Moving Beyond Preventive Maintenance
Electric utilities are among the most asset-intensive industries and one of the most critically essential, where failure can mean catastrophe. Maintaining those assets has, therefore, always been a priority, often a costly priority, but an unavoidable one. Failure of most assets could mean outages or reductions of power, and failure of certain critical assets could have a domino effect, causing failure of assets down the line. Any failure could result in loss of revenue and regulatory censure; massive failure could mean extremely costly replacement costs. The goal of any maintenance system is the prevention of both total failure as well as any deterioration of performance. Today, however, utilities can go beyond preventive maintenance by adopting predictive maintenance, which can provide a model that allows for far greater accuracy in judging the continued viability of assets and can vastly increase the efficiency with which they are maintained.
Typical preventive maintenance is designed to help assure that an asset will not reach failure or serious performance deterioration. It prescribes a fixed time schedule or may be based on operational statistics and relies heavily on manufacturer and industry recommendations. It’s much like the oil-change recommendations in all car manuals: after so many miles or a specific time-frame you change the oil. It’s simple, but it’s also highly inefficient. How many cars will actually break down or suffer serious deterioration in performance or reliability if the oil isn’t changed according to those parameters? Although they don’t know what the probability of failure is, most drivers simply follow the recommendation. That’s fine when dealing with a simple and low-cost maintenance procedure, but it can be very wasteful when dealing with something far more costly and complex.
Predictive maintenance relies on two basic tools: machine-to-machine (M2M) technology and complex algorithms. The algorithms extract insights from the continuously gathered data to predict when equipment needs maintenance, repair or replacement. Statistics on the age, failure rates, hours of use, environmental conditions and actual inspections for each asset are compiled and the results are compared against a fleet of like assets. The patterns and trends that emerge help define a prescriptive maintenance plan.
Predictive analytics relies on software that uses historical operational signatures for specific assets and then compares that data to real-time operating data gathered from that component to detect any subtle changes in behavior or conditions before they become critical to performance. For a specific piece of equipment that data might be compiled from thermal infrared imaging or from vibration analysis, for example. It thus provides more time for analysis and to plan corrective action, should that be needed and avoids routine maintenance that might be irrelevant to actual performance. In other words, rather than relying on the calendar or a predetermined metric, predictive maintenance is performed according to the data collected from specific, individual assets. Repair or replacement can thus be limited to components that actually need it, when they need it.
Predictive analytics offers utilities enormous savings in maintenance expenditures, which can easily account for 20- to 30 percent of operating expenses. It also captures the kind of knowledge that experienced employees have provided in the past, which is particularly important as the work-force ages and valued employees retire. Utilities can now be free to develop maintenance schedules based not just on manufacturers’ suggestions or the expertise of senior employees. Rather, maintenance can be scheduled based on real-time performance data and highly sophisticated predictive models.
With its obvious advantages in streamlining maintenance and increasing reliability, one would expect predictive maintenance to be the norm for the power industry. It no doubt will be, but implementing this analytics-powered approach to maintenance is a complex, costly task. It requires a highly advanced data architecture and leading-edge information systems, as well as almost universal deployment of M2M communications equipment. Furthermore, only the largest utilities have anywhere near the historical data predictive maintenance requires. Transformers, for instance, have low failure probabilities. It would take years for an individual company to compile enough data about why and when transformers fail. Fortunately, companies need not rely on their own experience; they can purchase fairly inexpensive commercial failure databases or use models based on academic research to supplement their own historical knowledge.
Because the data sets and the algorithms used to interpret them are so complex, traditional data processing application software cannot handle them. Since both the hardware and the expertise may be beyond the capabilities of all but the largest electric utilities, at least at the outset of adopting predictive maintenance, much of the technological aspects may be outsourced to companies such as IBM, SAP, Siemens and others.
Whether operated entirely in-house or with an outside-vendor partnership, predictive maintenance offers enormous, tangible benefits. It results in greater reliability, which garners greater customer satisfaction; reduces the total cost of ownership of assets; avoids unexpected equipment failure, which compromises crew efficiency; and it improves the utility’s overall safety and regulatory compliance. Those are powerful arguments for adopting it, and today, when the industry is quickly transforming from a staid, utility model to a dynamic, interactive one, the argument for its adoption becomes particularly compelling.