2026 Design Contest using Phytec’s phyBOARD®-RT1170 Development Kit

GridSense™ – Intelligent Energy Orchestration Platform

GridSense™ is an embedded controller designed to manage and coordinate multiple energy sources including utility power, solar PV, battery storage, generators, and controllable loads. The objective is to determine energy demand, evaluate available resources, and deploy energy where it is needed while prioritizing renewable generation whenever possible.

Using the PHYTEC phyBOARD®-RT1170, the system will communicate with meters, inverters, battery management systems, and generator controllers using industrial protocols such as Modbus and Ethernet. A custom decision engine, OptiFlow™, will continuously evaluate system conditions including load demand, battery state of charge, renewable production, and source availability to make real-time energy allocation decisions.

The prototype will demonstrate renewable-first energy management, battery optimization, load prioritization, automatic failover during outages, and a web-based dashboard showing real-time energy flows and system status.

The long-term goal is to develop a scalable platform that can be deployed in residential, commercial, and industrial applications to improve energy efficiency, resiliency, and utilization of renewable resources.

Thermal Imaging Diagnostic scanner (Wand)
A thin scanner with medical thermal cameras used to track inflammation, skin infections, or circulatory blood flow blocks without touching the patient.

The vision is to develop a diagnostic thermal imaging platform, an entire interconnected ecosystem that includes the scanner combined with advanced software, mobile apps, and data-management systems that allows to analyze, store, and report on the thermal data across multiple devices.

The scanner (called as wand) connects a digital thermal sensor grid, (the Melexis MLX90640) (a 32×24 thermal pixel array), via the I²C bus. This sensor internally converts thermal radiation to temperature values and streams the data digitally. The data matrix is small enough that the Cortex-M4 can fetch it rapidly.

The Cortex-M7 takes the raw temperature matrix, uses its 2D PXP Graphics Engine to smoothly interpolate the grid from a blocky 32×24 array into a high-resolution color heat map, and blends it on the screen over a normal camera preview from the MIPI-CSI camera port

The prototype for the thermal Imaging Diagnostic Wand will be developed using PHYTEC phyBOARD-RT1170.

RTCam - Industrial GigE Vision 2.2 Camera powered by i.MX RT1170

1. The future AI market demand
As AI infrastructure drastically increases, the demand for high-quality, uncompressed, and zero-latency machine vision system will skyrocket.
RTCam is a MCU alternative to the commercial industrial streaming cameras.
The goal is to build a fully compliant GigE Vision 2.2 industrial camera using the NXP i.MX RT1176, delivering deterministic real-time performance, instantaneous boot times, and low power consumption without the complexity and overhead of a Linux-based MPU.

2. Hardware & Acquisition
The project leverages the phyBOARD®-RT1170 Development Kit interfaced with a Sony IMX296LQR-C image sensor (Raspberry Pi Global Shutter Camera) via the MIPI-CSI interface.
The global shutter is mandatory for industrial automation to eliminate rolling-shutter artifacts on high-speed conveyor belts and robotic arms.

3. Software Architecture (Zephyr & LwIP)
The firmware will run entirely on Zephyr RTOS, utilizing the LwIP network stack optimized for the board’s Gigabit Ethernet port.
The core development focuses on two pillars of the GigE Vision standard:
GVCP (GigE Vision Control Protocol): Implementation of the UDP-based control plane. This handles device discovery and maps the GenICam XML registers, allowing the host to configure sensor parameters (Exposure, Gain, ROI) in real-time.
GVSP (GigE Vision Streaming Protocol): The performance bottleneck.
This handles the high-throughput UDP data plane, requiring strict management of IP packet fragmentation to stream uncompressed RAW/RGB video at Gigabit speeds with zero packet loss.

4. Technical Highlight: L1 Cache & eDMA Optimization
Raw CPU clock speed (1 GHz Cortex-M7) is not enough to sustain Gigabit network throughput from MIPI to the ENET_1G MAC.
To demonstrate a deep understanding of the ARM architecture, I will implement a “Zero-Copy” data pipeline using eDMA in tandem with extensive L1 D-Cache management.
MIPI RX buffers will be strictly aligned to the 32-byte cache line.
Precise Cache Invalidation (before read) and Cache Cleaning (after CPU write) routines will be implemented to ensure cache coherency during asynchronous DMA transfers.

5. Validation Strategy
Firmware robustness will be verified using industry-standard machine vision tools:
GVCP (Control Plane): Validated with Aravis (open-source framework) for device discovery, XML extraction, and register read/write stress testing.
GVSP (Data Plane): Validated with Pleora eBUS Player to certify stable, high-framerate streaming and zero packet-drop reliability.

6. What’s Next: The Path to Production
Once the firmware is validated on the development kit, the immediate next step is designing a custom, 4-layer carrier board tailored to the ultra-compact (30x30 mm) phyCORE®-RT1170 System on Module.
This board will only house the Gigabit PHY, power management (PoE), and mechanical connector for the C mount lenses.
The final result will be a production-ready, ultra-compact industrial cube camera.

Project Title: Mission-Critical ENIP/CIP to DECT NR+ Wireless Bridge

The Application (The Problem):
Modern industrial automation heavily relies on EtherNet/IP (ENIP/CIP) for deterministic control. However, transitioning these wired networks to wireless for mobile assets (like AGVs or robotic arms) often compromises determinism. Traditional Wi-Fi or Bluetooth lacks the Ultra-Reliable Low-Latency Communication (URLLC) required for mission-critical tasks. This project proposes an industrial wireless bridge that seamlessly translates standard L2 ENIP/CIP Ethernet traffic into a proprietary FTDMA-based DECT NR+ MAC stack, enabling wire-like reliability over a 5G non-cellular wireless link.

The Technical Fit (Why phyBOARD-RT1170):
The phyBOARD-RT1170 is the critical missing piece for this gateway architecture. Bridging deterministic industrial Ethernet to a time-slotted FTDMA wireless MAC requires immense real-time processing capabilities. The i.MX RT1170’s dual-core architecture (Cortex-M7 and M4) is perfectly suited for this: the M7 core can handle the deep packet inspection, IP/UDP header compression, and CIP object routing with near-zero latency, while the M4 core strictly manages the highly synchronized FTDMA scheduling and baseband interface. Furthermore, the board’s native Gigabit Ethernet interface with TSN (Time-Sensitive Networking) support provides the ideal physical entry point for industrial CIP traffic.

The Prototype (Expected Output):
The final prototype will demonstrate a functional L2 bridge. The phyBOARD-RT1170 will ingest standard ENIP/CIP packets via its Ethernet port, perform dynamic segmentation and adaptation, and interface with an external DECT NR+ RF module to transmit the data over FTDMA channels. The demonstration will showcase low-jitter, highly reliable wireless control replicating a wired industrial environment.

Quality lemon shorting system(QLSS)
Quality lemon shorting system(QLSS) is an embedded system which identifies the quality of Lemon in the given slot.It has three important elements as vision system,embedded controller and weight measuring scale.
Using the PHYTEC phyBOARD®-RT1170, the system will communicate with 5 different motor assemblies.It acts like an embedded controller.The vision system has a stereo vision camera with a GPU controller for vision processing.
The prototype will demonstrate the identification of quality lemons from the given slot.The long-term goal is to develop a scalable platform that can be deployed in lemon export industries.

This project proposes a compact Embedded Real‑Time Simulation and Control Module (RTSCM) based on the phyBOARD®‑RT1170. The module enables real‑time plant simulation, control algorithm execution, and hardware‑in‑the‑loop (HIL) interaction within a small microcontroller‑driven platform. It is designed for engineers who need a portable, deterministic, and efficient tool for validating control strategies or emulating system behavior without relying on large lab‑scale simulation systems.
Motivation:
Developers working in industrial automation, motor control, robotics, and power electronics often need real‑time testing environments. Large simulation tools can be expensive or impractical for rapid prototyping or field deployment. This project provides an embedded alternative that is lightweight, cost‑effective, and optimized for real‑time responsiveness.
Concept:
The module uses the dual-core i.MX RT1170 architecture to separate tasks efficiently:
• Cortex‑M7: runs simplified plant models (motor dynamics, converter approximations, robotic kinematics).
• Cortex‑M4: executes deterministic control loops (PID, state-space, FOC).
• Zephyr OS: provides predictable scheduling, modular task isolation, and secure system features.
• Industrial interfaces: Ethernet, CAN FD, UART, I2C, and USB allow real-time interaction with external devices.
This combination creates an agile platform capable of simulating responses, emulating sensors, and validating firmware under controlled timing conditions.
System Agility:
The RT1170-based module is agile in several ways:
• Fast startup and low latency: no Linux overhead, minimal jitter.
• Highly portable: small 30 × 30 mm SoM fits compact prototypes and test setups.
• Flexible architecture: simulation and control tasks split naturally between the two cores.
• Rich connectivity: supports industrial buses for various testing scenarios.
• Rapid model switching: allows developers to update or replace plant models quickly.
• Deterministic timing: supports tight control loops below 10–50 microseconds.
The system is ideal for quick experiments, onsite diagnostics, and rapid development cycles.
Realistic Capabilities:
The RT1170 is capable of meaningful embedded‑class real‑time simulation, including:
• Simplified electrical, mechanical, or kinematic models
• Sensor emulation (IMUs, encoders, current sensors)
• Closed-loop control testing
• Real-time communication via CAN FD and Ethernet
• Interfacing with external controllers and actuators
It is not intended to run large workstation-scale models or high-order numerical solvers. The system focuses on optimized, low-order models that fit microcontroller constraints.
Pros:
• Portable and field-ready
• Deterministic real-time performance
• Dual-core separation for simulation and control
• Rich industrial interfaces
• Lower cost and power usage than large HIL systems
• Works well with simplified models
• Supports sensor and actuator emulation
• Zephyr-based modular software environment
• Ideal for rapid prototyping and educational use
Cons:
• Not suitable for large or high-order models
• Limited memory compared to desktop systems
• Not intended for high-frequency switching converter HIL
• Requires simplified or optimized simulations
• Visualization limited unless MIPI-DSI is used
• Less numerical precision than FPGA-based simulators
Equipment / Components (Generic):
• RT1170-based simulation and control board
• Analog/digital I/O expansion modules
• Industrial communication adapters (Ethernet, CAN FD, UART)
• Sensor/actuator emulation hardware
• Signal isolation and conditioning modules
• Tools for model loading and parameter tuning
• Optional display for monitoring and adjustments
• Compact enclosure for industrial or field use
Target Applications:
• Motor drives and robotics actuator testing
• Power electronics control evaluation
• Industrial automation sensor emulation
• Closed-loop control algorithm development
• Portable field test benches
• IoT device behavior simulation
• Real-time modeling of simplified power systems
Impact:
The Embedded Real-Time Simulation and Control Module delivers a practical and agile platform for testing algorithms and embedded systems. It offers deterministic behavior, strong connectivity, and rapid prototyping benefits in a compact form. This approach enables faster development cycles, safer testing, and more accessible real-time experimentation for engineers working on next-generation industrial and IoT technologies.

Project Title:
AutoDefense Mobile Firewall

Project Description:
AutoDefense Mobile Firewall addresses an immediate need for the automotive market: a basic high-performance firewall and vehicle security appliance that also fulfills legally-mandated requirements for SAE 21434. As of 2024 vehicles are required to run detection and monitoring for networking run by the car’s telematics control unit (TCU), which is usually an LTE/5G modem. Aftermarket and older vehicles don’t have this at all, and many newer vehicles implement monitoring purely as an additional software task in a central SoC. I propose using the RT1170 as a device in between the TCU/modem and the rest of the vehicle to act as a fast stateful behavioral firewall. This appliance is a compact, low-cost network packet inspection device that sits between a vehicle’s always-on LTE/5G modem and Wi-Fi hotspot and the internal vehicle networks, inspecting and filtering incoming traffic to block attacks, scans, and compromised hotspot clients from reaching the internal part of the vehicle’s network. It brings network security to autos, RVs, and personal watercraft, protecting the growing cellular/hotspot attack surface without the cost, size, and added attack surface of a full Linux computer. I’m excited to leverage the 2 Etherent MACs as a transparent L2 bridge, third as a management tap, and lean on the dual core architecture for filtering through the Cortex-M7 (L2/L3/L4 stateful filtering, connection tracking, zero-copy DMA) and control/maintenance through the Cortex-M4 (anomaly logic, logging, config).

The RT1170 is an ideal use case for a deterministic networking appliance that is easy to model, simulate, and doe not have the timing jitter and security headaches of full-blown Linux. There are also some strong security appliances in hardware right on the RT1170 - secure boot, CAAM accelerator, TrustZone-M - all this is pretty strong for a security device and really justifies selecting this part for such an application.

I’m excited to take a shot at building this product using the RT1170 and would very much appreciate a kit such as this to prototype things on!

SecureEdge Predict:
===================
Industrial Predictive Maintenance & Edge-AI Condition Monitoring Platform

Project Overview:
=================
SecureEdge Predict is a secure, real-time Industrial Predictive Maintenance Platform designed for Industry 4.0
environments. The system continuously monitors equipment health, performs Edge-AI-based anomaly detection,
estimates Remaining Useful Life (RUL), and provides maintenance recommendations before failures occur. The
solution leverages the dual-core architecture of the PHYTEC phyBOARD-RT1170 and Zephyr RTOS to separate
deterministic real-time processing from AI analytics, networking, and security services.

Problem Statement:
==================
Traditional maintenance approaches are reactive or schedule-based, resulting in downtime, unnecessary
maintenance, and increased operational costs. SecureEdge Predict enables condition-based maintenance
through continuous monitoring and early fault prediction.

System Architecture:
====================
Machine condition data is acquired and processed on the Cortex-M4 core. AI analysis, health scoring, networking,
OTA updates, and security services execute on the Cortex-M7 core.

Cortex-M4 - Real-Time Monitoring Engine:
========================================
Sensor acquisition, signal conditioning, FFT processing, feature extraction, data timestamping, and deterministic
scheduling. The M4 guarantees consistent sampling and preprocessing regardless of network or AI load.

Cortex-M7 - Edge Intelligence and Platform Services:
====================================================
TinyML inference, anomaly detection, health scoring, Remaining Useful Life estimation, Digital Twin analysis, web
dashboard hosting, Ethernet/CAN-FD communications, OTA update management, and security services.

TinyML Workflow:
================
Feature Extraction -> Feature Vector -> TinyML Model -> Health Analysis -> Risk Classification -> Maintenance
Recommendation.

Security Architecture:
======================
Secure Boot verifies firmware authenticity at startup. Signed firmware and signed AI models prevent unauthorized
updates. Cryptographic validation ensures trusted execution.

Offline OTA Updates:
====================
Firmware is delivered from an engineering workstation over Ethernet. Updates are installed into an alternate slot,
verified, activated, and automatically rolled back on failure.

Key Capabilities:
=================
Dual-core architecture, Edge AI, Health Scoring, RUL estimation, Digital Twin monitoring, Industrial Ethernet,
CAN-FD integration, Secure Boot, Signed Firmware Validation, Signed AI Model Validation, Offline OTA Updates,
Rollback Protection, Fleet Dashboard, and Zephyr RTOS.

Expected Deliverables:
======================
Real-time dual-core processing demonstration, TinyML anomaly detection, health scoring, RUL estimation,
Ethernet dashboard, CAN-FD integration, secure boot, signed updates, secure offline OTA, and fleet-wide
monitoring.

Value Proposition:
==================
SecureEdge Predict goes beyond sensor monitoring by combining real-time acquisition, Edge-AI analytics,
industrial connectivity, cybersecurity, and secure lifecycle management into a deployable Industry 4.0 platform

EdgeInspect AI - A Modular Edge AI Platform for Smart Manufacturing Inspection
EdgeInspect AI is a modular edge AI platform powered by the NXP i.MX RT1170 that performs real-time automated optical inspection directly on embedded hardware. By detecting, classifying, and localizing manufacturing defects entirely on-device, the system demonstrates how low-cost edge AI can enable fast, reliable, and scalable quality inspection without relying on cloud connectivity. Although demonstrated using a simple PCB inspection task, the platform is designed to be easily adapted to a broad range of smart manufacturing applications.
———————————————————————————————————————————————
Project Overview
As manufacturing processes continue to evolve, there is an increasing demand for intelligent inspection systems that are both affordable and capable of making decisions in real time. While many existing vision-based inspection solutions depend on powerful industrial computers or cloud processing, EdgeInspect AI explores a lightweight alternative by bringing AI inference directly to an embedded edge device.
Built around the NXP i.MX RT1170, the system leverages the device’s high-performance crossover architecture to execute AI inference with low latency while maintaining the deterministic behavior expected from real-time embedded systems. Images captured by an onboard camera are analyzed locally to identify manufacturing defects, determine their locations, and generate immediate inspection results.
To demonstrate the platform, a custom PCB containing common electronic components will be used as the inspection target. However, the PCB serves only as a representative example. The underlying inspection framework is intentionally hardware-independent and can be adapted to different PCB layouts, electronic assemblies, or other manufacturing inspection tasks simply by retraining the AI model. This modular approach provides a practical foundation for future industrial, educational, and research applications.
—————————————————————————————————————————————-
System Architecture
EdgeInspect AI consists of three primary functional layers:
Image Acquisition
A camera captures images of the inspection target for real-time analysis.
Edge AI Processing
The NXP i.MX RT1170 performs on-device AI inference to detect manufacturing defects, classify defect types, and determine their locations without external computing resources.
Inspection Visualization
Inspection results are presented in real time, including defect localization, confidence scores, and an overall PASS/FAIL assessment, providing clear and immediate feedback for operators.
By executing the complete inspection pipeline on the RT1170, EdgeInspect AI demonstrates how embedded AI can deliver practical, low-latency, and cost-effective automated inspection while serving as a reusable platform for next-generation smart manufacturing solutions.

PrecisionSync – Deterministic Industrial Control Controller
PrecisionSync is a highly deterministic industrial controller designed to manage and synchronize time-critical operations across diverse manufacturing networks. The objective is to achieve precision synchronization and reliable communication, ensuring seamless interoperability between mixed-vendor equipment in next-generation industrial automation.
Using the PHYTEC phyBOARD®-RT1170 and Zephyr RTOS, the system will leverage Time-Sensitive Networking (TSN) via the i.MX RT1176’s ENET_QoS interface. To achieve this, the hardware will be adapted to bypass the standard onboard Ethernet PHY, allowing direct access to the TSN-capable MAC. The master controller will communicate with slave endpoints, such as an Industrial Raspberry Pi, to orchestrate strict real-time control loops.
The prototype will demonstrate robust real-time synchronization, low-latency deterministic data delivery, and mixed-vendor network integration. It will measure jitter and synchronization accuracy across the network, clearly comparing the TSN-enabled Zephyr environment against standard Ethernet communication.
The long-term goal is to develop a highly adaptable and scalable control platform ready for diverse and demanding applications, including advanced robotics, smart manufacturing, and universal industrial control systems.

Project Title:
IndustrAR Helmet - Wearable Real-Time Industrial HUD powered by i.MX RT1170

1. The Problem: The Technician’s Hands Are Full
Industrial maintenance and commissioning still follow a decades-old workflow: a technician stands at a machine holding a laptop or HMI panel in one hand, a multimeter in the other, and glances back and forth between the equipment and a screen. Head-mounted AR solutions exist (HoloLens-class devices), but they are expensive, fragile, battery-hungry Linux/Android computers with slow boot times and large attack surfaces that make them impractical on a plant floor. There is a gap in the market for a rugged, instant-on, deterministic heads-up interface that speaks native industrial protocols - Modbus, CAN FD, Ethernet - and simply overlays live machine data in the technician’s field of view, hands-free.

IndustrAR Helmet fills that gap: a safety-helmet-mounted, optical see-through HUD that turns the phyCORE(R)-RT1170 into a wearable industrial HMI.

2. The Concept: Optical See-Through, Not Video Passthrough
A compact MIPI-DSI AMOLED microdisplay is mounted inside the helmet brim, firing downward through a collimating lens onto a 45-degree semi-transparent combiner in front of the dominant eye. Because OLED black pixels emit no light, the UI is rendered as bright vector widgets on a black canvas - black becomes transparent on the combiner, giving true augmented-reality overlay with zero video compositing load. The wearer’s view of the world stays direct and unobstructed; the electronics only draw the data.

The HUD layout is discipline-driven: status arc on top (clock, link, battery, connected device), wearer telemetry on the left (heart rate, IMU speed/tilt), live machine telemetry on the right (process variables, register values, actuator states), and a deliberately empty center reserved for transient collision-proximity and alarm overlays. Green = nominal, amber = warning, red = alarm only. A glove or chin-strap joystick cycles pages; critical alarms preempt everything.

3. The Technical Fit: Why the phyBOARD(R)-RT1170
This application exercises nearly every headline feature of the platform, and the dual-core architecture maps onto it naturally:
- Cortex-M4 (400 MHz) - the fieldbus engine. Owns all deterministic I/O on fixed-period timers: Modbus RTU/TCP polling of PLCs (Zephyr’s native Modbus subsystem), CAN FD traffic, and I2C/SPI sensor acquisition (IMU, heart rate, ambient). Results are pushed into shared OCRAM via Zephyr IPC.
- Cortex-M7 (1 GHz) - the render engine. Runs LVGL with PXP 2D acceleration, updating only dirty widgets and blitting to the MIPI-DSI framebuffer. A stalled fieldbus device can never freeze the display - the cores are decoupled by design.
- MIPI-CSI camera (OV5640) serves as an auxiliary vision sensor, not the user’s view: it reads QR/DataMatrix asset tags so that looking at a machine auto-loads its Modbus register map - hands-free context switching.
- Zephyr OS end to end: the RT1170 is NXP’s Full Platform Support focus board for Zephyr, and every building block above (MIPI-DSI driver, video capture, Modbus, CAN FD, LVGL) exists upstream today.
- M.2 slot hosts a Wi-Fi module for untethered connectivity on the plant floor; secure boot and the CAAM protect a device that, by definition, gets trusted access to production PLCs.

4. Realistic Scope and Known Constraints
To keep the project honest and achievable in the contest window: this is a monocular, optical see-through HUD - not video-passthrough AR and not 3D-registered holography, which are beyond MCU-class silicon. The camera runs in YUV/RGB mode (respecting the RT1170’s MIPI-CSI RAW-mode erratum). The primary engineering risk is opto-mechanical (combiner alignment and focal distance), mitigated by using commercially available HUD combiner glass and a 3D-printed brim mount. The rendering budget is comfortable: HUD widgets at microdisplay resolution are a fraction of the LVGL benchmarks already demonstrated on this chip at 720p.

5. The Prototype (Expected Output)
The demonstration will show a wearer connected simultaneously to a Modbus TCP PLC simulator over Wi-Fi and a CAN FD node over wire, with live process variables, alarm preemption, IMU-based motion telemetry, and camera-triggered device switching - all overlaid on the real world through the visor, booting from cold to a functional HUD in under one second.

6. What’s Next: The Path to Production
After validation on the development kit, the 30 x 30 mm phyCORE(R)-RT1170 SoM moves onto a custom helmet-brim carrier board carrying only the display connector, battery management, IMU, and M.2 radio. The roadmap extends the sensor layer toward richer biometrics and eye-tracking input - a foundation for assistive interfaces, up to and including control surfaces for prosthetics and body extensions. The near-term product stands on its own: an instant-on, secure, deterministic industrial HUD at a fraction of the cost of AR headsets.

Vision Front-End for Flight Computer (VFFE)

Elevator Pitch
An intelligent image-processing module that works in front of a satellite Flight Computer (OBC). Its purpose is to analyze images onboard, reduce unnecessary data processing, and transmit only valuable information.

The RT1170 acts as an edge vision processor, enabling smarter use of limited satellite resources such as power, memory, CPU time, and communication bandwidth.
Earth observation satellites generate large volumes of image data. Small satellites have limited power and communication bandwidth.
Many captured images contain no relevant information.
Processing and transmitting unnecessary data wastes critical mission resources.

Proposed Solution:

The RT1170 performs intelligent onboard image analysis before sending data to the main flight computer or ground station.

Each image is evaluated based on:
-Image quality.
-Presence of relevant events.
-Scientific value.
-Transmission cost.

The system decides whether to Process further, Compress, Transmit and Discard.
Reduce workload on the main flight computer, reduce downlink bandwidth requirements, improve satellite energy efficiency and prioritize high-value scientific observations.
Potential Applications

-Wildfire detection.
-Smoke detection.
-Flood monitoring.
-Volcanic activity monitoring.
-Maritime surveillance.
-Space debris observation.
-Earth observation payload optimization.
-Autonomous satellite operations.

Main Capabilities

-Image quality assessment.
-Blur and sharpness estimation.
-Noise analysis.
-Contrast evaluation.
-Feature extraction.
-Region of Interest detection.
-Event detection.
-Image compression optimization.
-Intelligent data prioritization.

Get a Science Value Score (SVS)

Each image receives a score representing its scientific and operational value.

The score considers: Optical quality, presence of relevant events, temporal changes, novelty compared with previous images, estimated transmission cost.

Example: Science Value Score: 87/100

RT1170 Hardware Utilization

Cortex-M7

- Image processing algorithms.
- DSP operations.
- TinyML inference.
- Decision engine.
- Visualization.

Cortex-M4

- Camera acquisition.
- Sensor management.
- DMA handling.
- Real-time control tasks.

PXP (Pixel Pipeline)

- Image scaling.
- Color conversion.
- Cropping.
- Rotation.
- Hardware-assisted preprocessing.

Display + LVGL
Provides real-time visualization of:

- Image stream.
- Detected regions.
- Confidence level.
- Processing speed.
- CPU utilization.
- System status.

Algorithm Options
Possible algorithms:

- Edge detection.
- Sobel and Laplacian filters.
- Histogram analysis.
- Optical flow.
- Background subtraction.

TinyML Approach

Small neural networks for classification:

- Fire / No Fire.
- Smoke / No Smoke.
- Cloud / Ground.
- Water / Land.
- Debris / Background.

Target:

- Small memory footprint.
- Real-time inference.
- Low power consumption.

Validation Tests
Image Quality Assessment
- Sharp images.
- Blurred images.
- Low-light images.
- Noisy images.

The Vision Front-End does not replace the satellite Flight Computer. It acts as an intelligent perception layer that filters and prioritizes information before expensive processing and transmission.

The key concept of the proyect:

Do not transmit more data. Transmit only the data that matters.
A small satellite cannot afford to process everything. The Vision Front-End enables autonomous decision-making by transforming raw images into prioritized mission information directly onboard.

Project Title: SentryVolt-RT™ – Acoustic Arc-Flash & Coolant Leak Sentry for Next-Gen EV Charging Stations
Category: Power and Energy

1. Project Abstract & The Problem
Next-generation ultra-fast electric vehicle (EV) charging stations operate at high voltages (800V to 1000V+) and utilize pressurized liquid-cooling systems (such as water-glycol or dielectric fluids) to handle high-current heat dissipation. Traditional safety mechanisms—like thermal cameras or current-shunt monitors—typically detect faults only after an arc-flash has physically charred a connector, or when a coolant leak has caused a massive drop in pressure, resulting in expensive downtime, fire hazards, or catastrophic station failure.

SentryVolt-RT™ is an intelligent edge-sensing platform that detects the pre-fault stage of electrical failure and pressurized cooling leaks. It uses a digital MEMS microphone array to continuously capture high-frequency acoustic and ultrasonic emissions (up to 80kHz). By analyzing these spatial soundscapes, SentryVolt-RT™ can isolate the acoustic signature of micro-arc discharges (the precursors to full arc-flashes) and pin-hole fluid leaks, even inside the highly noisy environment of a charging station packed with heavy cooling fans and ambient highway noise.

2. Why the phyBOARD®-RT1170 is the Perfect Fit (Technical Deep-Dive)
The high-performance, real-time, and secure architecture of the NXP i.MX RT1176 on the phyCORE®-RT1170 SoM is uniquely suited to solve this challenging edge-AI audio-processing problem without the overhead, high power consumption, and security vulnerabilities of a Linux-based MPU.

• Dual-Core Workload Separation (Cortex-M4 vs. Cortex-M7):
- Cortex-M4 (400 MHz) - The Real-Time Audio Engine: This core handles high-speed deterministic tasks. It interfaces with the digital MEMS microphone array via the I2S/SAI or PDM interfaces using eDMA. It performs high-speed continuous Fast Fourier Transforms (FFT) and spatial filtering (beamforming) in real time to filter out the steady-state white noise generated by the station’s cooling fans. The resulting clean spectrogram is written to a shared On-Chip RAM (OCRAM) buffer.
- Cortex-M7 (1 GHz) - The Edge-AI & Control Engine: This core reads the spectrogram buffer and executes a lightweight convolutional neural network (CNN) classifier built with TensorFlow Lite Micro and optimized via the NXP eIQ toolkit. If an anomaly is classified (e.g., an active micro-arc or a pressurized liquid spray), the M7 takes immediate safety actions.

• Industrial Connectivity & Hard Real-Time Action:
Since safety relies on sub-millisecond response times, once a fault is classified, the Cortex-M7 triggers an emergency physical circuit trip or relays a high-priority “Power-down” command using the board’s native CAN FD or Gigabit Ethernet port (such as Modbus TCP over Zephyr’s network stack) directly to the charger controller.

• Advanced Hardware-Based Security:
As EV charging stations are critical grid assets prone to cyber threats, SentryVolt-RT™ implements cryptographic protection. It leverages the i.MX RT1170’s CAAM (Cryptographic Acceleration and Assurance Module) and SecureBoot to ensure that only signed firmware and verified TinyML models are executed, while encrypting remote communication.

• Multimedia Visualization:
For on-site maintenance technicians, the platform utilizes the 2D Pixel Pipeline (PXP) Graphics Engine to render a smooth, real-time waterfall spectrogram on a screen via the MIPI-DSI interface, allowing technicians to visually diagnose charger health on-site.

3. Prototype Deliverables & Demonstration Plan
Using the phyBOARD®-RT1170 Development Kit and Zephyr RTOS, the prototype will demonstrate:
1. Multi-core synchronization over Zephyr: The Cortex-M4 capturing and preprocessing microphone audio, while the Cortex-M7 executes the ML inference in parallel.
2. Active Noise Filtering: A live demonstration showing how the M4 filters out high-decibel fan noise, keeping the AI classification highly accurate.
3. Rapid Safety Response: Simulated fault triggers (using spark-gap and pressurized air valve simulations) demonstrating a CAN FD safety packet transmission latency of under 2 milliseconds.
4. On-Device Diagnostic HUD: A real-time audio waterfall display showing spectral anomalies using the MIPI-DSI interface and PXP engine.