What Building Automation's History Portends About Building AI

A Short History of Building Automation (by Decade)

The Advance(s) of Electronics

The switch to electronics also brought the ability to network. Now devices could send their data on local area networks (LANs), which were daisy-chained together to get the data to the BMS, which would review it and send instructions back out. 

The next attempt at automation was to add a layer of algorithms, called an auto-tuner. These were either embedded into the BMS or as a black box solution on top of it, depending on the site. The goal of the auto-tuner was to make controls work “better.” The idea was sound enough: tuning algorithms are technically a viable solution. Unfortunately, they were unrealistic in practice. To begin with, too few operations teams knew how to validate these algorithms, so they were rarely set up correctly. Also, the algorithm also didn’t adapt fast enough to changing conditions. So instead of keeping pace with changes, their lag created an additional layer of problems and reduced control for operations staff. In the end, few properties kept up their auto-tuners, and they faded into the sunset with the Sony Walkman and the floppy disk. 

In the ‘90s, the development of sophisticated computer architecture resulted in two major changes: 1) eliminating the building’s dependency on the BMS to monitor and supervise controls, and 2) lowering upfront and ongoing labor costs of the BMS. 

But it wasn’t all puppies and roses. All these early computer systems were proprietary, which meant high costs and little customer sway. You couldn’t even buy a thermostat off the shelf and use it: you could only use the specific (and expensive!) part from your supplier. The arrival of BACnet and the subsequent standardization in communication protocols and parts eventually transformed what had been a very customized environment into something far more open and interchangeable. 

Suddenly, building automation could be easier and less expensive than ever before. But it didn’t achieve that much more.

The problem was that the cost of data storage was still high. Although the automation systems at this time were capable of producing fountains of building energy data, they had to store the data on-site, which jacked up the cost of the system. So instead, these systems produced only a trickle of operational data because the vendors wanted to keep the cost of purchasing their systems low. The result was that the cost of troubleshooting problems down the line stayed high and didn’t create any value for the building operators, who went back to operating the systems manually.

The Cloud's Silver Lining

The advent of cloud computing brought a drop in the costs as well as the difficulty of analyzing large data sets like building energy usage. This development exposed a level of detail we couldn’t know before, making it much easier to visualize causes and effects. On the negative side, it also inherently produces a firehose of inaccurate data consisting of a little signal and lots of noise. 

Happily, software engineers embraced this Big Data challenge quickly, using relational databases and developing new modeling techniques to sift through the data and pull out what’s meaningful. At the same time, we lose “in building” dashboards. Today, through careful human training and validation, large data models can be taught to recognize patterns, which enable predictive controls and recommendations that have vastly improved the capabilities of building automation. 

Today, we are entering the generative AI phase, where statistical models are trained to make predictions from training data about a dynamic situation without supervision. This is a significant departure from previous practices.

We need to remember what generative AI is – a statistical model based on probabilities found in the training data. Therefore, its results are directly impacted by the quality and appropriateness of the data and its suitability for the chosen building. Will the training data be continuously updated to accurately reflect the target assets? That would be a first.

The advent of AI has also allowed the “auto-tuners” from the past to rebrand themselves. Now they remotely analyze and then push new algorithms back into the BMS, which is dangerous in central utility plants and large office buildings and further distances building operators from their buildings.