Reliability and predictive maintenance

Why it matters

Preventive maintenance changes parts on a schedule whether they need it or not. Predictive maintenance (PdM) measures the machine's condition and acts when the data shows a problem developing. With sensors wired to PLCs and historians, controls technicians increasingly set up, read and act on this data. Catching a failing bearing weeks early turns a breakdown into a planned 2-hour job.

The P-F curve

Most failures do not happen suddenly. A defect starts at some point, and the machine's condition then declines:

Condition
  |‾‾‾‾‾‾‾‾‾‾‾‾\
  |             P   (potential failure: first detectable)
  |               \
  |                 \   vibration rises, then noise, then heat
  |                    \
  |                       F  (functional failure)
  +---------------------------------> Time

The P-F interval is the time between when you could first detect a problem and when it fails. Vibration analysis usually detects bearing defects earliest, then oil analysis, then audible noise, then heat. Inspections must be more frequent than the P-F interval, or you will miss the window.

Vibration basics

Vibration sensors (accelerometers) are mounted on bearing housings. Readings are taken on routes with a portable data collector, or continuously with wired or wireless sensors feeding a PLC or monitoring system.

• Overall level: a single number summarizing the vibration, usually velocity in mm/s or in/s RMS. Simple and good for trending.
• RMS (root mean square): square every sample, average them, take the square root. For a pure sine wave, RMS = peak / √2 ≈ 0.707 × peak.
• Crest factor = peak / RMS. About 1.41 for a smooth sine; higher values point to impacts, such as early bearing defects.
• Spectrum (FFT): breaks the signal into frequencies, which is how analysts identify the cause: 1× running speed usually means imbalance; 2× often misalignment or looseness; bearing defect frequencies are non-whole multiples of speed.

Worked example. A velocity signal is a sine wave with a 4.2 mm/s peak. RMS = 4.2 / √2 ≈ 2.97 mm/s. Using example limits of 2.8 (alert) and 7.1 (danger), the reading is in alert. Severity limits depend on machine size, mounting and type; ISO 20816 (which replaced ISO 10816) gives guidance, and many sites set their own limits from baselines.

Trends beat single readings

A single reading of 2.9 mm/s may be fine for one machine and alarming for another. What matters most is the change from that machine's own baseline. Weekly readings of 1.8, 1.9, 1.9, 2.1, 2.4, 2.9, 3.6 mm/s show an accelerating rise: plan a bearing inspection now. A least-squares slope gives the average rate of change, and a rising slope is a strong signal even below the alert limit.

Thermal data

Infrared thermography finds heat that should not be there:

• Electrical: loose or corroded connections, overloaded or unbalanced phases, failing breakers.
• Mechanical: hot bearings, misaligned couplings, rubbing belts, blocked cooling.

Compare temperatures with similar components under similar load (phase to phase, bearing to bearing) and with ambient. Worked example. A motor's terminal connections read L1 41 °C, L2 43 °C, L3 68 °C. L3 is about 27 °C hotter than the coolest phase, a strong sign of a loose or corroded connection on that phase. Sites use severity tables (for example from NETA standards or their insurer) to decide how urgently to repair.

Thermal scans of electrical equipment are usually taken with covers open on energized gear, or through IR windows. That is energized work requiring a qualified person, PPE and an energized work procedure under NFPA 70E. Do not open panels to practise.

Other PdM techniques

• Oil analysis: wear metals, water, viscosity and particle counts.
• Ultrasound: compressed air leaks, electrical discharge, bearing lubrication condition.
• Motor circuit analysis and current signature: winding and rotor problems.
• Process data from the PLC: rising motor current, longer cycle times, pump pressure falling over weeks.

Reliability metrics

From the maintenance lesson: MTBF = operating time / failures, MTTR = repair time / repairs. Availability ≈ MTBF / (MTBF + MTTR). With MTBF 500 h and MTTR 4 h, availability ≈ 500 / 504 = 99.2 percent. PdM aims to raise MTBF and turn unplanned repairs into short planned ones.

Common mistakes

• Reacting to one reading without a baseline.
• Comparing vibration in different units (peak vs RMS, mm/s vs in/s) as if they were the same.
• Taking readings in different places on the housing each time, so the trend is noise.
• Collecting data that nobody reviews or acts on.

Beyond the platform

Vibration analysts commonly certify through ISO 18436-2 based programs, for example via the Vibration Institute (https://www.vibinst.org/); reliability professionals often pursue SMRP's CMRP or CMRT (https://smrp.org/). Thermographers certify through similar programs. Checked on 2026-10-23.

Recap

• Act inside the P-F interval: measure more often than failures develop.
• RMS = √(mean of squares); sine RMS ≈ 0.707 × peak; crest factor = peak / RMS.
• Trends and baselines matter more than single readings.
• Thermal: compare similar components and ambient; energized scans need qualified people.

Try it

Open the Do tab and finish a condition monitoring calculator by writing the RMS formula.