AI Integration with Ignition for Andon Light Integration

Ignition excels at capturing Andon signals—machine stops, quality calls, material requests—through its SCADA tags and MES event framework. However, these signals often lack context: a red light could mean a mechanical jam, a safety gate open, or a missing tool. An AI layer, connected via Ignition's scripting engine or REST API, analyzes the surrounding context. It cross-references the signal with real-time OEE data, recent maintenance logs from a connected CMMS, the operator's skill level, and even live video feeds to classify the issue, predict its severity, and determine the optimal response path before a human even reviews the alert.

The implementation wires an AI agent into Ignition's notification pipelines. When an Andon tag triggers, Ignition's system.tag.write or an event script packages the event with contextual data (machine state, product SKU, shift) into a payload for an AI service. The model returns a structured classification (e.g., issue_type: "mechanical_jam", severity: "high", recommended_skill: "Level_3_Technician"). Ignition then uses this to automate the next steps: dynamically updating alarm priorities in the HMI, creating a detailed work order in the CMMS via its database bridge, and dispatching an alert with pre-populated troubleshooting steps to the right team's mobile device via SMS or Teams integration.

Governance is built into the feedback loop. Every AI-suggested action and its outcome are logged back to Ignition's SQL database. This creates a closed-loop system where the model's predictions are continuously compared to the actual resolution documented by technicians. This audit trail is crucial for validating ROI, managing change control, and retraining models. Rollout starts with a single, high-impact line—like a final assembly station—where signal volume is high and resolution data is plentiful, proving the workflow before scaling to the entire plant floor.

The integration architecture treats Ignition's Andon system as a real-time event stream. AI models connect to Ignition's Tag Historian or Transaction Groups to consume light state changes (e.g., red, yellow, green), associated Alarm Status tags, and contextual metadata from the SQL Bridge (like work order, station ID, operator). This raw signal is enriched with live production data—current machine cycle time, recent quality checks, operator experience level—to create a rich context payload for the AI.

A dedicated AI Inference Service, deployed as a containerized microservice or within Ignition's Gateway Scripting environment, processes this payload. It uses a classification model to interpret the event (e.g., 'material jam' vs. 'tool breakage'), then executes a multi-step workflow: 1) Automated Escalation via Ignition's notification system (email/SMS) or direct API calls to a CMMS like SAP PM to create a work order. 2) Intelligent Dispatch by querying a maintenance skills database to route the alert to the best-suited technician. 3) Resolution Logging where the AI drafts a summary of the issue and solution into a Knowledge Base Tag in Ignition, creating a searchable memory for future similar events.

Governance is managed through Ignition's Audit Profile system, logging all AI inferences and actions. A Human-in-the-Loop (HITL) approval step can be configured for high-impact classifications before dispatch. The model is retrained using feedback loops from resolved work orders and technician notes, which are pulled back into the AI service via Ignition's database connections. This creates a closed-loop system where the Andon integration becomes more accurate and context-aware over time, reducing mean time to repair (MTTR) without replacing the existing PLC or HMI logic.

Secure Data Flow & Model Governance: The integration architecture must enforce strict access controls. AI models should never have direct, unmonitored write access to Ignition tags or SQL databases. Instead, inferences are passed through a secure middleware layer (e.g., a dedicated microservice or Ignition's built-in scripting modules) that validates the output against business rules before creating a work order in the CMMS or updating an escalation status. All model calls, input data (e.g., Andon signal context, machine state), and recommended actions are logged to a separate audit database with full traceability for compliance and model performance review.

Phased Rollout Strategy: Start with a monitor-and-alert phase in a single production cell or line. Configure the AI to analyze Andon signals and generate internal alerts or recommendations visible only to a pilot team of engineers and supervisors via a dedicated Ignition Perspective screen. This validates accuracy without changing live processes. Next, move to an assisted escalation phase, where the system suggests the correct support team and resolution steps to dispatchers, who retain final approval. Finally, after achieving a high confidence threshold, enable automated dispatch for specific, high-frequency issue types (e.g., 'Material Jam - Line 1'), automatically creating tickets in the connected CMMS like Fiix or SAP PM while notifying the assigned technician.

Continuous Feedback & Model Retraining: Governance doesn't end at deployment. Implement a closed-loop feedback system where technician resolution notes and time-to-repair from the CMMS are fed back into the AI model's training pipeline. This allows the system to learn from corrections and improve its classification and recommendation accuracy over time. Establish a regular review cadence with operations leadership to evaluate key metrics: reduction in Andon response time, escalation accuracy, and the rate of human overrides, adjusting the automation scope and model confidence thresholds accordingly.