Image source: http://www.computerweekly.com/news/450304522/Australia-adopts-British-internet-of-things-framework

What is IoT?

Internet of Things is now being regarded as the biggest revolution in technology space after the invention of Internet. With an estimate of the number of ‘things’ to be connected by 2020 to be crossing 50 billion, the rate at which data is going to be generated will be increasing on full throttle.  With storage costs coming down and concurrent advancement happening on the hardware side (sensors, storage, connectivity, data transfer etc.) the industry adoption may only be expedited.

Self-driving cars, which have hogged the limelight for most part of 2016, have the attention of almost every big company in Silicon Valley (Google, Apple, Tesla, Uber). A self-driving car is a quintessential IoT system with an array of complex sensors like laser range finders, gps, cameras, lighting, weather sensors among others.

Aviation is another industry, with companies like General Electric having made big bets, where an aircraft has no less than tens of thousands of sensors generating about a terabyte of data in every flight. Close monitoring of each of its components is extremely critical to avoid fatal crashes and to reduce fuel consumption.

In addition to transportation, other enterprise level applications of IoT can be found in manufacturing, smart homes, smart cities, oil & gas, healthcare and energy management.

What is AoT?

IoT generates data, AoT analyzes the data and instructs IoT making things happen in real-time without human intervention.

Data not touched and analyzed is lost opportunity. Imagine a self-driving car that collects data from all its sensors, but does not analyze and take any action. With the rise of AI development, it is one of the most important examples to demonstrate the power and advantages of IoT, wherein real time decisions are to be made to identify the direction and take a turn or to stop the vehicle in case of an accident (as happened recently with Tesla).

As IoT begins to permeate every industry, enough emphasis should also be given to developing new innovative algorithms or improving existing ones to analyze the data generated by these things. This has led to coining the term Analytics of Things (AoT), which deals with analysing raw data generated by the things of IoT.

What are the challenges in AoT?

Data generated and decisions to be made based on its analysis are real time in IoT space, and hence most of the analysis involves time series. Anomaly detection using time series involves flagging spikes in the graph and attributing those flags to occurrence of an event. This detection needs to be instantaneous and highly accurate. As there are several sensors which generate data from various components of the entity, correlations can be drawn based on the analysis and some relations can be attested to functioning of different components of the machine.

A simple example for this is the data generated from sensors of a milling machine. We collect data related to the speed at which it is being operated, temperatures at various parts of the tool and surface properties of the material. With a temporal plot of the variation of surface temperature and material properties like surface hardness, we can predict when the tool is about to breakdown and warn for maintenance or replacement beforehand. Relation of surface temperature with the speed of operation helps us in achieving an automated optimal performance of the machine.

Challenges in AoT

  • Processing for Algorithms
    • The data spewed out by sensors is not pretty, and needs lot of processing.
    • Unlike bank or retail spaces, sensors generate lot of noisy data in addition to useful information, which needs to be filtered before it can be stored. This step needs to be optimized to least time.
  • Variable data frequency
    • The frequency at which each sensor sends the data might be different.
    • Since data is collected from different components, data collection frequency could be different based on the physical conditions and requirements.
  • Diverse data formats
    • In the above example, owing to the different magnitudes of temperatures at different points on a milling machine, thermocouples of different scales are used to measure temperatures, which need to be normalized.
  • Unlabeled data sets
    • Data related to anomalies in various conditions is not available, and requires efforts to train models in an alternate way like bootstrapping until enough labeled data is available to train supervised algorithms.
  • Data security
    • As devices seamlessly integrate into our lifestyles, living rooms, office spaces, grocery stores and cities, we run into the risk of data generated or collected put to misuse. For successful IoT applications, algorithms need to be built for keeping security and privacy in check.
  • Data Interpretability
    • As data is collected from many sources (sensors), the dimensions have to be reduced by some dimensionality reduction methods (like PCA) for ease of computation, at the cost of interpretability of the new variables.
    • In anomaly detection, we need to identify if a spike in the analysis is due to an actual failure of the component, or due to erroneous signals sent by the sensor.

We have already seen IoT move into our homes and cities and with the future being ripe for disruptions with Artificial Intelligence, we are going to see some of that right in our living rooms. In fact 2017 is going to be a lot more interesting for AoT as much as 2016 was for IoT.

If you wish to learn more about what we are working on at Predera in the space of AoT, drop us a note at hello@predera.com