My adventures in Implementing IoT - Part I Setting up the hardware and data collection

Hardly a day goes by when you don't hear about IoT and how it will change the world. The CISCOs and the McKinseys of the world have impressive stats about the proliferation of IoT and how it would create Trillions (with a T) of dollars of value in all sectors of the economy..reminded me a lot of 1999 and how everyone used to intone:  'Internet would change the world.' Well it did! It took almost a decade. Is IoT all hype? Will it change the world, and would it have a faster or slower gestation period than the Internet of people? I realized IoT had become business mainstream when Harvard Business Review last fall had it as a cover article on it by the strategy guru Michael Porter. But it was all very theoretical and I was dying to figure out how to do something practical to feel it and see it work for myself. I had learned that connected devices provide value in stages:

a) As a monitoring device
b) As a controlling device
c) To optimize
d) To Automate

I needed to experience IoT firsthand. I would be very happy if I just built a monitoring device that could send data that I could monitor anywhere in the internet. I felt, over time I would collect data, filter the data, build some heuristics and develop a control strategy to improve process performance, leading to better results. There were a lot of books on IoT with value propositions, data architectures etc. but nothing practical/easy to do for someone who is not an embedded systems developer.

 The PiOT workshop

Seek and you shall find! In March I came across a mailing for a workshop on 'IoT-101' organized by Initial State a data services startup here in Nashville. I quickly signed up for the next available slot. Run by David, Rachel and Raymond, the workshop managed to get a whole lot done in a very short time. In a 2 hour period we learned about Raspberry PI a favorite platform for building IoT devices. We learned about using Python. I have written 10's of thousands of lines of code in Fortran and Visual Basic, so programming concepts were not new to me. What I was struck with was, the ease of use of Python to interact with the hardware system components to collect data, as well send control signals. I was hooked - or shall we say connected. During the workshop we had used a breadboard and hooked up switches and sensors to:

a) Connect a temperature probe and live streamed the readings to the Initial State web site
b) Turning on and off LEDs
c) Detecting whether a door was shut or open using a magnetic switch

I was excited! On the spot I bought an 'advanced' Raspberry Pi kit with a LCD display, temperature probe, breadboard and a bunch of LEDs and resistors and took it home to begin my adventures in IoTland.

Adventures at Home

I was able to follow the instructions and turn on and bootup my Raspberry Pi. I then realized that the neat connector setup that Initial State guys used with the probe during the workshop was not in my kit. A trip to their office yielded free connectors. Also after reading their online blogs I was able to figure out how to connect my Raspberry PI to a HDTV so that I didn't have to squint at a 3"x2" LCD display. That made my life a lot easier. By day three, I was able to replicate all exercises in IoT workbook.  Along the way Rachel from Initial State was hugely helpful, responding to my queries all hours of the day including 10 pm on a Sunday night!

Raspberry Pi connected to a temperature probe (inserted in the ice bath)

Time to take it to the plant

During the day I am the President/CEO of Porcelain Industries. We are the largest independent enameling facility in the US and have a large number of spray lines, dryers, washers and furnaces. I invited Jamie Bailey (CEO) and Raymond Jacobs (VP Bus. Dev.) from Initial State to our manufacturing plant.  Jamie and Raymond met with myself and Jim Hutzel our VP for Operations. We saw a lot of potential in embedding sensors in the plant to measure everything from furnace temperature profiles to whether (and at what rate) the lines were running and what was the pH and temperature of the washer. As a pilot, to keep things simple we agreed that it would be good to measure the humidity and temperature in the plant and make sure that the rheology/density of our enamel dispersion was optimum for those conditions. So our initial goal was to collect temperature/humidity and enamel properties and send them to a data depository.

Beside the real time temperature and humidity measurement and recording, our plan was to create an iPhone app so that the data could be entered by the operator (after he/she measures it on an analog instrument) every 30 minutes to an hour. Turns out it is really expensive to get a simple data entry app made (like $10-15 K). Since we were on a shoestring budget we decided lets just collect temperature / humidity data and manually enter the enamel property data so that we can at least build a model to see if there is a correlation before we invest $15 K in a seldom used iPhone app.

We needed a robust hardware set up which was better than a temperature probe connected to the Raspberry Pi through a breadboard. Jamie came up with a a suggestion - use a sensor kit from Dexter Industries called 'GrovePi' which came with a modular attachment to the Raspberry Pi. One could plug and play a variety of sensors for measuring or detecting  temperature/humidity, sound and light. It also came with a Python library of functions to allow one to interact with these devices in Python.After some trial and error I got the GrovePI to work with my Rasberry Pi, and soon I was able to stream temperature and humidity data from my office to the internet on the Initial State website.

The next challenge was to make this operational in the plant. We decided that WiFi is too noisy and we really should use a wired Ethernet drop to ensure good reliable network connectivity. One afternoon after the operating shift had gone home I connected the Raspberry Pi to the Ethernet drop and lo and behold I was transmitting plant data to the internet. Big milestone!

Streaming Temperature (Orange) and Humidity (Purple) Screen on Initial State Web Site
showing Raspberry Pi output from a 48 hour period

Software Victories

Now that I had a prototype version working and verified that it sends data measured in the plant  I had to operationalize the set up so that it functions on its own. So I decided that I was going to take a morning off my CEO job and actually code/hack to figure out how to make this work. First thing I needed was to get remote access to my Raspberry Pi so that I didn't need to carry a keyboard and monitor to the dusty plant environs. I was able to download and install PUTTY and PSFTP programs from the internet, they act as remote terminals for the Raspberry Pi. After some search through google I was able to fix the network setup on my Raspberry Pi so that it had a fixed IP address. Finally, the big victory of the day was installation of SMTP (Simple Mail Transfer Protocol) and figuring out how to send text messages to my Verizon phone when certain events occur on my sensor (for e.g. humidity exceeds 30%) from Python code. It all worked!! - my iPhone beeped and I received a message that the humidity had exceeded 30 %. By the end of the day I was bleary eyed but felt 20 years younger knowing I could still hack it like when I was in my 20s.

Messages on my iPhone from my Raspberry Pi

Finding a Home

I wanted to put the sensor in a location in the plant by a spray line but still keep it out of reach, have a case that will allow air flow and protect from dust/spray of enamel. Jim Hutzel suggested a bird house. So during lunch I went to Lowe's and bought a cedar bird house and unscrewed the bottom and put the Raspberry Pi inside. The small opening for blue birds in the front was just enough to provide an opening for the sensor, network and power wires. So far it has been in place for over a week and I am collecting valuable data of temperature and humidity over time.

Raspberry Pi and GrovePi inside the 'bird house'

The bird house assembled - the humidity/temperature sensor is in blue/white

Next Steps

I feel that I have come a long way. At least now I know how the connected hardware and data collection works. Now it is time to learn from the data, create heuristics and device a control algorithm in changing our operating variables so that they are more in line with the humidity/temperature conditions in the plant. We also have created a test bed to test the robustness of the hardware setup and its ability to stay up for extended periods of time in the harsh plant environment where it can approach 100 F during summer on the floor.

Key Takeaways

Here are some key takeaways.

1. Kudos to Initial State for organizing PiOT-101 workshops to create awareness and increase actual implementation of IoT.
2. The hardware for IoT is increasingly cheap and affordable. Our entire setup including sensors, GrovePI and Raspberry Pi was under $300 dollars (what you would pay for a spare part in a plant).
3. There are increasingly sophisticated online tools/APis available such as that from Initial State that will allow us to store and manipulate data at a very large scale - somewhat similar to what AWS has done for distributed computing.
4. There is a huge amount of know-how on Github and internet on Python programming and the Raspbian (Unix) on the internet.
5. You have to have the mindset of a 10 year old to succeed. Lots of time I tried things that lead to Epic Fails! but every time I learned something from them and the satisfaction to get it right is huge!
6. Collecting data is just the first step, turning it into knowledge and eventually wisdom is where the real value is created.

I will continue to post more blogs as we make progress.


Stay Tuned!
hemant