Why use WEDA?
Edge AI development involves more than training a model. Getting from a working prototype to a production fleet—reliably, at scale—requires solving five recurring engineering challenges. This is what WEDA is designed for.
1. Bridging the Deployment Gap
The Challenge: A model trained on an NVIDIA workstation cannot run directly on edge devices with different hardware (NXP, Qualcomm, Intel). Each platform requires different drivers, OS configurations, and container setups. This "deployment gap" can consume weeks or months of engineering time.
WEDA's Solution: Ready-to-Dev Containers provide pre-configured environments for major hardware platforms with GPU/NPU access already enabled. Developers use a consistent workflow regardless of target hardware, reducing time from PoC to production.
2. Managing Devices at Scale
The Challenge: Manual deployment works for a pilot with 5–10 devices. At hundreds or thousands of devices across multiple locations, traditional approaches break down—on-site configuration, firewall-blocked remote access, and uncoordinated AI model updates become unmanageable.
WEDA's Solution:
- Auto-Provisioning: WEDA Node automatically registers devices to WEDA Core via MAC address—no manual field setup required.
- Virtual TCP Tunnel: Secure reverse tunnel provides direct remote access (SSH, VNC, RDP) without complex firewall configuration.
- Centralized Model Management: Deploy and update AI models across all devices from WEDA Core, with support for staged rollouts and offline device synchronization.
- Model Protection: Built-in mechanisms to prevent unauthorized model extraction or replication.
3. Unified Data from Scattered Sources
The Challenge: Industrial systems generate data from multiple sensors, cameras, and I/O points across different devices and protocols. Building a unified view requires custom integration code for each data source—fragile and error-prone.
WEDA's Solution: The "Logical Device" concept in WEDA Core aggregates multiple physical data points into a single logical entity (e.g., "Production Line A"). WEDA supports Numeric, Image (Binary), and Datapack (JSON) formats, enabling sophisticated analysis and digital twin simulation without custom integration code.
4. Simplifying Hardware Integration
The Challenge: Advantech hardware—motherboards, I/O modules, cameras—each has different C-language drivers and APIs. Bridging these with AI applications written in Python or C# is time-consuming. For custom sensors, developers also need to understand Digital Twin protocols to communicate with WEDA Core.
WEDA's Solution:
- Device Library: Abstracts hardware differences so developers can control Advantech peripherals—GPIO, sensors, cameras, I/O modules—using Python or C# without learning device-specific C APIs.
- WEDA SubNode Framework: Eliminates the need to understand Digital Twin protocols. Developers implement sensor integration logic; the framework handles all protocol complexity.
5. Proactive Monitoring and Continuous Improvement
The Challenge: Production deployments require continuous health monitoring, fast troubleshooting, and iterative model improvements—all without on-site visits or manual workflows.
WEDA's Solution:
- Real-time Health Monitoring: Continuous monitoring with automatic alerts for CPU, temperature, disk, and memory across all devices.
- Remote Configuration with Shadow Technology: Modify device configurations remotely via WEDA Core APIs. Shadow technology ensures changes persist even when devices are temporarily offline.
- MLOps Integration: WEDA integrates with Edge Impulse and other MLOps platforms, enabling a continuous loop of data collection → model training → deployment → inference—without manual export/import workflows.