Civil Engineering Reference
In-Depth Information
the usefulness or stability of fixed gain controllers. In these cases adaptive control may be applied to
maintain stability and improve performance.
Adaptive Machine Tool Control
Traditionally the control of a machine tool, such as a grinder, consists of servoing the tool position and
feed velocity. However, there are many important grinding process variables and attributes that affect
part quality. The process forces, applied power, and the material removal rate during grinding must also
be considered to effectively produce parts without damage (as discussed earlier). Another consideration
is that grinding is a noisy and time-varying process that is affected by changing properties and conditions
in the grinding wheel, material, and existing surface profile.
A real-time grinding process model and adaptive controller are needed to provide a robust force controller
design suitable for creating precision surfaces in a timely manner. Control of the normal grinding force will
allow manufacturers to attain higher production rates on precision parts without damage, and give the
ability to grind for finishing with reduced part deviations from deflections induced by force variations.
Recursive parameter estimation techniques employing forgetting and windowing may be used to
integrate multiple sensor inputs and optimize the parameter estimation. The resulting parametric model
is used with a suitable adaptive control scheme to reduce the effects of the identified process variation.
As previously discussed, the basis for modeling a machining operation is a power relationship between
forces, speed, process variables, and the material removal rate (MRR). Most machining processes (milling,
turning, drilling, grinding) have this type of relationship or model for estimating the MRR from other
process variables. This is a necessary step in the design of a successful adaptive process controller.
Adaptive controllers generally follow a similar approach to control a system. Estimation of the current
or future states is used to determine updated controller gains based on either a design calculation or in
reference to a model. Thus adaptive controllers can be divided into two basic types, direct and indirect
methods (Astrom, 1986). Direct and indirect refers to the updated controller gains that are determined.
A common representative of a direct adaptive controller is the model reference adaptive controller
(MRAC). Here a model is used to tune the system gains to perform like the model. In general, the MRAC
updates the controller parameters directly, (no intermediate calculations are required).
The self-tuning regulator (STR) is an indirect method, as a design procedure is used to determine
controllers parameters. The STR uses model parameter estimation to determine current states for use in
the real-time design calculations to adjust the controller parameters. It should be noted that there are
similarities between the MRAC and STR techniques. The widely used explicit MRAC scheme of Landau
(1979) for discrete SISO plants is based on the perfect cancellation of stable plant poles with controller
zeros. The STR is the approach demonstrated in this text. The estimated grinding process model is used
to update the controller design and subsequently select new gains, providing a fast and robust system.
It should be noted that the use of MRAC and other adaptive methods have been successfully applied
in milling and turning. However, the limitations of these systems involve the speed the servo update of
the complete system with a force loop. Indeed, servo update rates for adaptive control have been low. For
example Kolarits (1990) had a servo update rate of only 9 Hz and Fussell (1988) had 8 Hz update, both
for adaptive force control for end milling. Much of the process noise is still visible in these works especially
in geometry transitions. This is probably attributable to the relatively slow force control update rates. New
hardware speeds and the parallel processor designs of the control and estimation architecture permit a
force control loop with a servo rate greater than 500 Hz. This faster force servo loop update rate yielding
better force regulation, than those of previous efforts in milling and turning.
Process models have been successfully utilized in model reference adaptive control schemes in milling
and turning, and as a mechanism for tool wear monitoring (
i.e.
, Tarn and Tomizuka 1989). Both fixed-
gain and parameter adaptive controllers have been implemented to maintain a constant cutting force in
milling and turning. Past research relied on slower computers, which resulted in a slow force control
response. Closed-loop force control system sampling frequencies are generally low (frequencies of less
than 3 Hz are not uncommon (Pein, 1992)).
