Skip to main content

Steel Production Line

No-Wait Flow Shop Scheduling

The Problem

Why steel production is a no-wait scheduling challenge

Continuous Casting & Hot Rolling

In continuous casting and steel production, molten steel slabs must pass through a series of processing stages — heating in a furnace, rolling in a mill, and controlled cooling — without any waiting between stages. The metal cannot be allowed to cool down between steps; if a slab sits idle between the furnace and the rolling mill, it loses temperature, resulting in defective products, wasted energy for reheating, or scrapped material entirely.

This operational constraint maps directly to the No-Wait Flow Shop Scheduling Problem (Fm | prmu, no-wait | Cmax). The objective is to find the optimal sequence of steel slabs through all stages that minimizes the total production time (makespan), while ensuring that each slab transitions immediately from one machine to the next.

Problem Classification Fm | prmu, no-wait | Cmax

n steel slabs, m processing stages (machines)
Each slab j has processing time p[j][i] on stage i
No waiting allowed: slab must move immediately to next stage

Complexity: NP-hard for m ≥ 3 machines

Stages

  • Furnace (Reheating)
  • Rolling Mill
  • Cooling Line

No-Wait Constraint

A slab finishing in the furnace must enter the rolling mill immediately. The start of a slab on a later machine is completely determined by its start time on the first machine, creating contiguous processing blocks.

Try It Yourself

Interactive no-wait flow shop solver for steel slab scheduling

Steel Slab Scheduling

5 slabs · 3 stages (Furnace, Rolling Mill, Cooling) · Click any cell to edit
Edit processing times below
Values in minutes
Slab Furnace Rolling Mill Cooling

The Algorithms

How no-wait flow shop heuristics work under the hood

Delay Matrix Computation

The key data structure in no-wait flow shop scheduling is the delay matrix D[j][k]. This gives the minimum time gap between the start of job j and the start of job k (on machine 1) such that no-wait is preserved for job k on all machines.

Delay Matrix Formula D[j][k] = max over r = 1, ..., m of:
    ( ∑i=1r p[j][i] ) - ( ∑i=1r-1 p[k][i] )

This ensures that on every machine r, job k does not start
before job j finishes on that machine. The no-wait constraint
forces job k to start at exactly D[j][k] time units after job j
begins on the first machine.

Because D[j][k] is generally not equal to D[k][j], the delay matrix is asymmetric. This makes the no-wait flow shop reducible to an Asymmetric Traveling Salesman Problem (ATSP), where the delay values serve as arc costs between jobs.

Nearest Neighbor Heuristic

A greedy algorithm analogous to the nearest-neighbor heuristic for TSP, adapted for the no-wait constraint using the delay matrix.

  1. Start with the first unscheduled job (or try all starting jobs in multi-start variant).
  2. At each step, pick the unscheduled job k with the smallest delay D[last][k] from the most recently scheduled job last.
  3. Append k to the sequence and repeat until all jobs are scheduled.
Complexity O(n2) — for each of the n positions, scan n remaining jobs

NEH-NW (Nawaz-Enscore-Ham for No-Wait) Heuristic

The celebrated NEH constructive heuristic adapted for the no-wait variant. Widely regarded as the best constructive heuristic for PFSP, extended here using delay-matrix-based makespan evaluation.

  1. Sort all jobs in descending order of total processing time (i p[j][i]). Jobs with longer total processing times are considered first.
  2. Begin with the first job in the sorted order as the initial partial sequence.
  3. For each remaining job, try inserting it into every possible position in the current partial sequence.
  4. Evaluate the no-wait makespan at each insertion position using the delay matrix. Keep the position yielding the minimum makespan.
  5. Repeat until all jobs are inserted.
Complexity O(n2 × m) — n insertions, each testing up to n positions,
each position requiring O(m) for delay-based makespan

Gangadharan-Rajendran Heuristic

A priority-index heuristic proposed by Gangadharan & Rajendran (1993), specifically designed for no-wait flow shops. It uses position-dependent weights that emphasize earlier stages more heavily.

  1. Compute a weighted processing time index for each job: W[j] = ∑i=1m (m - i + 1) × p[j][i], giving higher weight to earlier stages.
  2. Sort jobs in descending order of W[j] to form the initial sequence.
  3. Apply pairwise interchange improvement: for each pair of adjacent jobs, swap them if it reduces the makespan.
  4. Repeat the improvement pass until no further swaps reduce the makespan.
Complexity O(n2 × m) — sorting O(n log n), pairwise improvement O(n2 × m)

Real-World Complexity

Why industrial steel scheduling goes beyond the textbook model

Temperature Constraints

Slabs must maintain specific temperature ranges at each stage. Temperature loss during transfer is non-linear and depends on slab thickness, alloy composition, and ambient conditions.

Equipment Maintenance Windows

Rolling mills require periodic maintenance shutdowns. Scheduling must account for planned downtime, roll changes, and calibration periods that create unavailability windows on machines.

Time-of-Day Energy Costs

Electricity prices vary significantly throughout the day. Furnace operations are energy-intensive, creating incentives to schedule heating during off-peak hours, adding a cost-aware objective.

Quality Grades & Multiple Casters

Different steel grades (structural, automotive, stainless) require distinct processing paths and temperatures. Plants with multiple continuous casters add parallel machine routing decisions.

References

Key literature on no-wait flow shop scheduling

Röck, H. (1984). "The three-machine no-wait flow shop is NP-complete." Journal of the ACM, 31(2), 336–345.
Gangadharan, R., & Rajendran, C. (1993). "Heuristic algorithms for scheduling in the no-wait flowshop." International Journal of Production Economics, 32(3), 285–290.
Allahverdi, A. (2016). "A survey of scheduling problems with no-wait in process." European Journal of Operational Research, 255(3), 665–686.

Need to optimize your production line?

Reach out to discuss how operations research can improve your manufacturing scheduling.

Connect on LinkedIn Back to Portfolio GitHub Repository

Data shown is illustrative. Processing times are simplified for demonstration. Real steel production involves additional constraints not modeled here.

ESC