The Theory of Operations Management Presentation

Order Full Rate, Leadtime Variability, and Advance Demand Information in an Assemble-To-Order System Preview • Assembly-to-order system – – – – • Each product is assembled from a set of components, Demand for products following batch Poisson processes, Inventory of each component follows a base-stock policy Replenishment leadtime i.i.d. random variables for each component. Model as a M / G / ∞ input stream X – – queue, driven by a common multiclass batch Poisson Derive the joint queue-length distribution, Order fulfillment performance measure. Model • • • M different components, and F = {1,2,…,m} are the component indices. Customer orders arrive as a stationary Poisson process, {A(t), t>=0}, with rate λ . Order type K: it contains positive units of component in K and 0 units in F\K. – – – • K An order is of type K with probability qk, ∑ K q = 1 K K Type K order stream forms a compound Poisson process with rate λ = q λ K K A type-K order has Q j units for each component j, QK = (Q j , j ∈ K ) has a known discrete distribution. For each component i, the demand process forms a compound Poisson process. Model • • • Demand are filled on a FCFS basis. Demand are backlogged (if one or more components are missing), and are filled on a FCFS basis. Inventory of each component is controlled by an independent base-stock policy, – – – • Where si is the base-stock level for component i For each component i, replenishment leadtimes, Li , are i.i.d., with a cdf of Gi Net inventory at time t, I i ( t ) = s i − X i ( t ), i = 1, … m , where Xi(t) is the number of outstanding orders of component i at time t. Immediate availability of all components needed for an arriving demand as the “off-the-shelf” fill rate. – Off-the-shelf fill rate of component i, f i = P[ X i + Qi ≤ si ] K K – Off-the-shelf fill rate of demand type K, f = P[ X i + Qi ≤ si , ∀i ∈ K ] – Average (over all demand types) off-the-shelf fill rate, f = ∑ q K f K K Performance Analysis X (t ) = ( X 1 (t ),… , X m (t )) • Derive the joint distribution and steady state limit of vector – Each component i, the number of outstanding orders is exactly the number of jobs in service in an M iQi / Gi / ∞ queue with Poisson arrival λi and batch size Qi The m queues are not independent. Given the number of demand arrivals up to t, the X i (t ) s are independent of one another. – – Performance Analysis • • • • • Proposition 1: X (t ) = ( X 1 (t ),… , X m (t )) has a limiting distribution. Derive the generating function of X. In the special case of unit arrival, Qi ≡ 1 , the generating function of X corresponds to a multivariate Poisson distribution. For each i, X i is a Poisson variable with parameter λi i = (Σ K∈ℜi λ K ) i The correlation of the queue is solely induced by the common arrivals. If the proportion of the demand types that require both i and j are very small, the correlation between Xi and Xj is negligible. Level of correlation is independent of the demand rate. Reducing the variability of leadtime or batch sizes will result in a higher correlation among the queue lengths of outstanding jobs. Response-time-based order fill rate 1) 2) 3) 4) 5) f K ( w) is the probability of having all the components ready within w units of time. Di (t , t + u ] := Di (t + u ) − Di (t ) Total number of departures from queue i in (τ ,τ + w) = X i (τ ) + Di (τ ,τ + w] − X i (τ + w) I i (τ ) + { X i (τ ) + Di (τ ,τ + w] − X i (τ + w)} ≥ 0 X i (τ + w) − Di (τ ,τ + w] ≤ si , i ∈ K QiK 6) 7) X i (τ + w) = X (τ ) + ∑1{Lni > w} + X i (τ ,τ + w] w i Demand at n =1 can be supplied by τ + w iff QiK X iw (τ ) + X i (τ ,τ + w] − Di (τ ,τ + w] ≤ si − ∑1{Lni > w}, i ∈ K n =1 Yi := X iw − Yi w Qi ⎡ ⎤ K n f ( w ) = P Y + 1{ L > w } ≤ s , ∀ i ∈ K Order fill rate of type-K demand within time window w, ⎢ i ∑ i ⎥ i n = 1 ⎢ ⎥⎦ ⎣ Mean: ⎛ ⎞ ℑ ℑ E[Yi ] = ⎜⎜ ∑ λ E (Qi ) ⎟⎟ ( i − w) K 8) 9) ⎝ ℑ∈ℜi ⎠ Connection to advance demand information • • Suppose each order arrival epoch is known w time units in advance, where w>0 is a deterministic constant. Suppose a type-K order arrives at , and this information is known at , we can fill this order upon its arrival with probability, • Advance demand information improves the off-the-shelf fill rate: • Compare f AK (0) with that of the modified system, fˆ K (0) , where leadtime is reduced from Qi ⎧ ⎫ ˆf K (0) = P ⎪⎨ X w + 1[ Lˆn > 0] ≤ s , ∀i ∈ K ⎪⎬ ≤ f K ( w) ∑ i i i n =0 ⎪⎩ ⎪⎭ K • Li to Lˆi = [ Li − w]+ Knowing demand in advance (by w time units) is more effective, in terms of order fill rate, than reducing the supply leadtime of components.