Bring combined sensor and SaaS technology to industries that have been slow
to benefit from the rapid pace of technology innovation over the past
decade.
“Analog” industries such as Agriculture, Forestry, Fishing, Construction,
Trucking, Manufacturing, Oil and Gas, and many others are yet to fully benefit
from technological innovation in the way that information-based and
service-based industries have. These “analog” industries require physical
hardware to interface with the software world. But hardware development is
expensive, complex, and iteration takes too long compared to software
development.
But now hardware development is changing. Prototyping cycles have shortened,
electronic components can be ordered in small quantities quickly and cost
effectively, and there has been a surge in simple-to-program microcontrollers.
The production cost for a simple wireless sensor device can be less than $25,
and hardware development time can be reduced from a year to months or even
weeks! The plethora of Internet-of-Things consumer devices is one result of
these changes (fitness trackers,
wifi-connected light bulbs, smart refrigerators). However the path to
profit for Internet-of-Things consumer devices is not very clear. Are people
really willing to pay a monthly service fee to monitor their refrigerator? Or
install a new phone app for every device in their house?
On the other hand, there is a clear value proposition for Internet-of-Things
innovations for businesses, and more specifically, “analog” businesses.
Internet-of-Things devices and services can help them respond to customer
service problems faster, mitigate business risks, reduce labor expenses, reduce
energy expenses, improve coordination with suppliers, and improve understanding
of equipment health. The list goes on and on.
Analog gauge reader for residential and commercial heating and hot water
propane tanks. While products exist in this market, they are expensive, slow
to setup, overly complex, and aren’t updated frequently in response to
customer feedback – a perfect market conditions to disrupt!
A friend’s family member asked “Why isn’t there a way for me to know the
real-time level of my customers’ tanks? It seems like this should be easy to
solve.” He runs a propane distribution company and has to put a driver in a
delivery truck to check and, if low, fill-up customer tanks. And while he has
software to estimate customer tank levels based on historical consumption and
weather, the software requires accurate data-entry and cannot account for
unexpected changes in consumption from visiting guests, vacations, kids going
off to college, babies being born, interior renovations, etc.
But I ran in to one big problem while doing market research. Propane is not used
much in Washington state relative to other states. According to the U.S. Energy Information Administration,
Washington state is ranked 34th in its use of propane as a percentage of all
non-vehicle fuels used. Additionally, of the 3.2 billion BTUs of propane
consumed in the United States in 2013, just six states consume 74% of it, and I don’t live
anywhere near those six states: Texas, Kansas, Lousiana, Iowa, Vermont, and New
Hampshire. It doesn’t make much sense to start a business to serve a market
that’s hard for me to access.
Restaurants, hospitals, grocery stores, school cafeterias, any place that
stores food has refrigeration needs. Moreover, those refrigerators need to
be monitored so that they meet state health codes and food doesn’t
spoil.
Geographically, commercial refrigeration would be a much easier market for a new
small business to access. Target customers exist in almost every town and city
across the country.
However, the multitude of target customers combined with the relative simplicity
of temperature sensing hardware and low margins of food-serving establishments
(mostly restaurants) means there is not a lot of room for a profitable product
line. The customer acquisition cost alone for a hardware product like this would
be untenable, not to mention the ongoing SaaS maintenance costs. There would be
little margin left for continued R&D, which is a large part of my value proposition in using a SaaS
pricing model with hardware.
Build a field-testable prototype.
A family member working in oil and gas for over a decade heard about Gauge
Reader and thought the idea had legs but could be applied to much more than
propane.
Better understanding the opportunity for a product like this, I focused
full-time on building a field-testable prototype, planning out all the aspects
necessary for a functioning product.
Requirement Definition ➜ System Architecture ➜ Proof-of-Concept ➜ Financial
Modeling ➜ Project Plan
I wanted to quickly understand the feasibility of the product, the costs
involved, and the knowledge I would need to build a prototype. This included
extensive research in
Following this research, I sourced the minimally necessary components and built a prood-of-concept.
In one week, I taught myself Electronic Design Automation software and the
Electrical Engineering concepts I needed to create a printed circuit
board.
This was very much a just-in-time learning process, and I quickly
iterated between periods of research and production.
Within two weeks, I had a circuit board design sent off to a PCB fabricator.
During the two week turn-around time I had plenty of other work to keep me busy,
including tooling setup, component sourcing, and the server-side software with
which Gauge Reader would communicate.
A lot of tooling is needed for hardware prototyping. Some of this I already
owned; some I needed to purchase.
Over a couple weeks I outfitted an electronics workbench with the tools that
would allow me to efficiently prototype and iterate on hardware design.
Sourcing electrical components can be a full time job by itself. Multiple
variables need to be compared to find the components with the optimal
balance of characteristics.
I compared hundreds and sometimes even thousands of surface mount passive
components like resistors and capacitors, board-to-board connectors, battery
holders, waterproof enclosures, and mounting hardware.
For each of these components, variables such as price, physical size, voltage
and current tolerance, operating temperature range, temperature coefficient,
heat dissipation, equivalent series resistance, and quiescent current needed to
be evaluated at the same time. Additionally there is an art to being able to
quickly read and extract the necessary information from electrical component
datasheets.
Computer Vision enabled a key component of Gauge Reader: the ability to
flexibly translate an analog gauge reading to a digital value.
Computer Vision allows a computer to understand information about the physical
world – or in my case, allows a computer to read various types of
analog gauges.
Just like electronic hardware design and production, Computer Vision was an area
about which I knew very little. After reviewing multiple academic papers1, 2,
3,
4 on using
Computer Vision with analog gauges, I began using Python and OpenCV to design an algorithm that could
quickly and accurately read the image of an analog gauge.
Within three weeks I had an algorithm that could read many analog gauges in
addition to a web-based tuning UI so that the system could quickly be adapted
to new gauge faces. Pretty awesome to see the messy physical world translated to
the binary world.
By this time I had iterated twice on the design of the PCB during
respites from the mind-bending gymnastics of Computer Vision matrix
mathematics.
With a beautiful, well-functioning, power efficient PCB, it was time to integrate all the
pieces and begin testing.
Verify all initial design requirements and stress the hardware before
deploying it in a customer installation.
Beyond the basic functionality, I needed to ensure the hardware would stand up
to the harsh conditions that are normal for industries it was designed to serve.
Oil and gas, construction, forestry all require functioning equipment regardless
of the natural (and man made conditions) thrown at it.
Founder, Product, Software, Hardware
