SensiML Add Predictive Maintenance in Smart Building Devices Instructions

SensiML Add Predictive Maintenance in Smart Building Devices

Agenda

Pre-work: Users to have installed Simplicity Studio and SensiML Analytics Toolkit in advance

• Host Introduction – 5 minutes
• Introduce concepts and goal for lab – 10 minutes
• “Real-time” execution of step-by-step procedure for model creation – 60 minutes
  • Flash SensiML compatible data collection firmware to the Thunderboard Sense 2 (TBS2)
  • Configure and connect TBS2 to SensiML Data Capture Lab
  • Capture ‘slide demo’ data with bare board (users won’t have Fan kits)
  • Label data and save and sample project (we won’t be using for the remainder of the course though)
  • Invoke Analytics Studio (at this point, users will be working from pre-collected TBS2 fan demo dataset)
  • Work through the steps for model building the fan state detection model
  • Create a Knowledge Pack
  • Optional: Flash model to TBS2
• Smart Building Applications demo video – 5 minutes
• Q &A – 10 minutes

SensiML Introduction

• SensiML is a B2B software tools company for AI at the IoT edge
  • Enables developers to create ultra-compact ML sensor models without data science expertise
  • Models as small as 10KB!
  • Former Intel Curie/Quark MCU AI software tools team, left to form SensiML in 2017
• Silicon Labs and SensiML Solution
  • Bringing power efficient ML to the EFR32/EFM32 MCU family
  • Rapid smart IoT application prototyping with Thunderboard Sense 2
• SensiML has stability and worldwide support
  • Acquired in 2019 by QuickLogic Corp; setup and run as wholly independent software subsidiary (based in Portland, OR)
  • Established channel partners (Avnet, Future Electronics, Mouser, Shinko Shoji)
  • Sales/Support offices in UK, US, Japan, Taiwan, China

Opportunities For TinyML in Smart Buildings

Challenges with Existing Smart IoT Sensor Application Development

Cloud-Centric AI

• High Network Traffic Load
• High Latency
• Less Fault Tolerant
• Unknown Risk of Data Security
• Concerns of Privacy

Deep Learning

• Large training data requirements
• Large memory footprint
• High processing workload
• High power consumption
• Poor endpoint battery life

Hand-Coded Endpoints

• Slow and labor-intensive
• Unknown code size upfront
• Scarce data science expertise
• Complex AI/ML code libraries
• Not scalable/competitive

TinyML = IoT Edge ML + AutoML

• IoT Edge ML: Autonomous endpoints
  • Trivial network throughput and long wireless battery life
  • No cloud processing or network dependencies
  • Real-time responsiveness
• AutoML: Optimize Without AI Expertise
  • Auto-optimizer selects best model for the data provided
  • Classic machine learning (ML) up through deep learning
  • SensiML TinyML yields models as small as 10KB!
• Hand-coding not required
  • Model code auto-generated from ML training datasets
  • Saves months of development effort, and data science expertise
  • Developer can change any aspect of the AutoML code as desired

Model Building Workflow

Capture Data

• Time: Hours to Weeks* (Depending on application data collection complexity)
• Skill: Domain Expertise (As required to collect and label events of interest)

Note: We’ll be leveraging some previously collected data to accelerate this step for the workshop

Build Model

• Time: Minutes to Hours (Depending on degree of model control exerted)
• Skill: None (Full AutoML)
  • Basic ML Concepts (Advanced UI tuning)
  • Python Programming (Full pipeline control)

Test Device

• Time: Minutes to Weeks (Depending on app code integration needs)
• Skill: None (Binary firmware with auto generated I/O wrapper code)
  Embedding Programming (Integration of SensiML library or C source with user code)

Workshop Goals

• Introduce SensiML’s TinyML toolkit and the model building process on Silicon Labs Thunderboard Sense 2
• Experience with data-driven supervised ML sensor algorithm development
• Learn the workflow from data collection through validation and on-device testing for building IoT models
• Build a working HVAC predictive maintenance model start-to-finish
• Address questions you may have about the TinyML model creation process

A Working HVAC Predictive Maintenance Application

• For purposes of our hands-on portion, we’re going to be building a smart fan-monitoring device
• Fans used ubiquitously in building HVAC systems: Blowers, active cooling of equipment, air handlers, ventilation ducting
• Failure or degradation can cause loss of efficiency, increased energy consumption, HVAC failures
• We’ll construct a simple monitoring device that can detect multiple normal and abnormal fan states:
  • Fan off / on
  • Loose mounts
  • Fan guard obstruction
  • Partial or fully blocked airflow
  • Blade impingement
  • Excess vibration

Let’s Begin the Process

“Real-time” workshop step-by-step procedure for model creation – 60 minutes

• Flash SensiML compatible data collection firmware to the Thunderboard Sense 2 (TBS2)
• Configure and connect TBS2 to SensiML Data Capture Lab
• Capture ‘slide demo’ data with bare board (users won’t have Fan kits)
• Label data and save and sample project (we won’t be using for the remainder of the course though)
• Invoke Analytics Studio (at this point, users will be working from the pre-collected TBS2 fan demo dataset)
• Work through the steps for model building the fan state detection model
• Create a Knowledge Pack
• Flash model to TBS2

Demo Video

Copyright © 2021 SensiML Corporation. All rights reserved.

References

• SensiML Smart Building IoT Sensing Solutions with Silicon Labs

Read User Manual Online (PDF format)

Read User Manual Online (PDF format)  >>

Download This Manual (PDF format)

Download this manual  >>