The prospect sounds beguilingly attractive: a slickly-designed window pops up on your smartphone, alerting you to some critical piece of equipment that needs your immediate attention. Or does it? In most cases, that’s your job to figure it out. You’ll most likely have to trudge over to said piece of equipment to investigate. With some experience, knowing that this is just a false alarm, you may just shake your head, mutter something and swipe right or whatever to make the alert disappear. Now, amplify this scenario by a thousand, and you can start to see that this could get really irksome if the majority of the alerts are just rules crying wolf when no action is actually required. With the rapid proliferation of “smart” systems and sensors and the resulting firehose of “big data”, this is a growing reality in many industrial and commercial settings.
Alerts: too much of a good thing
While it is critical to have machines self-report critical errors, such alerts are only as robust as the analytics that generate them. Analytics for machine-generated data need to be sophisticated enough to flag only the important and persistent issues. The power of algorithms and software should be deployed to generate such insights automatically. The point of an effective Human-Machine Interface (HMI) system is to reduce the amount of work that humans are doing, not to increase it exponentially. Ultimately, accurate insights are more useful than a barrage of misleading alerts. In actual use, it is common for people to pore over the underlying data to validate the alerts since they don’t trust these systems. Vendors have responded by offering better data visualizations. Sadly, beautiful visualizations and charts – however much they may cause Edward Tufte to applaud – are besides the point. Needing battalions of analysts to manually validate and review the output of analytics systems is in fact counterproductive and costly.
In a recent example, a system to “help” operators of a building generated more than a hundred thousand alerts in the first month. For the exasperated operators of that system, you can bet that that all those alerts were pure noise. It’s easy to imagine their newly issued smartphones beeping plaintively in a trashcan somewhere. Unfortunately, this situation and the resulting “alarm fatigue” is all too common.
At what point do a swarm of alerts devolve into noise? Apparently humans can only pay attention to around four things simultaneously. Such limitations of working memory are an important consideration in human factors design, ranging from automobile dashboards to the control panels used for nuclear plants. These advances in user interface design were spurred in part by well known disasters such as Three Mile Island accident, where one of the contributing factors was the noisy design of the control panel.
So, in your quest to integrate systems in the Internet of Things world, beware of dashboard merchants bearing alerts. However slickly designed such systems may be, if the underlying analytics do not support the identification of a true root cause, it’s going to be a nuisance at best and a dangerous meltdown at worst.
The next time you hear someone promising you alerts, you should be alarmed.
NOTE: This is Part 1 of a 3-part series describing the state of the industry for each of Ecorithm’s 3 building optimization steps (outlined in the previous post) and how Ecorithm uniquely overcomes these challenges – and other pervasive issues – to deliver highly accurate intelligence.
1. Identify and Fix Issues
2. Maintaining Health
3. Optimizing for Peak Performance