Abstract
As manufacturing transitions toward Human-Cyber-Physical Systems (HCPS), the complexity of managing heterogeneous hardware and data grows. This paper introduces XiUOS, an open-source "Ubiquitous Operating System" (UOS) specifically designed for the Industrial Internet of Things (IIoT). XiUOS employs a modular four-layered architecture that integrates sensing, networking, AI-driven data analysis, and PLC-based control. By providing a flexible, POSIX-friendly alternative to proprietary platforms, XiUOS aims to lower the barrier for small-to-medium-sized factories to adopt intelligent manufacturing technologies.
Introduction
The evolution of industrial environments into HCPS necessitates a new class of operating systems capable of managing the "ubiquitous" nature of modern manufacturing. Traditional systems often struggle with the diverse requirements of IIoT, where devices range from simple sensors to complex robotic arms. Currently, the market is dominated by closed-source, proprietary platforms like Azure IoT or Siemens MindSphere, which can be prohibitively expensive or rigid for smaller enterprises dealing with legacy equipment. XiUOS addresses these challenges by offering an open-source, extensible framework tailored to industrial needs.
System Architecture
XiUOS is built upon a four-layered architecture designed to maintain real-time performance even on resource-constrained devices. This structure ensures a clean separation between hardware-specific details and high-level industrial logic.
- Hardware Layer: Utilizes a Hardware Abstraction Layer (HAL) to support ARM and RISC-V processors. It integrates industrial communicators (5G, LoRa, NB-IoT) and controllers such as Siemens and Mitsubishi PLCs.
- Kernel Layer: A minimal, POSIX-friendly kernel that balances abstraction with efficiency. It features real-time task scheduling and strict isolation between kernel and user modes to ensure system stability.
- Framework Layer: Acting as the system's "brain," this layer is organized into four functional pillars: Sensing, Networking, Knowing (AI), and Controlling.
- Application Layer: The top-most level where specific industrial logic, such as power monitoring or object recognition, is implemented.
The Four Pillars of the Framework Layer
The core innovation of XiUOS lies in its Framework Layer, which standardizes IIoT requirements into four distinct pillars:
- Sensing: Employs a physical-quantity-oriented programming model to simplify driver management and data parsing.
- Networking: Supports various protocols, including MQTT and wireless mesh technologies, to ensure robust connectivity across the factory floor.
- Knowing: Features a micro AI engine and speedup library for on-device data analysis, such as identifying instrument panels or detecting anomalies.
- Controlling: Provides a dedicated library for Programmable Logic Controllers (PLCs), enabling accurate and remote machine control.
Deployment and Results
XiUOS Version 1.0 was released in May 2021, supporting seven development boards and the QEMU emulator. The system's efficacy has been demonstrated through real-world deployments in two distinct industrial scenarios:
- Electric Power Monitoring: Successfully implemented at Qiantang River Electric Co.
- Workshop Environment Monitoring: Deployed at Zhejiang Great Shengda Packaging Co. to manage environmental data.
These deployments highlight the system's ability to handle practical industrial tasks while providing a roadmap for further integration of AI and advanced control features.
Conclusion and Future Work
XiUOS represents a significant step toward an open-source ecosystem for the Industrial Internet of Things. By releasing the system under the Mulan PSL v2 license, the developers aim to empower "digitally-lagging" factories to upgrade their infrastructure without vendor lock-in. Future development will focus on enhancing the Hardware Abstraction Layer, implementing Over-the-Air (OTA) updates, and integrating a lightweight real-time database. Additionally, the "Knowing" framework will be expanded to include more complex pattern recognition for defect detection and human-machine interaction.
Summary updated