In 2026, smart manufacturing is no longer well described as a path toward a fully unmanned factory. For most small and mid-sized injection molders in Taiwan, the practical questions are much more basic: What is each machine producing right now? Has the actual cycle time drifted away from normal? Where should an urgent order be inserted? If a customer complaint occurs, can the factory retrieve the work order, mold, and production records from that period?

Once these questions can be answered reliably, AI, predictive maintenance, and automated scheduling become much more realistic.

Author: Jush Li

1.1 Taiwan Plastics Manufacturing Overview (2025 Full-Year Data)

1.1.1 Industry Scale and Structure

Because 2026 is not yet a complete calendar year, the most appropriate full-year baseline is 2025 rather than an older market figure presented as if it were current.

Based on Taiwan's industrial production statistics, the 2025 production value of the plastics products manufacturing industry was approximately NT$389.7 billion. Plastic sheets, plates, pipes, and related products accounted for about NT$56.4 billion; plastic film and bags about NT$63.9 billion; plastic housings and accessories about NT$46.9 billion; and other plastic products about NT$222.5 billion.

These categories should not be treated as the production value of injection molding alone. Plastic products are made by many processes, and a statistical category can contain products made by different methods. It is more useful to look at where injection-molded parts are actually used.

Injection-molded components appear across electronics, automotive parts, appliances, medical and inspection equipment, sporting goods, industrial components, and consumer products. This wide product mix is one reason injection molders often operate in a high-mix environment with frequent mold changes and material changes.

Figure 1.1-1: Plastic injection molding equipment

1.1.2 What Is Changing in Injection Molding?

Injection molding itself is not a new process, but order patterns have changed. Long production runs with one mold fixed on one machine still exist, yet more factories also have to handle pilot runs, smaller lot sizes, urgent orders, and frequent product changes.

The biggest impact is not that the machines are incapable of producing the parts. The bigger issue is scheduling complexity. Machine tonnage, mold availability, material preparation, mold condition, changeover time, and due dates all have to be considered together.

This is also why high machine utilization does not automatically mean smooth delivery. If materials, molds, operators, or downstream operations are not ready, a machine can appear busy while the overall production flow is still losing time.

1.1.3 Energy and Carbon Management

Energy remains a significant cost for injection molding. In March 2026, Taiwan's electricity-rate review kept the average electricity price at NT$3.7823 per kWh. Actual factory electricity costs still depend on tariff category, contracted demand, time-of-use pricing, and peak/off-peak consumption.

For that reason, looking only at the monthly electricity bill is not enough. A factory gains more useful information when energy can be related to a machine, product, work order, or production period.

2026 is also the first year in which Taiwan's carbon-fee system moves into actual filing and payment based on 2025 emissions. The initial regulated group covers power, gas supply, and manufacturing facilities whose direct emissions plus electricity-related indirect emissions reach at least 25,000 tCO2e per year.

1.1.4 CBAM Is in Its Definitive Period, but Plastics Need the Right Interpretation

The EU Carbon Border Adjustment Mechanism (CBAM) entered its definitive period on January 1, 2026. Its current direct scope primarily covers cement, iron and steel, aluminium, fertilisers, electricity, and hydrogen. General plastic injection-molded products are not a broad direct CBAM category.

For injection molding suppliers, the more immediate impact is often customer-driven. Once a brand owner or major supply-chain customer starts building product-carbon data, upstream material and processing suppliers may be asked to provide activity data.

In practice, carbon management for many injection molders currently behaves more like a supply-chain data requirement than a direct CBAM charge.

1.1.5 Policy Direction: Digitalization and Net-Zero Together

Taiwan has been promoting digital transformation, AI pilot lines, equipment automation, energy management, AIoT, and process improvement for traditional manufacturing sectors including plastics and rubber. The direction increasingly combines digitalization with net-zero management.

This is close to what factories need in practice: first acquire reliable machine data, then identify which work order and product the data belongs to, and only then use it for quality, energy, or scheduling analysis.

1.2 Five Common Challenges in Injection Molding

Injection molding problems rarely occur in isolation. More urgent orders mean more schedule changes and mold changes. More changeovers mean more setup and trial time. Unstable conditions can then affect yield and delivery.

Figure 1.2-1: Digital transformation in plastics manufacturing

These problems can be handled manually when the production mix is simple. As product variety increases, delivery windows shorten, and customers ask for more traceability, the time required to collect and reconcile information becomes a cost by itself.

1.3 What Smart Manufacturing Means for Injection Molding

Smart manufacturing in injection molding does not have to begin with AI. For many factories, the practical starting point is to place equipment, production, and management data into one traceable process.

The injection molding machine already knows its cycle time, cycle count, and operating state. ERP knows the work order, but it does not necessarily know which physical cycles were produced at 10:32 a.m. MES fills that gap by linking the order to the machine, product, mold, operator, and production period.

Smart Manufacturing Does Not Remove Experienced Technicians

Many molding decisions still depend on production experience. A mold may have just been repaired, a material lot may behave differently, or an assembly problem may only appear downstream.

A more realistic approach is to preserve the part of that experience that can be recorded: historical cycle time, molding conditions, mold history, common defects, and corrective actions.

AI Should Come After the Data Structure

If historical records cannot reliably identify the product, machine, or work order, adding AI usually does not solve the underlying problem.

Our preferred sequence is to make the data structure reliable first, then decide which problems are suitable for algorithms. Scheduling with many constraints can use optimization methods, while predictive maintenance should wait until enough equipment-health and failure-history data exists.

Measure Value with Production KPIs

There is no universal rule that smart manufacturing will improve yield by a fixed percentage or reduce labor by a fixed amount. Every plant starts from a different baseline.

A more useful approach is to establish a baseline before implementation and then compare practical questions such as:

• How long does manual scheduling take?
• How long does it take to replan after an urgent order is inserted?
• Can the work order be accurately matched to the machine production interval?
• How long does it take to retrieve production data after a customer complaint?
• Are downtime reasons classified, or does the system only know the machine stopped?
• Can current cycle time be compared with a normal historical baseline?
• Can energy use begin to be measured by product or machine?

Smart Manufacturing in 2026 Is Increasingly “Digital + Net-Zero”

Machine connectivity is no longer only about OEE. The same machine and work-order data can also support energy analysis, carbon data, quality history, and product traceability.

A small or mid-sized injection molder does not have to deploy a complete smart factory at once. Starting with one machine, one product, or one frequently problematic process and proving the data flow first is usually easier to maintain and expand.