MANUFACTURI NG P ROCESSES AND P ROCESS CONTROL

David E. Hardt
Laboratory for Manufacturing and Productivity
Massachusetts Institute of Technology
February , 1996

The following paper outlines a basic modeling paradigm for

manufacturing process control. Once the model is defined, three The equipment inputs u are separated as a subset of the parameters
distinctly different modes of process control are described based that are accessible, certain and “manipulable” in a “reasonable”
on this model The model then leads to a control taxonomy for time frame relative to the basic process execution time.
manufacturing processes.
Y = Φ(α, u)

Model Definition The vector α can be further broken down into two categories:

All manufacturing processes have but two outputs: • Material Parameters

• Equipment Parameters
• Geometry (macroscopic shape of the product)
Within both equipment and material parameters, we are interested
• Properties (all intrinsic material properties) in the thermodynamic state and the constitutive properties of
each. For example, the equipment states will be the power pairs:
These two outputs completely define the performance of the force-velocity, pressure-flow, temperature- entropy (or heat flow)
product, and the design specifications that it must meet. voltage-current, and chemical potential-extent of reaction.
Material states can be the same quantities within the material. By
All processes also involve the transformation of material from contrast, the equipment properties govern how and when the
an initial geometry and set of properties to the final outputs. This energy is applied. Thus, the geometric, mechanical, thermal and
transformation is accomplished through the application (or electrical properties of the equipment determine its constitutive
removal) of energy, distributed about the surface or volume of the properties. Constitutive properties for the material are the well-
material. The source of this “directed energy” is the manufacturing known quantities such as stiffness, yield point, melting point,
machine or equipment. Thus, we can first define a manufacturing viscosity, etc.
process as the interaction of equipment with a material t o
transform the material to the desired outputs geometry and
properties. This model is shown in block diagram form in Fig 1.

Outputs
Equipment Material
Energy Geometry
Properties

Fig. 1 The Relationship of Equipment and Material in a

Manufacturing Process

Since all transformations are driven by and governed by the

equipment, the only control over the process (other than changing
the material itself) is through the equipment. Thus, the control
inputs to the process are those equipment inputs that modulate the
intensity and distribution of the energy input to the material. In
other words, during the operation of the process, the only
accessible means of controlled change is the equipment inputs.
This leads to the process model shown in Fig. 2

Machine Outputs
Equipment Material
Inputs Energy Geometry
Properties Fig. 3 Development of A Process Model for Control Y=
process outputs α = process parameters Φ= the
process transformation function. The parameter
Fig. 2 Process Model with Equipment Inputs Shown
vector α is progressively broken down into distur-
bances (Δα) and inputs (u)
To help define internal variables in the process as well as the
inputs and outputs, the basic output causality of the process model
is shown in Fig. 3 using a simple functional relationship between In general, the states of the equipment and material change over
the course of the process as energy is applied, whereas the
the process output vector Y and the parameters of the process α.
constitutive properties tend to remain unchanged. However, the
energy focused on the material often causes significant changes i n
Y = Φ(α) the constitutive properties. Indeed the process outputs as defined
David E. Hardt 2/2/04 1
above can be thought of as the terminal mechanical states and is minimized. (Note that no inherent feedback loop exist here;
constitutive properties for the material. thus changes in Δα will not be directly compensated for.)

Modes of Process Control

There are several ways in which processes are controlled, ranging
from off-line sensitivity reduction to real-time output control. In
all cases, the objective is to minimize the effect of disturbances
(i.e. Δα ) on the output Y . This basic objective is captured by the
first order variation equation: Fig. 4 Optimization of αο to minimize the effect of Δα

∂Y ∂Y
ΔY = Δα + Δu
∂α ∂u Statistical Process Control
where: SPC is actually a process diagnosis tool that tries to determine if
process disturbances (Δα ) that are non-random in nature exist.
Δ Y = variation of the output This is done by examining the statistics of output data sampled
from the process. If such disturbances are found, SPC provides n o
∂Y prescription for action, but implies that the disturbance should be
= disturbance sensitivity of the process
∂α eliminated. This is equivalent to detecting and eliminating Δα
(the disturbances) in the above variation equation. (This obviates
Δα = parameter disturbances taking any control action via Δ u.)

∂Y
= input-output sensitivity or “gain” State Control using Feedback Systems
∂u

Δ u = equipment input changes Since it is parameter variations that are responsible for variations
in the process output, it is advantageous to try to reduce this
variation. A common and very powerful method for reducing
To minimize the basic variation of Y we can: uncertainty is feedback systems, based on direct measurement of
the equipment or material states. Examples of this include
equipment force, velocity, temperature, pressure, or flow control.
• design and operate the process to have low disturbance It may also involve control of material temperatures, pressures or
sensitivities (minimize ∂Y/∂ α ). This is the goal of p r o c e s s displacements, for example. This method is shown in Fig.5
optimization.

• design or control the equipment to minimize parameter

variations Δα This is the goal of S tatistical Process Control
(SPC)

• counteract Δα by appropriate changes in Δ u, most typically

through the use of feedback control to minimize Δ Y over an
appropriate frequency range

Sensitivity and Parameter Optimization

Fig. 5 Feedback Control of Process States.
One means of characterizing the properties of a process is t o
quantify the effect of variations in the parameters on the outputs. There is an important distinction to be made between equipment
This “generalized sensitivity function” takes the matrix form: state and material state control. In the former we have a much
∂Y� closer coupling between the inputs (equipment inputs) and the
controlled variable (an equipment state). When the control i s
∂α� applied to the material state, the equipment and all its static and
dynamic properties are included in the control loop as well as the
If such sensitivity functions are known, the process can then be often uncertain energy interaction “port”
tuned or “optimized” so as to minimize these functions. This
control method is shown schematically in Fig. 4 In this method, we place certain parameters (a subset of the
material states) α e of the system under feedback control to reduce
In this control method the objective is to select nominal values of their variation or to render their values more certain. Thus, the
parameters α ο such that the sensitivity of the output to effect of parameter variations on the outputs:
disturbances
∂Y� ∂Y
Δαe
∂α e
∂α�
David E. Hardt 2/2/04 2
€
is reduced by reducing Δα e . Note that the outputs of the process
and the disturbance effect on these are still outside the control Why are these processes so vastly different in our control
loop. framework; one easily controlled and made responsive, yet slow,
while the other is total unresponsive within a part cycle
framework, yet very fast? The differences are many, and in fact the
only similarity is that each uses mechanical energy on metals.
For starters, machining induces shape change by removing
Process Output Feedback Control material. Forging does so by deforming a fixed mass of material.
Machining uses part -independent tooling that applies the
Ultimately the only means of insuring proper output target values transformation energy to a local region, creating the shape by
and minimizing variation is to use feedback control directly on the moving the tool along a specified trajectory. Forging uses a part
outputs. As shown in Fig. 6, this strategy automatically specific, “shaped tool” that creates the part in a single
encompasses all influences on the processes, provided a true unidirectional stroke of the forging press, applying mechanical
output measurement is available. In practice this is seldom the energy in a highly distributed fashion.
case and the measurement process itself can add time delays and
errors to the process. This simple example suggests at least one means of classifying
processes for control; one that begins with the basic physical
(Note that we have added the measurement process as a separate mechanisms of shape change, and then considers whether the
problem, and indeed this problem can dominate in output control change occurs locally or globally in time. Finally, to determine
problems. ) the physical processes that must be analyzed, modeled, measured
and manipulated, delineating the dominant energy domain of the
OUTPUTS (Y)
Desired MANUFACTURING
material transformation process will be necessary.
CONTROLLER
Process PROCESS
Outputs
PARAMETERS (α)
MATERIAL
State and Properties
MACHINE
Process Classification for Control
State and Properties

DISTURBANCES (Δα)
Based on this discussion, a rudimentary hierarchical system of
Δ MATERIAL classification is suggested and is illustrated in Fig. 7
ΔMACHINE

Transformation Methods
measured or estimated outputs

Fig. 6 Direct Feedback Control of Process Outputs

Serial Interaction Parallel Interaction

Finally, it is clear that many specific factors in the process

physics and processing equipment design affect the ability t o Dominant Energy Domain Dominant Energy Domain
implement each of these types of control. The problem i s
Mechanical Mechanical
discussed in detail in Hardt [1991], but the following helps t o Thermal Thermal
overlay the above framework on the universe of all manufacturing Chemical Chemical
Electrical Electrical
processes.
Fig. 7 Hierarchy of Process Classification for Control

At the head of this hierarchy is the transformation process or the

A Taxonomy for Control of Manufacturing actual mechanism of shape change. Such methods include:
Processes
• Material Removal
The objective of process control is to force the process outputs of • Material Addition
geometry and properties to match certain target values, and t o • Plastic Deformation
create a “responsive” process. Ideally we would simply measure • Solidification
these outputs and adjust the appropriate controls, but most
processes require a further breakdown of these objectives into In each case a workpiece material is altered to create the desired
control “sub problems”. To help define these sub problems, it i s shape, and within each category, all material types can be
first necessary to classify manufacturing processes in a manner included, and different energy sources can be used. In addition, the
that highlights the dominant control issues. transformation energy can be either applied locally and moved in a
serial manner to create or follow the part contour, or applied
For example, both machining and closed-die forging have clear everywhere on the part in parallel. It is clear that the details of the
geometry and property outputs, but what determines these outputs physical phenomena involved will depend upon the material type
for each process is quite different. It is this difference that makes and the energy source used for the transformation. This
machining an easy process to control in real-time and one that can distinction is shown schematically in Fig. 8
respond quickly to command or target changes, but also one with a
rather slow production rate. On the other hand, the geometry For example, a metal can be cut by imposing large local shear
output of closed-die forging is nearly impossible to change in- forces, causing the material to separate at the tool point.
process since complete tooling redevelopment is required each However, we can also cut this material by moving a concentrated
time a new target shape is needed. However, forging is often an heat source along the desired cut line. This heat can in turn come
order of magnitude faster in production rate than machining for from a gas torch, a plasma arc or a laser beam. Concentrated
parts of similar shape complexity.1 shearing is also used to cut large parts from sheet, but rather than
doing so at a point, or in a serial fashion, matched contoured cut-
1
One extreme example of this is the production of high performance
landing gear for commercial aircraft. The part starts as an isothermal the part must be machined, to refine the shape and improve dimensional
forging, which might take a few hours to complete. However, once finished accuracy, but this machining can take over 200 hours!
David E. Hardt 2/2/04 3
ting tools are used to cut (or “stamp”) all portions of the part i n be changed in-process. As a result, this type of process often has
parallel. While all of these processes are clearly “removal” very long time constants for change.
processes, they differ significantly in the other areas of the
classification hierarchy, and these differences will directly affect In some cases there are boundary conditions such as global forces,
upon our ability to control them. temperatures and pressures that can change the relationship
between the tooling shape and the final part shape, and while
these can be changed rapidly, their effect is still global in nature.

v(t) As for control, serial processes typically are more controllable,

because local changes do not affect the global part, and changes i n
time correspond to spatial changes over the extent of the part.
Since all the "action" is easily located in the region of the local
process, measurement is simplified, and models can be developed
a) Serial Process (e.g. laser cutting) that are valid locally without as much concern about the global
accuracy. With serial processes gross changes in geometry can be
effected simply by changing the trajectory and the process i n
K(s) y(t) concert. Little or no fixed tooling is used (except to fixture the
parts prior to processing)

Parallel processes are not well modeled by conventional feedback

control methods, but some research has uncovered useful
application of system theory to certain problems. In most cases,
the only means of process control are statistical in nature and
b)Parallel Process (e.g. sheet stamping) cannot exert significant control authority. Such processes are
typically quite inflexible, since the time constant of change i s
Fig. 8 Classification of Manufacturing Processes Note that very long compared to the part processing time
for the serial process geometry is determined by the
time varying velocity vector while for the parallel In most categories of material transformation methods, such as
process the curvature distribution K(s) is the primary those listed in Table 1, there are both serial and parallel processes.
determinant of geometry As will be demonstrated in each of these categories, our ability t o
effect control will vary in direct relation to whether the process i s
serial or parallel.
Finally, the dominant energy used in the transformation has
One of the best examples of how process design can be used t o
significant impact on the speed and accuracy of the process. In
improve control properties is the recently invented process of
general mechanical and electrical processes are extremely fast and
stereo lithography. (Kodama, 1981, Hull, 1986) This process
these energy forms can be applied to very small or very large
(listed in Table 1 as a solidification/serial/thermal process) is i n
areas, whereas thermal and chemical energy process are diffusive
fact the serial version of more conventional polymer molding
by nature and tend to have long time constants. Processes
processes. By selectively solidifying the material, the need to a
dominated by these domains are typically rate limited by heat
model is eliminated, and the outer shell of the part as it is produced
transfer or reaction rate limits, and even if energy is locally ap-
becomes the mold. This eliminated fixed tooling, and converts
plied it quickly diffuses into the material (e.g. welding).
the process control problem into one of primarily trajectory
control, since the local polymerization process is well behaved
As a first attempt at applying such a classification scheme, many
and can operate well in an open-loop fashion. Other processes
of the common processes in use today are listed in Table 1
such as laser sintering (Deckard, 1989) and 3-D printing (Sachs et
according to this scheme.
al., 1990) are examples of taking parallel process physics and
converting it to a serial process.
Serial and Parallel Processes
Summary
After categorization by transformation method, the serial/parallel
distinction is perhaps the most important from the point of view This paper presents a simple model to help understand the universe
of control since it will have the greatest impact on the of manufacturing processes in the context of their control. By
controllability and time frame of control. Serial processes are defining the function of a process as geometry change of the
those that modify geometry and properties by moving a local workpiece, and delineating the role of the machine or equipment,
process along a prescribed trajectory, making changes i n the actual means of control for any process can be defined. This i s
sequence. These processes include most machining, laser addressed with the "Variation Equation" that illustrates the
processing, welding, rolling and many assembly processes. The respective roles of three dominant process control methods:
set of control problems in serial processes always includes an Statistical Process Control, Process Optimization and Feedback
equipment displacement or trajectory control problem. Also, most Control. Finally, the limits on control are expressed in a process
robotic processes are by definition serial processes, and, although taxonomy that categorizes processes according to how easily and
tooling is present, is it usually not specific to any one part. rapidly the output can be changed by manipulating the equipment.

Parallel processes are those that affect large regions of a

workpiece simultaneously. These processes include all that use
shaped tooling (forming, forging, casting, molding, etching,
ECM or EDM) and additive processes such as powder based, spray
or plating processes. In most cases any “trajectory” that i s
involved is a simple single axis movement and has little effect o n
part geometry. For parts with shaped tooling, the geometry is es-
sentially completely determined by this tooling, and seldom can i t
David E. Hardt 2/2/04 4
References
[Link], C.R., (1989), “Method and Apparatus for Producing
Parts by Selective Sintering”, U.S. Patent # 4,863,538, Sept.
2. Hardt, D.E., "Modeling and Control of Manufacturing
Processes: Getting More Involved”, ASME J. of Dynamic
Systems Measurement and Control, 115, June. 1993, pp 291-
300.
[Link], C.W., (1986), “Apparatus for Production of Three
Dimensional Parts by Stereolithography”, U.S. Patent #
4,575,330.
[Link], H., (1981), “Automatic Method for Fabricating a Three
Dimensional Plastic Model with Photo-Hardening Polymer”,
Rev. Sci. Instrum., 52,m no. 11, Nov.
[Link], E., Cima, M., Cornie, J., Brancazio, D., and Curodeau,
A., (1990), “Three Dimensional Printing: Ceramic Tooling and
Parts Directly from a CAD Model”, Proc. National Conference
on Rapid Prototyping, Dayton, June.

Continued on the next page.............

David E. Hardt 2/2/04 5

 Table 1 Process Taxonomy for Control

Transformation MATERIAL REMOVAL

Control Mode SERIAL PARALLEL
Energy Source Mechanical Thermal Chemical Electrical Mechanical Thermal Chemical Electrical
Cutting Laser ECM EDM Die Stamping ECM EDM
Cutting
Grinding "Flame" Photo-
Cutting lithography
Broaching Plasma Chemical
Cutting Milling
Polishing
Water Jet

Transformation MATERIAL ADDITION

Control Mode SERIAL PARALLEL
Energy Source Mechanical Thermal Chemical Electrical Mechanical Thermal Chemical Electrical
3D Printing Laser Painting E-Beam HIP Sintering Diffusion
Welding Bonding
Ultrasonic Sintering Arc Welding Inertia
Bonding
Welding Resistance
Welding

Transformation MATERIAL FORMATION

Control Mode SERIAL PARALLEL
Energy Source Mechanical Thermal Chemical Electrical Mechanical Thermal Chemical Electrical
Plasma Stereo- Casting LPCVD
Spray lithography
DBM Molding Plating

Transformation MATERIAL DEFORMATION

Control Mode SERIAL PARALLEL

Energy Source Mechanical Thermal Chemical Electrical Mechanical Thermal Chemical Electrical
Bending Line Heating Drawing Magneto-
forming
Forging Forging�
(open die)� (closed-die)�
Rolling�

TABLE KEY
Transformation Method of Geometry Change
Control Mode Local or Global Change
Energy Source Primary Energy form used for Change

David E. Hardt 2/2/04 6