In today’s world of Internet of things(IoT) everywhere, the real benefit of IoT comes not just with inter
connecting more and more devices, but in our ability to harness this data to take quicker decisions leading
to accurate and timely actions by leveraging Advanced Analytics.
Advanced Analytics has become a key driving factor to define the success of a company irrespective of either
the Business is a Corporate or a Boutique startup.
Imagine the days of no online shopping or convenience of calling a cab via a mobile app on a rainy day or unable
to check the weather for the day online or find the route to drive home. Internet of things (IoT) has
advanced in endless areas of our lives like Manufacturing, Healthcare, Social media, Commerce etc., more so in areas that
impact our Personal and Social lives like Weather, Connected homes and soon Connected cars, Connected stores and even
Connected Elevators. After all, what will we do without our Smart phones on our side with its array of sensors to
wirelessly connect us to this world.
Right now there is a huge gap in our capability to harness this potential and so an exciting opportunity to Analyze
this Data flood.
Businesses sit on top of piles of Digital data with no way to unearth the nuggets of wealth that lay hidden.
The challenges faced include how to effectively apply Advanced Analytics on this overwhelming data and also how to
identify and prioritize their application areas.
Let’s have a look at the areas of applications and it effectiveness vis-a-vis the effort invested.
IoT implementation patterns can be seen from a Systems perspective cutting across applications. The Applications are
far too many and impacts all areas starting from data inception, consumption to mining foresights from the data.
Before we start discussing, we need to distinguish between Business intelligence systems and Analytics systems. More
so because of the looseness in which these terms are used interchangeably these days.
Objective of Business intelligence systems are to churn out Management Reports in the form of Adhoc reports,
Structured or Fixed / Canned reports that can be consumed by Management or Operations. Reports will also include Push
or pull notifications like SMS, email etc.,
On the contrary, Analytics starts where Business Intelligence ends with a key distinguishing factor being complexity
of calculations performed by leveraging mathematical based models and its outcomes.
Broadly, the application areas of Advanced Analytics in IoT can be seen as below in the order of the journey of the
data from source to consumption and its complexity
- Analytics location
- Rule based deviation detection
- Anomaly detection
- Change point detection
- Predictive system behavior
- Edge Analytics
Advanced Analytics for IoT can be classified based on:
- Nature of deployment in a System: That is, where exactly in the Data flow of the system is the Calculation logic or Engine deployed
- Complexity of Advanced Analytics calculations: How complex or advanced these algorithms used for the calculations are,
Rule based deviation detection: This is a basic application of Analytics in IoT where in a
predetermined relationship between a set of parametric values are known to indicate a specific outcome of another
parameter within the system. These parametric values are nothing but data read from sensors mounted on various IoT
subsystems like Temperature, Pressure etc.,
These rules are knowledge acquired by Subject matter experts (SMEs) about the system which is translated and embedded
as Software code for raising an alert or alarm when the conditions are met.
On demand Rule based deviations can be detected using BI reports or Applications with UI to configure these rules.
Anomaly detection: is a step up where in using Statistical techniques, a deviation from the normal
operating ranges of an equipment is identified and alerts are raised. Anomaly can be due to a single parameter or a
combination of parametric values which are out of wack
Next higher in the order of complexity is Change point detection wherein a statistically driven
algorithm detects a Change in performance of the system. Change point detection is often confused for Rule based
deviation detection. The primary distinction is that Change point detects any permanent departure in the performance
of a system while a Rule based change detection detects a point in time change which might fall back to normal range
after a few moments of spikes
One of the advanced applications of Analytics in IoT is Predicting systems behavior. These can be
either Forecasting future state of a parameter using Time series models or a combination of Time series forecasting and
Multivariate event driven Predictive models. The Target for Prediction can be linear values (like Temperature) or binary
conditions (Failure Yes/No condition)
Based on where the Analytics computation takes place it can be either
- Server or Database
- Client – App
- Edge layer
The rules engine can run either in the Server, Client/ App or the Edge depending upon the time critical nature of the
outcomes. The time criticality to act on the prediction results varies based on the application area like say,
- Predicting response of certain demography to Promotion campaigns
- Suggest a restaurant to a person filling up fuel on a highway
- Initiate change in parameters in-process for a Robo arm undertaking a Plasma metal spray operation on a Aero engine part
One of the most complex areas to implement Advance Analytics Predictive model is Real time control of automated
processes in Manufacturing like Welding, Spray painting etc.,
Some of the factors that influence the use of Analytics in the Edge layer are,
- Edge devices like Switches, Routers, Gateways have limited resources for computation intensive Analytics applications
- Historical data required for building and using a Predictive model is stored in Servers that are several layers downstream from the edge. This hinders Real time action based on Predictions done closer to the Data layer
- The layer at which Predictive Analytics model is run are seldom hard wired to the controllers in the edge layer, which in turn can send control signals to change course while a process is underway
While Advanced features like Edge Analytics may sound glamorous from commitment to modernizing an Organization perspective,
it is recommended to consider a very detailed evaluation to justify the need for such deployments.
The more closer one moves towards the Edge layer, more tighter coupling between the Software and Hardware platforms
are needed. Thus, any deployments on the Edge requires a overhaul of the Hardware controlling the equipment, and
hardwiring them to Analytics capable Software systems.
In terms of process, it will also mean, “To what extent a Decision to Action translation can be allowed in an
automated system without human intervention to judge validity of such Machine generated decisions.?”