RoboFork: digital twin and AI-based PHM for in-plant logistics fleets

Changes to in-plant logistics, such as new routes, fleet size or station layout, are risky to test directly on the factory floor, and battery or motor failures of logistics robots are usually noticed only after they occur. This result combines a digital twin of the factory logistics environment with a predictive health management (PHM) layer. The twin models the real site plan and runs a multi-vehicle fleet driven by task packages derived from real logistics records, so that what-if scenarios can be evaluated before they go live.

The PHM layer estimates battery state of charge and state of health, motor condition and cell imbalance, and detects anomalies by combining rule-based checks with LSTM models. A locally hosted language model explains the health indicators to operators in natural language, without data leaving the site. The result was developed in the Sa4CPS in-factory logistics use case.

• Factory digital twin built from CAD (DWG/DXF) site plans, with multi-vehicle navigation and collision avoidance • Conversion of real logistics records into simulation task packages • Fleet-wide PHM: SoC, SoH, motor prognostics, cell imbalance • Hybrid anomaly detection (rule-based + LSTM) • Natural-language explanation of PHM indicators with a locally hosted LLM • Live telemetry exchange with monitoring platforms over MQTT

• Unity-based simulation environment • MQTT (ISO/IEC 20922) and REST/JSON interfaces; Docker (OCI) deployment • Edge devices and IoT gateways with MQTT or REST; vehicle data via CAN bus (ISO 11898) • Standard PHM API for CPS and IoT platforms

Manufacturers with in-plant logistics, operators of electric vehicle and robot fleets, and industrial IoT platform providers looking for a predictive maintenance component.

Commercial license. The PHM library and APIs can be licensed separately; the digital twin is offered as a project-based service.