WIN 2017/18

HVAC/P - HVAC and Plumbing Product News magazine - is the go-to resource for the latest products in the industry.

Issue link: https://hvacpproducts.epubxp.com/i/919673

Contents of this Issue


Page 16 of 38

BY SUNDEEP SANGHAVI A nyone who has had their HVAC equipment go down at home in 90+ F heat knows that shortages or failures can be a real annoyance. However, in a large-scale setting, this can be a life or death situation for businesses. Downtime costs for HVAC systems can range from $5,000 per hour up into the millions depending on how critical the system is to the business. Moreover, when HVAC equipment is not properly maintained, it can pose a great risk to workers and the general public. Traditional HVAC units are a common cause of fires in both resi- dential and industrial environments, and caused $207 million in damage to houses in 2010 alone. However, in tightly packed settings, the human and financial cost can be much higher. In today's digital age, though, HVAC contractors and facility managers can avoid both the human and financial costs associated with system failures by employing the help of anomaly detection and prediction systems, rather than rely- ing on the reactive approach of focusing on the degradation of equipment. The traditional approach of fault anal- ysis focuses heavily on air handling units, which make up the primary equipment of an HVAC system, as well as checking equipment (e.g., heat exchangers, fans and sensors) for damage or misconfigu- ration. Most companies take a sample of data from this equipment and build a model to see when machinery will fail — which means companies are only looking at generalized models, which do not give accurate insights. However, this method is not enough — nor the best way — to ensure an HVAC system is protected from failure. The traditional approach tries to fit a single behavioral pattern model to the data for time series analysis, which mostly serves for detection of extreme outliers defined by thresholds. The simple detection criteria also set off many false alarms that make fault detection a more difficult and time-consuming process. Often, there are so many false alarms that managers miss the most important red flags. "These manual and automatic rule- based systems and dashboards can flag when a problem has already happened and alarms have been generated but, depending on the design, it may not help in understanding "why" a certain issue took place," writes Nick Ismail, reporter for Information Age. Moreover, these methods are purely reactive, giving managers no option to recognize and solve problems before they occur. As a result, they're often left to clean up a mess they didn't see coming. With such reactive measures in place, HVAC contractors and facility managers will often not have a clear idea in advance of when issues will arise. As such, they cannot effectively communicate the needs to maintenance workers, which translates to unnecessary time and risk spent searching for problems that may or may not exist. As problems go unsolved — or worse, ignored — equipment faces additional and unnecessary damage that could easily be avoided. The issue of deferred mainte- nance effectively shortens the life of an HVAC system and can cost a business thousands of dollars to repair or replace. And that's not even the most costly part when considering the cost of downtime. While the preventive maintenance strategy of changing oil and parts every six months is a step in the right direc- tion, it is often wasteful and inefficient. Under this strategy, managers don't know what actually needs to be done or how much longer the parts could last because the decision is not based on hard data coming from IoT sensors within the machines themselves. In order to ensure an HVAC system operates reliably and with low-main- tenance cycles, managers must move from a reactive approach of anomaly detection to a predictive and cogni- tive one. Cognitive computing driven anomaly detection and prediction relies on the same sensors already con- nected to HVAC systems, but radically changes the analysis process to better anticipate problems and prevent them before they arise at higher levels of speed and accuracy. Cognitive computing powered anomaly detection does this by identifying patterns in seemingly unconnected data, diagnosing the root cause of the problem and implementing self-correction mecha- nisms. The technology is also capable of analyzing weather reports to optimize building temperatures and reduce the cost of electricity, coupled with actionable advice on how to reduce energy consump- tion without compromising comfort. Unlike the traditional method of anomaly detection, cognitive comput- ing enables the monitoring of an HVAC system under a more flexible model that better tracks the system's various behaviors. This means there will be fewer false alarms, and more accurate detection of data anomalies that, while falling within a normal range, actually indicate a critical issue. Using this method, problems are pre- dicted in advance and assigned corre- sponding importance levels that HVAC contractors and facility managers can use to appropriately deal with the situa- tion. In turn, the cognitive system offers clear instructions for the best course of action, thereby avoiding the timely and costly trial and error maintenance of traditional methods. Industry leaders like GE and Sam- sung, which specialize in HVAC systems and have hundreds of smart factories around the world, have already begun to implement this technology to great success. Their connected HVAC eco- systems create a huge amount of data, which they can then leverage to more effectively predict breakages and main- tenance problems in the exact machines used throughout the country. In fact, according to research, machine learning and cognitive predictive mainte- nance have been used to produce a 93 per- cent accurate anomaly prediction model that could predict issues at time periods of one day, three days and one week. With the rise of increasingly so- phisticated HVAC technology, such as variable refrigerant flow systems, the need for proper and timely mainte- nance has become even more crucial to avoid system failure and the ensuing costly downtime. For this reason, HVAC contractors and facility managers must take the steps now to implement cogni- tive computing as a predictive anomaly detection method, or run the risk of wasting precious business resources. Sundeep Sanghavi is CEO of AI-driven ma- chine learn ing company DataRPM. Why anomaly detection, prediction is a must for HVAC DIFFERENT APPROACH TO MAINTENANCE HVAC contractors and facility managers can avoid both the human and fi nancial costs associated with system failures by employing the help of anomaly detection and prediction systems. HVACPproducts.com HVAC & Plumbing Product News \ Winter 2017/2018 14

Articles in this issue

Links on this page

Archives of this issue

view archives of HVACP - WIN 2017/18