“Please take your shoes off.” we were instructed, as we entered the Control Room. Cleanliness was apparently paramount.
It was cold, our Marketing Director noted as the floor robbed warmth through the fabric of his socks.
The room was spotless. White walls were neatly lined with desks full of displays and panels filled with buttons. Men in office clothing gathered around one monitor engaged in serious discussion. Busy day.
EnCo (short for Energy Complex) is a group of buildings that serve as office spaces for government agencies and companies in the energy industry in Thailand. These include the Ministry of Energy and some subsidiaries of PTT Public Company Limited.
Commercial buildings such as EnCo require HVAC (Heating, Ventilation, Air Conditioning) facilities to maintain comfortable atmospheres. These facilities make use of industrial assets such as chillers, condensers, and the like. What drives all these machines? That’s right: motors. And this is why we were there that day.
EnCo asked us what we could come up with if they gave us a couple of hours in their facility. Our system is currently meant for long-term installation to collect enough historical data for prediction. Similarly, EARS is designed to be left unattended for extended durations in order to achieve valuable results. In a previous post, we shared our experiment with PTTEP exploring a handheld approach to our AI system. We decided we’d do the same for this project. We used an uncalibrated device – a mobile phone – to collect acoustic samples.
From the control room, we were led to the actual chiller plant. It was an expansive room filled with pipes and vents that ran from different pieces of equipment and disappeared into the ceiling. The gleaming floor reminded us that cleanliness was fundamental to their operations.
After having been given a tour of the facility, we began. While doing recordings, we chilled with the EnCo team. They were quite interested in our solution, and we discussed it in detail. It was a little difficult to challenge the humming emitted by all the equipment. The sound enveloped the room and would drown out one’s voice if not speaking close enough in a conversation.
When all was said and done we managed to monitor a total of 7 motors and took a few samples from each. It took quite some time as each sample was 2 minutes long. We ran each sample through our AI system to detect any faults. Learning from our previous experiment, we tweaked the process and managed to gain a ~71% increase in accuracy!
The results were promising enough that the EnCo team requested a long-term installation of our EARS.
Moving forward, we will continue to explore this potential solution and figure out how we can further improve.