8 Elements that Make IoT Data so Different
Equipment manufacturers are facing an ambivalent challenge. On one hand, they have access to large amounts of rich IoT data. On the other hand, they struggle to efficiently handle it. The Internet of Things has nuances that require specific analytics tools that the traditional/general-purpose big data, web or mobile analytics suites can’t provide.
The tools used to analyze IoT data require a different approach entirely. In this blog post, we take an in-depth look at 8 elements that make IoT data so unique.
Diversity & Volume of IoT Data Sources
IoT data is derived from a plethora of devices. Often times, the type of data and its cleanliness make it impossible to process. Let’s use the example of rotation speed from magnetic bearings, or temperature, flow, and pressure from industrial compressors. These are all analog and generated by custom devices that are sometimes more than twenty years old.
In addition, a single device can have a dozen data sources from a variety of sensors. The IoT data ingestion pipeline needs to provide all the tools necessary to prepare the data and correlate it. It also needs to be able to create a time-series based graph of the observed behaviors.
Because of the diversity & volume of IoT data sources, the models themselves need to be able to adapt. And that basically means having a data model for every tenant.
You can have thousands of enterprise implementations in mobile analytics. However, they will still be tracking the same thing – taps, clicks or logins. In other words, they will track actions from the end-user.
There may be differences between a gaming application and a travel application. But ultimately, the data model and the outcomes you are trying to achieve will be very similar. But if we compare this to getting 3-axis vibration signals from a magnetic bearing versus flour/water/oil levels and usage from a bread maker… it’s very different!
In fact, in the IoT, you end up having a different data model for every implementation.
It is critical to spend time building an infrastructure that can handle a different data model for each tenant.
Figure 1.0 Mnubo SmartObjects Data Modeller
Data models need clean data to work. Indeed, having a higher data quality increases overall productivity. Unfortunately, when it comes to IoT data, cleanliness is not a given.
Everyone underestimates how ‘dirty’ IoT data is. Biases and outliers exist everywhere in the data. For instance, the location or the way the sensor is placed will actually introduce bias.
This is a concept unknown to web/mobile analytics. Indeed, all the ingested data comes from a human (ok, maybe some bots) clicking/tapping on a page/app.
As you can imagine, this poses a problem for data engineering, machine learning, and the necessary iterative re-training.
But once you’ve effectively cleaned your data, you can start analyzing it. And as you can expect, with over 100 devices getting connected to the internet everyday, there is a wide array of potential IoT data use cases.
Variety of Use Cases
In mobile and web, analytics and metrics are very well defined: user sessions, page loads, clicks and taps, time spent. Regardless of the website or mobile application, the metrics are universal.
In IoT, each use case has its own set of custom metrics to be analyzed. For instance, the charge time of electric cars, the cooling time for a plastic molding machine or the deflation rate for tires. Each use case will gather custom sensor data to form a coherent view of the problem to analyze.
An IoT Analytics platform must be able to assemble the right metrics and sequences to generate insights on a per device or per use case basis.
Figure 2.0 Mnubo SmartObjects EV Charger Demo
Because of its very nature, IoT data cannot be analyzed like web or mobile data. Indeed, aggregates are not sufficient.
Aggregate vs. Individual Data Analysis
With most web/mobile analytics implementations, it is sufficient to look at ‘aggregates’. For instance, the total number of ‘actions’ every 15 minutes / hour / 7 days can give you great insights. However, the individual actions are less important.
In IoT, one sensor reading can be the key to detecting a weakness or future failure. In other words, your system actually needs to index and keep all data (at least until it is deemed useless). Moreover, the algorithms needed in IoT are diverse: optimization, classification, time-series predictions, etc.
On the other hand, the web/mobile algorithms are most often about clustering end-users.
Figure 3.0 Aggregate view in Mnubo’s SmartObjects Platform
Figure 4.0 Individual view in Mnubo’s SmartObjects Platform
Because of how diverse IoT data is, and how diverse its use cases are, it caters to many different types of personas.
In web and mobile analytics, the user personas are well defined. They are digital marketing professionals, product managers, etc. In IoT Analytics, though, they serve several cross-functional personas who require insight into a variety of metrics or KPIs.
A product manager will want to see device usage. A field technician will need to get anomaly detection, alerts and maintenance history. An operations manager will want predictive maintenance recommendations.
In IoT, very few data scientists are available. This makes the importance of out-of-the-box insights (focused on IoT use cases) critical.
However, having cross-functional users doesn’t necessarily mean that all domain experts will access the platform. And that’s why there needs to be an emphasis on collaboration.
An IoT analytics expert / data scientist must recognize the value of domain expertise. An IoT platform needs to enable easy sharing and collaboration between domain experts and data scientists. In mobile/web, typically the user is a digital marketing professional who gets both the ‘analytics’ piece and the ‘domain expertise’.
This is not the case in the IoT where horizontal analytics platform ingest and manage data across verticals.
IoT data is unlike any other type of data. It is more diverse, and less structured. It also produces much larger volumes than regular web or mobile data. For all those reasons, specific tools need to be used when it comes to analyzing it.
Stay up to date
You might also like
Consumer IoT, Industrial IoT
Why IoT needs Machine Learning to (Really) Take Over the World
Putting the Internet of Things and Machine Learning in the same sentence is like playing…
5 IoT Analytics Myths
AI and the IoT have snowballed in attention over the past little while – garnering…
Consumer IoT, Industrial IoT
How 5G will change everything
What is 5G and how is it going to change the world we live in…