Green box

Green Box: The 7.5ms Decision Engine. Autonomous Edge AI that predicts failure before it happens, on any plane , ship ,port , airport , factory and remote site. Renewable-powered. Arm NPU speed

Comment

About the Project: Green Box The Green Box project is a radical redefinition of industrial maintenance and safety, built upon the foundation of Arm's efficient AI architecture and the principle of absolute autonomy.

What Inspired Us The inspiration was twofold: the staggering economic drain of unplanned downtime ($1.4 Trillion annual loss globally) and the critical failure of infrastructure during emergencies. We realized that current cloud-based solutions are slow and lack an automated closed-loop logistics system. We don't just predict failure; we designed a system that makes a 7.5ms decision and instantly places the order for the replacement part.

How We Built the Project The core of the build focused on maximizing performance within a minimal power budget (2–5 Watts), linking that performance to the business model:

Foundation and Smart Application: We selected the Arm-based NXP i.MX 8M Plus processor to run Edge AI. The models are specialized CNNs/Autoencoders for vibration analysis.

Arm Efficiency: Models are trained in the cloud, then quantized to INT8 and deployed using TensorFlow Lite (TFLite) for ultra-fast, power-optimized execution on the Arm NPU.

Closing the Loop via oneM2M: The 7.5ms AI decision is encapsulated into a unified message and sent to the Marketplace (acting as an Application Entity - AE), ensuring the system instantly understands which spare part is needed.

Engineer Interface (The Smart Assistant): The Smart Assistant functions as a crucial Application Entity (AE) for field engineers. It receives the critical decision via oneM2M, presents it to the engineer graphically, provides diagnostic context and historical data analysis, and allows for human override or confirmation of maintenance actions, ensuring smooth coordination.

Logistical Activation: The Marketplace, upon receiving the oneM2M notification, activates the logistical response. The Drone is utilized as an immediate delivery tool, ensuring the required spare part reaches mobile assets (ships/aircraft) at their next stop, completely preventing unscheduled downtime.

Challenges We Faced The most significant challenge was seamlessly integrating high-speed edge performance with communication reliability:

Transmission Challenge: Ensuring the oneM2M message carrying the critical AI decision reaches the Marketplace (Cloud) reliably, even with automatic switching between intermittent channels (LoRa to Satellite Modem). This required building a Store-and-Forward software layer.

NPU Optimization: Guaranteeing the INT8 models utilized the Arm NPU with 100% efficiency to maintain the 2–5 Watt power budget, which required deep tuning of the TFLite framework.

What We Learned We learned that the true value of Edge AI lies in its ability to become Actionable:

The Winning Formula: We proved that combining the Arm NPU/TFLite stack with RTOS is the only reliable solution to achieve 7.5ms deterministic latency.

Value in the Chain: Green Box's ultimate value isn't just detecting the fault; it's oneM2M's ability to notify the Marketplace, which instantly triggers supply via the Drone, closing the loop and converting prediction into profit.

Autonomy Means Trust: Designing the system to perform analysis and delivery entirely independent of traditional infrastructure is the ultimate guarantee

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Green box

Built With

Share this project:

Updates