Abelon Systems News Stories and Articles

Machine Learning (ML) is traditionally associated with large servers and enormous amounts of computing power. However recent improvements in both the efficiency of the algorithms and processing platforms mean that it is now possible to include ML functionality on much smaller embedded systems.

This article covers some of the advantages and challenges of distributed machine learning applications, discusses some of the suitable ML methodologies for use in an embedded environment, and finally looks at some of the pitfalls for the unwary.
For the last 15 or 20 years Machine Learning has become one of the key building blocks of IT systems and has become part of all of our lives, albeit mostly a hidden one. The ever increasing amounts of available data require increasing sophisticated processing techniques and the advances in ML over this period have been quite mind blowing.

Things ain’t what they used to be …

The emergence of reliable, low-power, low-cost ‘Internet–of–Things’ (IoT) devices has made it economical to deploy advanced Building Management Systems (BMS) into small to medium - sized buildings, but the relative simplicity of those devices (basic sensors, actuators and communications stacks) could expose those buildings to the threat of cyber-attack, with potentially catastrophic consequences.

An integrated BMS monitors, manages and controls the critical systems within a building - HVAC, lighting, security, fire alarms, access control and more; a sophisticated attack could disrupt, disable or damage these vital support systems.
Deployment of these systems is growing rapidly; according to Gartner around 45% of the 1.1 billion connected things in 2015 were devices in smart buildings and homes and this total is set to explode to around 21 billion devices in a global smart building market estimated to be worth $36bn by 2020.

Connecting things together is what humans do.  Combine that skill with the need to eat and drink gave rise to the first soft-drink vending machine connected to a campus network.  No more need to leave your seat to find out if there was a cold Coke waiting for you.  That was back in the 1980s and embedding computing power into real-world devices and connecting them to the Internet is now called the “Internet of Things” or IoT.  In just a few years there will be billions of enabled devices connected to the Internet.

Joining the IoT party challenges product manufacturers to become service providers.  A consumer buys the product, plugs it into the ‘net and subscribes to the service provided.  There is a clear upside for the makers of things.  They can establish a lasting customer relationship through the data collection and the subscription. That relationship can be carefully deepened and leveraged.  

Building Smart Metering products is complex.  The device must safely and securely measure energy consumption as meters have done for decades, but a Smart Meter must also accurately calculate bills, be a part of the energy supplier’s systems and be a part of the customer’s information network. 

The Smart Meter can be seen as part of the on-going Internet-of-Things (IoT) and cloud revolution.  It is also a key part of the national smart grid infrastructure. You won’t yet be controlling your Smart Meter with your phone, but you will have access to your bill, your energy use and innovative Smart tariffs that can meter energy at special rates and times.  Smart metering is all about communications, but not those prevalent in a typical cloud.

A common discussion point with many of our clients, particularly the smaller ones, is whether they should try to buy an “off-the-shelf” hardware platform, or design a bespoke system from scratch.

The usual answer is “it depends” and while I’m not claiming that we have the definitive design solution I’d like to share some of the lessons that we have learned over the last decade regarding the most cost effective way to develop products for our customers.