Supply Chain Optimization KUBO Mikio Agenda Definition of Supply Chain (SC) and Logistics Decision Levels of SC Classification of Inventory Basic Models in SC Logistics Network Design Inventory Production Planning Vehicle Routing Def. of SCM Council of SCM Professionals Supply chain management encompasses the planning and management of all activities involved in sourcing and procurement, conversion, and all logistics management activities. Importantly, it also includes coordination and collaboration with channel partners, which can be suppliers, intermediaries, third party service providers, and customers. In essence, supply chain management integrates supply and demand management within and across companies. Def. of Logistics Council of SCM Professionals Logistics management is that part of supply chain management that plans, implements, and controls the efficient, effective forward and reverses flow and storage of goods, services and related information between the point of origin and the point of consumption in order to meet customers' requirements. What’s Supply Chain IT(Information Technology+Logistics =Supply Chain Real System, Transactional IT, Analytic IT brain 解析的IT 処理的IT nerve Analytic IT Model+Algorithm= Decision Support System Transactional IT POS, ERP, MRP, DRP… Automatic Information Flow Real System=Truck, Ship, Plant, Product, Machine, … muscle 実システム Levels of Decision Making Strategic Level A year to several years; long-term decision making Analytic IT Tactical Level A week to several months; mid-term decision making Operational Level Transactional IT Real time to several days; short-term decision making Models in Analytic IT Supplier Plant Retailer DC Logistics Network Design Strategic Multi-period Logistics Network Design Tactical Operational Inventory Production Safety stock allocation Inventory policy optimization Lot-sizing Scheduling Transportation Delivery Vehicle Routing Models in Analytic IT Supplier Strategic Plant Retailer DC Logistics Network Design Multi-period Logistics Network Design Tactical Operational Inventory Production Safety stock allocation Inventory policy optimization Lot-sizing Scheduling Transportation Delivery Vehicle Routing Models in Analytic IT Supplier Plant Strategic Retailer DC Logistics Network Design Multi-period Logistics Network Design Tactical Operational Inventory Safety stock allocation Inventory policy optimization Production Lot-sizing Scheduling Transportation Delivery Vehicle Routing Inventory=Blood of Supply Chain Inventory acts as glue connecting optimization systems Supplier Raw material Plant Work-in-process DC Retailer Finished goods Time Classification of Inventory In-transit (pipeline) inventory Inventories that are in-transit of products Trade-off: transportation cost or production speed ->Logistics Network Design (LND) Seasonal inventory Inventories for time-varying (seasonal) demands Trade-off: resource acquisition or overtime cost -> multi-period LND Trade-off: setup cost -> Lot-sizing Cycle inventory Inventories caused by periodic activities Trade-off : transportation (or production) fixed cost -> LND Trade-off: ordering fixed cost-> Economic Ordering Quantity (EOQ) Lot-size inventory Cycle inventories when the speed of demand is not constant Trade-off: fixed cost ->Lot-sizing, multi-period LND Safety inventory Inventories for the demand variability Trade-off: customer service level >Safety Stock Allocation, LND Trade-off: backorder (stock-out) cost ->Inventory Policy Optimization In-transit (pipeline) Inventory Inventory that are in-transit of products Trade-off: transportation cost or transportation/production speed ->optimized in Logistics Network Design (LND) Seasonal Inventory Inventory for time-varying (seasonal) demands Trade-off: resource acquisition or overtime cost -> optimized in multi-period LND Trade-off: setup cost -> optimized in Lot-sizing Demand Resource Upper Bound Period Cycle Inventory Inventory caused by periodic activities Trade-off : transportation fixed cost -> LND Trade-off: ordering fixed cost -> Economic Ordering Quantity (EOQ) Inventory Level demand Cycle Time Lot-size Inventory Cycle inventory when the speed of demand is not constant Trade-off: fixed cost ->Lot-sizing, multi-period LND Time Safety Inventory Inventory for the demand variability Trade-off: customer service level ->Safety Stock Allocation, LND Trade-off: backorder (stock-out) cost ->Inventory Policy Optimization Classification of Inventory Seasonal Inventory Cycle Inventory Lot-size Inventory Safety Inventory In-transit (Pipeline) Inventory Time It’s hard to separate them but… They should be determined separately to optimize the trade-offs Logistics Network Design Decision support in strategic level Total optimization of overall supply chains Example Where should we replenish pars? In which plant or on which production line should we produce products? Where and by which transportation-mode should we transport products? Where should we construct (or close) plants or new distribution centers? Trade-off in Facility Location Model: Number of Warehouses v.s. Number of warehouses 輸送中在庫費用 • • • • • Service lead time ↓ Inventory cost ↑ Overhead cost ↑ Outbound 輸送費用 transportation cost ↓ Inbound transportation cost ↑ Trade-off: In-transit inventory cost v.s. Transportation cost 輸送中在庫費用 In-transit inventory cost 輸送費用 Transportation cost Multi-period logistics network design model Decision support in tactical level An extension of MPS (Master Production System) for production to the Supply Chain Treat the seasonal demand explicitly Demand Period (Month) Trade-off: Overtime v.s. Seasonal Inventory Cost 資源超過ペナルティ 作り置き在庫費用 Overtime penalty Seasonal inventory (残業費) Demand Resource Upper Bound Period Constant Production Inventories Overtime Variable Production Model: MIP+Concave Cost Minimization BOM or Recipie × 3 Safety Stock Cost Warehouses Customer Gropus Plant s Suppliers Product ion Lines Safety Stock Allocation Decision support in tactical level Determine the allocation of safety stocks in the SC for given service levels 安全在庫費用 Safety Stock サービスレベル Service Level +統計的規模の経済 +Statistical Economy of Scale (リスク共同管理) or Risk Pooling Basic Principle of Inventory Economy of scale in statistics: gathering inventories together reduces the total inventory volume. -> Modern supply chain strategies risk pooling delayed differentiation design for logistics Where should we allocate safety stocks to minimize the total safety stock costs so that the customer service level is satisfied. Lead-time and Safety Stock Normal distribution with average demand μ, standard deviation σ Service level (the probability with no lost sales) 95%->safety stock ratio 1.65 Lead-time (the time between ordering and arrival of item) L Max Inv. Volume = L+Safety Stock Ratio L The relation between lead-time and (average,safety,maximum) inventory 3000 2500 2000 Average Max. Safety 1500 1000 500 0 0 5 10 Lead-time 15 20 Safety Stock and Guaranteed Lead-time Guaranteed Lead-time (LT):Each stocking point guarantees to deliver the item to his customer within the guaranteed lead-time Safety stock =2 days 2 Guaranteed LT of upstream point =1 day = Entering LT 1 Guaranteed LT to downstream point =2 days 2 Production time=3 Stocking point An example: Serial multi-stage model Average demand=100 units/day Standard deviation of demand=100 Normal distribution (truncated), Safety stock ratio=1 Guaranteed lead-times of all stocking points =0 Production time 3 days Inventory cost 10$ Safety stock cost 1732 $ 2 days 1day 1day 20$ 30 $ 40 $ 2828 $ 3000 $ 4000 $ Total 11560 $ Optimal Solution Guaranteed LT=3 Entering LT=2 Safety stock=3-(2+1)=0 day Production time 3 days Guaranteed LT 0 day Safety stock cost 1732 $ 2 days 1 day 1 day 2 days 3 days 0 day 0$ 0$ 8000$ Total 9732$ (16% down) Algorithms for Safety Stock Allocation Concave cost minimization using piecewise linear approximation Dynamic programming (DP) for tree networks Metaheuristics (Local Search, Iterated LS, Tabu Search) A Real Example: Ref. Managing the Supply Chain –The Definitive Guide for the Business Professional –by Simchi-Levi, Kaminski,Simchi-Levi 15 x2 37 5 28 Part 4 Malaysia ($180) 37 3 Part 5 Charleston ($12) 58 4 Part7 Denver ($2.5) 29 58 37 8 Part 6 Raleigh ($3) Part 2 Dallas ($0.5) 39 37 15 17 Part 3 Montgomery ($220) Part 1 Dallas ($260) 30 15 15 30 Final Demand N(100,10) Guaranteed LT =30 days 43,508$ (40%Down) What if analysis: Guaranteed LT=15 days ->51,136$ Inventory Policy Optimization Decision support in operational level Determine various parameters for inventory control policies 品切れ費用 Safety Stock 安全在庫費用 Lost Sales Classical Newsboy Model 発注(生産)固定費用 Cycle Inventory サイクル在庫費用 Fixed Ordering Classical Economic Ordering Quantity Model Economic Ordering Quantity (EOQ) Model Given d (items/day): a constant demand rate. Q (items): order quantities. K (yen): a fixed set-up cost of an order. h (yen/day・item): an inventory holding cost per item per day. Find the optimal ordering policy minimizing total ordering and inventory carrying cost over infinite planning horizon. Inventory d Q Cycle Time (T days) Cost over T days = f(T)= Cost per day = Time Find the optimal ordering quantify Minimize f(T) positive So f(T) is convex. By solving f’=0, we get: EOQ (Harris’) formula Newsboy Problem inventory cost backorder (lost sales) cost demand of newspaper (random variable) Distribution function of the demand Density function Expected Value of Total Cost Expected cost when the ordering amount is s: Optimal Solution First-order differentiation: Second-order differentiation : is convex! Base-stock Policy Base stock level=Target of the inventory position Inventory position= In-hand inventory+In-transit inventoryBackorder Base stock policy: Monitoring the inventory position in real time; if it is below the base stock level, order the amount so that it recovers the base stock level Base Stock Policy (Multi Stage Model) n serial inventory stocking points demand point is 1 final supplier is n+1 that has enough inventory Notations (1) time index local stock at the i-th point backorder at the i-th point net inventory at the i-th point Notations (2) inventory on order inventory in transit (transit inventory) Notations (3) inventory ordering position inventory transit position Notations (4) :lead time :demand between time interval (s,t] :base stock level :backorder cost :inventory cost Inventory Flow Conservation Equation base stock level s’i By using ITP’i(t) =>random demand L’i IN’i(t+L’i) Recursive Equation :equilibrium value of stationary demand during lead time Using i+1 i can compute B’ from n+1 to 1. =>cannot compute the opt. base stock levels Echelon Inventory Model :echelon inventory at the i-th point :system backorder :net echelon inventory Echelon Inventory Model Notations (Cont’d) :echelon inventory ordering position :echelon inventory transit position :echelon base stock level at the i-th point Echelon base stock policy: Order the amount so that the inventory ordering position recovers the echelon base stock level. Echelon Inventory Flow Conservation Equation :echelon inventory cost at the i-th point Flow conservation equation for echelon inventory: Recursive Equation :equilibrium value of stationary demand during lead time =>can compute net inventory from n to 1 Objective Function Local inventory model Echelon inventory model Derivation of Optimal Solution (1) :expected cost for 1 to i points when INi+1 is x :expected cost for 1 to i points when INi is x :expected cost for 1 to i points when ITPi is y =>Convex Function Derivation of Optimal Solution (2) expected cost for 1 to i points when INi is x The minimum cost to the i-1st point when the echelon net inventory at the i-th point is x i i-1 =>Linear+Convex=Convex Derivation of Optimal Solution (3) expected cost for 1 to i points when ITPi is y y=ITPi The minimum cost to the i-th point when the echelon net inventory is y- Di =>random demand Di L’i IN => Expectation of convex functions => convex Derivation of Optimal Solution (4) expected cost for 1 to i points when INi+1 is x i+1 i Echelon net inventory Minimum cost when x =INi+1 =y Derivation of Optimal Solution (5) Echelon base stock level: C is convex Since echelon base stock level is non-decreasing, The optimum local base stock level: where is Basic Formula of SCM Is convex Basic formula of SCM (Q,R) and (s,S) Policies If the fixed ordering cost is large, the ordering frequency must be considered explicitly. (Q,R) policy:If the inventory position is below a re-ordering point R, order a fixed quantity Q (s,S) policy:If the inventory position is below a re-ordering point s, order the amount so that it becomes an order-up-to level S Periodic Ordering Policy Check the inventory position periodically; if it is below the base-stock level, order the amount so that it recovers the basestock level Order Mon. Tue. Wed. Thu. Demand Arrival of the order of Mon.(Lead-time=1day) Algorithms for Inv. Policy Opt. Base-stock,(Q,R), and (s,S) policies ->DP Periodic ordering policy -> Infinitesimal Perturbation Analysis During simulation runs, derivatives of the cost function are estimated and are used in non-linear optimization Lot-size Optimization Decision support in tactical level Optimize the trade-off between the set-up cost and the lot-size inventory 段取り費用 Lot-size Inv. Setup Cost 在庫費用 Basic Single Item Model (1) Parameters T : Planning horizon (number of periods) dt : Demand during period t ft : Fixed order (or production set-up) cost ct : Per-unit order (or production) cost ht : Holding cost per unit per period Mt: Upper bound of production (capacity) in period t Basic Single Item Model (2) Variables It : Amount of inventory at the end of period t (initial inventory is zero.) xt : Amount ordered (produced) in period t yt : =1 if xt >0, =0 otherwise (0-1 variable), i.e. , =1 production is positive, =0 otherwise (it is called “setup variable.”) Basic Single Item Model (3) Formulation Lot-sizing (Basic Flow) Model Production x(t) Inventory I(t-1) I(t) t Demand d(t) Week formulation x(t)≦ “Large M” ×y(t) [set-up variable] I(t-1)+x(t) = d(t)+I(t) 0-1 variable Valid Inequality Then the inequality (called the (S,l) inequality) is valid. Valid Inequality,Cut,Facet Inequality of week formulation Facet (valid inequality) Relaxed solution x* Solution x Integer Polyhedron Cut Extended (Strong) Formulation Notations Xst : ratio of the amount produced in period s to satisfy demand in period t ( ) The cost produced in period s to satisfy demand in period t Formulation Facility Location Formulation => Strong formulation; it gives an integer hull of solutions Lot-sizing Model Facility Location Model Ratio of the amount produced in period s to satisfy demand in period t Xst s t Xst ≦y(t) s t Xst = 1 d(t) Extended Formulation and Projection is a formulation of X = Q is an extended formulation of X Facility Location Formulation and Projected Polyhedron Extended Formulation (Facility Location Formulation) Projection Integer Polyhedron of Original Formulation Comparison of Size and Strength Standard Formulation # of var.s # of const.s O(T ) Facility Location Formulation O(T ) Week formulation # of var.s O(T 2 ) # of const.s 2 O(T ) added const.s (S, l) ineq.s O(2T ) cut Strong formulation Strong formulation linear prog. relax. =integer polyhedron T: # of periods Dynamic Programming for the Uncapacitated Problem Upper bound of production (capacity) Mt is large enough. F(j) : Minimum cost over the first j periods (F(0)=0) O(T2) or O(T log T) time algorithm Silver-Meal Heuristics Define: Let t=1. Determine the first period j (>=t) that satisfies: (If such j does not exist, let j=T.) The lot size produced in period t is the total demand from t to j. Let t=j+1 and repeat the process until j=T. Least Unit Cost Heuristics Let t=1. Determine the first period j (>=t) that satisfies: (If such j does not exist, let j=T.) The lot size produced in period t is the total demand from t to j. Let t=j+1 and repeat the process until j=T. Example: Single Item Model Period (day,week,month,hour):1,2,3,4,5 (5 days) setup production Setup cost: 3 $ demand : 5,7,3,6,4 (tons) Inventory cost : 1 $ per day Production cost : 1,1,3,3,3 $ per ton Comparison with ad hoc methods Product at once: setup (3)+production(25)+inventory(20+13+10+4)=75 Just-in-time production:setup(15)+prod.(51)+inv.(0)=66 Optimal production:setup(9)+prod.(33)+inv.(15)=57 Comparison with heuristics Silver-Meal heuristics Determine the lot-size so that the cost per period is minimized. setup(9)+prod.(45)+inventory(7)=61 Least unit cost heuristics Determine the lot-size so that the cost per unit-demand is minimized. setup(9)+prod(51)+inventory(14)=74 Algorithms for Lot-sizing Metaheuristics using MIP solver Relax and Fix Capacity scaling MIP neighbor local search Scheduling Optimization Decision support in operational level Optimization of the allocation of activities (jobs, tasks) over time under finite resources What is the scheduling? Allocation of activities (jobs, tasks) over time Resource constraints. For example, machines, workers, raw material, etc. may be scare resources. Precedence relation. For example., some activities cannot start unless other activities finish. Solution methods for scheduling Myopic heuristics Active schedule generation scheme Non-delay schedule generation scheme Dispatching rules Constraint programming Metaheuristics Vehicle Routing Optimization Customers earliest time latest time Customer Depot waiting time service time Routes service time Algorithms for Vehicle Routing Saving (Clarke-Wright) method Insertion method Guided Local Search Iterated Local Search History of Algorithms for Vehicle Routing Problem Approximate Algorithm Genetic Algorithm AMP Tabu Search Local Search Simulated Annealing Sweep Method Generalized Assignment Construction Method (Saving, Insertion) (Adaptive Memory Programming) Location Based Heuristics Route Selection Heuristics GRASP (Greedy Randomized Adaptive Search Procedure) Exact Algorithm Set Partitioning Approach State Space Relax. Cutting Plane K-Tree Relax. 1970 1980 1990 2000 Hierarchical Building Block Method Conclusion Decision Levels of SC Classification of Inventory Basic Models in SC Logistics Network Design Inventory Production Planning Vehicle Routing