Ted Gifford Director – Research Group Schneider National Inc Facility Location Problems At...
-
Upload
arnold-mccoy -
Category
Documents
-
view
216 -
download
1
Transcript of Ted Gifford Director – Research Group Schneider National Inc Facility Location Problems At...
Ted GiffordDirector – Research GroupSchneider National Inc
Facility Location ProblemsAt Schneider Logistics
General Background
• Subsidiary of Schneider National– Largest truckload carrier in North America,
2001 Revenue $2.4 B, 14,000 tractors/drivers, 42,000 trailers
• SLI operates in 38 countries, employs 1,200 associates
• Freight Management Services– Global, Multi-Modal Transportation - Manage $2.5 B purchased transportation
• Engineering Services– Network and Route Design– Facility Location– Supply Chain Engineering
• Payment Services & Bid Management – Pay $7.5 billion third party invoices– Combined Value Auctions
Facility Location Design at SLI
• 12-15 distinct problems / year• 8 – 10 engineers participate, 30 engineers total• Problem size Average Max
– Mfg facilities 2-10 20– DC sites 15-30 50– Customers 200-300 500– Products 5-10
25• Major Industries/Clients
– Consumer Products – KC (US & Europe)– Manufacturing – Otis Elevator (US & Europe)– Automotive - Ford (US & Europe) & GM– Paper products – Polyone– Home Delivery (food) – Schwan’s
Problem Types
PLANTSUPPLIER DC CUSTOMER
P-median / coverage
DC Location:Fixed ChargeMultiple CommoditiesSourcing
Indirect CostsSupplier SourcingTransportation impact
Warehouse: Transportation Cost DC fixed Cost DC Variable Cost
Plant: Union Taxes Wages / Education Transportation Cost Variable Cost
Operational Issues
• Most problems are one-off -- A specific client request which often has unique complications or side constraints.
• Time & budget constraints often argue against sophistication or pursuit of a mathematically elegant solution.
• Limited availability or poor quality of data is often the main challenge and greatest consumer of effort.
• Realistic cost models are often either complex or ill-defined. Estimates of actual costs are highly variable.
• Service requirements are often ambiguous. Cost/service tradeoffs are unclearly specified.
• Forecasts of expected demand exhibit considerable uncertainty.
Current Process
• Access Database / Excel with VB Add-ins– Basic user interface– Navigation per problem taxonomy– Data import & maintenance
• AMPL w/ CPLEX– MIP models
• Set Covering• P-center• Fixed Charge• P-median• Special algorithms
What is the underlyingdistance network?
Spatial(planar or rectilinear
distances)
Single ormultiple
locations?
Rectilineardistances orstraight line?
Single
Rectilinear
Rectilinearlocationproblem(easy)
Fermat-Weberproblem
(numericalmethods -
Excel solver)
Straight Line
Multi-sourceWeber
problem(randomized
search)
Multiple
Graphical(arcs and nodes)
ServiceCoverage or Cost
Objective?Service Cost
Fixedcoveragedistance?
Set CoveringProblem(AMPL,
reductionmethods)
Yes
Minimize maximum distance
Number ofeschelons
Single Multiple
Fixed chargeson locations?Cover all
demands?
Yes Not necessary
MaximumCoveringProblem(AMPL,
LagrangianRelaxation )
P-CenterVertex
Problem(Special
algorithms)
Solutionon nodes?
Nodesonly
Arcs andNodes
AbsoluteP-CenterProblem(Special
algorithms)
Number ofcommodities
Capacitieson locations?
Yes
No
CapacitatedFixed ChgProblem(CAPS,AMPL)
Yes No
UncapacitatedFixed ChgProblem
(AMPL w/Addins)
P-MedianProblem
(AMPL w/Addins)
Single
HierarchicalLocationProblems(AMPL,iterative
menthods)
Multiple
GeneralSupply Chain
Design(All methods)
Given a set of customers with different demands:Choose a set of locations to serve those demands based
on the answers to the following questions
SLI has applied, formulated, and solved models addressing all of these problems
Easy for mostpractical problems
ModeratelyDifficult
Possibly VeryDifficult
Problem Taxonomy - 1
underlying distancenetwork?
Graphical(arcs & nodes)
Service Coverage or CostObjective?Service
Cost
Fixedcoveragedistance?
Set CoveringProblem(AMPL,
reductionmethods)
Yes Min Max distance
Number ofeschelons
Single
Fixedcharges
on locations?
Cover alldemands?
Yes Not necessary
MaximumCoveringProblem(AMPL,
LagrangianRelaxation )
P-CenterVertex
Problem(Special
algorithms)
Solutionon nodes?
Nodesonly
Arcs andNodes
AbsoluteP-CenterProblem(Special
algorithms)
Capacitieson
locations?
Yes
CapacitatedFixed ChgProblem
(CAPS, AMPL)
Yes No
UncapacitatedFixed Chg
Problem (AMPLw/
Addins)
(Emergency Service) (depends on tightness)
Problem Taxonomy - 2Underlyingdistancenetwork?
Spatial(planar or rectilinear
distances)
Single ormultiple
locations?
Rectilinear orstraight linedistance?
Single
Rectilinear
Rectilinearlocationproblem(easy)
Fermat-Weber
(numericalmethods -
Excelsolver)
Straight Line
Multi-sourceWeber
(random-ized
search)
Multiple
Cost
Number ofeschelons
Single Multiple
Fixed chargeson locations?
Number ofcommodities
No
P-MedianProblem
(AMPL w/Addins)
Single
HierarchicalLocationProblems(AMPL,iterative
menthods)
Multiple
GeneralSupplyChainDesign
(Allmethods)
Service Coverage orCost Objective?
Graphicalnodes & arcs
(relaxation close)
(10 %) (90 %)
(agricultural centroid)
Use of off-the-shelf software
• Some experience with CAPS Logic Tools, SLIM 2000, I2 Strategist
• Common difficulties– Underlying model and MIP Formulation not
visible to user– Limited flexibility – rigid parameter options – Often necessary to “trick” the system for
some simple problems– Poor performance for problem structures not
anticipated by the model• Example - CAPS several hours• Excel add-in 15
seconds• High cost per use – not general enough to
handle all problem types
Factors which tend to complicate problems
• Ill-defined criteria • Inconsistent or missing data• Large number of candidate facilities• Single sourcing requirements• Disjunctive constraints• Reverse logistics – container flow back• Echelon skipping• Minimum flow constraints• Piecewise linear (or worse) cost functions
Current Development Activities
• Monte Carlo Simulation – Risk Analysis– Probability Distributions for Cost & Demand
• Constraint Programming– ILOG Solver / Dispatcher /OPL– Local Search Heuristics– Constraint Propagation / Domain Reduction– Index variables over enumerated sets– Special Purpose Algorithms for Global
Constraints• Conversion of AMPL to OPL
– Convert Excel Solver routines– Utilize global & set constraints– Hybrid optimization – Cooperative Solvers