Scheduling in PSE:Before and After the State-Task Network
Pedro M. Castro
His most successful articles
July 18, 2019 2In Honor of Professor Roger Sargent
Number of citations per source
July 18, 2019 3In Honor of Professor Roger Sargent
Web of Science
49
Scopus
Google Scholar
#5 of all time CACE#3 excluding reviews
Number of citing articles from each country
July 18, 2019 4In Honor of Professor Roger Sargent
39 countries
4624 are mine#6 in authorโs list
#1-Ignacio#2-Christos
Impact of articles citing seminal paper
July 18, 2019 5In Honor of Professor Roger Sargent
Total citations CACE articles in 2018: 13,224
Before the STNโข Review article of Rippin (CACE โ93)
โ General considerationsโข โCustomersโ requirements will be more specific and more demanding
in terms of specification, quality and delivery, requiring manufacturers to be much more flexible- a goal thatโฆ may be more readily achieved in batch rather than in continuous production.โ
โข โAdditional requirements in batch operationโฆ allocation of production tasks to equipment items and the sequencing and timingof the production of products through the plant.โ
โข โโฆthe treatment of variable time and capacity requirement for tasks as a function of equipment or batch size is by now relatively routine.โ
โ Short-term scheduling as a type of batch processing problemโข โIt seems likely that future systems will provide a synthesis between
algorithmic and heuristic methodsโฆ The best balanceโฆ will depend upon the nature of the particular batch processing environment.โ
โข Trend : โcomparison between exact algorithms and heuristicsโ
July 18, 2019 6In Honor of Professor Roger Sargent
Multiproduct plant design & scheduling
โข Network flowshop configuration (Birewar & Grossmann โ90)โ Multiple products with same recipeโ Task-unit allocation to be decided by
optimization (same for all products)โข Possible to use equipment in parallel
โ Merging of tasks is possibleโข Same for all products
โ Continuous-time MINLP modelsolved with DICOPT
โข Upgrade of single product approach (Yeh & Reklaitis โ87)
โ Batch size dependentprocessing times
โ Heuristic solution procedure
July 18, 2019 7In Honor of Professor Roger Sargent
Mixing1
Reaction2
Crystallization3
Drying4
Multiproduct scheduling with storage
โข Flowshop plant (Ku & Karimi โ88)
โ Single unit per stageโ Same product sequence
in all unitsโข Permutation schedules
โ Continuous-time MILP modelโข Heuristic solution approach preferred
โ Storage policy affects the makespanโข Unlimited intermediate storage (UIS)โข Finite intermediate storage (FIS)โข No intermediate storage (NIS)
July 18, 2019 8In Honor of Professor Roger Sargent
Unit/stage1
Unit/stage2
Unit/stage4
Unit/stage3
Storage unit
107
UIS
111
NIS
FIS
107
Scheduling in a multipurpose plantโข Precedence network
structure (Rich & Prokopakis โ86)
โ Product ๐๐ obtained from one or more products ๐๐โข Stoichiometric factors
โ Amount of ๐๐ required for one unit of ๐๐
โ Product-unit assignment is knownโ Aggregate task per product ๐๐
โข Variable quantitiesโ Multiple runs ๐๐, ๐๐๐๐,๐๐ batches
with fixed size/time (๐ก๐ก๐๐)
โ Multiple intermediate due datesโ Continuous-time MILP with
general precedence variablesโข Big-M disjunctive constraints
July 18, 2019 9In Honor of Professor Roger Sargent
5
1
2
3 4
Unit #2
Unit #1
Unit
Unit
Unit
๐๐๐๐,๐๐๐ โ ๐๐๐๐,๐๐ + ๐๐ 1 โ ๐ฆ๐ฆ๐๐,๐๐,๐๐,๐๐๐ โฅ ๐๐๐๐,๐๐๐ก๐ก๐๐๐๐๐๐,๐๐ โ ๐๐๐๐,๐๐๐ + ๐๐ ๏ฟฝ ๐ฆ๐ฆ๐๐,๐๐,๐๐,๐๐๐ โฅ ๐๐๐๐,๐๐๐๐ก๐ก๐๐
Most general algorithm before the STN
โข SRSP program (Egli & Rippin โ86)โ Products with alternative routes
โข Intermediates & final productsโ Units shared by different productsโ Time-dependent resource
consumptionโข Specified for a time period ๐๐ relative
to the start of the taskโข E.g. electricity & steam demand
โ Sequence-dependent times/costsโ Preemption over weekendsโ Multiple intermediate due datesโ Production scheduled on a hourly
basis over 20 daysโข Algorithm, not a mathematical model
โ Ready for re-scheduling
July 18, 2019 10In Honor of Professor Roger Sargent
Seminal article of the State-Task Network
July 18, 2019 11In Honor of Professor Roger Sargent
AbstractA general framework for handling a wide range of scheduling problems arising in multiproduct/multipurpose batch chemical plants is presented.Batch processes involving a variety of complexities are represented using a state-task network. The novel feature of this representation is that both the individual batch operations (โtasksโ) and the feedstocks, intermediate and final products (โstatesโ) are included explicitly as network nodes. Processes involving sharing of raw materials and intermediates, batch splitting and mixing and recycles of material, can be represented unambiguously as such networks.The short-term scheduling problem is formulated as a mixed integer linear program (MILP) based on a discrete time representation. Flexible equipment allocation, variable batch sizes and mixed intermediate storage policies involving both dedicated and multipurpose storage vessels are taken into account. Limited availability of raw materials, both at the start and during the time horizon of interest, is accommodated. Product deliveries may take place at any time during the horizon, and the amounts involved may be either fixed or variable. The use of utilities by the various tasks may vary over the task processing time, and may be constant or proportional to the batch size. The availability and/or cost of utilities may vary over the time horizon of interest.The objective function is the maximization of a profit function involving the value of the products, and the cost of raw materials, utilities and material storage.The formulation may result in MILPs involving large numbers of binary variables. Issues pertaining to the efficient solution of these problems are discussed in Part II of this paper.
State-Task Network (STN) (Kondili, Pantelides & Sargent โ93)
โข New representation modelโ Superstructure featuring all
alternatives (Sargent & Gaminibandara โ76; Grossmann & Sargent โ78)โข Removes ambiguities of recipe
networks (Reklaitis โ91)
โ Allows for complex recipes, multiple processing routes, shared intermediates, recyclesโข Material states as circlesโข Tasks as rectangles
โ Process units and task-unit suitability not shown explicitlyโข Other resources also not shown
โ STN not necessarily connected graphs (disjoint sub-graphs)
July 18, 2019 12In Honor of Professor Roger Sargent
State-Task Network (STN) (Kondili, Pantelides & Sargent โ93)
โข New discrete-time MILP formulationโ A first in PSE literatureโ Easy to handle time-dependent profiles/costsโ Drawback
โข โEven the solution of a small example using a state-of-the-art generic MILP solver was found to require substantial amounts of computationโ
July 18, 2019 13In Honor of Professor Roger Sargent
12 |T|-1
t=|T|3 4 |T|-2|T|-3
time pointsft1 ft2 ft3 ft4 ... ft|T|ft|T|-1ft|T|-2
time of each time point is known a priori
ฮด
...
uniform slot size (time units)
๐๐๐ ๐ ,๐ก๐ก = ๐๐๐ ๐ ,๐ก๐กโ1 + ๏ฟฝ๐๐
๏ฟฝ๐๐
๏ฟฝ๐๐=0
๐๐๐๐
(๏ฟฝฬ ๏ฟฝ๐๐๐,๐ ๐ ,๐๐๐ต๐ต๐๐,๐๐,๐ก๐กโ๐๐ โ ๐๐๐๐,๐ ๐ ,๐๐๐ต๐ต๐๐,๐๐,๐ก๐กโ๐๐) + ๐ ๐ ๐ ๐ ,๐ก๐ก โ ๐ท๐ท๐ ๐ ,๐ก๐ก โ๐ ๐ , ๐ก๐กMaterial balances (multiperiod)
Consumption Batch size Raw-material supply & product demand
Material state availability Production
๏ฟฝ๐๐๐
๏ฟฝ๐ก๐ก๐=๐ก๐ก
๐ก๐ก+๐๐๐๐โ1
๐๐๐๐๐,๐๐,๐ก๐ก๐ โ 1 โค ๐๐ 1 โ๐๐๐๐,๐๐,๐ก๐ก โ๐๐, ๐๐, ๐ก๐กEquipment allocation constraints (Big-M)Assigns start of task ๐๐ to unit ๐๐ time ๐ก๐กProcessing time
๏ฟฝ๐๐
๏ฟฝ๐ก๐ก๐=๐ก๐ก
๐ก๐กโ๐๐๐๐+1
๐๐๐๐,๐๐,๐ก๐ก๐ โค 1 โ๐๐, ๐ก๐กFewer & tighter constraints (Shah, Pantelides & Sargent โ93)
After the STN: representation improvements
โข mSTN (Barbosa-Pรณvoa & Macchietto โ94)
โ Design and schedulingโข Explicit location of material states to
identify connection between unitsโข Unambiguous representation of
recipe/flowsheet/transfer information
โข Resource-Task Network (Pantelides โ94)
โ Unified treatment of production resources (states, units, etc.)
โ Tasks pre-assigned to unitsโข Multiple tasks for alternative units
โ Structural parameters linktasks & resourcesโข Excess resource balances
July 18, 2019 14In Honor of Professor Roger Sargent
Hh_C1Cast_Gg_CC1
Duration=154 min
Hh
PW ENCC1
Hhยด_C1 Hhยด
After the STN: handling of timeโข Periodic scheduling
(Shah, Pantelides & Sargent โ93)โ Wrap-around operator
โข Continuous-time, single grid (Zhang & Sargent โ94)โ Mockus & Reklaitis (โ95)โ Schilling & Pantelides (โ97)
โข Also handles continuous tasksโ Castro et al. (โ01, โ04)โ Maravelias & Grossmann (โ03)
โข Continuous-time, multiple time grids(Ierapetritou & Floudas โ98)โ Giannelos & Georgiadis (โ02)
July 18, 2019 15In Honor of Professor Roger Sargent
12 |T|-2 |T|-1
slot 1
3
time slot 2 slot |T|-2 slot |T|-1
event points t=|T|
T1 T2 T3 T|T|-2 T|T|-1 T|T|
timing variables to be determined by optimization
Sucessful applications in industryโข Discrete-time RTN model
โ Dow Chemical(Wassick โ09, Wassick & Ferrio โ11)โข Drumming facility for 15
businessesโข Liquid-waste treatment network
โ ABB (Castro et al. โ09)โข Handling time-dependent
electricity costs in a cement plant
โ Johnson Controls(Rawlings et al. โ18)โข Heat recovery system providing
heating and cooling to Stanford University
July 18, 2019 16In Honor of Professor Roger Sargent
individual silo
multiple silos
individual silo
grinding mill
grinding mill
grinding mill
Final stage of cement manufacturing
cement clinker
cement grade
cement grade
cement grade
Storage
truck
boat
train
Highlights of contributions in schedulingโข State-Task Network process representation
โ Engineers can solve a scheduling problem without knowing about mixed-integer linear programming
โข Same concept of process simulators
July 18, 2019 17In Honor of Professor Roger Sargent
โข STN-based discrete-time formulationโ Covers a wide variety of scheduling problemsโ Recommended approach when facing a new problem
โข Essentially the same formulation
โข STN-based continuous-time formulationโ Far less applicable than its discrete-time counterpart
โข Significant developments have occurred since seminal paper
โข Still, each real-life scheduling problem has its own specific constraints, so we are not done just yetโ Generalized Disjunctive Programming
helps to derive computationally efficient formulations