Scheduling Continuous-Process Equipment in Food Plants

Continuous process scheduling for food means telling the engine which stations produce a countable stack of pieces and which produce a flowing stream, because the two need different overlap rules. A cartoner hands off 50 finished units. An extruder hands off product that is simply moving. EDGEBIC by User Solutions carries both models and picks between them with one flag on the work center, so a single routing can run a discrete filler, a continuous oven and a discrete case packer without any of them pretending to be something else.

For the mechanism in general terms, read lot streaming explained. For this sector's broader constraint picture, see food and beverage production scheduling, and for the map of which capability fits which plant, how different industries use EDGEBIC.

Two overlap models, one flag

Without overlap, every step waits for the one before it to finish the whole order. That is the default and it is often badly wrong, because most of the material is ready long before the last piece is.

EDGEBIC resolves overlap two ways.

The selector is IsContinuousProcess on the work center. Set it true and that station always uses the lag model, whatever the routing says about transfer batches. That is deliberate: the flag describes physical equipment, so it belongs to the machine rather than to the product running on it. A pasteuriser is continuous whether it is running juice or milk.

If you want the piece-count concept on its own, what a transfer batch is covers it.

Why the lag model is right for continuous equipment

The lag is a start-to-start relationship measured in hours, and it is independent of quantity, run time and setup:

next step earliest start = upstream start + lag hours

That independence is the point. On a continuous line the condition for starting downstream is not "N units exist", it is "material has been flowing long enough to reach the next station and stabilise". A cooling tunnel accepts product roughly the same number of minutes after the oven starts whether the order is 500 kilograms or 5,000. A piece-count rule would stretch the wait with quantity and get the physics backwards.

Here is the documented example, using a spray and cure line, which behaves identically to an extruder and cooling tunnel or a cooker and filler.

Setup. 500 panels through a spray booth flagged continuous process. The booth runs four hours in total. The lag is set to 1 hour. The curing oven is next.

Spray booth: 08:00 to 12:00 (4 h)
Lag: 1 h
Curing oven may start: 08:00 + 1 h = 09:00
Without streaming, the oven waits until 12:00.
Compression on this handoff: 3 hours

Nothing ran faster. The plan simply stopped forcing the second station to wait for a condition that was never physically required.

Where the piece-count model still belongs

Most food plants are hybrids, and it is worth being precise about which stations are which.

Discrete stations, where the transfer batch is right: fillers with countable output, cartoners, case packers, labellers, tray sealers, hand-pack cells.

Continuous stations, where the lag is right: extruders, ovens and dryers, cooling tunnels, pasteurisers and homogenisers, coating and enrobing lines, mixing and blending vessels, spray and cure lines. A kibble plant is mostly the first entry on that list, and extrusion and changeover scheduling for pet food works through what modeling it as continuous does to a whole line.

For a discrete station the piece-count formula is explicit:

flow time = setup + min(transfer batch, order quantity) x per-piece hours

Two details save planners from surprises. The batch is capped at the order quantity, so a transfer batch of 100 on a 30-piece order becomes 30 rather than waiting for a hundredth piece that never exists. And a transfer batch equal to or larger than the order quantity behaves exactly like no streaming at all, which the anomaly report flags so you can lower it for that order.

Capacity on a rate-driven line

Continuous and semi-continuous food lines are usually described in units per hour rather than hours per unit, and the model supports that directly. A work center can measure capacity in hours, in pieces, or in both, where both means whichever constraint produces the longer duration.

For a pieces-based line the rate the plan uses is the effective one, not the nameplate:

effective pieces per hour = base rate x efficiency factor

A bottling line rated at 120 bottles per hour that holds 90 percent of that offers 108 bottles per hour to the schedule, and 108 is what its routing steps should carry. The converter beside the hours value on the step turns a rate you know in pieces per hour into the decimal hours per piece the step stores. Choosing "both" is the honest setting for a line that is sometimes rate-limited and sometimes time-limited, because the engine then takes the longer of the two answers rather than the more flattering one.

Worked through on a filler: a line rated at 120 bottles per hour that sustains 90 percent of that offers 108 per hour. A 500-bottle order therefore needs about 4.6 hours of run time. If the routing also carries a time-per-unit figure and the work center is set to measure both, the engine computes the duration both ways and takes the longer one, so the plan is limited by whichever constraint actually binds that day. Setting a line to "both" is the honest option when the answer changes with the product.

The two configuration traps

Both are documented and both have anomaly checks, which means the report tells you rather than the schedule quietly misbehaving.

A transfer batch on a continuous-process station. The piece count is ignored and the lag applies. The schedule is correct, but a planner reading the routing sees a piece count that is doing nothing. The C1 check flags the combination so you can clear one field or the other.

A lag and a queue time on the same step. The lot-streaming result replaces the queue-adjusted end rather than adding to it. If you set a lag of one hour and a queue time of two hours expecting three, you get one. The F2 check flags the pair. The fix is to use the transfer delay field for the physical wait, because that value is added on top of the lot-streaming time rather than replacing it. In practice: lag for the flow condition, transfer delay for cooling or handling, queue time only when no streaming applies. The composition rules are covered in more detail in queue and transit times explained.

A worked hybrid routing

A snack line with three stations shows the models working together.

Step 30 is the interesting one. Once the routing reaches a station that produces countable output, the piece-count rule takes over and the plan starts reasoning in cases again. Nobody had to split the order or the routing to make that happen: each station simply used its own physics.

Getting it configured

1. Walk the line and label each station continuous or discrete. Ask one question at each: does the next station wait for a quantity, or for flow to establish? The answer is the model.
2. Set the continuous-process flag on every continuous station in the work center editor.
3. Set the lag in hours on the routing step feeding each continuous handoff. Use the number the line operators give you, which is usually the observed minutes between starting one station and being able to start the next.
4. Set transfer batches only on discrete stations, sized to the pallet, tray or case quantity that actually moves.
5. Add transfer delays where a physical wait exists on top of the flow condition, such as a cooling hold before packing.
6. Run the anomaly report and clear any F2, F4 or C1 entries. Each one names a specific field on a specific step.
7. Re-run the schedule and compare finish dates. On a plant with three or four continuous handoffs the compression is usually visible immediately.

None of this requires a global switch or a restart. Transfer batches default to zero, which means no existing routing changes behavior until a planner sets a value on a specific step.

One question decides most of these settings, and it is worth asking it out loud at each station with the person who runs it: what has to be true before the next station can start? If the answer names a quantity, you want a transfer batch. If it names a condition ("once it is running", "once the tunnel is up to temperature and product is flowing"), you want the continuous flag and a lag. If the answer is "we wait for the whole batch because of the process", then neither applies and full-lot handoff is correct. Recording that answer per station takes an hour and settles the configuration for good.

What it buys

Two things, and the second is the durable one.

Shorter promised lead times, because the plan stops inserting waits the process does not require. On a routing with several continuous handoffs the compression compounds across steps.

More importantly, a plan that matches how the line actually runs. Supervisors already know the cooling tunnel starts before the extruder finishes. When the schedule says otherwise, they stop trusting the schedule and start running from memory. Modeling the handoff correctly is what keeps the printed plan and the floor telling the same story.

Continuous equipment is one of three constraints food plants usually need. The others are covered in allergen and flavor changeover sequencing and scheduling around shifts, sanitation and holidays. Plants that also run reactors or blending vessels will recognize the same problem in chemical manufacturing scheduling. Bakeries meet it at the oven, which holds a bake for its full cycle: see bakery oven and line scheduling.

Bring one hybrid routing and its real handoff times to a demo of food manufacturing scheduling software, and we will model the line the way it runs.