What do you expect from predictive maintenance?

• Maintenance cares about day-to-day operations – Reduced downtime
• Operations and IT look at the bigger picture – Improved operating efficiency
• Engineering groups get product feedback – Better customer experience
• Upper management wants to drive growth – New revenue streams

Why MATLAB and Simulink for predictive maintenance?

• Get started quickly
• Reduce the amount of data you need to store and transmit
• Deliver the results of your analytics based on your audience
• Create training data for your algorithm in the absence of real failure data

You can watch Branko’s full talk and download related resources from MathWorks Videos and Webinars.

Consulting firm Cadmus provides full-spectrum energy-efficiency support services to energy utilities throughout North America. These services include studies of energy use that require extensive data collection and analysis.

To make the most of the opportunities presented by the IoT, Cadmus engineers used MATLAB^® and the ThingSpeak IoT platform to develop a workflow for collecting, storing, analyzing, visualizing, and interpreting data from sensors in homes and businesses distributed across wide geographic areas.

“In just a few months, we implemented a new service that measures and analyzes temperature and humidity changes in dozens—and soon hundreds—of homes,” says Dave Korn, vice president of engineering at Cadmus. “Without MATLAB and ThingSpeak, we would still just be talking about it. Instead, we’re already pitching this service to utilities. That is a huge competitive advantage for our company.”

Solution

Cadmus used MATLAB and ThingSpeak to develop and deploy two systems of cloud-connected sensors for the near-real-time measurement and analysis of energy data, while establishing a workflow for rapidly implementing similar systems.

The first system, designed for an energy efficiency study of residential homes, used custom sensors to send temperature, relative humidity, and device battery voltage measurements to ThingSpeak every five minutes. The second, designed to monitor loads of HVAC systems and large appliances, used off-the-shelf home automation hardware to send power usage data, captured at residential circuit breakers and individual outlets, every minute.

Using the ThingSpeak web application, the engineers created new channels to collect data from the sensors and to quickly verify that each new sensor added to the system was reliably sending data.

Working in MATLAB, the team analyzed the collected data, using standard statistical techniques to identify outliers and calculate means and standard deviations. For example, they calculated and plotted power load profiles by the hour and correlated the load data with weather data from an outside source.

They created a variety of data visualizations, including interactive maps, which they shared with clients and prospects.

Cadmus engineers are currently using MATLAB and Statistics and Machine Learning Toolbox to develop predictive algorithms based on machine learning and advanced classification and regression techniques. These algorithms are designed to predict and model load based on weather conditions and sensor data collected via ThingSpeak.

Results