Low-cost Environmental Sensors in a Smart City Environment. Dr. Jonathan Corey OTEC Presentation October 11, 2017

Transcription

1 Low-cost Environmental Sensors in a Smart City Environment Dr. Jonathan Corey OTEC Presentation October 11, 2017

2 Outline Sensing in Smart Cities The Sensor Package Usage Calibration Walkabout

3 Sensing in Smart Cities The Smart City environment is enabling greater connectivity between devices, people, vehicles, and services Ubiquitous wireless connectivity is one of the biggest benefits from a sensor standpoint Ubiquitous connectivity allows a swarm of sensors to continuously report back

4 Sensing in Smart Cities The Smart City environment is also a complex one Sensors may leave your communications horizon Communications can be complicated by multiple addresses/protocols/nodes Sensors may drift in and out of contact sending partial messages

5 Sensing in Smart Cities Low cost sensing is easier and more capable than ever Hobbyist and enthusiast sensors and platforms are freely available through Amazon and other e-tailers Preprogrammed libraries and code are also available for platforms such as Arduino and Raspberry Pi

6 The Sensor Package A Raspberry Pi handles Communications Synchronization of disparate sensors GPS location Kali Linux is used as the operating system Built in GPS support and logging With a compatible WiFi adapter MAC addresses can be collected with Kali tools

7 The Sensor Package An Arduino compatible microcontroller (Teensy 3.6) forms the base of the sensor system More computational power than Arduino Built in MicroSD card for data storage/backup Built in serial and network connectivity The Teensy is in charge of data logging because it tolerates changing the microsd

8 The Sensor Package

9 The Sensor Package Particle sensors Shinyei PPD42 Cheap ($25) Reports both PM 2.5 and PM 10 Detection floor ~1µm Shinyei PPD60 Moderately priced ($350) Reports PM 2.5 Detection floor ~0.4µm

10 The Sensor Package The difference in detection floor is very important Combustion particles have particle sizes with the upper tail of the distribution edging up to ~0.65 µm This means the PPD60 detects smoke and diesel exhaust directly

11 Usage One of the first major decisions to be made is how the sensor data will be used Trends analysis Statistical correlation Direct measurements Each usage scenario imposes different limitations on sensor deployments and programming

12 Usage Trends analysis is the least restrictive Sensor values are compared against previous values to determine whether values are increasing or decreasing Works best with repetitive routes or stationary sensors Longer aggregation intervals (15+ minutes) assist in smoothing out the data

13 Usage Statistical correlations, like Bayesian Neural Networks, can be impacted by Changing sensors Sensors moving between zones with different activity levels Changes in sensor movement/placement You may need several BNNs to account for different zone properties

14 Usage Calibrating the sensors to ensure accurate outputs is harder than it sounds Many people are changing how the sensors operate in their final devices Use fans instead of convection for airflow Mount the sensor in an orientation other than the one it was designed for Sample faster than air in the sensor changes or than the circuitry was designed for

15 Calibration We had an opportunity to directly calibrate a PPD60 while working on another project Testing showed we needed consistent airflow, therefore we added an air pump With the air pump, the resistive heater for convective airflow became superfluous and was removed This drastically altered sensor operation

16 Calibration Factory calibration curve Percentage of time interval particles were detected

17 Walkabout Group of smokers Passing smokers Smokey Truck