Where is all the AI in the land of industrial IoT?

• Read original article here

For years, there has been a buzz around AI and how it would revolutionize industrial IoT, changing our lives forever. Organizations would realize enormous gains in productivity, do much more with less, improve working conditions, and reduce waste. AI would usher in an age of reliable devices that do our bidding without issue.

Today, consumers interact with AI solutions every day, from Alexa to self-driving cars, wearable devices, and even “robo” financial advisers. This begs the question, given all the potential gains in the industrial space and the oceans of data being created, where’s the AI in the land of industrial IoT?

The answer is not as complicated as you might think.

In the beginning … very few organizations had the means or time to invest in hiring the armies of consultants and data scientists needed to deploy AI. Even less had meaningful data to use for these purposes. The early adopters who did make the investment took a top-down approach to change the organizations’ internal workflows through AI.

In many cases, this resulted in disjointed and overcomplicated solutions that didn’t align with the real-world needs of the front-line operators. The outcomes of these early initiatives ended up as either very expensive data experiments or zombie pilot programs that never made it into production.

The industrial IoT space is now at a point where data is plentiful and readily available. Almost every piece of industrial equipment now reports some form of output, location, or machine health data that can be used for predictive maintenance, outlier detection, or location intelligence.

AI technology has also come a long way since those early years. It is no longer a black box that can only be understood by a select few. It is no longer necessary to hire an army of outside help to get started. Organizations can now deploy AI on their existing infrastructure using off-the-shelf solutions.

If you have devices that report data, you have everything you need to get started. As we have mentioned in previous posts, it takes a short amount of time to train a predictive maintenance model when working with IoT data.

Let’s look at an HVAC unit in a building; you typically have five data points (discharge temperature, return temperature, humidity, setpoint, and room temperature)reporting to a controller that is then visually displayed on a dashboard or building management system. This data is collected once a minute and can help provide current environmental and machine health indications. You might even have rules in place that give an alert when something breaks or is trending in the wrong direction. This is good, but we can do much better.

Now take those same five data points and connect them to a Predictive Maintenance Template via RESTful API. Over the course of a few days, enough data has been created to train, select, and deploy an AI model that is best suited for that particular HVAC unit (or multiple models if multiple units are presented). Once deployed, the system immediately sends predictions back to the existing dashboard indicating the unit’s health using all the available data. These predictions provide an indication of future failure allowing the users time to take corrective action before a problem occurs.

This API-driven approach enables front-line operators to deploy AI using their existing infrastructure resulting in seamless integrations and fast improvements to the bottom line.

We have all lived through enough technology hypercycles to be skeptical of nascent technologies. The interesting thing is that AI in IoT has moved far beyond the hype and is now delivering real-world results.

According to PWC, manufacturers that deployed predictive maintenance on average:

Don’t let F.U.D. get in the way of your success. Sign up for free access to Elipsa APIs today!