For most Japanese manufacturing SMEs, the programmable logic controller (PLC) is the heart of the factory. It runs conveyors, presses, packaging lines and machining cells, often reliably for decades. So when the conversation turns to DX in manufacturing and AI, a natural worry follows: do we have to replace everything that already works?
The answer is no. The most cost-effective route is to keep the PLC doing what it does best, deterministic, safe, real-time control, and add AI alongside it to analyse data and support better decisions. This “PLC + AI” approach lets SMEs modernise existing lines step by step, with investment that scales as value is proven.
PLC and AI: different jobs, one system
It helps to be clear about the division of labour.
The PLC controls. It executes logic in fixed, predictable cycles, handles interlocks and safety-related functions, and responds in milliseconds. These are exactly the properties you want in machine control.
AI analyses and advises. AI models find patterns in large amounts of data: vibration trends that precede a bearing failure, visual signs of a defect, or combinations of settings that produce the best yield. AI outputs are probabilistic, which makes them powerful for prediction but unsuitable as the sole basis for safety-critical control.
The winning architecture keeps these roles distinct. AI informs, recommends and, where appropriate, adjusts non-critical setpoints within limits that the PLC enforces. The PLC remains the final authority on the machine.
A practical PLC + AI architecture
A typical, scalable setup for an SME has four layers.
- Field layer: existing sensors, drives and actuators, plus any low-cost additions such as vibration sensors, current sensors or cameras.
- Control layer: the existing PLCs. Japanese factories commonly run controllers from manufacturers such as Mitsubishi Electric (MELSEC), OMRON (SYSMAC), Keyence (KV series), Fuji Electric (MICREX), Panasonic and JTEKT (TOYOPUC), and all of these support ways to share data with higher-level systems.
- Edge layer: an industrial PC or edge device near the line that collects data from the PLCs, runs AI models locally, and returns results. Processing at the edge keeps latency low, reduces cloud costs and keeps sensitive production data on site.
- Application layer: dashboards, alerts, and integration with production management or ERP systems, on premises or in the cloud.
Connecting to the PLC. Data is usually read through standard industrial protocols. OPC UA is widely supported and vendor-neutral; many PLC makers also offer their own protocols and gateways. In Japan, the Edgecross Consortium, established in November 2017 by Advantech, OMRON, NEC, IBM Japan, Oracle Japan and Mitsubishi Electric, promoted an open edge-computing platform for manufacturing. In January 2025, however, it announced that it will end its promotion activities in FY2026, saying its original purpose had been fulfilled. Treat it as an existing option to check compatibility with, rather than a growing ecosystem; vendor-neutral standards such as OPC UA remain the safer long-term choice.
Starting with read-only data collection is strongly recommended. It lets you create value from data without touching control logic, which keeps risk very low.
High-value use cases for manufacturing SMEs
Predictive maintenance
Vibration, temperature, motor current and cycle-time data can reveal early signs of wear. An AI model trained on this data can flag equipment likely to fail, so maintenance happens during planned downtime rather than in the middle of a production run.
AI visual inspection
Cameras at the line combined with image-recognition models can detect scratches, dents, missing components or labelling errors at production speed. The PLC can then trigger a reject gate or stop the line according to rules you define. For SMEs facing a shortage of experienced inspectors, this provides consistency, while people handle borderline cases.
Process optimisation and quality prediction
By linking process parameters (temperature, pressure, speed, material batch) with quality results, AI can identify which settings produce the best yield and warn operators when conditions drift towards defects.
Energy monitoring
Tracking energy consumption by machine and by product helps identify waste, schedule energy-intensive jobs more efficiently and support decarbonisation reporting requested by larger customers.
Downtime and OEE analysis
Automatically capturing stoppage reasons and cycle times from the PLC gives an accurate view of overall equipment effectiveness (OEE), replacing paper logs and guesswork.
Keeping it safe and secure
Never let AI bypass safety functions. Emergency stops, guarding and safety interlocks must remain in dedicated safety systems and PLC logic, designed and validated under the applicable machinery safety standards.
Use limits and human confirmation. If AI recommends setpoint changes, enforce minimum and maximum limits in the PLC, and start with operator approval before any automatic adjustment.
Secure the network. Connecting operational technology (OT) to IT networks introduces cyber risk. Segment networks, control remote access, keep an asset inventory, and use frameworks such as the IEC 62443 series for industrial automation security as a reference.
Plan for model drift. Materials, tools and products change. Monitor model accuracy and schedule periodic retraining.
A phased plan for scalable automation in SMEs
Phase 1: Connect and see (1 to 3 months). Collect data from one line in read-only mode. Build a simple dashboard for OEE, stoppages and key process values. Many SMEs uncover quick wins at this stage without any AI at all.
Phase 2: Predict (3 to 6 months). Choose one use case, such as predictive maintenance on a critical machine or visual inspection of a high-defect product. Train a model, run it in “advisory mode” alongside current practice, and measure accuracy.
Phase 3: Act (6 months onwards). Once the model has proven reliable, integrate its outputs into operations: automatic reject gates, maintenance work orders, or operator prompts with limits enforced by the PLC.
Phase 4: Scale. Reuse the same edge architecture, data model and dashboards on other lines. Because the foundation is already in place, each additional line costs much less than the first.
Keeping costs under control
- Reuse existing PLCs and machines rather than replacing them.
- Start with one line and one use case that has a clear financial impact.
- Choose open standards such as OPC UA to avoid lock-in.
- Run AI at the edge where it reduces cloud and bandwidth costs.
- Check subsidy eligibility. As of September 2026, the main options are the New Business Expansion and Monozukuri/Commerce/Service Subsidy (新事業進出・ものづくり商業サービス補助金), which from FY2026 merges the former Monozukuri Subsidy and the SME New Business Expansion Subsidy (first call: 30 September to 30 October 2026), and the Digital/AI Introduction Subsidy. Recipients commit to multi-year targets, including growth in value added and wages, so read the official guidelines (公募要領) before applying.
- Access specialised skills flexibly. Data engineering and model development can be sourced from experienced partners, including offshore teams, while your own engineers retain ownership of control systems.
Conclusion
PLC + AI is one of the most practical forms of DX in manufacturing for Japanese SMEs. It respects the reliability of existing control systems, avoids large replacement costs, and delivers measurable benefits in uptime, quality and efficiency. Start with data, prove value on one line, keep safety firmly in the PLC, and scale what works.