The previous chapter focused on equipment interfaces, PMC Server, databases, and APIs. At the application layer, the question changes: can the data reduce the manual searching, copying, and reconciliation that people perform every day?

Not every function needs to be labeled “AI.” Some functions simply connect the work order to machine data, some compare current values with historical baselines, and some use optimization algorithms. If a function solves an actual production problem, that is enough reason for it to exist.

Author: Jush Li

4.1 Injection Molding Process Data Integration

Shop-floor data may be scattered across paper forms, Excel, ERP, machine controllers, and multiple databases. Each source contains useful information, but tracing a single work order often requires manual reconciliation.

The process-data platform puts those records into the same production context: which work order, which product, which machine, which mold, and what time.

4.1.1 Web Operation Interface

Operators do not need direct database access and should not re-enter machine values that the system can already acquire automatically. The web interface mainly establishes the production context and fills in information that the machine cannot know by itself.

Work Order and Mold Relationship

The system obtains work order, item number, product name, and mold ID from existing data sources. When the operator starts production, those identifiers create the production record.

Machine cycle time and manually entered defect counts can then be queried from the same work order.

Process and Material Information

Depending on the product, the system can record cavity count, part weight, runner weight, and other necessary conditions.

If pressure, temperature, injection speed, plasticizing time, or cooling time is available from the machine, that data should be acquired automatically. Only unavailable fields need manual input.

Quality and Notes

Some information does not fit a fixed numeric field. Notes such as “release for mass production only after first-piece dimensional confirmation” or a special setup condition still need a place in the production history.

Figure 4.1-1: Production record interface

4.1.2 RESTful API Between Frontend and Backend

The frontend does not operate ERP tables or low-level database tables directly. It requests required resources through RESTful APIs.

{  "id": 12,  "orderNo": "MD20250617",  "machineId": "NEX-V360",  "cycleTime": 12.34,  "parameters": {    "injectionPressure": 230.0,    "holdPressure": 45.6,    "barrelTempZone1": 240  },  "quality": {    "goodCount": 480,    "rejectCount": 2,    "remarks": "First piece OK"  },  "timestamp": "2025-06-17T08:10:00Z"}

JSON is easy for Web, Python, .NET, and other applications to process. More importantly, field definitions must remain stable: cycle-time units, timestamps, data types, and time zones should be explicit.

4.1.3 OpenAPI / Swagger Documentation

As the number of APIs grows, relying on verbal explanations quickly creates version mismatch.

OpenAPI defines endpoints, parameters, and response formats, while Swagger UI makes them visible and testable to frontend and test teams.

4.1.4 Summary

This layer solves data fragmentation. Once work-order records, manual records, and machine data share the same production interval, later quality analysis and anomaly monitoring have a reliable basis.

4.2 RWD Production Reporting

After machine connectivity is established, the system may know the current cycle time but still not know which work order is running.

Production reporting creates that missing link.

At production start, the operator links the work order, product, mold, operator, and machine. Data collected during that interval can then be associated with the correct production session.

4.2.1 Shop-Floor Reporting

Figure 4.2-1: RWD production reporting

Production Start

The operator selects the work order, item, mold, and machine. The system records the start time and creates the Production Session.

Cycle time, cycle count, and other connected data can then be matched to the work order through the machine and production interval.

Manual Data

The machine does not know whether a part contains black specks, flow marks, or other visual defects. Good quantity, reject quantity, and defect reason still come from the operator or QC.

Automatic and manual data should each handle the information they are best suited to provide.

Responsive Web Interface

The system uses an RWD web interface so PCs, tablets, and phones can access the same application without maintaining separate native apps.

4.2.2 Production Report and Query

Figure 4.2-2: Production daily report query

Management can query by date, item number, machine, operator, or other relevant conditions.

Manual reporting data and machine data can be viewed together, including production time, cycle time, good count, reject count, and remarks.

If more detailed process history is stored in the backend, the system can drill from the work-order interval into corresponding machine values.

4.2.3 Summary

The purpose of reporting is not to replace a paper form with an electronic form. Its most important job is to establish the time relationship between the work order and the machine data.

4.3 Cycle Time Deviation Alerts

Cycle Time is one of the easiest injection molding indicators to acquire and understand.

But “the current cycle is 35 seconds” has limited value by itself. The useful question is whether 35 seconds is normal for this product on this machine.

4.3.1 How the Alert Works

Figure 4.3-1: Cycle time outside the defined range is highlighted

1. Confirm What Is Running

The machine must first be linked to the correct item, work order, and mold. Comparing the current cycle with the wrong product baseline is meaningless.

2. Establish Historical Cycle Time

The system stores historical cycle data for the same product and machine as a comparison baseline.

Baseline-update rules matter. Abnormal production should not automatically become the new “normal.”

3. Compare Current and Historical Values

After a cycle completes, the current value is compared with the relevant baseline. Values outside the configured range are highlighted.

4.3.2 Return from the Alert to the Production Floor

Figure 4.3-2: Current machine-data monitoring

Longer cycle time can be caused by cooling changes, robot waiting, operator actions, materials, molds, or other production conditions.

The system's role is to identify which machine deserves attention first, not to replace the technician's diagnosis.

Historical trends can also show whether the deviation occurred for one cycle or continued over a longer period.

4.3.3 Practical Value

• Less manual machine-by-machine checking: Machines outside the baseline are highlighted first.
• Time-based history: The system can show when the deviation began.
• Work-order traceability: The deviation can be linked to the work order and product.
• Reusable baseline: The next production run has a historical reference.

4.3.4 Summary

Cycle Time deviation alerts are not technically complicated, but they are a practical first real-time anomaly function for an injection molder because the value is easy to obtain and easy for production personnel to understand.

4.4 GA Production Scheduling

Injection molding scheduling is difficult not only because there are many orders. Each order may be constrained by mold compatibility, machine tonnage, due date, changeover, material, and current machine conditions.

The system uses a Genetic Algorithm (GA) to generate an initial feasible schedule, then production planners adjust it according to actual floor conditions.

4.4.1 GA Pre-Scheduling

Figure 4.4-1: GA pre-scheduling Gantt chart

GA does not produce a mathematically guaranteed “perfect schedule.” It repeatedly generates and evaluates candidate schedules and retains better solutions according to the fitness rules.

A schedule can include work-order sequence, assigned machine, and start/end times. The scoring function can include:

• Due-date delay
• Total completion time
• Number of mold changes
• Machine availability
• Mold/machine compatibility
• Other plant-specific constraints

Selection, crossover, and mutation are repeated across generations, and suitable schedules are presented in a Gantt chart.

4.4.2 Manual Adjustment Through RWD and RESTful API

Figure 4.4-2: Schedule list and manual adjustment interface

The algorithm does not know every floor condition. A mold may be under repair, a machine may stop unexpectedly, or a customer may suddenly pull an order forward.

GA results therefore remain editable. Planners can change sequence, move a job to another machine, or insert an urgent order, then save the updated schedule.

4.4.3 Practical Value

The main difference from pure manual scheduling is that the system performs a large amount of repetitive constraint checking before the planner begins the final adjustment.

An urgent order can also be evaluated against the downstream impact on other jobs more quickly.

4.4.4 Summary

GA is an optimization tool, not a fully autonomous AI production planner. The system handles quantifiable constraints; production personnel add the conditions the system does not yet know.

4.5 Mobile Warehouse Web Interface

A common warehouse problem is that material or finished goods are physically in front of the operator, but the transaction still requires returning to a fixed desktop ERP terminal.

A responsive web interface allows warehouse personnel to query, receive, count, and register inventory near the actual material or rack.

4.5.1 Material Lookup and Traceability

Figure 4.5-1: Material lookup

The primary purpose of the QR Code is fast identification. It reduces manual entry of item numbers, batch numbers, or document numbers.

After scanning, the system queries ERP or WMS for the corresponding record. The QR Code itself does not need to contain the complete inventory record.

4.5.2 Warehouse Receiving

Figure 4.5-2: Warehouse receiving

The warehouse operator scans the label, verifies item, document, and quantity, and confirms the actual receipt before inventory is updated.

Because receiving changes inventory, authentication, permission control, and transaction history are important.

4.5.3 Cycle Counting and Variance Adjustment

Figure 4.5-3: Cycle counting

Actual quantity can be entered or scanned on site and compared with system inventory.

Differences should first be flagged and then adjusted through the company's authorization workflow rather than allowing any scan to change official inventory immediately.

4.5.4 Finished-Goods Receiving and Traceability

Figure 4.5-4: Finished-goods receiving

If the label already includes work order, item number, and quantity, scanning can populate the existing data rather than forcing the user to enter it again.

When the receiving record is linked back to production reporting, the factory can trace which production work order created that inventory lot.

4.5.5 Summary

The goal of the mobile warehouse interface is not to rebuild ERP. It brings a small set of warehouse operations to the place where the physical material is handled.

The process platform organizes the data, production reporting establishes context, Cycle Time alerts identify suspicious changes, GA supports scheduling, and the mobile warehouse links the post-production material flow back into the system.