Chapter 6｜Digital Factory Roadmap
Smart manufacturing does not have a fixed finish line. After a system goes live, the factory often begins to see gaps that were not visible before: missing data, poorly defined abnormal causes, and workflows that still depend on manual reconciliation.
The next stage should therefore focus on making the existing data relationships more complete before adding more models or automation.
6.1 Building a Digital Injection Molding Factory
Current machine connectivity can already provide selected equipment states, Cycle Time, cycle count, and process data. MES links those records to the work order, product, and mold.
The next stage of the Digital Factory is to connect more of the remaining production data to the same production timeline.
Future work will continue to connect OPC UA equipment data, work orders, production intervals, molds, material lots, quality results, and other available process records.
If the equipment provides pressure, position, or other process curves, those records can also be retained and compared with quality results.
The goal is not simply to accumulate more tags. The goal is to create relationships between the records.
If a dimensional problem occurs, the factory should be able to move from the product lot to the work order, machine, mold, production interval, and corresponding Cycle Time or process data.
Start Simulation from Real Historical Records
A digital factory can gradually add simulation functions, but actual equipment data and calculated model results should remain clearly distinguished.
Scheduling can simulate the impact of inserting an urgent order using current order, machine, and mold conditions. Product and mold development can use CAE to compare design and molding conditions.
These simulations can reduce manual trial calculations, but they do not replace actual mold trials and production validation.
6.2 Multi-Objective Scheduling and Decision Support
The current GA scheduler can generate an initial schedule from defined constraints, but real production decisions rarely have only one objective.
A schedule that minimizes completion time may increase mold changes. Reducing mold changes may cause some urgent orders to finish later. The next stage therefore includes evaluating multi-objective optimization.
Evaluating NSGA-II Multi-Objective Optimization
One planned direction is to evaluate NSGA-II or similar multi-objective methods as an extension to the current GA scheduler rather than assuming it should immediately replace the existing method.
Potential objectives include:
- Total completion time
- Tardiness
- Number or duration of mold changes
- Machine-load balance
- Energy-related indicators when reliable measurement data is available
These objectives often conflict, so the system may produce multiple useful solutions rather than one absolute “best” schedule.
Pareto Front: Compare Options, Do Not Remove the Planner
Multi-objective optimization is useful when it generates several trade-off solutions for the production planner.
One schedule may minimize tardiness but use more mold changes. Another may reduce mold changes while accepting a small delay on selected jobs. The planner can then choose according to the actual order and shop-floor situation.
| Scheduling Direction | Possible Characteristics | When It May Be Useful |
|---|---|---|
| Due-Date Priority | Reduces tardiness but may increase changeovers | High urgent-order pressure |
| Changeover Priority | Groups jobs with the same mold or similar setup | High mold-change cost or long setup time |
| Load Balance | Avoids concentrating too much work on a few machines | Multiple compatible machines share capacity |
| Balanced Plan | Trades off due date, mold change, and machine load | Normal daily scheduling |
6.3 Continuous Improvement
If a system goes live and the rules are never reviewed again, it can eventually become another system that only stores records.
Continuous improvement therefore has to return regularly to production KPIs and actual user problems.
Regular KPI Review
OEE, yield, Cycle Time, downtime, scheduling time, and energy can be reviewed on a regular basis.
However, a lower monthly KPI does not automatically mean performance deteriorated. Product mix, production volume, and equipment conditions must be comparable before month-to-month conclusions are made.
Update Baselines and Alert Rules
Cycle Time baselines, anomaly thresholds, and scheduling weights all require periodic review.
If a mold is modified and the normal cycle becomes shorter, an old baseline can create meaningless alerts.
Manual schedule changes can also be stored and later analyzed to determine which real constraints were missing from the algorithm.
Adjust System Architecture According to Maintenance Needs
As functions grow, production reporting, scheduling, machine data, warehouse, and other modules can be separated according to actual maintenance needs.
A complete microservices architecture should not be adopted simply because the term sounds more advanced. For an internal MES with a limited number of modules, clear API boundaries and data ownership can be easier to maintain than splitting the system too early.
Independent services become more reasonable when modules need separate deployment, different update cycles, or significantly different loads.
6.4 Training and Shop-Floor Knowledge Retention
Smart manufacturing is still used by people on the production floor. If the new system asks operators to enter data that nobody uses, or engineers do not trust the alerts, the system will not be adopted no matter how complete the software looks.
Different Roles Need Different Training
Training should not only teach which button to press. It should explain why a field is recorded and where the data is used later.
- Operators: Production reporting, abnormal reasons, reject quantities, and necessary notes.
- Process Technicians: Cycle Time, molding conditions, mold status, and abnormal records.
- Quality: Relationship between inspection results, work orders, and production time.
- Production Planners: Schedule results, reasons for manual adjustment, and delivery impact.
- Management: KPI trends and abnormal causes, not only a single number.
Data Does Not Replace Shop-Floor Experience
Many molding decisions still require experience. A longer Cycle Time may indicate equipment behavior, operator waiting, or a deliberate cooling-time change.
The more practical model is to let shop-floor experience and system data complement each other.
When a technician identifies the cause, the result can be recorded so the next person does not have to start from zero.
Keep Improvement Records Searchable
If every setup adjustment, mold repair, scheduling change, or quality improvement remains only in verbal discussion, the next similar problem may have to be solved again from the beginning.
Future system development should place more emphasis on recording three things: the condition before improvement, what was changed, and the result afterward.
The next stage of smart manufacturing is therefore not simply “more AI.” Data completeness, scheduling models, system maintainability, and the way production personnel use the system all need to improve together.
For Power Tech Mold, the direction remains practical: solve an actual shop-floor problem first, then decide which function deserves the next investment.