Deploy Multiple AI Edge Workloads with FactoryPulse on Qualcomm Dragonwing IQ 9075

[pigzippa47]

This blog post was originally published at Qualcomm’s website. It is reprinted here with the permission of Qualcomm.

How do you run Vision AI, GenAI with Retrieval‑Augmented Generation (RAG), and speech recognition together on a single edge platform without sacrificing performance?

The FactoryPulse solution showcases how to efficiently deploy multiple AI workloads – including Vision AI, GenAI with Retrieval-Augmented Generation (RAG), and speech recognition on the Qualcomm Dragonwing IQ-9075 processor, using Qualcomm Hexagon Tensor Processor (HTP) accelerators.

Developed as a collaboration between Qualcomm Technologies, Inc. and Qt Group, this solution highlights how Qt-based industrial applications can leverage Dragonwing IoT platforms to deliver real-world AI solutions in factory environments.

Explore the FactoryPulse Multi‑AI Edge Solution​

FactoryPulse is a complete solution for operating a factory with four production lines. It illustrates five common use cases found in industrial automation environments:

• Factory Management – overseeing activities to ensure efficiency, safety, and productivity.

• Defect Detection – identifying product imperfections using Vision AI.

• Worker Safety – monitoring protective equipment and restricted areas.

• Predictive Maintenance – monitoring equipment, anticipating failures, and assisting maintenance.

• Demo Mode – showcasing the system capabilities.

Figure 1. Digital twin view of the IAD Complex 01A factory — all four production lines are shown with conveyor segments color-coded by status. The solution renders this in 3D using Qt Quick 3D.​

Set up the Dragonwing IQ‑9075 Edge AI system​

The FactoryPulse solution setup includes:

• The Dragonwing IQ-9075 EVK

• 2 touch displays

• 2 cameras

• A speaker and a microphone

This configuration demonstrates how a single powerful embedded platform can manage multiple AI-driven workloads alongside real-time rendering of 2D and 3D graphics.

Figure 2. FactoryPulse physical setup — two displays running the operator dashboard and the digital twin, a ring-light conveyor rig with overhead camera for quality-check inference, and the Dragonwing IQ-9075 board.​

Build the multi‑AI software stack for edge deployment​

To implement AI across these use cases, the solution combines three categories of components:

• Runtimes – the Qualcomm Intelligent Multimedia SDK (IM SDK), the Edge Impulse development platform, and multiple AI runtimes running directly on the board.

• GenAI libraries – the Genie software library runs inference for GenAI models with RAG.

• Media integration – Qualcomm Technologies and Edge Impulse GStreamer elements support the Vision AI use cases; Whisper Automatic Speech Recognition (ASR) enables speech-to-text conversion.Learn more: Qualcomm Intelligent Multimedia SDK · Edge Impulse for Qualcomm

Start building with Qt + Qualcomm today: Request an extended evaluation license​

Monitor and control AI workloads with the Operator Dashboard​

The operator dashboard consolidates every active workload on a single screen: safety-area monitoring, line production metrics, resource utilization, and the defect-detection camera view. Operators can switch production lines, toggle individual subsystems, and enter Demo or Maintenance mode from the top bar.

Figure 3. Operator dashboard in dark mode – Vision AI defect detection, safety-area access log, resource-utilization trends, and real-time line telemetry shown side by side.​

Figure 4. Full view of the SCADA system and portable interface.​

Detect product defects with vision AI at the edge​

The solution includes AI models for detecting product defects:

• Edge Impulse’s YOLO Pro model

• Generated using Edge Impulse’s cloud service

• Deployed with Edge Impulse GStreamer elements for inference and overlay

This demonstrates how industrial vision applications can be developed in the cloud and deployed efficiently to Dragonwing edge hardware.

Enable real‑time worker safety monitoring with Edge AI​

Worker safety is supported through two TensorFlow Lite (TFLite) models:

• A model for detecting protective gear (safety helmet and vest)

• A model for tracking people near restricted areas

These models use Qualcomm GStreamer elements for on-device inference acceleration, enabling real-time safety monitoring without reliance on cloud processing.

Run GenAI with RAG for factory operations and maintenance​

For factory management and part-replacement scenarios, the solution integrates:

• Meta Llama 3.2 Instruct 3B model

• Compiled using Qualcomm SDKs for the Dragonwing IQ-9075 Hexagon HTP AI accelerator

Retrieval-Augmented Generation (RAG)​

• BAAI BGE-large model generates embeddings from user inputs

• Compiled using Qualcomm SDKs for HTP acceleration

• Documents and embeddings are stored in Chroma DB

This setup allows the system to process natural-language queries and retrieve relevant information efficiently at the edge.

Enable on‑device speech recognition with ASR​

For voice interaction, the solution uses:

• The OpenAI Whisper-small model

• The Qualcomm AI Hub Whisper-Base ASR model

This enables speech-to-text conversion directly on the device, supporting hands-free interaction within the factory environment.

Optimize AI workloads for Hexagon HTP acceleration​

All AI models used in the solution — Edge Impulse’s YOLO Pro for defect detection, two TFLite models for worker safety, Meta Llama 3.2 Instruct 3B for GenAI, BAAI BGE-large for RAG embeddings, and OpenAI Whisper-small for ASR – are optimized for the Hexagon Tensor Processor (HTP) accelerators. They run across the two HTP cores of the Dragonwing IQ-9075, demonstrating how multiple AI workloads can execute efficiently on a single embedded platform.

Render real‑time 2D and 3D interfaces with Adreno GPU​

The graphical user interface (GUI) on both displays is implemented using the declarative Qt Modeling Language (QML). The digital twin view of the factory on the first screen, together with the two live camera views for Vision AI and the live metrics shown on the second screen, are rendered at approximately 60 frames per second (FPS), ensuring a real-time, responsive interface.

Build multi‑AI edge applications with this architecture​

The FactoryPulse solution shows how developers can:

• Combine Vision AI, GenAI, RAG, and ASR in a single system

• Optimize models for Hexagon HTP accelerators

• Use Qualcomm SDKs and Edge Impulse to accelerate deployment

• Build Qt-based industrial applications powered by on-device AI

Start building with Qt + Qualcomm today: Request an extended evaluation license​

Ready to get started?​

Start building today: Request an extended evaluation license to build your own solutions with Qt and Qualcomm Technologies

Train and deploy: Explore Edge Impulse workflows for Qualcomm Technologies to train, deploy, and run defect detection using Edge Impulse on Qualcomm hardware.

Watch it in action: Check out our webinar, FactoryPulse: Deploying Vision AI and Generative AI at the Industrial Edge with Qt, Qualcomm Technologies, and Edge Impulse.

Get hands-on: Follow our step-by-step tutorial, Build Along: Building Smarter AI UIs with Qt on Dragonwing.

Explore the tech: Learn more about Qt and QML to dive deeper into the technologies powering this solution.

Coming soon: Stay tuned for our upcoming technical deep-dive posts on Vision AI and GenAI + RAG!

The post Deploy Multiple AI Edge Workloads with FactoryPulse on Qualcomm Dragonwing IQ 9075 appeared first on Edge AI and Vision Alliance.