Question 3 of 8: Why an Optimal Schedule Is Not Always Implementable

Question 3: Why an Optimal Schedule Is Not Always Implementable (20 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

(a) Factors That Can Prevent an Optimal Schedule from Being Implementable

A schedule computed as “optimal” is only optimal with respect to the deterministic, complete data the optimization model was fed; the shop floor is neither deterministic nor fully knowable at the moment the schedule is planned. Four representative factors expose this gap.

1. Unplanned equipment downtime. An optimal schedule almost always assumes every machine is available exactly as planned. In reality machines fail stochastically — a bearing seizes, a tool breaks, a controller faults — and every activity sequenced after the failed resource is immediately invalidated, not just delayed by the downtime itself but potentially resequenced entirely if the failure changes which resource is the bottleneck.

2. Variable processing and setup times. Optimization models are usually built from a single nominal (average or standard) time per operation, but actual cycle times vary with operator skill, material batch quality, and tooling wear, and setup/changeover times in particular are often far more variable than their average suggests. A schedule built on point estimates begins drifting out of sequence within hours of execution as small timing errors compound across the routing.

3. Labour availability and work-rule constraints. An LP or heuristic scheduler frequently assumes any qualified operator is available whenever the model needs one, ignoring absenteeism, shift-change handoffs, skill/certification requirements (only certified staff may run certain equipment), and union or overtime rules that were never encoded as explicit constraints. The plan can be resource-feasible on paper and completely unworkable against the actual roster on the day.

4. Material shortages, late deliveries, and rush orders. The schedule assumes every input material arrives on time in the planned quantity; a late supplier shipment forces either an idle station or an unplanned resequencing. Separately, a customer-priority rush order inserted after the schedule is finalized invalidates the sequencing logic the optimizer relied on, since the model never anticipated a new job entering partway through the horizon.

(b) Ways to Overcome Each Limitation

1. Unplanned downtime → preventive/condition-based maintenance plus deliberate schedule slack. Total productive maintenance (TPM) and condition monitoring (vibration, temperature, oil analysis) reduce the frequency of unplanned failures directly, while sizing schedule buffers from historical mean-time-between-failures (MTBF) and mean-time-to-repair (MTTR) data — rather than assuming 100% availability — keeps a single breakdown from cascading through the whole plan.
2. Variable processing/setup times → robust scheduling plus SMED. Build the schedule against a time estimate that reflects the process's actual variance (a percentile or a mean-plus-safety-margin), not just its average, so the plan is realistic rather than best-case; separately, apply Single-Minute Exchange of Die (SMED) principles to shrink changeovers and make their duration more consistent, attacking the variability at its source rather than only buffering around it.
3. Labour constraints → cross-training and explicit constraint modelling. Cross-train operators across more of the routing so the “only a certified operator can run this station” constraint binds less often, and encode shift patterns, overtime limits, and certification requirements as explicit constraints in the scheduling model itself, rather than solving an unconstrained optimum and discovering the labour conflict only on the floor.
4. Material shortages/rush orders → safety stock and rolling-horizon rescheduling. Carry safety stock or a vendor-managed-inventory arrangement on critical or long-lead-time items so a single late shipment does not stall the line, and replace a single static optimum computed once with a rolling-horizon policy that re-optimizes frequently against current shop-floor status (via real-time production data), so a rush order or a shortage is absorbed by the next re-plan instead of breaking the current one.