This Machine Learning (ML) application software pack, ML-Based System State Monitor, relies on Deep Learning and enables developers to develop and deploy neural networks on MCU-based systems for building smart sensing applications that use time series real-time sensed data for system state monitoring.

The example software package provides general enablement and was designed to show how to create a Fan Vibration State Monitoring Solution and also through an Application Note it provides details on how to validate and evaluate the performance of a model by running it through different inference engines on different evaluation kits.

To access this application software pack, click on the GitHub button to visit the NXP MCUXpresso SDK GitHub Repository and follow the steps found in the Lab Guide to evaluate the software solution.

Please visit section Support of this document for the extended list of associated resources.

1. Software
2. Hardware
3. Setup
4. Results
5. Support
6. Release Notes

• ML Applications provided by this package - host machine applications used to define the neural network architectures, train, and generate ML models
• MCU Applications provided by this package - software applications for targeted embedded devices (i.e., FRDM-MCXN947, MIMXRT1170-EVK, LPC55S69-EVK, FRDM-K66F) used to deploy, run, and evaluate the ML models

Toolchain:

• MCUXpresso SDK (revision MCUX_2.14.0)
• MCUXpresso IDE
• eIQ Toolkit
• Jupyter Notebook (Python, TensorFlow, Keras, etc.)

Required:

• Personal Computer
• Mini/micro USB cable
• Any of the following development boards:

FRDM-MCXN947	MIMXRT1170-EVK	LPC55S69-EVK	FRDM-K66F

Optional (functional scenario: run inference using with real-time data read from the sensor):

• Any of the following sensor shield boards:

FRDM-STBI-A8974	ACCEL-4-CLICK	FRDM-STBC-AGM01

• Arduino ProtoShield + 5V DC Fan

ProtoShield	Fan

Optional (functional scenario: read and log sensor data to be used for neural network training):

• SD card

You need to have both Git and West installed, then execute below commands to gather the whole APP-SW-PACKS/ML-State-Monitor delivery at revision ${revision} and place it in a folder named appswpacks_ml_state_monitor.

west init -m https://github.com/nxp-appcodehub/ap-ml-state-monitor.git --mr ${revision} appswpacks_ml_state_monitor
cd appswpacks_ml_state_monitor
west update

Replace ${revision} with any SDK revision you wish to achieve. This can be main if you want the latest state, or any commit SHA.

To build and run the application please refer to the Lab Guide or check the steps in Run a project using MCUXpresso IDE.

To configure the hardware settings please refer to the Lab Guide or follow the readme guidelines for the specific board and application that are located in the following path ml_state_monitor/mcu_app/boards/<board_name>/<application_name>/readme*.

This Application Software Pack provides support to develop and deploy neural networks on MCU-based systems for time-series data. It illustrates how to build and deploy on embedded devices neural network models capable to monitor the input sensor data and detect the state of a device.

Please refer to AN13562 document which demonstrates through a real use case the steps required to produce and assemble a dataset, define the architecture of a neural network, train, and deploy a model on an embedded board by leveraging the NXP’s SDK and the eIQ technology. This document also shows which metrics to use and how to evaluate the behavior of a neural network model at runtime on an embedded board, as well as benchmarks and performance results.

Please refer to the following resources list for extended support:

• ML-Based System State Monitor Website
• ML-Based System State Monitor Lab Guides
• Application Note AN13562 - Building and Benchmarking Deep Learning Models for Smart Sensing Appliances on MCUs
• Application Software Packs - Full portfolio of rapid-development easy-to-use software bundles
• MCUXpresso SDK - Full SDK delivery that provides access to other SDK examples and components
• eIQ ML Software Development Environment - ML workflow tools like eIQ Toolkit, along with inference engines, neural network compilers and optimized libraries

Questions regarding the content/correctness of this example can be entered as Issues within this GitHub repository.

Warning: For more general technical questions regarding NXP Microcontrollers and the difference in expected funcionality, enter your questions on the NXP Community Forum