Manufacturing Changeover Sequence Optimization: APS Implementation in Thailand

An APS project for manufacturing changeover sequence optimization needs more than a target to reduce setup minutes. The schedule must account for the direction of each product transition, material release, available molds and qualified people, promised delivery dates, and the period in which shop orders must stay fixed. This guide helps production, manufacturing and IT teams at Thai factories turn those needs into an RFP, a comparable proof of concept, and testable FAT and SAT criteria.

Start with an executable sequence, not the shortest setup total

A schedule with few changeovers cannot run if the required material has not passed receiving or inspection. Moving an urgent product to the front does not work when its dedicated mold is occupied on another line. Sequence optimization means choosing among feasible alternatives while making explicit tradeoffs among delivery, setup, overtime, work in process and schedule stability. Safety and quality prohibitions, resource eligibility, material release, operator qualifications and already issued orders should be hard constraints. The optimization objective comes after those constraints.

The planner should be able to see why an order was deferred: material not released, shared mold unavailable or an instruction inside the frozen window. A single weighted score without an explanation makes the result difficult for supervisors to trust. SAP’s PP/DS documentation describes a setup matrix with different times and costs for each transition; A to B need not equal B to A. Siemens describes sequence dependent changeovers, resource availability and material dependencies in Opcenter APS. These are descriptions of product capabilities. They do not establish the savings a particular factory will achieve. Test against the factory’s own orders and operating conditions.

Model the factors that actually change a production sequence

A directional setup matrix

One setup time per product is insufficient. The record needs the preceding product, the next product and the resource. Moving from a light color to a dark color may require a different cleaning procedure than the reverse. In resin, food, chemical, coating and printing processes, cleaning, drying, inspection and first article approval may need separate steps. State whether the time depends on the machine, tool, batch size, operator or quality procedure. Record these attributes rather than burying them in a comment.

The first job in the planning horizon has a predecessor even if that predecessor is outside the screen. Carry forward the last actual product, machine cleaning state and installed mold. Otherwise the first shift may incorrectly have zero setup load. Do not silently replace unknown transitions with zero. Use an explicit provisional value, its basis, an owner and a review date. Measure common and high impact transitions first, then expand the matrix as actual records arrive. Grouping similar products can reduce master data work, but verify that the grouping does not erase a critical cleaning or quality difference.

Materials, molds, operators and quality release

An idle machine is not necessarily ready to produce. Inventory may be reserved, awaiting inspection, due later in the day, or restricted to a customer. Oracle’s Production Scheduling documentation describes how supplies, demands and shortages affect the schedule. In an RFP, define when and for which order a lot can actually be used. A planned arrival date alone is not the same as usable stock. A shared mold cannot be assigned to two lines at once, and a changeover that requires a qualified technician must respect that person’s shift and absence. Prototype or quarantined material should not enter a mass production sequence without approval.

Trying to model every rare exception from day one can delay the project. Start with the bottleneck process and shared resources that alter real delivery decisions. Add further constraints when their absence causes a meaningful difference in the schedule. Keep a visible register of assumptions. When a constraint is temporarily relaxed for a simulation, label the result as a scenario, not as a released shop floor order.

Due dates, lot sizes and operation relationships

Grouping the same setup family can reduce changeover time but delay a small urgent order. Distinguish requested date, confirmed date, shipment cutoff and transport time. Decide which timestamp counts as late. A larger lot may save setup but increase work in process, storage pressure or downstream queueing. For shelf life or expiry controlled materials, an apparently efficient campaign could create unusable stock. Show these tradeoffs separately so a planner can choose a feasible business response.

Operations may have minimum or maximum waiting time. One product can proceed immediately while another must cool, dry or cure before its next process. SAP’s detailed scheduling documentation describes placing and rescheduling activities with resource and product availability. Use representative factory orders in the RFP to specify the allowable start and end windows for each operation. Record the difference between an impossible schedule and a feasible schedule with a late order.

Decide where spreadsheets end and APS begins

A spreadsheet and experienced planner can be adequate for a small number of lines with limited changes. APS becomes worth evaluating when one expedited order forces the planner to recheck several lines, molds, materials, people and downstream operations, and that manual coordination is repeated every day. The value is in maintaining consistency across constraints and comparing alternatives, not merely in displaying a more attractive Gantt chart.

APS will not repair uncertain master data on its own. Observe how the current planner decides, overrides and asks colleagues for confirmation. A working division may place orders, BOM, inventory and work orders in ERP/MRP; finite capacity and sequence decisions in APS; and actual start, completion, stops, scrap and setup completion in MES or a shop terminal. Product scopes overlap, so assign a system of record and update direction for each field rather than relying on software labels. For the broader migration path, see our guide to moving production planning beyond Excel. To compare vendors on shared test cases, use our production scheduler RFP and PoC guide.

Audit the data before configuring optimization

Begin with one changeover heavy process group, not every line in the plant. Select a normal week, a peak week and a week with a material delay. Collect orders, product and routing masters, standard and actual durations, machine calendars, molds, materials, confirmed due dates and the sequence that operators actually ran. Leave unknown fields visibly unknown and record their impact. An attractive minute by minute schedule based on arrival dates that are only known to the day is false precision.

For setup data, agree when timing starts and ends, whether parallel work counts and how outliers are recorded. Ask operators to explain the longest transitions. Measure frequent or costly pairs first. The data audit deliverable is a list of missing or unreliable fields, with an owner, provisional rule and correction deadline. If material receiving is recorded only as a date, choose a conservative usable time or integrate the warehouse release event.

When actual records and master data disagree, investigate before declaring either one correct. The operator may have pressed the start button late, performed two preparation tasks in parallel, or followed a procedure that has changed since the standard time was set. Factories also need one definition of setup time: including machine downtime at one site but excluding it at another makes cross-site rules unreliable. Document the measurement boundary and review representative records with the people who performed the work before changing the optimization model.

Data for materials and molds is managed not only by their current values but also by when they are expected to become available. Even if the scheduled arrival is updated, if the items remain pending inspection, planners need to see them as “arrived but unused.” For products that can use alternative materials, the procurement securing the substitute and the quality department approving its use are separate events. Molds are distinguished by storage location, installation destination, and repair status. In factories where moving molds takes time, even if a mold is available, it cannot necessarily be immediately installed on the next equipment. By handling these statuses along with timestamps, the system can explain on-site why a resource that appears “available” cannot actually be started. Conversely, if the necessary status cannot be obtained, the scope of the target process should be narrowed or a manual confirmation checkpoint established, with these decisions clarified during the RFP stage.

Improving data quality is not a one-time migration task. Each time a new product is launched, the number of changeover directions increases, and standard times change due to mold repairs or equipment modifications. It is necessary to decide which changes are automatically imported and which become effective only after manufacturing and quality approvals, minimizing the period during which outdated values remain in calculations. By reviewing weekly differences between plans and actual results, and re-measuring from product transitions with large discrepancies, operations become more sustainable than attempting to measure all combinations at once and stalling. User-friendly input screens for on-site reporting and mechanisms that indicate when corrections are reflected in subsequent plans are also design elements essential for maintaining data quality.

Specify objectives and a frozen window

“Optimal” has no common meaning across vendor demonstrations. Separate hard constraints from objectives, then rank the objectives. A factory may require zero violations of quality prohibitions, minimize missed confirmed dates, then reduce changeover time, overtime and schedule churn. If a product uses weighted scores, ask to see what happens to the proposed sequence as weights change. The list of late orders and reasons should be visible without manually tracing each operation.

Recalculating all orders whenever something changes can disrupt work already issued to the line. Define a frozen window and distinguish jobs started, material issued, molds being exchanged and jobs promised to customers. SAP documentation on stabilizing sequence dependent setup activities also describes conditions under which rescheduling can still occur. In factory practice, a freeze rule needs an exception owner, a notification path and an audit trail. Its duration depends on preparation lead time, shift change and the time at which orders are released. A window that is too long prevents recovery; one that is too short forces operators to chase a moving plan. Compare no freeze, same day freeze and process specific freeze scenarios using both setup time and the number of changed shop instructions.

Exceptions to fixed periods must not become “loopholes” that anyone can freely adjust on the screen. In case of equipment failure, maintenance registers the expected downtime, production management identifies affected orders, and quality and warehouse departments confirm the feasibility of alternative processes and materials, following a predetermined sequence. The rescheduling proposals presented by the APS must specify which approved instructions to cancel and which new instructions to issue. If materials have already been issued or molds are currently installed on-site, the reasons for making changes—even at the cost of wasting these preparations—must be documented. To prevent coexistence of old and new versions after approval, version numbers, distribution times, and receipt confirmations are recorded. After emergency responses, frequent exceptions are reviewed and freeze periods or input timings adjusted accordingly.

Make rescheduling a controlled workflow

Material delay, machine failure, quality hold, rush order, operator absence and mold repair are different events. For each, specify who records it, when the solver runs, who approves the new schedule and how operators receive the revised order. Automatic calculation does not imply automatic release. Keep simulation outputs separate from the approved production schedule.

The before and after view should show sequence, starts and completions, late orders, setup time, material wait, resource load and effects on frozen orders. When a schedule is impossible or includes lateness, show whether a rule was relaxed by configuration or an exception was explicitly approved. The ability to reconstruct that decision from a log is part of the acceptance criteria. Compare vendor systems with equivalent constraint settings; the label “constrained” alone may hide different behavior.

The next calculation is only as current as its inputs. Return actual starts, completions, stops, quantities, rejects and setup completion from MES or a shop terminal. Define what happens when integration fails: retry, duplicate suppression, time clock differences and manual corrections. A factory without MES can still define a minimum event feed from an existing terminal. Make someone responsible for reviewing late or missing events before a new schedule is released.

Write an RFP that can be tested

Avoid the line “must optimize changeovers.” State six areas: data, rules, user views, integration, operations and acceptance. Identify the factory and process, resources, planning horizon, time resolution, representative order volume and existing ERP/MES. Use actual factory values where possible and label vendor assumptions. A checked box on a feature sheet is less informative than a schedule produced from the same anonymized factory orders.

Specify directional setup time by product pair, resource, mold, cleaning and quality check; initial state; and behavior for undefined transitions. List material release, shared tools, qualifications, maintenance and interoperation waiting constraints. Rank due date, setup, overtime, work in process and plan stability. Give event cases for rush orders, breakdown and late material, and require the resulting difference and approval flow to be shown. For integration, identify the system of record, message direction, update frequency, retry, manual correction precedence and retention for each field. Thai holidays, overnight shifts and time zones must appear in test data, not only in a requirements paragraph.

Describe the screens in the order planners use them each day. After importing actual results in the morning, they need to review exceptional data, hold material that has not arrived, compare candidate schedules and release only an approved version to the shop. Gantt colors and display speed alone do not establish usability. The planner should be able to inspect why an order cannot start, how its setup duration was calculated, which constraint caused a delivery delay and what changed from the previous approved schedule. If a planner fixes an order manually, preserve the reason and release condition, and make it possible to check that a later calculation respected the fix.

Decide whether the same role may edit master rules and approve production plans. Someone can make a plan appear on time by shortening a setup master value, but the result may be impossible to execute. Link every rule change to approval, effective date, old and new values, and recalculation of the affected schedule. Check whether configuration can be exported and maintained by the factory after the project team leaves. Keep a way to consult the last approved shop instructions during an outage. These operating requirements distinguish a one-off demonstration from a system that the plant can continue to use.

Compare proof of concept results under identical conditions

Give each candidate the same anonymized orders, setup matrix, machine calendar and metrics. Include at least a normal sequence, a long reverse transition, late material, shared mold conflict, rush order and breakdown within the frozen window. Do not prescribe one perfect order list for every case. Instead check hard constraints, explanation of lateness and the planner’s ability to decide whether to release the result.

Include the labor needed to prepare data and configure the model. A vendor demonstration can look effortless if a specialist repairs the inputs by hand and leaves no method for the customer to repeat the work. Ask who can update rules, what regression checks follow a change and how Thai support handles incidents. If calculation time is measured, hold order volume, constraints, server conditions and start and end points constant. Vendor marketing performance figures are not measured outcomes for your factory.

If a proof of concept reveals that a proposed constraint does not reflect how the plant actually works, record that discovery. Identify missing data, exceptional procedures and differing assumptions across departments. If a mandatory rule changes, give the same revised case to every candidate. Otherwise a difference in output may come from unequal test conditions rather than the product. A system that explains why no feasible sequence exists can still be valuable. Do not reward a system for hiding a conflict and producing a superficially complete plan.

Use several success measures: late orders and lateness duration, setup time, overtime, rescheduled order count, planner overrides and agreement with released shop orders. If one improves while another worsens, document why and who accepted the tradeoff. When using current planning as a baseline, include informal changes that were communicated verbally. A simple before and after comparison across periods with different demand does not isolate an APS effect.

Separate FAT from SAT acceptance

FAT verifies agreed configuration and integration on test data before production use. Cases should cover directional setup, warning on missing values, material shortages, double booking a mold, frozen order protection, permissions, logs and ERP/MES exchange. For each case retain the input, initial state, steps, expected output, actual output, evidence and approver. Merely opening a screen is not evidence that the sequence is feasible.

SAT verifies the actual Thai factory environment: real masters, shifts, network, terminals and handoffs. Have planners run the morning update, approve an exception, send shop instructions, receive actual events and recalculate for the next day. Test late integration messages, incorrect inputs, connection loss, unauthorized edits, Thai holidays and overnight shift boundaries. Assign an owner and deadline to issues that remain open at handover.

Include supervisors, warehouse, maintenance and quality staff, not only planners. When a sequence changes, they need to know how an old instruction is withdrawn, who acknowledges the replacement and whether material and tooling can be ready in time. A plant that still prints job orders also needs a process for recalling outdated paper. SAT may reveal that a notification or morning meeting procedure needs correction even when the software itself is functioning as specified.

A contractual target such as “20% setup reduction” may change with product mix and demand even if the system is configured correctly. Define acceptance around reproducible behavior: zero hard constraint violations on the agreed dataset, the specified event producing an explainable difference, required logs retained and operators able to complete approval. Track business improvement separately with a documented baseline and evaluation period. Separating FAT and SAT helps distinguish configuration faults from local operating problems.

SAT test sheets record not only successful operations but also how recovery was achieved after failures. For example, they verify whether duplicate registrations occur when the same instruction is resent from the ERP, or whether on-site terminals temporarily offline confuse old and new instructions. When material inspection results arrive after plan approval, it is determined whether recalculation is necessary or if confirmation by the responsible person suffices. There is no need to risk production data to cause failures; tests are safely reproduced in test environments or limited time windows to confirm recovery procedures and responsibilities. Only after deciding who will continue any manual tasks identified during testing can the handover to the site be considered complete.

Compare full cost and ongoing ownership

License fees are only one part of an APS investment. Include data cleaning, setup measurement, ERP/MES interfaces, special rules, tests, training, Thai support and continued master maintenance. Ask vendors to price the same plants, lines, user counts, horizon, interfaces and hosting arrangement. Break the estimate into standard product, configuration, custom work, migration, training, FAT, SAT, go live support and maintenance. Identify who owns the setup matrix and who updates it when products or molds change.

Do not count saved setup minutes as financial benefit unless the released capacity can actually be used, given demand, material and downstream capacity. Measure late delivery avoidance, overtime, planner rework and work in process separately to prevent double counting. Include the internal labor to maintain rules. Thailand BOI guidance describes conditions for digital technology investment promotion, but eligibility must be checked for the specific project with BOI or an appropriate adviser; this article does not determine eligibility.

Assign operating roles at the Thai plant

When a Japanese headquarters sets policy and a Thai plant handles exceptions, define who publishes the plan, confirms material delays, updates setup values, releases quality holds and approves emergency orders. Thai planners need usable screens, training and a support route in a language they can work with. Put these requirements in the RFP rather than treating them as post launch details.

At regular reviews, classify reasons why actual production differed from the schedule: stale input, missing actual event, incorrect setup duration, unmodeled rule or a justified shop decision. A justified deviation is a chance to improve the model, not a reason to blame the operator. Track whether manual overrides fall, delivery promises become easier to answer and changes become easier to explain. This keeps the model useful as products, equipment and staffing change.

Summary

For manufacturing changeover sequence optimization, start with a directional setup matrix and combine it with materials, machines, tools, qualified people, due dates, operation relationships and a frozen window. An APS decision should be based on whether the system explains its sequence, handles disruptions without silently rewriting issued work, and receives accurate actual events. Put representative factory scenarios and data ownership into the RFP, compare vendors under the same PoC conditions, and test configuration and on site operation separately through FAT and SAT.

If your Thai plant is still organizing changeover records or defining the test cases for an APS RFP, contact TOMAS TECH with your target process and current planning workflow. The discussion can begin before a product has been selected.

Even after implementation, each plan should not be automatically treated as correct; the site must report exceptions and planners must be able to explain their causes. When adding new products or equipment, review the setup matrix, resource compatibility, material usage conditions, and fixed planning operations together. By deploying the data definitions and acceptance sheets obtained from the initial target process to subsequent processes, unintentional proliferation of differing rules across sites can be avoided. Rather than focusing solely on reducing changeover times, continuously verifying how delivery date responses, instruction changes, and on-site explanation burdens have changed is a condition for embedding APS into operations.

FAQ: Changeover sequence optimization and APS

Will a shorter setup sequence always improve delivery?

No. Grouping products may delay an urgent small order. Review delivery, material and downstream constraints alongside setup time.

Must every setup matrix pair be measured before a project starts?

No. Start with frequent and high impact transitions. Mark unmeasured pairs with provisional values and owners; never treat unknown as zero.

Does APS replace ERP or MES?

Their functions differ. ERP may own orders, BOM and stock, APS may own the approved finite sequence, and MES or terminals may return actual events. Define ownership per field.

What if a machine fails inside the frozen window?

Use a documented exception authority and notification path. Recalculate the affected instructions, show the difference and retain approval history.

How do FAT and SAT differ?

FAT tests configuration and interfaces on agreed test data. SAT tests the end to end process in the actual Thai plant with its people, shifts and systems.

Sources

• SAP Help Portal: Setup Matrix
• SAP Help Portal: Detailed Scheduling
• SAP Help Portal: Deactivation of Sequence-Dependent Setup Activities
• Siemens: Opcenter Advanced Scheduling
• Oracle: Consideration of Material Constraints
• Thailand BOI: Investment Promotion Guide 2025