Energy Management and Operations: from review of bills to predictive maintenance and AI-based facility control
Nowadays, energy data is commonly used to reduce energy cost but, in manufacturing, is this the best use of this data?
July 15, 2021 By Anatoli Naoumov
July 15, 2021 – Energy parameters reflect all processes happening at production floor—at every machine, at every second of every day—with an unbiased accuracy and precision no human can match. Analysis of this record can reveal hidden troubles, and even help convert them into opportunities when enough data is collected, and proper algorithms are applied.
The key words here are enough and proper.
This bounty of data—mostly untapped today—presents a tremendous opportunity to create value in many areas of manufacturing operations: from production planning and quality control to preventive maintenance, procurement, and emissions reductions, as well as traditional energy cost reductions.
Standing at the edge
Breakthroughs often happen when technologies developed for one area are applied to another—a phenomenon commonly called convergence.
So far, few high-tech technologies have been applied to energy management, and even less to management through energy data. In fairness, the high cost of sensors and data storage prevented or limited the availability of data; plus, processing vast volumes of data was further limited by the cost of computing power and the absence of algorithms.
Today, however, sensors, data storage and computing power become cheaper by the day, while algorithms gain power at a similar same pace. As such, lots of data becomes available, along with the tools with which to analyze that data.
As a result, we stand at the edge of a massive breakthrough in the use of energy data. Let’s look at what was, what is already here, and what is coming soon.
Archaic energy management: bill review
In pre-modern times—that is, a mere 10 years ago—energy management, in all but the most advanced companies, comprised a review of bills in the hope of uncovering tariff errors or misplaced decimal points.
It was accounting work… with a dash of engineering flavour.
As you can imagine, the data was not very granular. Bills provided a snapshot of the entire operation, and forecasts were performed on a monthly basis, at best. Errors took months to be noticed. Correction could take several months to implement.
Yesterday’s energy management: Dashboards and ratios
As meters came down in price, energy use monitoring came into fashion. Energy dashboards started to decorate lobbies and lunchrooms. Energy reports filled Inboxes and presentations. Dividing monthly kWh by widgets became a sign of modernity; displaying results on a graph became a cutting-edge energy management procedure.
Still, energy use and operations continued to be siloed, which severely limited value of energy data.
Several years ago, I attended an energy meeting at a major Ontario company where monthly kWh/ton ratios were presented for the past 12 months. Month-to-month values differed by 200%, yet no one so much as blinked, because energy management and operations were disconnected.
More than once, while standing by a flashy energy dashboard, I would have a conversation that sounded something like this:
Energy manager/engineer (proudly): Energy use is under control. You see, the plant used 10,977 MWh since Monday.
Me: How much should it have used?
Energy manager/engineer: It varies with production… which is what operations does.
When energy consumption is considered separately from operations, it has limited value to business. A sharp-eyed engineer may notice some big abnormalities, but this is more of an art than a science.
That said, energy management of yesterday provided somewhat more value:
• Granularity reached a minute level, even though the data was not analyzed as such.
• The level of detail moved down to the level of major machines. Individual machine efficiency could now be assessed.
• Correction time was reduced to days or weeks, but only for well-pronounced abnormalities, timeously noticed by a sharp-eyed engineer.
• Consumption forecasts remained at a month-to-month level, at best. Annual forecasts were more typically.
Most companies stay at this level today.
Today’s energy management: establishing the link between consumption and operations
Linking energy consumption to operations while considering the interconnection between processes opens a new horizon for energy management for creating value in areas other than energy cost reduction.
First, linking consumption to operations helps identify whether consumption is in line with what it should be. This is a fundamental change in our relationship with energy consumption: stepping away from reflecting on the past to predicting the future.
The implications of this seemingly modest change can be massive.
For example, at our Class A manufacturing client, the new plant manager noticed peak demand varied from day to day, even though production was predictably the same throughout.
He asked the maintenance manager to look into this disparity, and they discovered random instances of some loads running when they were not needed. By turning Off what was not needed, the plant decreased peak demand by 8% in just two months, with zero additional investment and zero negative effect on operations. This paved the way to active curtailing, which later saved the company over $150K a year.
Here are a few more examples of how a company benefited from linking energy use to operations:
• The analysis of natural gas consumption at a baking oven—in relation to the volume of baked bread—uncovered a deficiency in annual maintenance procedures, which caused inconsistent product quality atop of unnecessary gas consumption.
• At another company, the CFO gained the ability to accurately allocate energy costs to individual SKUs to ensure the pricing of products reflected their true cost.
Tomorrow’s energy management: integration of data sources for real-time automatic control, predictive maintenance, and more
Despite how far we have come, today’s approach to energy data in manufacturing remains primitive and slow as compared to how data is used in other industries.
As modern technologies arrive on the production floor (e.g. big data, machine learning, IIoT), the operations of entire plants and buildings will become automatically controlled in real-time to achieve operational efficiencies, enable predictive maintenance and procurement, as well as reduce energy costs.
Automation, sensors, and algorithms (human-written and machine-learned) will take on repetitive tasks by being our eyes, ears and, often, hands at every machine. This is much more than automation; through machine learning, algorithms develop strategies to take control of the equipment in a way that meets the requirements of predicted operational situations.
Here is a simple example of how this will work:
• Yesterday: HVAC always maintains set temperature.
• Today: HVAC maintains daytime and nighttime settings through On/Off operation.
• Tomorrow: AC turns On at the exact level and time required to meet desired temperatures at minimum cost in time for people’s arrival to each area of the building, based on current and forecast weather, occupancy forecast per area, electricity prices, and expected building behaviour.
The beauty and power of this approach is that algorithms can predict many future parameters by learning the past behaviour of machines, building and people. Also, the longer such a system runs in a facility, the better it can predict equipment failures.
This is not some far-off, futuristic vision: some of these technologies are already here.
For example, providers of peak demand prediction services employ big data and machine learning to predict when millions of energy consumers may create a peak, and how the reaction of hundreds of Class A companies may affect that peak.
And here are just a few more examples of how data-based technologies already work in different industries (where energy data is a part of the whole data package):
• Analysis of wind turbine operation using AI algorithms reduces operational costs at farms operated by Enbridge.
• AI-based production forecast for wind farms allows up to 20% revenue increase, as claimed by Michael Terrel, head of energy market strategy at Google.
• Ecopilot from Nova Scotia guarantees up to 40% HVAC savings in shopping malls and high-rise buildings by adding an AI-based control to the existing HVAC system.
• Otis offers an AI-based predictive maintenance system to ensure that elevators are always available.
• The IBM Maximo application suite improves asset reliability with condition-based maintenance based on asset health insights from operational data and analytics.
These examples illustrate an important trend in data-based operations management; that is, using data that emanates from numerous, diverse sources—commonly referred to as big data—to improve various aspects of operations.
Metering does not equate to energy management, although it makes objective energy management possible.
Energy data hides a vast and mostly untapped source of value for manufacturing companies, especially when it is integrated with other sources of data and processed using AI algorithms.
And the time to start collecting data is now, as most companies do not have the right data to begin training AI algorithms.
Using AI algorithms to control operational processes presents an opportunity to deploy the best operational knowledge at each plant. Companies that embrace AI and machine learning stand to build an unbeatable competitive advantage.
Meantime, those choosing to wait and see risk joining those businesses that have chosen to wait and see how that whole internet thing plays out.
Anatoli Naoumov, MBA, MSc, CEM, CMVP, is a managing partner and “chief energy waste buster” at GreenQ Partners, and has been involved in various areas of business analysis and development for over 25 years for companies in Canada, The Netherlands and Russia. Anatoli’s latest venture focuses on the use of energy data for creating business value in manufacturing. He can be reached at firstname.lastname@example.org.
Print this page