Using TensorFlow for Industrial IoT: What You Need to Know in 2018

A guide to commercial applications for machine learning and more

TensorFlow is a machine learning library that can be used for applications like neural networks in both research and commercial applications. Originally developed by the Google Brain team for internal use, it is now available to everyone under the Apache 2.0 open source license. TensorFlow has continued to be rapidly developed since its release in 2015, with over 21,000 commits so far, many from outside contributors.

In this post, we’ll take a look at how TensorFlow and machine learning in general can be applied to businesses in many different types of industries as well as some IoT applications of the technology.

Machine Learning In ContextMachine Learning Key Terms

As more and more devices get connected to the internet, it’s easier than ever to extract valuable data from them.

As a result, machine learning technology is accelerating quickly, enabling a wide variety of use cases across many different industries.

One particularly useful method of machine learning is called deep learning. Deep learning uses learning algorithms called neural networks to process information. This enables computers to identify patterns in data and define relationships between complex systems of inputs and outputs, among other tasks.

To put this into context, think about the neural networks in the human brain. When certain neurons are activated by stimuli such as light or sound, they send an electrical signal that processes those stimuli into vision and hearing.

In fact, biological neural networks inspired the design of artificial neural networks!

All About TensorFlow

TensorFlow’s strong support for deep learning as well as machine learning algorithms is just one of the many advantages of the framework.

Another example is Tensorflow’s architecture. It allows high performance numerical computation to be deployed across a variety of platforms as well as computers and mobile devices, servers, and edge devices.

TensorFlow Architecture

Image Source

Additionally, the framework offers a variety of programming languages, as well as functionality for both experienced and beginner-level developers.

Finally, besides being the most developed serving platform compared to other machine learning frameworks, Tensorflow’s significant backing from Google has lead it to become a proven market leader.

All of these features have enabled TensorFlow to be in the best position to deliver a shift from machine learning in the lab to uses in the real world.

Some Commonly Cited Uses For TensorFlow

TensorFlow is used for a variety of different applications from language detection, to image recognition, and time series analysis. There are a number of often cited examples that show the capabilities of the product and illustrate just how varied the real world applications of deep learning can be.

The Famous Case of the Japanese Cucumber Sorter

Cucumber Sorter

Image source

One of the most memorable examples of TensorFlow in action is Makoto Koike’s cucumber sorter. Makoto used TensorFlow’s image recognition capabilities to sort the cucumbers by quality, saving hours of work for his family each day.

Of course, image recognition is useful for many different industries and applications. Here’s a list of some of the real world examples of this technology in action:

  • Social media companies are using object recognition algorithms for photo tagging.
  • Scientists are using deep learning to facilitate image analysis for imaging applications like microscopy.
  • Retail stores can use object detection to create a smarter checkout experience, similar to Amazon Go stores.

There are many other industries that are beginning to use image and object recognition including healthcare, aviation, automotive, and space exploration.

Google’s Energy Bill Savings Plan

You may be shocked to hear that Google is concerned about the costs of their energy bills. But the reality is, power usage in data centers is a big problem for lots of larger companies, costing them hundreds of millions of dollars every year.

But imagine if you could reduce your power consumption by 15%.

Google did. How?

Power Usage - Google's Data Center

Well, along with other best practices, they used machine learning to leverage existing sensor data to model performance and improve energy efficiency.

They designed neural networks to look at data from around 120 different variables in the data centers and work out the most efficient methods of cooling, pump speeds, temperatures, and more to optimize their operations.

As shown in the chart above, Google’s Data Center Power Usage Effectiveness (PUE) has decreased significantly since the efforts began in 2008. Google’s data centers are now among the most efficient in the world.

TensorFlow and Industrial IoT: Predictive Maintenance and Beyond

factory-sensors

The ability to work backwards from an end goal without needing to specify the variables needed to achieve that goal is one of the most valuable applications of machine learning technology.

To put this in context, think about the many elements and variables that can affect the processes involved in complex manufacturing and other industrial operations.

Things like temperature, humidity, power usage, and fan speed, among others can all have an effect on the system’s operation.

With machine learning companies can set a goal, for example, power reduction, and the algorithm will be able to figure out which elements and variables are important in achieving that goal. This reduces the need for domain knowledge in sorting through the datasets to find the most important ones.

Additionally, TensorFlow (and machine learning in general) can offer a peek into the future by looking at the data it collects and predicting future events based on that information. The predictions will also get more accurate as time passes and more data is assimilated.

The ‘self-improvement’ capabilities of these algorithms alongside connected sensors can provide valuable insights to companies in all types of industries especially when it comes to predictive maintenance.

In fact, according to a report from McKinsey, machine learning capabilities like predictive maintenance will help companies save $630 billion by 2025.

Unlimited Uses, Capabilities, and Benefits

Industrial IoT Factory

Leveraging TensorFlow and IIoT together can allow users to:

  • Foresee the possibilities of a device failing
  • Predict the remaining life of equipment
  • See the causes of failure
  • Reduce maintenance time
  • Inform real-time process adjustments
  • Detect anomalies
  • …and more

Imagine a world in which you could schedule maintenance of your machinery before it breaks, removing the element of surprise and keeping downtime to a minimum.

Imagine being able to actually extend the lifecycle of a piece of equipment by running it at its most optimal settings.

Imagine how much these things could reduce the costs of running your operations.

From cucumber counters to object detection to predictive maintenance, the possibilities, applications, and use cases of TensorFlow are nearly limitless.

Curious about how TensorFlow, machine learning, and IIoT can help your business?  Contact Temboo today to learn more about our Industrial IoT solutions for companies of all sizes. And make sure to follow us on Twitter, Instagram, Facebook, LinkedIn, YouTube, and Medium for IIoT news, updates, stories and more!