Download document () of 20

AC anomaly detection cools off a hot issue

Among all of our home appliances, the air conditioning system is one of the most energy-intensive devices we own. However, improving the energy efficiency of AC can be a challenging task. Faults in AC systems can result in consuming 15-20% more energy, leading to increased cost for the consumer and negative health and safety impacts. The Center for Intelligent Power developed a process to detect power anomalies, enabling prompt action to address failures and ensure optimal performance of the system.

The efficient operation of air conditioning (AC) systems is essential for both energy conservation and the safety and health of occupants in a building. The early detection of complex failures in AC systems presents a significant challenge in the field of HVAC systems. 

CIP employed a combination of statistical and machine learning techniques to detect power anomalies in real-time and forecasting potential issues, in order to minimize downtime and optimizing energy consumption.

Research

The project started with gaining a better understanding of the factors that could inhibit air conditioner performance.  This research phase included identifying what conditions could impair performance of a unit.  Things like fan motor issues, frozen evaporator coils, blocked air filters or vents, dirty condensers, and compressor cycle issues. With these understood as the most common problems with AC function, the team turned to collecting appropriate data that could lead to fault detection and identification (see figure below).

Analysis

Our data scientists used Eaton Smart Breakers data to first establish performance profile for an AC unit.  These included: 

  • Electrical spike: is the AC experiencing significant power spikes over ‘normal’ on a regular or consistent basis?
  • Long durations: Is the AC working non-stop more than 15-20 minutes (the expected cycle for a properly working air conditioner)
  • Seasonality: We expect ACs to work harder in the summer, when temperatures are higher. But Is the AC working harder this year, when compared to prior years?  If so, it may indicate a need for maintenance.
  • Descriptive statistics: Here we calculate the mean, median, and standard deviation of the electrical current data using the Sliding Window* method. Unusual mean/standard data could indicate a fault.

Any combination of these issues could then be mapped against the known possible faults established in our initial research.  This summary could then be sent to the consumer or manufacturer to address before the anomaly lead to a critical failure.

Implementation of this data-driven solution has benefits for consumers and manufacturers including energy conservation, increased safety, improved reliability, and increased comfort for consumers. 

AC anomaly detection.png

*What is the "sliding window'" technique?

Sliding window technique is a method used to efficiently solve problems that involve defining a window or range in the data and then moving that window across the data to perform the analysis of that data. Each “window” is a fixed size, and each data set within a window is processed independently.