4.4: Production-Inventory Planning: Multiple Products · ISE 754, Fall 2026 · in class 27 October 2026
A recap answers what students found most difficult in their pre-class review of the lecture, along with anything the instructor adds. Blue Lecture panels quote the lecture; orange Another view panels look at the same idea from another side.
The questionWhat a switchover cost is, and when it is charged.
“Switching a production system from one product to another requires changes, retooling, recalibrating and revalidating, and cleaning if the process is continuous, and their costs are incurred once the system is set up to produce the product. Unlike the setup cost of Lecture 4.3, incurred in every period a product is produced, a switchover cost is a one-time cost: as long as the system keeps producing that product and no other, it is not incurred again, and the setup carries over.”
“In the following model, the switchover cost is separate from the production cost in the objective function. There are no setup costs in this model; instead, there are switchover costs. They differ from setup costs in that the switchover cost is incurred only in a period when production switches to a different product at a given stage. As long as the same product continues to be produced in the next period, no additional switchover cost is incurred. This differs from a setup cost, which is incurred every period a stage produces, as in Lecture 4.3.”
Example 4 is a plant making two products in three stages, each stage used by one product at a time, over six months. Its switchover costs, and where it starts:
“A switchover costs $2,800, $800 and $400 at stages 1 to 3 for the first product, and $3,200, $600 and $300 for the second. The plant starts with 25 tons of the second product as finished product and every stage set up for the first product, and ends with no inventory.”
“The stages it sets up for them in each month, \(z_{ij1}\) and \(z_{ij2}\), are:”
A row for each stage and a column for each month: \(z_{ij1} = 1\) where stage \(i\) is set up for the first product in month \(j\), and \(z_{ij2} = 1\) where it is set up for the second.
“Determine the switchovers at each stage, and their cost, from the setup indicators.”
“A stage switches over to a product in a month in which it is set up for the product and was not the month before, its setup before month 1 the initial setup: \(v_{ijk} = 1\) where \(z_{ijk} = 1\) and \(z_{i(j-1)k} = 0\), the paid switch of row (g). Every stage starts set up for product 1, so month 1 has no switchover. Reading each stage’s two rows of setup indicators across the months:”
“Stage 1 is set up for product 2 in months 2 and 3 and for product 1 from month 4 on, its last month idle but still set up for product 1: switchovers to product 2 in month 2 and to product 1 in month 4.”
“Stage 2 is set up for product 1 in months 1 and 2, product 2 in month 3, product 1 in months 4 and 5, and product 2 in month 6: switchovers in months 3, 4 and 6.”
“Stage 3 follows stage 2, month for month: switchovers in months 3, 4 and 6.”
“That is eight switchovers for $9,000. Stage 1, whose switchovers cost the most, switches over the fewest times, as in Ex. 3.”
The setup indicators start as Example 4’s: \(z_{ij1}\) for the first product and \(z_{ij2}\) for the second, a row for each stage and a column for each month, with each stage’s setup before month 1 in the start column. Press an indicator to switch that setup on or off. A stage is set up for one product at a time, so switching a stage on for one product in a month switches it off for the other. Below them are the switchover indicators the setups give, \(v_{ijk} = 1\) where \(z_{ijk} = 1\) and \(z_{i(j-1)k} = 0\), each with its cost.
| stage | start | 1 | 2 | 3 | 4 | 5 | 6 |
|---|
| stage | start | 1 | 2 | 3 | 4 | 5 | 6 |
|---|
| stage | 1 | 2 | 3 | 4 | 5 | 6 |
|---|
| stage | 1 | 2 | 3 | 4 | 5 | 6 |
|---|
Product 1 is blue and product 2 orange, as in the lecture’s Fig. 5, and a switchover is outlined in blue, as in Fig. 7; a setup changed from Example 4’s has a red edge. The indicators count switchovers only: whether a set of setups can meet the demand is the question Model 4 answers in Example 4(c).
“What is wanted is another indicator, \(v\), the switchover indicator, which is 1 only in the period the stage switches to a product and starts producing it. In the next period, if the stage continues to produce the same product, no switchover cost is incurred.”
“In order to ensure that a switchover occurs only the first period that production switched to a different product, the constraint \(-v_j + z_j - z_{j-1} \le 0\), row (g) for one stage and product, can be used.”
| \(v_j\) | \(z_j\) | \(z_{j-1}\) | \(-v_j + z_j - z_{j-1}\) | Case |
|---|---|---|---|---|
| 0 | 0 | 0 | 0 | no setup |
| 0 | 0 | 1 | \(-1\) | setup ends |
| 0 | 1 | 0 | 1 | unpaid switch |
| 0 | 1 | 1 | 0 | setup continues |
| 1 | 0 | 0 | \(-1\) | extra cost |
| 1 | 0 | 1 | \(-2\) | extra cost |
| 1 | 1 | 0 | 0 | paid switch |
| 1 | 1 | 1 | \(-1\) | extra cost |
“Two combinations are the ones to check. The one that must be feasible is a stage set up for the product in a period when it was not set up for it the period before, with the switchover indicator 1: \(-1 + 1 - 0 = 0\), which satisfies the constraint. The main one that must not be is the same switch with the switchover indicator 0, a switch with no switchover cost charged: its left side sums to 1, so it is always infeasible.”
“They satisfy the constraint, its left side at most zero, so row (g) lets them through. The model never selects one, because it only minimizes cost and will not add a positive cost to a solution for no benefit.”
“Row (g) looks at \(z_{i(j-1)k}\), the setup in the period before, so for the first period the model has to be told what each stage is currently set up for.”
The second quoted paragraph is about the three extra costs: a switchover paid in a period with no switch.
“Without that cost, a plan would probably switch between the products more frequently; with it, it is often to the plan’s advantage to keep a stage producing the same product a period longer and avoid a switchover, depending on the switchover cost’s relative magnitude. Inventory is what makes that possible: more of a product can be produced while a stage is set up for it, and the inventory then covers its demand over a longer stretch in which it is not produced.”
Example 2 is a plant of three stages making two products over 13 weeks with no switchover cost; Example 3 is the same plant with one, $400, $90 and $50 at stages 1 to 3 for product 1 and $600, $110 and $70 for product 2.
“Counted at each stage, they are 3 switchovers at stage 1, 4 at stage 2 and 6 at stage 3, the fewest where they cost the most.”
“In short, indicator variables require MILP instead of LP, and lumpy resources require indicator variables to include setup and switchover costs. For shared resources, coupling constraints are used to ensure total output does not exceed available capacity, and the model can still be an LP; linking constraints are used to ensure only one activity per period.”
Lecture 4.4, Sec. 1.
Each row of the table is a problem from the two lectures, in the words of their tables of problems, Table 3 in Lecture 4.3 and Table 2 in Lecture 4.4. Press one, and the formulation below it becomes that problem's model. It is one formulation for all of them: the rows every model shares stay plain, the rows the chosen model adds or changes are highlighted, and the rows only other models use are dimmed, each naming the models that use it. Under each of the chosen model's rows are the lecture's own words for it.
| Products | The resource and its capacity | Problem | Solved by |
|---|---|---|---|
| One product, Lecture 4.3 | Set outside the model: \(u_{ij}\) given, \(0\) where unavailable | no setup cost, a storage limit binds (Ex. 4) | linear program, Model 1; the greedy plan breaks the limit |
| no setup cost, Model 2’s assumptions hold (Ex. 3, 7) | greedy procedure, Model 2; its plan is optimal, costing the same as the linear program’s | ||
| Gated inside the model by indicators \(z_{ij}\) | setup cost, stages with capacity (Ex. 5, 6) | mixed-integer linear program, Model 3 | |
| Several products, Lecture 4.4 | None: each product has its own resources | products planned one at a time (Ex. 2 without its linking rows) | LP for each product, Lecture 4.3’s Model 1 |
| Combined: products share a resource’s capacity, by coupling constraints | shared capacity given, no setup cost | LP, Model 1 | |
| trucks and trailers shared, a setup cost for each load (Ex. 1) | MILP, Model 2, setups an integer count | ||
| Exclusive: a resource set up for one product a period, by linking constraints | no switchover cost (Ex. 2) | MILP, Model 3 | |
| switchover cost when a stage changes product (Ex. 3, 4) | MILP, Model 4 |
subject to
| (a) | Flow balance | \(\displaystyle -x_{ij} + x_{(i+1)j} - y_{ij} + y_{i(j+1)} = 0,\) \(i = 1, \dots, m-1,\ j = 1, \dots, t\) | |
| (a) | Flow balance | \(\displaystyle -x_{ijk} + x_{(i+1)jk} - y_{ijk} + y_{i(j+1)k} = 0,\) \(i = 1, \dots, m-1,\ j = 1, \dots, t,\ k = 1, \dots, g\) | |
| (b) | Flow balance | \(\displaystyle -x_{mj} - y_{mj} + y_{m(j+1)} = -d_j,\) \(j = 1, \dots, t\) | |
| (b) | Flow balance | \(\displaystyle -x_{mjk} - y_{mjk} + y_{m(j+1)k} = -d_{jk},\) \(j = 1, \dots, t,\ k = 1, \dots, g\) | |
| (c) | Capacity | \(\displaystyle x_{ij} \le u_{ij},\) \(i = 1, \dots, m,\ j = 1, \dots, t\) | |
| (c) | Capacity | \(\displaystyle x_{ijk} \le u_{ik},\) \(i = 1, \dots, m,\ j = 1, \dots, t,\ k = 1, \dots, g\) | |
| (c) | Capacity | \(\displaystyle x_{ij} \le u_i\, z_{ij},\) \(i = 1, \dots, m,\ j = 1, \dots, t\) | |
| (c) | Capacity | \(\displaystyle x_{ijk} \le u_{ik}\, z_{ijk},\) \(i = 1, \dots, m,\ j = 1, \dots, t,\ k = 1, \dots, g\) | |
| (d) | Initial and final inventory | \(\displaystyle y_{i1} = y_i^0,\) \(i = 1, \dots, m\) \(\displaystyle y_{i(t+1)} = y_i^{t+1},\) \(i = 1, \dots, m\) | |
| (d) | Initial and final inventory | \(\displaystyle y_{i1k} = y_{ik}^0,\) \(i = 1, \dots, m,\ k = 1, \dots, g\) \(\displaystyle y_{i(t+1)k} = y_{ik}^{t+1},\) \(i = 1, \dots, m,\ k = 1, \dots, g\) | |
| (e) | Storage | \(\displaystyle y_{ij} \le u_i^y,\) \(i = 1, \dots, m,\ j = 2, \dots, t\) | |
| (e) | Storage | \(\displaystyle y_{ijk} \le u_{ik}^y,\) \(i = 1, \dots, m,\ j = 2, \dots, t,\ k = 1, \dots, g\) | |
| (f) | Coupling | \(\displaystyle \sum_{k=1}^{g}\, x_{ijk} \le u_{ij},\) \(i = 1, \dots, m,\ j = 1, \dots, t\) | |
| (f) | Coupling | \(\displaystyle \sum_{k=1}^{g}\, x_{ijk} \le u_i\, z_{ij},\) \(i = 1, \dots, m,\ j = 1, \dots, t\) | |
| (f) | Linking | \(\displaystyle \sum_{k=1}^{g}\, z_{ijk} \le 1,\) \(i = 1, \dots, m,\ j = 1, \dots, t\) | |
| (g) | Switchover | \(\displaystyle -v_{i1k} + z_{i1k} \le z_{ik}^0,\) \(i = 1, \dots, m,\ k = 1, \dots, g\) \(\displaystyle -v_{ijk} + z_{ijk} - z_{i(j-1)k} \le 0,\) \(i = 1, \dots, m,\ j = 2, \dots, t,\ k = 1, \dots, g\) | |
\(\displaystyle x_{ij},\ y_{ij} \ge 0\) | |||
\(\displaystyle x_{ijk},\ y_{ijk} \ge 0\) | |||
\(\displaystyle z_{ij} \in \{0, 1, \dots, \bar{z}_{ij}\}\) | |||
\(\displaystyle z_{ijk} \in \{0, 1\}\) | |||
\(\displaystyle v_{ijk},\ z_{ijk} \in \{0, 1\}\) |
where, new in this model:
where, new in this model:
where, new in this model:
where, new in this model:
where, new in this model:
“Constraints (a) and (b) in the model are the flow-balance equations for each node, while the remaining constraints represent lower- and upper-bounds on the decision variables \(x\) and \(y\).”
“The procedure is exact only under its assumptions. A storage limit, an upper bound on the amount of inventory that can be carried from one period to the next, can break it: in Ex. 4 the latest plan would hold more at a stage than it can store, and only the linear program finds the plan that does not.”
Model 2 is a procedure, not a math program. Shown here is Model 1, whose plan it finds when its assumptions hold; its assumption (c), “no storage limit binds”, sets the storage row aside.
“Model 3 generalizes Model 1 with the number of setups at each stage in each period. A machine is set up at most once in a period, so for a machine the number is the indicator; for trucks it is a count.”
“Model 1 generalizes Model 1 in Lecture 4.3: each flow carries a product index, and the products at a stage share its capacity in each period, which is given, as Lecture 4.3’s capacities are.”
“Model 2 generalizes Model 1 as Model 3 in Lecture 4.3 generalizes Model 1 in Lecture 4.3: the shared capacity, given in Model 1, is the capacity per setup times the number of setups, determined in the model, each setup at a cost.”
“Model 3 generalizes Model 1 in Lecture 4.3: each product has its own network, its capacity at a stage gated by whether the stage is set up for it, and the linking rows join the networks.”
“Model 4 generalizes Model 3 by adding a switchover indicator for each stage, period and product, its cost to the objective, and the rows (g) that set it.”
Lecture 4.3's models make one product, so its rows carry two indices, \(x_{ij}\); Lecture 4.4's add the product, \(x_{ijk}\), and the shared rows change notation with the lecture. A highlighted term in the objective is the cost the model adds.