Traditional alerts provide limited lead time between identifying potential failure and functional failure. Potential failure denotes that point at which components are in need of repair. Functional failure is that point at which equipment no longer is delivering its intended business value. (All images courtesy SmartSignal)

Condition-based and predictive-maintenance activities use software and hardware tools to analyze equipment performance, delivering value by identifying equipment faults and focusing maintenance efforts on those components needing repair. The value of these analytical processes, however, reaches a limit when existing tools focus solely on the equipment itself rather than the entire equipment system. The most current advances in equipment-reliability management use advanced algorithms to improve understanding of equipment health across varying operating modes, thereby increasing equipment availability and reliability.

Reliability engineers typically use a “P-F curve” to compare the cost of managing maintenance and repair activities against that of equipment failure. Key points on the curve represent “potential failure” (P) and “functional failure” (F) points (Figure 1). Potential failure occurs when events lead to component damage that needs repair. Functional failure occurs when equipment performance no longer meets design conditions and must be shut down for repair.

Where systems have relatively predictable life cycles, engineers schedule time-based maintenance work to an interval shorter than the statistically confident time-interval between points P and F. In the oil and gas industry, however, due to widely varying operating conditions, equipment life cycles generally aren’t predictable. Engineers therefore are challenged when attempting to conduct failure mode and effects analysis (FMEA) of the potential faults. The shape of the P-F curve, including the total life between startup and functional failure, keeps changing. This leads facility operators to use process alerts to focus analysis, decision, and planning efforts.

Engineers strive to identify the potential failure point of equipment life as early as possible. However, they face trade-offs when setting alert levels for process variables. Alerts need to provide meaningful notice of potential failure, but also must have tolerance to the normal noise of processing transients and equipment loading variations. Without advanced software algorithms, the critical planning and decision-making time between points “A” and “F” reach a maximum limit where false alarms begin.

Reliability management evolution

Reviewing reliability management advances helps pinpoint its limitations. For one, signal processing has moved frequency-domain, vibration-spectrum analysis to portable devices. Technicians can make immediate vibration-signature analyses without a trip back to the central processor and workstation. Vibration signatures can be stored in a variety of data historians.

Process alerts without advanced analytics must balance detection sensitivity with the need to eliminate false alarms.



The downside of this is that historical data can be compiled at a dizzying rate. Once collected in an historian, the associated alarm-management process spawns an “analyze everything” activity mode, which expends as much effort as any alarm-management process, but with newly added analysis complexity.

A second advance is application of expert, rules-based systems intended as productivity aids for engineers. However, given custom-engineered equipment systems typical to the oil and gas industry, it can require as much effort or expertise to maintain the expert systems as it may take to keep highly trained specialists on staff.

Statistical computing packages, once the domain of mainframe or high-end workstation computing, have given engineers tools to crunch large quantities of data, but the same trap remains as with expert systems. Statistical tool kits may require very deep mathematical skills to assure their outputs aid decision-making in the real world.

While the technology associated with equipment-reliability management has advanced, so has the associated management discipline. The practice has roots in the aviation industry, which undertook critical views of all activities and costs associated with maintenance and overhaul and weighed them against the requirements of safety and operating economics. The oil industry has firmly embraced these advances and codified them into API Standards 617, 670, and 678.

Finally, support-related businesses that focus on inspection, equipment health, maintenance, and overhaul also developed in the oil industry. Original-equipment manufacturers and independent companies provide inspection technicians and overhaul consultants. Industry-recognized certification programs in vibration and thermography are the industry norm. And Internet connectivity has made remote-monitoring and diagnostic services more readily available.

Changing reliability management

Despite technology and best-practices improvements, changes in the core work processes for reliability management have been less dramatic. Field work-management and planning processes saw the biggest changes, with the focus shifting to cost definition and control. Fault analysis has been directed inward toward equipment with faster, cheaper, and more abundant data, plus analytical horsepower.

A model of rotating equipment that includes the processing equipment of which it is a part can deliver a more complete picture of equipment health.



There are still “craft” divisions between engineers who focus on reliability and maintenance, on the one hand, and those who concentrate on managing and optimizing processes. The result is a barrier between those looking deeply into rotating equipment versus those looking at how equipment ought to operate under a variety of conditions.
In other words, some machines experience higher vibration under low load, while others do so under high load. The key point is that generally speaking there is a need for better understanding of equipment condition in the context of its operating range.

As it pertains to the rotating equipment example, vibration and bearing temperatures are the top dogs. Typically, diagnostics begin after the specialty staff receives notice of an alert. Often a “fire-fighting” mindset and the need to eliminate false alarms means alerts are set at levels that occur only when equipment has sustained damage.
The consequences of a short planning horizon leading to an unplanned shutdown are well known:
• Downtime is longer due to resource and scheduling constraints;
• Expediting parts and labor increases costs; and
• Risk of catastrophic failure is greater since equipment will have passed its potential failure point, and time till functional failure is uncertain.

Operators need earlier detection of potential failures and to overcome the current limitations of existing alarms- or rules-based systems. The key is to move away from the inward focus of failure detection and fault diagnosis toward the bigger picture of equipment health in the context of varying load conditions. Use of advanced pattern-recognition and predictive algorithms can help achieve this desired equipment-centric view.

Advanced predictive algorithms

A predictive-analytics package must build on a firm’s existing technical foundation and be a logical next step in extending work processes. It cannot be seen as an individuals science project or an unknown quantity prior to actual implementation.

The predictive algorithm system needs to meet the following criteria:
• Ability to describe the operating profile of each piece of equipment across known operating contexts and ambient conditions;
• Delivers, via its field instrumentation, an equipment health profile sufficient for decision making;
• Instrumentation has industrial-quality precision; consistent, repeatable measurements are more important than test-stand precision;
• Data infrastructure collects and records instrument readings at five- or 10-minute intervals. Many process-control historians collect and record data at significantly higher frequencies;
• Fault-tolerance to failed instrumentation is possible;
• Performs as a fit-to-purpose product — rather than a tool kit — and has incorporated user-driven features and enhancements over time;
• Visualization of events and incipient alerts is possible;
• Deployment is based on use of models empirically derived based on known performance, rather than first-principles comparisons to test stand performance. These configurable and extendable models enable rapid buildout for both common and unique machines; and
• Ease of use provides myriad deployment, maintenance, and monitoring options, including full service, to flexibly meet the needs and cultures of users.
SmartSignal’s second-generation product, EPI Center, for predictive-analytics within the oil and gas industry, meets the criteria above. It can identify events and minor equipment damage well before that point at which traditional alerts trigger in-depth equipment-diagnostic activities.

Modeling rotating equipment together with processing equipment yields a more complete picture of equipment health. A straightforward example would be evaluation of compressor rotor axial position in the context of differential pressure, flow, speed, and gas temperatures. This gives reliability engineers a way to see whether events may have caused minor damage, such as labyrinth rubs, or if there are slow-acting changes in evidence, such as rotor fouling.

Case study in event detection

One of the most common potential failures in regard to the performance of compressors is fouling, whether of the axial air compressor section of gas turbine engines or gas processing compressors. In either of these cases, predictive analytics can detect the events and chronology that lead to cumulative fouling. In a gas compressor, performance is monitored as part of a system that includes all of the process components.

Predictive-analytic results from the suction cooler outlet on a booster gas compressor. Heat transfer events provide clues to liquid carryover that can affect compressor performance.



SmartSignal compared all of the operating conditions of the compressor and surrounding system and identified periods when the suction coolers were not able to keep up with the needed heat transfer. In all, there were 11 periods of probable liquid carryover, ranging from 6 hours to 17 days.

Eventually, the repeated episodes of liquid entrainment resulted in rotor fouling that could be measured directly in compressor performance (Figure 3). Extending the compressor-performance model to include associated equipment such as the suction coolers readily enabled the root-cause analysis of the compressor fouling problem. Predictive analytics generated an accurate chronology of the events that led to the fouling, which could then branch the analysis to the inlet separator performance or to actual field operating conditions.

In more general terms, predictive analytics and pattern-recognition capabilities can increase the span of diagnosis and planning time that falls between an event and the consequent functional equipment failure (Figure 4). In some cases, equipment operators can completely avoid functional failures by taking early corrective action. They now can move to the next step in the evolution of work processes through use of advanced predictive algorithms to prioritize their use of engineering and field resources.

Comments