Integrating RTN scheduling models with ISA-95 standard. Pedro M. Castro Ignacio E. Grossmann Iiro Harjunkoski

Transcription

1 Integrating RTN scheduling models with ISA-95 standard Pedro M. Castro Ignacio E. Grossmann Iiro Harjunkoski

2 Motivation EWO aims to simultaneously account for KPI across multiple business units Integration of supply chain management, production control, planning & scheduling Need to efficiently transfer data and information between different systems Focus on production management system and scheduling solution ISA-95 standard can act as data-exchange platform Goal: Integrate with generic scheduling framework to cover wide variety of problems March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 2

3 Generic scheduling framework Resource-Task Network formulation (Pantelides 94) Key success drivers Modeling paradigm easily understood by business stakeholders Flexible approach, easily modified when new information becomes available Discrete-time representation Better at finding good solutions in short computational time Processing times pp ii (h) rounded up to multiples of δδ ττ ii = pp ii /δδ (# time slots) Parameter used by model δ uniform slot size (time units) T -3 T -2 T -1 t= T time points ft 1 ft 2 ft 3 ft 4... ft T -2 ft T -1 ft T time of each time point is known a priori ffff tt = (tt 1) δδ 0 H Challenge Develop RTN models that map ISA-95 information Need for RTN Handbook Time horizon March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 3

4 Key model entities: Tasks (ii) Characterized by: Extent variables Binary NN ii,tt Continuous ξξ ii,tt Structural parameters (from production recipe) Binary interaction μμ rr,ii,θθ Continuous interaction νν rr,ii,θθ Arrow indicates production and/or consumption Relative time index θθ Measure unit: # time slots Events at start of task: θθ = 0 Events at end of task: θθ = ττ Intermediate events: θθ {1,, ττ 1} Fixed amount μμ AA,ii,0 = = 160 μμ BB,ii,0 = = 240 Variable amount, fixed duration D A B Equipment (U) μμ UU,ii,0 = 1 μμ UU,ii,ττ = 1 Task ii νν DD,ii,0 = 1 νν EE,ii,ττ = 1 Variable amount & duration Task ii 1 μμ FF,ii2,3 = Task ii 2 μμ FF,ii3,4 = 500 Task ii 3 NN ii,tt = 1 Task ii μμ FF,ii1,2 = 300 E F 1 C Batch size=400 kg μμ CC,ii,ττ = = 400 NN ii,tt = 1 Batch size [300,500] kg ξξ ii,tt [300,500] NN ii1,tt + NN ii2,tt + NN ii3,tt = 1 Size (kg) Duration March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 4

5 Key model entities: Resources (rr) Characterized by: Excess resource variables Continuous RR rr,tt Unit assigned to task RR rr,tt = 0 Idle unit RR rr,tt = 1 Parameters Initial availability RR rr 0 Typically RR rr 0 = 1 (equipment units) RR rr 0 = nn (if there are nn equivalent units) Time-dependent availability (dedicated storage capacity) RR mmmmmm mmmmmm rr,tt RR rr,tt RR rr,tt Shared storage (multiple materials rr simultaneously in unit uu) vv mmmmmm uu mmmmmm rr RR rr,tt vv uu External interactions ππ rr,tt Supplied to (+) & removed from (-) system Shared storage (one material at a time) Define one storage task per material All consume unit uu (RR uu 0 = 1) Hide material resources rr temporarily (one time slot) νν AA,iiAA,0 = 1 νν CC,iiCC,0 = 1 RR AA,tt = RR BB,tt = RR BB,tt = 0 tt A νν AA,iiAA,1 = 1 B νν CC,iiCC,1 = 1 C Storage Task ii AA Storage Task ii BB Storage Task ii CC ττ = 1 Task/resource interactions Excess resource balances U μμ UU,iiAA,0 = 1 μμ UU,iiAA,1 = 1 μμ UU,iiCC,0 = 1 μμ UU,iiCC,1 = 1 RR rr,tt = RR 0 ττ rr tt=1 + RR rr,tt 1 + ii ii θθ=0(μμrr,ii,θθ NN ii,tt θθ + νν rr,ii,θθ ξξ ii,tt θθ ) + ππ rr,tt rr, tt March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 5

6 Sequence dependent changeovers Need to disaggregate equipment resource One for each operation mode U,A U Replaced by U,B U,C Include easy clean (e.g. A to A) in processing task Avoids the need for clean and dirty states Same mode can immediately follow Deduct easy clean from real values Takes advantage of intermediate events Real values Time A B C A B C RM U,A A-B U,B θθ = ττ AA cccc AA AA Make A A-A θθ = 0 θθ = ττ AA A RM Make B B B-B Model values Time A B C A B C March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 6

7 Time-based dependencies Define auxiliary timing variables Starting time TTTT ii = tt ffff tt NN ii,tt Ending time TTTT ii = tt (ffff tt + ττ ii δδ) NN ii,tt Assumes execution of a single instance of task ii during time horizon ANSI/ISA , Figure B-6 No constraint needed TTTT BB = TTTT AA TTTT BB TTTT AA TTTT BB TTTT AA + TT TTTT BB TTTT AA + TT TTTT BB TTee AA TTTT AA TTee BB TTTT BB = TTee AA TTTT BB TTee AA TTTT BB TTee AA + TT TTTT BB TTee AA + TT March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 7

8 Minimizing makespan Option 1 (Maravelias & Grossmann 03) Iterative procedure (more efficient) Start with short horizon (infeasible) 1 Optimize auxiliary objective E.g. Maximize production Increase # slots until feasibility 1 st solution is makespan optimal LB= δδ #Batches Iterations Makespan CPUs 5 (1,1,1) (2,2,2) (3,3,3) (3,3,3) (3,3,3) (5,5,5) (5,5,5) Option 2 (Castro 01) Define sufficiently long horizon Minimize starting time of last task Consumes all product demand A B C 400 μμ AA,ii,0 = Last task ii LP relaxation Lower bound (LB) Returns solutions before optimum δδ #Batches Makespan CPUs 5 (1,1,1) (2,2,2) (3,3,3) (3,3,3) (3,3,3) (5,5,5) (5,5,5) March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 8

9 RTN in C# Gantt scheduling program Import files from XML files with ISA-95 info RTN added to list of heuristics/algorithms Select tuning parameter δδ Makespan strategy Pharma problem with 40 orders; δδ=5 min; Makespan=27 h; 95 CPUs March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 9

10 Conclusions Guidelines for modeling scheduling problems with RTN discrete-time formulations Makespan minimization can be challenging Efficient handling of sequence dependent changeovers RTN model integrated with ISA-95 standard New feature in ABB s Gantt scheduling program Rule-based heuristics (e.g., ASAP) Important features still missing Time-dependent interactions Intermediate events of a task (beyond start and finish) Interaction with system boundaries (e.g. electricity prices) Pre-emptive tasks Key for completing after the weekend (e.g., maintenance) March 15, 2017 Enterprise-Wide Optimization Meeting: ABB Project Overview 10