Last update of the README on 8th of July 2018

Software Infrastructure or framework for managing environmental sensors and data aquisition

The controller is based on Raspberry Pi for functionality and easy block building reasons. The bus used for sensors are: USB (serial and I2C), GPIO (SPI) and I2C. The scripts are written in Python 2.

Visual feedback is provided with led/button (power On/Off) and optionally a tiny Oled display from Adafruit.

MySense in data concentrator mode has the possiblity to collect these measurements data from e.g. the TTN MQTT dataconcentrator. The bus used for sensors are: UART (serial), I2C and GPIO. The scripts are written in (embedded) micro Python. Micro python has more functionality as the language C used with Arduino boards.

Visual feedback is provided with RGB led and optional a tiny Oled display from Adafruit.

Provide a generalised dynamic Open Source based infrastructure to allow:

• environmental measurements with sensors
• data acquisition
• dynamic transport of data to other data systems: e.g. databases, mosquitto, Influx,...
• data storage and archiving
• access for free visualisation
• free availability of the data
• free availability of all software (GPLV4 license)

MySense supports calibration of every single sensor. Sensor values will differ between the sensors within a branche and between branches. Correlation software is included. Advised is to calibrate the sensors regularly for a test period of several days (conditions should vary in the test period).

Dust measurements are done by counting the particles. The most common dust sensor is the Nova SDS011. The Plantower PMS7003 is however 1/3 in size and counts more classes of particles as well provides also the raw values. Both have a fan and laser which are powered off in idle state. Dust measurments are influenced by humidity. A correction algorithm to enable to compare the dust measurements with reference sensor equipment (e.g. BAM1020) is in beta test (April 2018).

MySense sensor kits examples

• Create MySense user e.g. ios and login as this user.
• Install the software on e.g. the Raspberry Pi 3 on a new user e.g. ios in the directory e.g. MySense. Use INSTALL.sh to install all dependencies and startup scripts.
• Configure MySense.conf using MySense.conf.example as a lead.
• Test one by one the input and output scripts in the python debugger as standalone e.g. pdb MySDS011.py. Once this is tested, go to the next step.
• command line start and control ( followed by kill %1 will stop the process): Run Mysense as follows python MySense.py and you will see all output on your screen.
• If you use a tiny display: start the display server: python MyDisplayServer.py start
• Start as daemon process. Start up MySense: python MySense.py start When first started make sure you configure console output channel, logging to stdout or stderr and logging debug level.

If needed See the README files and documentation files in docs for more detailed info.

If you installed a led switch (controlled by /usr/local/bin/poweroff:

• Pressing the switch longer as 20 seconds will poweroff the Pi
• Pressing the switch longer as 10 seconds will reboot the Pi
• Pressing the switch 6 seconds will restart a search for wired or wifi internet connectivity.
• If the Pi is powered off a disconnect and connect of the adapter will boot the Pi.

Without internet connectivity the MySense software will not be started on a reboot.

The @reboot /home/ios/MySense/MyStart.sh in the ios crontab table will automatically start MySense on a reboot. Comment this out in the test phase.

The main sensor kit case carrying the Raspberry Pi and all sensor/GPS modules is build from PVC roof gutter pieces: gutter end pieces for keeping the air in and the rain out, and overflow gutter box as casing. The case has a poweroff button and small window to show a tiny display with current measurements. The sensors are fixated on a Lego plate to allow flexibility of sensor changes.

All scripts are written in Python 2. Python 3 is supported but not tested well. Scripts have been tested on Raspberry Pi (2 and 3) running Wheezy, Jessie and Stretch Debian based OS. Scripts have a -h (help) option. With no arguments the script will be started in interactive mode. Arguments: start, status, stop.

• MyLed.py: control the Pi with button to power off and put it in wifi WPA mode. Pi will set up a wifi access point MySense if no internet connectivity could be established via wifi or LAN.
• MyDisplayServer.py, a display service: messages received will be shown on a tiny 128X64 oled (I2C) display.

The main python script is MySense.py. It acts as intermediate beween input plugins and output channels. It uses MySense.conf (see MySense.conf.example) to configure itself. The MySense configuration file defines all plugins available for the MySense.py command.

• input (modules) plugins: temperature, dust, etc. sensor device modules and brokers
• output (modules) channels: console output, (MySQL) database, (CSV/gspread) spreadsheets, and brokers (mosquitto, InFlux, ...).

Try ./MySense.py --help to get an overview.

On the command line the option --input and --output plugins can be switched on (all other configured plugins are disabled).

MySense starts with a configuring phase (options, arguments, reading configuration, loading modules), whereafter in the readsensors() routine it will first access the input modules to obtain measurement values, combine them into an internal buffer cache per output channel, and finaly tries per output channel to empty the queued records.

The output of sensor values to an output channel will always on startup to send an identification json info record. Each configurable interval period of time MySense will send (input) measurements values to all configured output channels. For each output channel connected via internet MySense will keep a queue in the case the connection will be broken. If the queue is exceeding memory limits the oldest records in the queue will be deleted first. If the configured interval time is reached it will redo the previous loop.

If switched on and configured an email with identification information will be sent to the configured user. Make sure one obeys the Personally Identifiable Information ([PII]http://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-122.pdf) privacy rulings.

MySense.conf is the configuration/init file from which plugin or modules are imported into the MySense process. See the MySense.conf.example for all plugins (sections) and the plugin options.

For every plugin module there is an README.plugin with explanations of the input/output plugin. The input from sensors is read asynchronous (in parallel) via the module MyTHREAD.py. If needed it can be switched to read only in sync with the other input sensors.

A working example of MySense script in todays operation:

          remote access             |  INTERNET (wired/wifi, wifi-G3/4 mobile)
          syst.mgt.     webmin -----||_ wifi AP -- webmin/ssh system mgt
                    ssh tunnel -----||_ BlueTooth -- terminal access
   TeamView/Remot3 (Weaved)IoT -----|
                                    |
                                    |    
    INPUT PLUGINs                   |        OUTPUT CHANNELS    GATEWAY/BROKER
                                  __|__
    DHT11/22-meteo ---GPIO --->| ///|\\\ |>- CSV                _____
    GPS-locator -Uart USB  --->|=MySense=|>- console           ///|\\\  
    RSSI-wifi signal-strength >||  Pi3  ||>- MYSQL           |=MySense=|>-gspread
                               ||Pi ZeroW|
    Dylos-dust -USB-- RS232--->||Stretch||>- Mosquitto pub-->|| Debian||>-MySQL
    Grove-loudness ---GPIO---->| \\\|/// |>- HTTP-Post       || Linux ||>-CSV
    BME280 -meteo ---- I2C --->|    |    |>- email info      | \\\|/// |>-console
    BME680 -meteo+gas--I2C --->|    |    |                   | server  |
    SHT21/31 - planned-I2C --->|    |    |                   |         |
    PPD42NS -dust-Arduino-USB->|    |    |>- InFlux publish  |_________|>-InFlux pub
    Nova SDS011 -dust -USB --->|    |    |>- oled display SSD1306 (SPI/I2C)
    Plantower PMS7003 -USB --->|    |    |>- Google gspread (alpha, deprecated)
    O3,NO2,CO SPEC UART USB -->|    |    |   (beta test April 2018)
    NH3 - AlphaSense - I2C --->|    |    |   (planned Jun 2018)
    Adafruit rain -----GPIO -->|    |    |   (planned Aug 2018)
                               |    |    |    
    LoRaWan (TTN MQTT) ------->|    |    |>- broker? (planned)
    Mosquitto sub ----server ->|    |    |>- LoRaWan (planned TTN)
    InFlux subscribe -server ->|    |    |>- Bluetooth (planned)
    LoRa TTN MQTT ----server ->|    |    |>- Luftdaten.info databases
                                    |
                                    |>-raw measurement values -> InFlux server or file
                                           calibration

MySense LoRa air quality measurement kit:

          Arduino/Atom/Makr WiFi/USB --|-- WiFi / BlueTooth
                                    ___-__________
    DHT11/22-meteo ---GPIO---->|   / Marvin        \
    BME680 -meteo+gas--I2C --->|= <  PyCom LoPy     >|-LoRa TTN MQTT >-< MySense >
    BME280 - meteo ----I2C --->|   | PyCom WiPy     >|-SigFox IoT (planned)
                                   |               |                 
    Nova SDS011 -dust -Uart -->|   \ ESP           /
    Plantower PMS7003 -Uart -->|    --------------
    Grove GPS ---------Uart -->|       |
                               |       |
    commands - LoRA TTN     -->|       |
                                       |
                                       |> SSD1306 128X64 oled display

LoRa TTN is also used e.g. to change sample timings or to force information (e.g. location) to be send from the sensor kit.

See MySense.conf.example for an example of MySense.conf`.

Use for configuration of plugins/outputchannels the section (plugin name in lowercase) and section options. The option input = True or False and output = T/F will define resp input plugin and output channel to be imported and to be switched on or off. Input plugins as for gas and dust (particle counts) will have a configurable sample time (time to get vales) and interval time (time (interval minus sample) to wait before the next sample). The MySense main loop has an own interval time within input plugin sensor values will be collected (sliding average from sample values) and push values to output channels.

See for an example the file Config.py. Make sure useXXX and the pins are defined and wired correctly.

Interaction with plugins and output channels is done in json datastructure: Example of json to display a measurement on the console (and others):

     { "time": UNIXtimeStamp,
        "temp": 23.2,
        "rh": 30.2,
        "pm": 234.2,
        "o3": None }

At the startup MySense.py will start with an identification record providing details of the version, the location if available, a unique identifier, sensor types and measurement unit, etc. This information will define eg the first row of a spreadsheet or the database table with all sensor info (called Sensors).

Towards a broker the output will consist of an (updated e.g. GPS location) combination of the data json record and the infomration json record:

    { "ident": id-record, "data": data-record }

See for an example the file: testdata/Output_test_data.py

The input sensor plugins provide (sliding window of a per plug definable buffer size)) averages in a per input plugin defined interval time in seconds. The output is done on a general interval period timing using the average time of input timings.

Typical input rate from a sensor is 60 seconds (can be tuned) and for brokers it is 60 minute interval (can be tuned).

MySense can act either sensor manager or as input from broker manager to a set (dynamic) of output channels.

Available input plugins:

• Dust: Dylos DC1100 or 1700 via serial interface, Shinyei GPIO (e.g. Grove dust sensor), Nova SDS011, Plantower PMS5003/7003.
• Temperature/humidity: Adafruit DHT11/22, AM3202 and Grove variants, Bosch BME280 or BME680 (has indoor aq gas sensor), Sensirion SHT31-D.
• RSSI (strength of wifi signal): via the platform
• Location: GPS (GPS Ultimate from Adafruit/Grove) via TTL serial interface

The Pi allows to install a wifi connectivity with internet as well a virtual wifi Access Point. A backdoor configuration is provided via direct access to webmin and ssh (Putty), as well via a proxy as ssh tunneling and/or using the proxy service of Weaved (https://www.remot3.it/web/index.html).

If no access to Internet either via LAN or WiFi is obtained a WiFi AccessPoint is started (SSID MySense-XYZ, and default password) which enable you to use ssh ios@192.168.2.1 command to obtain shell access. Correct in /etc/wpa_supplicant/wpa_supplicant.conf the SSID and psk password phrase to your local access point.

Sensors have a hardware interface to I2C, GPIO: those sensors are tested on RaspBerry Pi (and Arduino Uno) Sensors with USB serial are tested on Linux Debian platforms which run Python.

The GrovePi+ shield is used to ease hardware installation by just using 4-wired connectors and avoid mistakes. No soldiering, nor DuPont wires which are easily get disconnected. The GrovePi+ shield has 3 I2C connectors.I2C connectors are all in parallel. Use eg a Grove I2C 4-port connector to extent the amount if needed.

Install GrovePi+ Dexter libraries as user pi with the following command:

    curl -kL dexterindustries.com/update_grovepi | bash

and reboot/poweroff the pi. Install the shield and proceed.

We use small USB cables with a 90 degrees connector and/or USB hub with 4 USB connectors with 10-15 cm wires to ease fixation of the wiring. As well use a Lego board and Lego stones to fixate all modules and sensors on the Lego board.

See README.pi for installation of the Raspberry Pi platform. MySense plugins: Use the shell file INSTALL.sh [DHT GPS DB plugin ...] to download all dependent modules.

The sensor plugins, and output modules can be tested in standalone mode, e.g. for BME280 Bosch chip, use python MyBME280.py. Or use the Python debugger pdb in stead. See the script for the use of sync and debug options at the end of the script to test.

If the Pi supports BlueTooth one could also use BlueTooth terminal access by installing BlueTooth terminal service: ./INSTALL.sh BLUETOOTH. This type of accesss is not recommanded.

See the README's and docs directory for descriptions how to prepair the HW, python software and Pi OS for the different modules.

CONTENT.md will give an overview of the files and short description.

See the various README/docs directory for the plugin's and modules for the status of operation, development status, or investigation.

Failures on internet connectivity and so retries of access is provided.

Use the following first if one uses MySense for the first time: test each sensor input or output channel one at a time first. Use the Conf dictionary to set configuration for the test of the module.

The sensor plugin as well the output pugin channels all have a __main__ test loop in the script. This enables one to test each plugin (one at a time) in standalone modus: pdb MyPLUGIN.py. Use for the sensor input plugins Conf['sync']=False (to disable multithreading) and switch debug on: Conf['debug']=True. Set the python debugger pdb to break on break getdata (input plugin) or break publish for stepping through the script. Failures in configuration are shown in this way easily.

After you have tested the needed input/output modules: To test the central script MySense.py use first the Python debugger pdb. The main routine after the initiation and configuration phase is sensorread, in pdb use break sensorread. Continue to this break point and use print Conf to show you the configuration settings. Step to the first getdata call or publish call to go into the input or output module. Note that the getdata() input routine may need some time in order to allow the module to collect measurement(s) from the sensor.

By default MySense uses a so called lightweight process (multithreaded) to allow sensor data to be collected asynchronously. Input is tested with serial, I2C-bus and GPIO sensors (meteo,dust,geo,audio, (gas in September 2017). The focus is to allow Grove based sensors (easier to plugin to the MySense system) and weather resistent cases for the system.

The gas sensor development (NO2, O3, NH3, CO) is just (Febr 2017) started, Aug 2017 alpha tests.

Calibration of dust counters like Shinyei, Nova SDS011 and Dylos is started in May/June 2017. Outdoor correlation tests started Sept 2017. Indoor calibration tests with Plantower PMS7003, Nova SDS011 and BME280/BME680 were done in April 2018.

The use of the DHT22 has been depricated after a 3 month period beginning of 2018 with 10 sensots kits equipted with Marvin LoRa/DHT22/SDS011 sensors. The DHT22 differ too much from one to the other. Are much influenced by higher rel. humidity. As well the I2C bus (e.g. BME280) seems more reliable and is easier to use.

The SDS011 (and probably PMS7003) are heavily influenced by rel. humidity of 80% and higher: exponential overestimated dust densities. In study with RIVM is a recalculation scheme to correct the values.

Calibration of Alpha Sense gas sensors is a problematic area. Probably Sept 2017. First tests show Alpha Sense O3, CO2 are OK, NO2 not successfull, NH3 prosponed.

To facilitate measurements for calibration purposes all sensor plugins are optionaly (set raw option to True for the particular sensor in MySense.conf) able to output on file or to an InFlux DB server the raw measurements values, as follows:

    raw,sensor=<type> <field1>=<value1>,<field2>=<value2>,... <nano timestamp>

This is an InFlux type of telegram, where the UNIX timestamp is in nano seconds. Example for database BdP_02345pa0:

    raw,sensor=bme280 temp=25.4,rh=35.6,pha=1024 1496503325005000
    raw,sensor=dylos pm25=250,pm10=15 1496503325045000

E.g. download the serie for eg correlation calculation from this server or into a CVS file (awk maybe your friend in this). Or use a file, say MyMeasurements_BdP_02345pa0.influx.

    # send the file to the InFluxdb server via e.g.
    curl -i -XPOST 'http://localhost:8086/write?db=BdP_02345pa0&u=myname&p=acacadabra' --data-binary @MyMeasurements_BdP_02345pa0.influx

InFlux query reference manual:

• https://docs.influxdata.com/influxdb/v1.2/query_language/

Using the Influx CLI (command line interface) one is able to convert the columnized output into whatever format, e.g. to CSV:

    influx --format csv | tee InFlux.csv
    >auth myname acacadabra
    >use db_name
    >show series
    >select * from raw order by time desc limit 1
    >select * from raw where time > now() - 2d and time < now() - 1d order by time desc
    >quit

After the correlation calculation set for the sensor the calibration option: e.g. calibration=[[25.3,-0.5],[13.5,63.203,0.005]] for here two fields with a linear regression: <calibrated value> = 25.3 - 0.5 * <measured value> for the first field values. The second field has a 2-order polynomial as calibration.

To avoid outliers the MySense input multi threading module will maintain a sliding average of a window using the buffersize and interval as window parameters. Python numpa is used to delete the outliers in this window. The parameters for this filtering technique are default set to a spread interval of 25% (minPerc MyThreading class parameter)) - 75% (maxPerc). Set the parameters to 0% and 100% to disable outlier filtering. Set busize to 1 to disable sliding average calculation of the measurements.

For calibration the Python tool statistics/Calibration.py has been developped. The script uses pyplot and is based on numpy (numeric analyses library). The calibration uses values from two or more database columns, or (XLSX) spreadsheets, or CSV files as input and provides a best fit polynomial (dflt order 1/linear), the R square and shows the scattered plot and best fit graph to visualize the difference between the sensors. Make sure to use a long period of measurements in a fluctuating environment (a fixed indoor temperature measurement comparison between two temp sensors does not make much sense).

The DHT meteo sensors show intermittant lots of read errors. Humidity: outdoor use of the sensor will show after a while 99.5% rel. humidity all the time. Allow the sensor to dry. The meteo sensor BME280/680 might be a better alternative. Tests show a linear correlation between this sensor and the DHT. However the chip seems to build up heat and shows a higher temperature as it should be. Airpressure seems very reliable. The current focus however is on the Sensirion SHT31 chip. Which has promissing specifications.

Calibration test results (April 2018) with 3 sensor kits, indoor (temperature and humidity does not vay much) test of 32 weeks with 5 minute samples:

1. reference kit with BME680
2. BME680:
   • gas R2=0.85, correction 3.192e2, 5.650e-1 (56% of ref BME680)
   • temp R2=0.843, correction -4.703, 8.528e-1 (85% of ref BME680)
   • humidity R2= 0.972, correction -2.99e0, 0.519e-1 (5% of ref BME680!)
   • pressure R2=0.99, correction 1.238e1, 9.820e-1 (-2%)
3. BME280:
   • temp R2=0.98, correction -1.662e0, 9.310e-1 (93% of ref BME680)
   • humidity R2=0.625, correction -4.824e0, 1.404e0 (14 X BME680!)
   • pressure R2=0.9779, correction 1.291e2, 8.753e-1 (only an offset, -13% of ref) In short: air pressure values correlate fine (high R2) and need some correction. R2 for temperature and humidity are just ok. But the SDS011 need corrections to the BME680. Previous tests with DHT22 show a far lower R2 and higher (linear) corrections.

The Shiney PPD42NS (tested 3 sensors) gave lots of null reading on low PM10 values. The sensor values are not stable enough in comparison with newer sensors from Nova and Plantower as well the bulky Dylos handhelt.

Due to airflow the sensors need to be cleaned periodically. The Plantower sensor is hard to clean as it cannot be opened.

Plantower dust sensor measures also PM0.3, PM0.5, PM1 and PM5 as PM counts.

Both Plantower and Nova dust sensors use USB bus. The values are provided in mass values. The conversion from particle count to mass is not made public.

All these PM counting based sensors show an exponential overcalculated value on higher rel. humidity. Research is going on to correct it to values which are compatible with e.g. BAM1020 reference sensors.

Calibration test results (April 2018) with 3 sensor kits, indoor test of 2 weeks with 5 minute samples:

1. reference kit with PMS7003
2. PMS7003:
   • PM0.1 R2=0.9664, correction 8.635e-3, 1.096e0 (no difference)
   • PM2.5 R2=0.9638, correction 5.626e-1, 1.092e0 (no difference)
   • PM10 R2=0.9564, correction 6.630e-1, 1.174e0 (not much difference)
3. SDS011:
   • PM2.5 R2=0.9473, correction -5.141e-1, 2.615e0 (ca twice PMS7003)
   • PM10 R2=0.8805, correction 8.967e-1, 2.704e0 (ca twice PMS7003) In short: about no difference (high R2), correction between the Plantower is about none (PM0.1 1%, PM2.5 1%, PM10 17%). Plantower with Nova SDS011 differ not much (high R2), correction is linear (PM2.5 260%, PM10 270%).

Tghe Alpha Sense gas sensors have a high cost level (ca 80 euro per gas). NH3 is hard to test and still planned. NO2 give too many errors in the field. The sensors have a very limited time.

The Grove GPS sensors is applied via USB bus connection and the standard Debian GPS deamon. The location is not precise enough. The wait is for the Galileo GPS sensors availability.

The tests are done with the Raspberry Pi 3. With the GrovePi+ shield and the big V5/2.5A adapter it gets bulky. The new Raspberry Pi Zero V1.3 is half size, uses far less power and costs only 25% of the Pi3. We expect the Zero might be applicable.

There is no funding (costs and development time is above personal budget level). Costs at start are high due to failures on tests of common sensors (Arduino is skipped due to too low level of functionality; Shiney and DHT sesnors failures, application of smaller adaptors, etc.). Money is lacking for sensors research and travel expenses coverage to meet other initiatives.

July 2017: local government is asked to subsidy operational phase: distribution of sensors kits and maintenance.

FSF GPLV4 Feedback of improvements, or extentions to the software are required.

• Copyright: Teus Hagen, ver. Behoud de Parel, the Netherlands, 2017

A list of references for the documentation and/or code used in MySense.py:

• Open Data Stuttgart ESP8266 controller oriented: https://github.com/opendata-stuttgart
• MIT Clairity CEE Senior Capstone Project report V1 dd 15-05-14
• https://www.challenge.gov/challenge/smart-city-air-challenge/ Smart City Air Challenge (2016, USA GOV) See also: https://developer.epa.gov/air-pollution/
• http://opensense.epfl.ch/wiki/index.php/OpenSense_2
• http://mysensors.org
• http://opensensors.io
• http://mydevices.org (Cayenne)
• https://waag.org/nl/project/urban-airq Waag Society Amsterdam Smart Citizens Lab Urban AirQ
• http://www.citi-sense.eu/ Citi-Sense EU project
• http://waag.org/nl/project/smart-citizen-kit Smart-Citizen-Kit Waag Society
• http://smartemission.ruhosting.nl/ Smart Emission, Maps 4 Society Nijmegen
• https://github.com/guyzmo/polluxnzcity Pollux NZcity, NZ
• https://github.com/HabitatMap/AirCastingAndroidClient AirCasting on Android Client
• https://mosquitto.org/ Mosquitto (MQTT) broker
• https://docs.influxdata.com/influxdb/v1.2/ documentation from InFluxData.com
• https://cdn.hackaday.io/files/21912937483008/Thomas_Portable_Air_Quality.pdf interesting overview of sensors