 |
Claims  |
|
|
We claim:
1. In a shop having a number of shop resources for doing jobs by performing
operations on workpieces, each workpiece following a path through at least
two shop resources, said path being specified by a work order schedule
list characteristic of said job and maintained within computing means and
specifying the sequence, location and time of a predetermined set of
operations on a workpiece in a predetermined set of at least two shop
resources, said computing means including a data processing system
comprising WOM means for scheduling operations in at least two shop
resources, based on data received from a resource set of at least two BRO
means, each BRO means being associated with a shop resource, the method of
ordering a work order schedule list for at least one job in a shop, in
which:
for at least one job, said WOM means passes a call specifying an initial
resource operation for said job to an initial relevant BRO sub set of at
least one BRO means for an initial resource operation;
for said initial resource operation at least one BRO means returns a bid to
said WOM means specifying at least one suggested time slot for said
initial resource operation associated with that BRO means, thereby forming
a set of suggested time slots for each resource for said initial resource
operation;
said WOM means selects one bid for said initial resource operation in
accordance with a predetermined strategy, thereby scheduling an operation
time for said selected initial resource operation;
said WOM means then repetitively passes calls to a subsequent relevant BRO
sub set of at least one BRO means for each other operation in said
predetermined set of operations and selects bids returned from said
subsequent relevant BRO set, thereby defining a set of scheduled time
slots for said at least one job; and
said computer means then calculates a completion date for said job,
characterized in that:
said method of ordering a work order schedule, including said steps of
defining a set of scheduled time slots, is performed in a planning mode;
said set of scheduled time slots are contained within a corresponding set
of contract time slots having an extent in time at least as great as said
set of scheduled time slots and a first BRO means reacts in an operations
mode to shop events occurring in its associated shop resource by moving a
scheduled operation time within a corresponding first contract time slot
from an ineligible scheduled time slot to an eligible time slot, whereby
for a first class of shop events said BRO means can adjust the operations
of said associated shop resource without affecting the operations of other
shop resources or of other jobs; and
for shop events having a schedule impact greater than the extent of said
first contract time slot, said system causes that BRO means associated
with the next operation in said work order schedule list to move the
scheduled time slot of said next operation within its corresponding
contract time slot, whereby for a second class of shop events the
execution of other jobs is not affected; and for shop events having a
schedule impact greater than the extent of the contract time slots of
associated BRO means, said system identifies a conflict set of jobs
affected by said shop event and causes said WOM means to pass calls to and
select bids from those BRO means associated with operations in said
conflict set, whereby said system reschedules those jobs affected by said
shop event.
2. A method according to claim 1, further characterized in that: at least
some of said BRO means contain means for maintaining an input queue of
workpieces ready to be worked on and said BRO means selects, in said
operations mode, a next workpiece based on a price function dependent on a
delay quantity representative of the difference between an initially
scheduled finish date and a currently estimated finish date.
3. A method according to claim 1, further characterized in that said system
further comprises assumption means for maintaining a set of current
assumptions about work order schedule lists;
at least some of said BRO means contain means for maintaining an input
queue of workpieces ready to be worked on; and
said BRO means selects, in said operations mode, a next workpiece based on
data obtained from said set of current assumptions.
4. A method according to claim 2, further characterized in that said system
further comprises assumption means for maintaining a set of current
assumptions about work order schedule lists; and
said BRO means selects, in said operations mode, a next workpiece based on
data obtained from said set of current assumptions and included in said
price function.
5. In a shop having a number of shop resources for doing jobs by performing
operations on workpieces, each workpiece following a path through at least
two shop resources, said path being specified by a work order schedule
list characteristic of said job and maintained within computing means and
specifying the sequence, location and time of a predetermined set of
operations on a workpiece in a predetermined set of at least two shop
resources, said computing means including a data processing system
comprising WOM means for scheduling operations in at least two shop
resources, based on data received from a resource set of at least two BRO
means, each BRO means being associated with a shop resource, the method of
ordering a work order schedule list for at least one job in a shop, in
which:
for at least one job, said WOM means passes a call specifying an initial
resource operation for said job to an initial relevant BRO sub set of at
least one BRO means for an initial resource operation;
for said initial resourse operation at least one BRO means returns a bid to
said WOM means specifying at least one suggested time slot for said
initial resource operation associated with that BRO means, thereby forming
a set of suggested time slots for each resourse for said initial resourse
operation;
said WOM means selects one bid for said initial resource operation in
accordance with a predetermined strategy, thereby scheduling an operation
time for said selected initial resource operation;
said WOM means then repetitively passes calls to a subsequent relevant BRO
sub set of at least one BRO means for each other operation in said
predetermined set of operations and selects bids returned from said
subsequent relevant BRO set, thereby defining a set of scheduled time
slots for said at least one job; and
said computer means then calculates a completion date for said job,
characterized in that:
said method includes both a planning mode and an operations mode, in which
planning mode said steps of developing a schedule are performed;
including the steps of scheduling through a bottleneck shop resource in
which a bottleneck operation is performed by first scheduling for each job
having said bottleneck operation an operation time having a bottleneck
start time and a bottleneck finish time through said bottleneck and
subsequently scheduling earlier operations in said work order list to meet
said bottleneck start time and scheduling later operations in said work
order list to begin after said bottleneck finish time; and
said WOM means contains at least two strategies for scheduling bottleneck
operations and selection means for selecting a second strategy in the
event of failure of a first strategy to schedule jobs within a set of
applicable constraints.
6. A method according to claim 5, further characterized in that said system
includes bottleneck identification means for identifying bottleneck shop
resources as a function of scheduled operations above a predetermined
threshold, whereby said WOM performs bottleneck scheduling on a
time-varying set of bottleneck shop resources.
7. A method according to claim 6, further characterized in that said system
further comprises assumption means for maintaining a set of current
assumptions about shop resources and work order schedule lists; and said
bottleneck identification means identifies bottleneck operations using
data obtained from said assumption means.
8. A method according to claim 5, further characterized in that said method
of ordering a work order schedule, including said steps of defining a set
of scheduled time slots, is performed in a planning mode; said set of
scheduled time slots are contained within a corresponding set of contract
time slots having an extent in time at least as great as said set of
scheduled time slots and a first BRO means reacts in an operations mode to
shop events occurring in its associated shop resource by moving a
scheduled operation time within a corresponding first contract time slot
from an ineligible scheduled time slot to an eligible time slot, whereby
for a first class of shop events said BRO means can adjust the operations
of said associated shop resource without affecting the operations of other
shop resources or of other jobs; and for shop events having a schedule
impact greater than the extent of said first contract time slot, said
system causes that BRO means associated with the next operation in said
work order schedule list to move the scheduled time slot of said next
operation within its corresponding contract time slot, whereby for a
second class of shop events the execution of other jobs is not affected;
and for shop events having a schedule impact greater than the extent of
the contract time slots of associated BRO means, said system identifies a
conflict set of jobs affected by said shop event and causes said WOM means
to pass calls to and select bids from those BRO means associated with
operations in said conflict set, whereby said system reschedules those
jobs affected by said shop event.
9. In a shop having a number of shop resources for doing jobs by performing
operations on workpieces, each workpiece following a path through at least
two shop resources, said path being specified by a work order schedule
list characteristic of said job and maintained within computing means and
specifying the sequence, location and time of a predetermined set of
operations on a workpiece in a predetermined set of at least two shop
resources, said computing means including a data processing system
comprising WOM means adapted for scheduling operations in at least two
shop resources, based on data received from a resource set of at least two
BRO means, each BRO means being associated with a shop resource, the
method of ordering a work order schedule list for at least one job in a
shop, in which:
for at least one job, said WOM means passes a call specifying an initial
resource operation for said job to an initial relevant BRO sub set of at
least one BRO means for an initial resource operation;
for said initial resource operation at least one BRO means returns a bid to
said WOM means specifying at least one suggested time slot for said
initial resource operation associated with that BRO means, thereby forming
a set of suggested time slots for each resource for said initial resource
operation;
said WOM means selects one bid for said initial resource operation in
accordance with a predetermined strategy, thereby scheduling an operation
time for said selected initial resource operation;
said WOM means then repetitively passes calls to a subsequent relevant BRO
sub set of at least one BRO means for each other operation in said
predetermined set of operations and selects bids returned from said
subsequent relevant BRO set, thereby defining a set of scheduled time
slots for said at least one job; and
said computer means then calculates a completion date for said job,
characterized in that:
said system generates a time map for said job containing representations of
at least some of said resource operations, including specification of a
start time and an end time and linking information connection sequentially
ordered resource operations, and including at least one path through said
time map from a job start time to a job end time;
said system further adds to said time map resource operation constraints on
said work order schedule list for said job, said constraints having an
uncertainty specified by a lower time and an upper time and being linked
to an associated resource operation;
said system further includes a search routine that examines paths through
said time map and selects the most specific path, having the least
uncertainty, whereby additional information including delays, changes in
resource may be entered into said system by putting more specific times
into said time map; and
said system may extract information by searching for that path through said
time map having the least uncertainty, whereby timerelated information may
be extracted from said system.
10. A method according to claim 9, further characterized in that said
system further comprises assumption means for maintaining a set of current
assumptions about work order schedule lists and said assumption means are
linked to said time map, whereby said time map reflects said set of
current assumptions.
11. A method according to claim 9, further characterized in that said
system includes subordinate time maps, each associated with a shop
resource and containing associated shop resource data and being linked to
a predetermined location in said time map, whereby said search routine
searches from said predetermined location through said subordinate time
map and returns to said time map, thereby reflecting data in said
subordinate time map. |
|
 |
|
 |
Claims |
|
|
|
Description |
|
|
TECHNICAL FIELD
The field of the invention is that of distributed problem solving, in
particular a distributed, capacity constrained scheduling system that
schedules a series of operations on greatly varying jobs typically having
small production quantities and subject to constraints in a finite
capacity environment. The essential functions of the system are to
estimate the potential completion times of work and to subsequently
schedule that work.
BACKGROUND ART
It has been found that a workpiece moving through a job shop, one in which
small orders are scheduled independently of one another, as contrasted
with a production shop in which production runs are much longer and more
repetitive in nature, spends a small fraction of its time actually being
operated on, perhaps 10%. The remainder of the time is spent waiting its
turn for the next operation. The benefits of improvement in inventory
costs and shop throughput are thought to be potentially very large, but
efforts to date have had only limited success in handling the very
difficult computational problems associated with scheduling.
Extensive work has been done on this problem, some of which is summarized
below:
In the Integer Programming approach, characteristics of routes (like
set-ups and lead times) and interactions between routes (like capacity and
lot-size) are represented using integer variables. Though integer programs
provide powerful models, they require detailed information up front and
are computationally intractable for large scheduling problems.
An approach known as Hierarchical Decomposition of Integer Programs is an
extension of linear programming which decomposes the problem into two
stages: a higher level capacity balancing stage and a lower level route
sequencing stage. This approach requires forecasts of capacity
requirements up front and usually assumes fixed lead times. This approach
is based on infinite capacity.
Simulation techniques allow detailed modeling of the shop under dynamic
conditions. They provide a good assessment of release and lot-sizing
strategies but fail to regard the schedule with respect to performance
measures (for instance, tardiness). With no performance measures it is
difficult to assess improvements when making operational decisions.
MRP (Manufacturing Resource Planning) is characterized by infinite capacity
backward scheduling. More recently, extensions have provided incremental
revisions to the plan based on changes in capacity. But MRP does not alter
lead times dynamically based on changes in lot sizes and capacity. This
tends to produce highly inflated lead times. An advantage of MRP for large
manufacturing enterprises is excellent database support and the
corresponding ability to communicate the plan throughout the organization.
Queuing Simulation--Queuing models are good at predicting throughput based
on static forecasts of demand. They do not seem appropriate in shops where
demand and shop conditions are unpredictable.
OPT, a commercial program available from Synchronous Manufacturing, is a
finite scheduling system which focuses on critical shop resources
(typically bottlenecks) to constrain capacity. OPT attempts to decrease
inventories and increase throughput based on changing lead times. However,
OPT cannot be easily extended by its users and thus provides limited `seat
of the pants` decision support.
DISCLOSURE OF INVENTION
The invention relates to an improved shop scheduling system referred to as
the Cooperative Scheduling System (CSS) in which a central routine, the
Work Order Manager (WOM) interacts with a set of subroutines representing
shop resources comprising one or more machines to first set a planned
schedule allowing for finite shop capacity at "bottlenecks" in a planning
mode and then, in an operational mode, to correct and modify the schedule
to accommodate for inevitable delays, machine breakdowns, changes in
priority, etc.
In any shop, the life cycle of an order is characterized by continually
increasing knowledge and changing assumptions about fabrication, capacity,
demand and management priority. Due to this, the execution of promises and
schedules vary from those originally planned. In an experimental shop,
variations can be dramatic since exact fabrication and resource needs are
not known until the job is complete. Planning capability is, of course,
essential to making good promises but it is also necessary to provide
tools to react to conditions which cause variance from the predicted
schedule. Accordingly, the system provides two modes: a planning mode for
predicting schedules and an operational mode for reacting to changing
conditions.
Planning Mode--The planning mode produces an estimated target date for each
work order and operation within the work order and feeds those dates to
the work order tracking system.
The goal in planning is to make a feasible promise based on the customer's
want date, resource capacity, and the throughput desired by the shop. The
essential ingredient to meet this objective is that schedules for each
work order, and consequently its routes, be measured against the goal.
This requires that the "goodness" of each possible schedule for a work
order be measured in some quantifiable term; this quantifiable term is
generally called the objective function and in Cooperative Scheduling it
is the weighted tardiness function. Many functions may be used for this
purpose. The one employed in this embodiment is the sum of the product of
the priority and the square of the number of late days.
Planning mode must produce these estimates as constrained by resource
capacity. In Cooperative Scheduling, the philosophy that throughput should
be throttled by the most critical shop resources is adopted. Planning mode
uses this assumption to constrain schedules by identifying the critical
shop resources and scheduling all routes that pass through them first.
Bottlenecks may be identified by stored data in the database, by operator
input, or by the system itself dynamically classifying resource centers as
bottlenecks in response to the projected load. All other schedules of work
order routes are then constrained by the capacity for the critical
resource for which a schedule is already determined.
Planning mode also considers relationships among work orders. Work orders
which are part of the same project may be combined in work order packages.
These packages are used to determine critical paths in the work order
relationships. This ability then leverages initial priority decisions;
work orders outside the critical path will typically be given a lower
initial priority.
A job is entered along with a sequential list of the operations to be
performed on the workpiece; the list is displayed to an operator along
with a first list of time windows (earliest start-latest completion) for
the various steps that is automatically generated from stored process
information. This information will be derived from a time map to be
described later.
The list may be interactively updated by the operator to accommodate
scheduling through shop bottlenecks by imposing tighter constraints (later
start and earlier end) on the bottleneck operations.
The time windows in the initial analysis are transmitted to another
routine, called a Resource Broker (BRO), that returns at least one "bid"
for each step. A bid is a target start and complete time within the
"window" set by the WOM. The bids are generated with regard to the finite
capacity and previous scheduling commitments of the resources.
The WOM accepts one of the offered times for each step, thereby
establishing the final planned schedule, having some slack time between
the completion date of step n and the start time of step n+1.
Operational Mode--Operational mode uses the weighted tardiness measure
established in the planning mode to support decisions in the short term.
There are two types of decisions which must be supported: releasing
decisions and reacting decisions.
Releasing decisions answer the question: what job should be released next
to this work center? Planning mode produces a list of work predicted for
each work center. At execution time, this list is sequenced according to
the weighted tardiness measure; the route which will contribute to an on
time shop the most goes first.
Reacting decisions are made when new information affecting a planned
schedule is received or when assumptions have changed. Most reacting
decisions involve reordering the list of work to be done in the best way.
Again, the best way is measured according to the weighted tardiness
measure.
In the operations mode, the broker routine reacts to an input delay in the
start time because the part has been affected by a schedule change in an
earlier fabrication step to adjust the start time and still meet the
target start time for the next operation. If this is not possible, the BRO
routine will notify the WOM, which will, in turn, call the next BRO
routine on the schedule for that part to see if it can use its slack to
accommodate the remaining delay. The effect of a delay thus ripples along
the list of steps until it is absorbed or until the system fails to make
adjustment.
The priority decisions as to which of the waiting jobs will be done first
is made by use of the weighted tardiness function described above, which
computes for each waiting job, its completion date and selects the job
that will increase the total delay in passing through the shop by the
least amount.
A feature of this invention is that both the final scheduling decision and
the first attempt at schedule recovery are: "delegated" to the BRO
routines, which may be running on local workstations within an overall
network.
Another feature of the invention is that a time map is generated for each
work order parts and accessed by the WOM in the course of scheduling.
The time map carries the work order scheduling information; is updated to
reflect changing circumstances; and may be accessed by an operator such as
a shop manager and/or a work center foreman to display current information
and to answer "What if" questions that show the consequences of
alternative actions.
Other features and advantages will be apparent from the specification and
claims and from the accompanying drawings which illustrate an embodiment
of the invention.
BRIEF DESCRIPTION OF DRAWINGS
FIG. 1 illustrates an overall block diagram of the system.
FIG. 2 illustrates the different operating modes of the system and the data
flow.
FIG. 3 illustrates the data flow between the WOM and the Broker routine in
planning mode.
FIG. 4 illustrates the flow of the planning mode.
FIG. 5 illustrates a time map.
FIGS. 6a, 6b, and 6c illustrate a simplified time map of a work order.
FIG. 7 illustrates a time-map path and the "window" allocated to different
operations.
FIGS. 8A through 8G illustrate the steps in forming a time map.
FIG. 9 illustrates the process of scheduling an operation.
FIGS. 10-13 illustrate portions of FIG. 9 in more detail.
FIG. 14 illustrates the operation of the Broker routines.
FIGS. 15A and 15B illustrate different routines that respond to a call from
the Work Order Manager.
BEST MODE FOR CARRYING OUT THE INVENTION
Referring to FIG. 1, there is shown a block diagram of the overall system.
The top block, referred to as the master scheduler, is the overall driver
program that communicates with the others. This central point is a
convenient place to enter global parameters such as, the priority of
different classes of jobs, the rules governing priority and both on the
overall shop level and within a smaller group referred to a resource.
Conventionally, the standard priority is to meet the due dates that have
been assigned to an individual job. Other priorities, such as reducing the
amount of work in progress, may also be implemented. Different priorities
and goals may be assigned to different areas of the shop at this central
point.
One of the main functions of this block is to provide control, either
manual or automatic, for problems that cannot be handled by lower levels.
A common example will be those in which the shop is overloaded and not all
due dates can be met. The set of jobs that are affected, i.e., delayed, by
a change to make a high priority job come out on time is referred to as
the conflict set. In a case where a high priority job has bumped other
jobs, the conflict set will be passed to the master scheduler which will
then rearrange and arrange for the rescheduling of the jobs. This may be
done, for example, by setting relative priorities for the jobs and then
passing that information down to the work order managers for automatic
rescheduling in light of those priorities.
This central point is also a convenient place to present to the person
responsible for the shop qualitative information, such as, a rough
estimate of the length of time for a job, so that a customer's inquiry may
be responded to by someone having access to this overall program.
The master scheduler supports the following seven functions in the planning
mode:
1. Setting the Scheduling Horizon, which is the time for which CSS plans
schedules.
2. Identifying critical resources and capacity bottlenecks.
3. Assigning work orders to the proper work order managers.
4. Assigning resources to the proper resource brokers for finite
scheduling.
5. Providing an outlook reflecting the optimism of the shop manager
regarding the ability to meet schedules based on known shop conditions may
be provided.
6. Passing information about appropriate strategies to the work order
managers and resource brokers.
7. Recommending that another shop be used when no capacity exists for
incoming work.
The scheduling horizon will typically be the shop lead time but may be any
period of time which the scheduler chooses. The shop manager may set the
scheduling horizon through a menu provided by the user interface of CSS
with the aid of capacity charts, etc. The outlook reflects the shop
manager's expert opinion about the ability of the shop to meet due dates
generally. The outlook is entered through the user interface as a fraction
between 0 and 1. This outlook is then used in generating schedules as a
measure of optimism. Simply, if the shop manager is pessimistic, the work
order manager will assume it will take longer to do the job.
The shop manager knows about these policies and goals and through his
expertise may know how a work order manager or resource broker should go
about its task to contribute to them. The shop manager can provide this
knowledge through strategies. For instance, if the policy is to reduce
inventories, the master scheduler can impose a "minimize wip" (work in
process) strategy on a work order manager. But since the overall objective
is to get the job out on time, the strategy may be "minimize tardiness
first and then minimize wip."
Strategies are built by the shop manager through the user interface editing
screens and may later be selected dynamically through other screens
provided by the interface. Strategies may be selected globally, per work
order, or per work center as deemed appropriate by the shop manager.
The next level down is that of the work order manager routines, which may
be a single routine operating on different data or may be a set of
different data structures or data objects. For the purposes of this
application, the word "routine" or "means" in the appropriate context will
mean a routine, data structure or object, depending on the language that
is used to implement the invention. In a distributed computing
environment, there may be several routines operating in parallel.
The work order manager has the task of arranging for the scheduling of a
work order, which is a series of tasks to be performed on a part or on a
set of parts. Often, a project or piece of equipment will require a number
of parts that are related in that they must all be finished before the
system can be assembled. The work order lists will be grouped in a work
order package and dealt with as a unit, in order to identify the critical
path for the package, so that some parts do not need to wait longer than
necessary for the others.
The work order manager starts out with an order from a customer for a part
or parts. It accesses the data base to pull out the standard list of tasks
and processes required for such a part (or accepts as input the list of
processes and associates them with standard times. It then sends out
calls, which are a set of parameters, to the relevant resource brokers
which handle the machines that can perform the operations on the standard
list. These parameters specify the desired start date and finish date of
each operation, and also, the standard process time, which the operation
is assumed to take. These dates are set with a large enough margin so
there is a reasonable chance of finding an open slot on a machine within
that window. Each broker then searches the data base to find the
parameters of the part, such as the physical size and so on to select
which of the machines it controls that can be used. The broker then scans
the time reservation list for each machine to find at least one open
machine window, either on one machine or more than one machine. The broker
formulates bids that contain the proposed start and finish times for the
operation and an indication of the cost of this option in terms of
weighted tardiness. Cost accrues both from the tardiness of the operation
to be scheduled and the impact that scheduling this operation has on
previously scheduled work. The broker selects bid windows to be returned
to the WOM according to a local strategy that may not be the same as the
global strategy. Typically, the local strategy is to minimize the weighted
tardiness function which is a calculated quantity which measures the
impact of delay of this part on the total shop throughput. The work order
manager then scans through the list of bids from each broker and selects a
set of award windows (one window for each operation) that will do the job.
The choice will be governed by the global strategy which is usually
getting parts out on time.
The work order manager then allocates the slack time, which is the time
between scheduled job operations, to the different brokers. An example is
illustrated in FIG. 7 showing on the first line a set of five operations
A-E having different bid windows in which they could be performed over a
period of twenty-two days. The actual processing time for each operation
is not indicated on this Figure but will be considerably shorter than the
open windows. The lower line referred to as B illustrates the allocation
of slack, the final window is called the "contract window." This window
provides the broker with information about the earliest and latest times
the operation can be run given the scheduled times of its predecessor and
successor operations. This information is useful in operational mode. In
this case, operation D is a bottleneck operation so that it is extremely
unlikely that D will be performed earlier in time than is indicated on
FIG. 7 and is quite likely to slide to a later time. Thus, the window D'
is skewed to a later time. The window C' is also skewed to a later time,
since it is not likely at all that it will be possible to move operation D
up in time. The significance of this slack allocation will be explained
later. The work order manager also compares the total start and finish
times to the initial assumptions when the order was accepted. Quite often,
it will be found that it is not possible to meet the due date that was
promised. This is a common situation in which the order takers or sales
people tend to be optimistic. In the particular case of bottleneck
operations, they invariably slide and are delayed. If the due date cannot
be met, the simplest thing is simply to report failure, i.e., that the job
will be delayed. As an alternative, the system is capable of having the
work order manager instruct the resource brokers, some or all of them, to
drop low priority jobs from their schedule and then to reschedule the
higher priority jobs.
When an operational event (also referred to as a shop event), such as a
machine breakdown, sick machinist, etc. occurs, there are potentially
hundreds of operations affected. This is the conflict set. The master
scheduler must direct work orders from the conflict set back into CSS for
automatic rescheduling or bring them to the attention of the shop manager.
Project Trade-offs--when a conflict occurs that cannot be handled locally
by a resource broker or WOM and also cannot be rescheduled by the system,
it is likely that there are two projects that together exceed the shop
capacity. In this case, the shop manager will have to intervene to assign
greater priority to one project.
Transactions are the basic information packet transmitted between CSS
components. Transactions contain the identification of the sender and
receiver, a sequence number and other data items required to ensure
communications integrity. Transactions have a "content" and a "type" that
determine their meaning and communicate the intent of the sender to the
receiver. Depending on their content and type, transactions represent
calls, bids, and awards, as well as other queries or assertions
communicated through the system. The transaction types are: initialize,
shutdown, query, assert, receive response, and evaluate.
The representation language and the temporal reasoning utility provide a
language in which queries can be expressed throughout CSS. The format is
similar to the pseudo first order predicate calculus language used in
Prolog. Typical queries are expressed as a list whose first element is a
predicate and whose remaining elements are terms. Inference is performed
on these queries by backward chaining. An important extension to the
reasoning typically done in expert systems is that the queries allowed in
CSS are temporal in nature; they mention and reason about time explicitly,
using the time map representation described below. The query language is
supported at many levels throughout the system: queries can appear in
rules, in program code, or they can be entered through the developer's
interface. Queries also appear as the content of transactions as discussed
above.
The system may use a standalone, local, finite scheduling resource broker
as illustrated in FIG. 15A or a more general rule-based object that
employs an inference engine to apply a set of rules. The advantage of this
latter approach is that artificial intelligence techniques may be used to
translate the expertise of experienced managers to the system.
The system will employ a routine to calculate the weighted tardiness
measure or other figure of merit for the schedule. In the preferred
embodiment, Lagrangian Relaxation techniques, a well known variety of
mathematical modeling, were employed to quantify the effects of capacity
constraints on the schedule. Less coding may be expended if a simpler
technique were used, such as summing the tardiness function calculated
separately for different work orders, rather than using the Lagrangian
technique to calculate an optimum change for the whole set of affected
work orders.
Due date performance is measured by the difference between operation due
dates and scheduled operation completion date. Provision is made to weight
job lateness by the relative importance of the job and to place a higher
price on a job as its lateness increases.
Determination of the price (which is measured in days, rather than dollars)
is directed by four constraints:
1. A precedence constraint forcing work order operations to be performed in
sequence.
2. A resource constraint ensuring that resource usage cannot exceed
capacity limits.
3. A time constraint that requires a work order to arrive at the operation
before the processing of that operation can begin.
4. A time constraint that an operation takes a prescribed amount of time to
complete.
The prices determined by the weighted tardiness function are saved in the
knowledge base so that decisions in the operational mode can be made in
light of these prices. Prices are saved for each resource type and each
work order route.
The WOM can schedule work orders according to different goals:
1. Schedule Around Bottleneck--Bottleneck work centers which the master
scheduler has identified are scheduled first and the rest of the work
order is scheduled around them. To schedule the bottlenecks, the WOM
identifies all unscheduled work orders that go through a bottleneck
workcenter. All of the work orders that pass through the same bottleneck
are gathered together and scheduled through the bottleneck at once using
optimization. Conflicts generated by this process can be automatically
renegotiated, manually renegotiated or left unscheduled. After the
bottleneck operations have been scheduled the remaining operations are
scheduled around them.
2. Reduce Lead Time--Work orders are scheduled to finish as soon as
possible.
3. Reduce Lead Time and WIP--First, the WOM reduces lead time on the job
and then compresses the release times of the upstream operations to reduce
the WIP.
No matter what the goal, the call-bid-award communications work the same
way. Handling of various goals occurs in the selection method invoked when
bids are returned.
When the resource brokers are unable to return bids within the call window
the WOM must generate a recovery action. Options are:
1. Leave the work order unscheduled. If most of the operations have failed
to schedule, the work order should be left unscheduled and brought to the
attention of the shop manager.
2. Commit to the operations which have acceptable bids and leave the
remaining operations unscheduled. This option may be preferable when most
of the operations have received acceptable bids.
3. Renegotiate the failing operations allowing more impact on the schedule.
This option applies to very high priority jobs.
4. Modify or relax constraints. The WOM can ask the master scheduler to
relax the release or want date of the work order. This change may affect
other work orders in the work order package.
A bottleneck resource is defined as any of the following:
a) A resource at or over 100% capacity for a period of time in which
capacity on that resource is demanded by new or ongoing work.
b) A point where many part numbers come together for processing or
assembly.
c) A point from which routes commonly diverge to many different parts of
the shop.
d) A resource which has particular desirable characteristics and is often
in high demand.
Overall capacity is constrained by the capacity of bottleneck resources. In
other words, the shop can never finish work orders at a faster rate than
the bottleneck can finish operations, given that the work order passes
through the bottleneck. The master scheduler analyzes the set of work
orders to be scheduled and the current bottlenecks to determine if any
work orders pass through the bottlenecks or if new bottlenecks are
created. The master scheduler then performs the following steps: a)
packages the bottleneck work orders; b) passes the work order package to a
work order manager which computes quick, roughcut schedules for
non-bottleneck routes in those work orders; and c) selects a strategy to
guide the work order manager in negotiation with the resource brokers
which manage the bottleneck workcenters.
Depending on the configuration of the CSS, the resource broker may do one
of the following: a) add the new operations to the existing set of
bottleneck operations and optimize the entire bottleneck schedule or; b)
provide options to the work order manager which reflect the impact of
adding the new operations and let the master scheduler guide the
scheduling process or; c) finite schedule the bottleneck automatically
with the guidance of the weighted tardiness measure and the knowledge
base.
After the resource brokers schedule the bottlenecks, the master scheduler
then passes the package of work orders to a work order manager to further
schedule the non-bottleneck routes. In this way, the CSS in planning mode
produces a predicted schedule constrained by capacity at the bottleneck.
Once work orders are received by the work order manager, there are two ways
to schedule them:
A pure optimization that chooses among work order scheduling options
through the weighted tardiness function.
A negotiation that consults not only the weighted tardiness function but
also the knowledge base, the resource broker, and possibly the master
scheduler to determine a schedule which satisfies a variety of
constraints.
Option one is provided and selected for speed. It will be useful for
providing estimates for large numbers of work orders at a time in planning
mode and quickly resolving large numbers of conflicts in the operational
mode.
Option two is likely to be used most often. In option two, the work order
manager identifies the resource brokers which manage each of the work
centers that a work order is expected to pass through. The work order
manager writes a contract for each operation in the work order and sends a
"request for bids", also termed a "call", to the resource brokers
identified. The brokers return bids trying to maintain a good work center
schedule. The work order manager selects the best bid from each broker
trying to build a good work order schedule. This process (as well as the
option 1 process) is guided by the weighted tardiness objective. The
call-bid process may go on a number of times before a satisfactory
schedule is determined. At the end of the process, the work order manager
"awards" work to the broker based on the best bids.
Planning mode queues are positioned at work centers and are filled with
awards made during the planning mode. The primary function of the planning
mode queue is to keep predicted schedules in waiting before releasing them
to specific resources in the work center. "Release" means that this job
will be performed next. As up to date information about the work load is
| | |
