Equipment Diagnosis – The Doctor will see you now.
Author: Michael Santucci
What if knowledge of typical equipment failure modes could be stored and combined with historical operating data? The potential for
engineers to easily identify the probability of early-onset failure modes provides key insights into the risk associated with continued
operation. Cutting-edge diagnostic technology in Predict-It™ reveals subtle changes in operating data and allows these symptoms to be
combined with on-site observations; arriving at an accurate diagnosis of the failure’s root cause. Using optimal prescriptive maintenance
actions helps target evidence-based diagnoses and achieves the lowest maintenance costs.
Let’s examine how maintenance has evolved and see how Predict-It™ leverages diagnostic insights for the highest potential savings.
Types of Maintenance
Reactive maintenance – With this method, repairs to equipment are made post-breakdown as a restorative solution. Fix it when it breaks.
This makes sense for some assets where there is redundancy, for multiple parallel pumps running at low capacity factors, or for assets with
little to no impact on plant output. Both the cost of instrumentation and the infrequency of common failure modes do not justify the cost of
the sensory package required to monitor it.
Preventive maintenance – Changing your engine oil every 5000 miles or replacing your furnace filters once a quarter are forms of
preventative maintenance. Original Equipment Manufacturers (OEMs) generally supply conservative recommendations for maintaining their
equipment such as when to replace parts, service fluids, clean, etc. This type of servicing can be expensive and may also lead to premature
removal of completely operational parts, all of which to avoid a less than likely catastrophe.
Predictive Maintenance – Predictive maintenance digitally monitors equipment health to identify possible failures long before traditional
monitoring techniques. This technology enables a tailored maintenance approach rather than scheduled maintenance during overhauls. The primary
goal of predictive maintenance is to precisely target conservation and repair efforts to avoid costly downtime. Advanced Pattern Recognition
(APR) algorithms are deployed to identify sensors that are not operating as expected. These anomalies can be mapped to symptoms, the outward
effects of common faults.
According to Allied Market Research, a market research and advisory company of Allied Analytics LLP, the global predictive maintenance market
size was valued at $2.8 billion in 2018 and is projected to reach $23 billion by 2026, characterizing the importance that predictive maintenance
offers to the industry, both now and—even more so—for years to come.
While the value of predictive maintenance is clear, a challenge is presented in how to distinguish between trivial issues and those of potentially
catastrophic proportions. Power Generation companies and other heavy industries like the Oil and Gas sector employ Operational and Diagnostic
Monitoring Centers (ODC’s) to monitor critical assets and digest asset anomaly data for the fleet. In monitoring centers, Subject Matter Experts
(SME) are called to interpret the anomalies and provide a differential diagnosis of what failure mode(s) or mechanism may be the cause of the
failure. However, these diagnoses can be somewhat subjective and lack a calculated estimation of fault likelihood.
Experts with years of industry experience have an empirical knowledge base to understand different types of alarms. This allows them to differentiate
between a combination of symptoms to pinpoint a root-cause for many typical problems. As these experts retire or move up in an organization, the
holes in collective knowledge leave an operation exposed to the learning curve of less-experienced engineers.
Predict-It’s Diagnostic Reasoner captures the knowledge of the SME and provides a uniform approach to interpreting fault probability. Prior occurrences
of a given failure, statistical symptoms, and the strength of the causality between fault and symptom are the basis for determining the likelihood of
the fault. This way, every instance of a likely asset fault is evaluated with the same metrics.