schematic of the actuator can be seen in )igure <.. located in section <.../. There are two cylinders, which work together to transfer the hydraulic pressure into injection ram motion. )rom the figure, it can be seen that as volume 5 is filled, the ram moves forward, injecting polymer into the mold. 0hen this is reversed and pressure is placed on side %, and side 5 is released to the Tank, the ram moves backwards. *n addition to this mechanism for moving the ram, there was also a hydraulic motor "not shown here# which rotated the auger when the polymer is being plasticised. This work did not involve the plasticising of the polymer to be injected, the hydraulic motor will not be analy$ed here. To determine the pressure which was created in the actuator, a balance of flow in the front cylinder was done. The total flow into the cylinder, 42, would have to e!ual the sum of the change in volume, the change in pressure and the flow leaking past the seals as such:
dV (x ) Va (x ) dP a Q2 = a + + C1 P a dt dt
or
V ( x )dP a Q2 =Aa vram + a + C1 P a dt
../4
0here 5a is the area of the actuator cylinder, vram is the velocity of the ram, 0here 8a is the volume of the actuator cylinder "this term is a function of the ram position#. 5lso is the bulk modulus and - / is the leakage
coefficient, which is modeled as constant. 'olving this e!uation for the derivative of the pressure in the actuator, 3a, gives:
dP dVa ( x ) a =Q2 C 1 P a dt dt Va ( x )
or
../<
dP a (Q2 = Aa vram C1 P a ) dt Va (x )
*ntegrating the entire e!uation leads to an e!uation for the pressure in the actuator:
P a =
dVa ( x ) C 1 P Q2 a Va ( x ) dt
or
../.
P a =
Va ( x )
(Q2 Aa vram C 1 Pa )
as
The ram position during the injection phase of the molding cycle is found using a force balance e!uation: F1ma. *n this case, the sum of the forces is F1)actuator )friction, which results in:
P A Ff dV = a a dt m
../:
of which two were identified and modeled in the simulation were: /. The directional nature of coulomb friction poses a problem because, depending on the direction of the movement of the ram, the coulomb friction acting on it will be of a constant amplitude, but opposite sign, causing a discontinuity at $ero velocity. The nonlinear 'imulink models accounts for this by taking the force calculated for the friction and multiplying it by the sign of the velocity. 2. 5t very low velocities the force needed to overcome the static friction is higher than the kinetic frictional force, stick slip frictional behavior can cause problems such as chattering or speed hunting. =.,Cheng. 0!1 Modeled this stick slip type friction. +is model implies that for small velocities "less than =v# the system can be assumed to have no movement. The velocity re!uired for the object to begin moving is called the breakaway velocity "=v#, and once that is reached, the friction force is reduced, as static friction is no longer acting. The 'imulink model incorporates this idea by
DEPARTMENT OF MACHANICAL ENGINEERING
having a function which inputs the current velocity, and the breakaway velocity, if the velocity is less than = v, the output velocity is $ero, otherwise, it is the same as the input velocity. The friction function uses the input velocity to determine if the frictional force should include both the static and kinetic friction terms, or just the kinetic friction term. 6. 5nother part of the friction in a fluid system is the viscous friction. 'ince the velocities are rather low, and polymers heated to their proper temperatures are much less viscous than cool polymers, it was decided that viscous friction should be neglected in the model. The final model for friction was: F f 1 signum"8ram#D)c 0here 8ram is $ero for values less than =v. 3utting this into the e!uation ../::
dx 2 = P signum (vram ) Fc a A a dt 2
../9
../; %y integrating this e!uation "adding an additional pole at $ero in the Eaplace domain transfer function# the e!uation for the ram position was found to be:
x=
..2@
Pa Aa ( signum ( vram ) Fc ) ms 2
%im#lin4 mo el
-+ The ,olta"e to the relie$ ,al,e wa! !et at a con!tant ,olta"e& the !tea%' !tate ,olta"e rea%in" $ro# the pre!! re tran!% cer wa! then recor%e%+ In "eneral the tran!ient re!pon!e o$ the pre!! re loo5e% li5e an n%er6%a#pe% -n% or%er !'!te#& with a ti#e %ela'+ 7+ The pre!! re wa! then increa!e% in incre#ent! which loo!el' corre!pon%e% to a pre!! re increa!e o$ *11 p!i 0a! %eter#ine% b' the #icroproce!!or8! open loop ,olta"e co##an%! to the a#pli$ier4& an% !tea%' !tate ,olta"e wa! recor%e% a"ain+ 9+ Thi! proce!! wa! %one $or both -. an% /. $low rate!& o,er the ran"e o$ 1+:99. to ;+;* .& corre!pon%in" to a pre!! re ran"e o$ *+1<e3 to *+</e< Pa+ The re! lt! o$ the!e te!t! can be !een in Table -+ (sing the results, a plot of the pressure response of the valve from @ ;.; 8 was created. This function was then simplified into two linear continuous functions which were used as the calibration between voltage input and steady state pressure in both the modeling and the e,perimental controller as well. The functions used are as follows:
3lots of the e,perimental results and the lineari$ed functions are found in )igure /:.
)rom the tests run, it was clear that the pressure could be controlled more accurately at low pressures when the flow rate was lower "the lower flow rate in the tests was around @.@@@@< l?sec which corresponded to a 2 8 input#, however, the machine is unable to create pressures above 6.4e. 3a with this low flow rate. 'ince the cavity pressure of the packing phase is normally around 9e. 3a for this si$e mold, the machine could not be operated at such
DEPARTMENT OF MACHANICAL ENGINEERING
a low flow rate. *n order to be able to generate higher pressures, the flow rate had to be set higher, such as the @.@@@. l?sec corresponding to a < 8 input. 5t this higher flow rate, the machine was able to generate pressures as high as those generated with the flow valve completely open "/@ 8 output#, and still have acceptable control of low pressures.
The transient response of the pressure valve was modeled by a second order transfer function, the coefficients of the transfer function were determined empirically using plots of the e,perimental pressure response. The damping coefficient and natural fre!uency chosen from the e,perimental results were tweaked using the final model to make the simulated pressure react in the same way that the e,perimental pressure did.
Simulink model