Summary of implementation of a Hierarchical Production scheduling and

1
Inventory Management system at Scotsburn Dairy Group

Background and Objective:

Scotsburn Dairy Group is the largest Atlantic Canadian dairy producer producing milk, butter,
cottage cheese, sour cream, and frozen dairy desserts. Its main plant, located in Truro
produces approximately 30 million litres of frozen products per year and more than 300 SKUs on
its three production lines. It produces ice cream under its own brand name and also under private
label for major retailers. Given the number of products, it implemented a hierarchical production
planning approach to improve scheduling at the plant which led to minimizing costs while
meeting expected demands.

Ice cream production:

Ice cream production planning requires the ability to plan for demand that has different seasonality
for different products. Workforce and warehouse utilizations depend on the product size, mix type,
and brand of each product. Ice cream production process is a two-stage process: In the first stage a
base-mix is produced which requires the mix to be heated for 30 minutes to kill bacteria. Once the
mix has been homogenized and cooled, it is stored overnight to age in preparation for the following
morning’s production. The second stage involves the addition of flavours and inclusions. Next the
mix is frozen and whipped to add cold air after which the final product is called ice cream. This ice
cream is pumped into packaging tubs and lids are applied. The tubs are then coded, scanned for
metal and shrink- wrapped. Before being sent to the warehouse, the tubs are sent through a spiral
freezer.
Truro has to produce multiple products per day on each line to meet demand, while keeping its
storage costs low. A significant amount of mix is lost each time a base mix changeover occurs
because the lines must be completely flushed before a new mix can be used. Unexpected line
downtime or a packaging-size changeover can result in a significant amount of product loss. As per
law, a production run cannot cross over from one day to the next. Production scheduling is also
constrained by factors such as allergens and the availability of buttermilk and cream.

The Hierarchical System:

An integrated series of models that planned for production on monthly, weekly, and daily
time frames was developed. Once solutions were implemented at one level, they constrained the
model decisions at the next level. Linkages between the outputs of the higher-level models and
the constraints imposed on the lower-level models kept the system results feasible from one stage
to the next.
 Summary of implementation of a Hierarchical Production scheduling and
2
Inventory Management system at Scotsburn Dairy Group

1. A central database was designed to provide Scotsburn with a single location for storing data on
ingredients, costs, recipes, reports on projected ingredient requirements, product costing, labour
costs, mix schedules, aggregated forecasts, holding costs and sequencing restrictions. This
database provided the models with up-to-date data.

2. Monthly Production Planning Model: A monthly inventory and staffing model based on a
mixed-integer programming (MIP) formulation and marketing forecasts was developed. The
model determined labour levels and anticipated monthly production output; with an objective to
minimize total production, holding, shipping, and labour costs. The monthly model selected an
appropriate labour schedule and seasonal stock levels per month to meet forecasted demand for
all products over a one-year horizon. Capacity was limited by the processing speeds of the 3
production lines, the blast freezer and the size of the 4 warehousing facilities.

12 potential labour schedules were worked out based on a production month containing 4 or 5
weeks. For e.g. one schedule was a five day week with an 8 hour shift per day on all the 3
production lines. Fixed charges were not taken into account as they are not affected by changes in
schedules. The model was designed such that only one schedule was selected for each month in
the planning horizon.

The 300 plus SKUs were aggregated into 26 smaller products groups based on similar production
rates, holding costs and packaging costs. For each product group the production rates and the
holding costs were estimated using managerial inputs. To calculate the forecasted demand, the
effective demand for each product within a group was calculated. Effective demand represented
the production required per period to cover actual demand once the initial on-hand inventory for the
product has been depleted. Total effective demand for a product group was then calculated by
summing the effective demand for all products within the product group. Scotsburn normally ran
the model every three months and this provided them with flexibility to respond to discrepancies
between the actual sales over the previous quarter and the changes to the sales forecasted in the
upcoming year. The solution included the labour schedule selected for each month and the actual
production needed in each period.

Prior to the implementation of the model, Scotsburn increased production in January in anticipation
of the summer demand. However, by the end of August, it had not sold all the scooping ice cream
it produced; thus, it had to hold the inventory for months. With the implementation of the monthly
 Summary of implementation of a Hierarchical Production scheduling and
3
Inventory Management system at Scotsburn Dairy Group

model, the company delayed the increase in output until May and recommended increasing
staffing in June and July to meet the summer demand. Inventory now accumulated more slowly in
the months leading up to the demand peak in June and July.

3. Weekly Production Planning Model: The purpose of the weekly model was to provide a cost-
effective, detailed scheduling plan on a short-term basis to maintain the production and inventory
plan that the monthly model developed. It used current inventory status, together with the labour
schedules and production utilization recommended by the monthly model, to determine the product
groups and quantities to be produced in the coming week. Grouping of products was done such
that when they were produced in sequence, the setup time and costs were minimized. Since
switching from one base mix to another or one packaging from to another required downtime and
additional labour, it was desirable to group products based on shared packaging types and base
mixes. The model determined which SKUs should be produced in the following week and the
number of future weeks, the production run should cover.

The problem was modeled as a network where nodes represented the periods and arcs
represented the cost and production requirements associated with meeting demand from node to
node. The labour schedule that monthly model produced dictated the capacity of the facility each
week. Calculating family lot sizes again required the use of the effective demand concept. To
calculate effective demand, orders expected to ship next week were deducted from the initial
inventory. Then effective demand was calculated for the week following next week and from that
point onward. Safety stocks were maintained to dampen the effect of fluctuations in final
shipments. If a decision was made to run a family and produce enough to cover demand for the
next n periods, then only the products in the family in which the effective demand for n periods was
greater than zero were produced. Once the possible family lot sizes were established, their
associated production times, production costs, and holding costs could be calculated. To maintain
feasibility within the hierarchical structure and produce a practical solution, the weekly model was
constrained by the plan selected by the monthly model. The total weekly scheduled production had
to fall somewhere between the recommended production capacity utilization (minimum) and the
total available capacity (maximum).

The production runs were actually to replenish inventory for product expected to be required for the
following week. This is because shipments went out throughout the week, and trying to produce
product to meet required shipments in the same week was too difficult to coordinate. If the actual
sales differed dramatically from the expected demand, the model would not be able to meet
 Summary of implementation of a Hierarchical Production scheduling and
4
Inventory Management system at Scotsburn Dairy Group

demand without incurring overtime. Truro ran this mixed-integer linear programming (MILP) model
weekly using a 13-week rolling horizon.

4. Daily Production Planning Model: The Daily model had to take the recommended production
plan from the weekly model and develop a daily schedule. A production schedule for each line for
each day of the week was designed, to minimize the overall changeover costs, given the planned
production for the week and the available production capacity of each line. This schedule was
constrained by the production speeds of each line, changeover times, line constraints, allergen
constraints, and mix batch sizes. The speeds of the three production lines were essentially the
same; however, all products could be produced on all lines. The goal was to assign products to
each line, and sequence the products to minimize the mix and ingredient-loss costs that resulted
from each changeover. For e.g. once a product with an allergen had been produced on a line, that
line was contaminated with that allergen for the remainder of the day. The only products that could
be produced for the remainder of that day had to also contain the same allergen.

The daily model was a MIP model which consisted of binary variables that indicated whether a
particular product would follow another product on the same line, on the same day. When these
binary variables were set to one, production and setup times were incurred and used to calculate
the start time for each job. Production quantities for the SKUs were converted to production times
based on the speed of each line, setup times between jobs were estimated based on observed
data, and then each SKU’s run time was treated as a job to be scheduled.

Results of the Hierarchical System:

The implementation of the monthly and weekly models resulted in a substantial reduction in the
number of products that were run each week. Scotsburn did not have very accurate data on the
mix losses and subsequent costs caused by the changeover from one particular mix to another
on a given line; therefore, it had decided not to directly implement the daily model. The daily
schedule was manually done by the scheduler who was aided by the central database of
information created.

Subsequently however when the number of SKUs rose considerably it needed a Daily Model. A
two stage daily model was developed. It was formulated as a simple problem of assignment of
each product to a given line and day, ignoring all sequencing issues. The objective was
minimizing the maximum number of mixes used on any one day, subject to a limit on production
capacity each day. Approximate setup times were used to account for the total production times
 Summary of implementation of a Hierarchical Production scheduling and
5
Inventory Management system at Scotsburn Dairy Group

on each line each day with a limit on the number of allergen products that could be run each
day. With this assignment of products to days, the single five day problem effectively became
five single day problems. This was implemented by bounding the time variables for each job to
occur within the appropriate day start and end times and by removing all the variables that
linked two jobs that occurred on different days.

Hierarchical production planning provided the ability to capture the different levels of production
decision management, and also maintain some form of connection and integration. By solving
each level of the hierarchical structure on its own rolling horizon, rather than modeling the entire
system at once, only meaningful results were computed at each stage. The monthly model
provided a snapshot of how to meet demand across a seasonal schedule while providing a
yearly labour schedule that minimized aggregate production and storage costs. Production
capacities were chosen from alternative predefined labour schedules that met Scotsburn’s
current accepted labour practices for the Truro plant. The weekly model created a production
schedule to meet demands for upcoming weeks. By aggregating products based on common
base mixes and packaging, products with shared setup costs were naturally produced together.
By solving the daily model in two parts – by first assigning products to days and then optimizing
each daily sequence – the company was been able to generate a schedule for the week very
quickly.

The results include improved warehouse utilization, reduced setup costs and product loss,
better use of capacity, and less stockouts. In addition, production plans extended further into
the future, which can lead to improved raw material procurement and early identification of
potential problems.

Source: Scotsburn Dairy Group Uses a Hierarchical Production Scheduling and Inventory
Management System to Control Its Ice Cream Production by Eldon A. Gunn, Corinne A. MacDonald,
Andrea Friars and Glen Caissie