Preventive maintenance triggers on time or runtime; predictive maintenance triggers on condition data. The right CMMS supports both, and the right wastewater plant uses both. The mistake is treating them as competing strategies. In practice they are different tools for different failure profiles.
This guide explains the two approaches in plain language, shows which asset classes belong on which strategy, walks through the transition from preventive to predictive, and quantifies the return. If your utility is choosing between "traditional PM" and "predictive analytics", the answer is almost always "both".
Getting the definitions right
| Approach | Trigger | Example |
|---|---|---|
| Reactive | Failure has occurred | Pump has seized, replace bearing |
| Preventive (time or runtime) | Calendar interval or accumulated runtime | Bearing greased every 90 days |
| Predictive (condition based) | Sensor data crosses a threshold | Vibration signature indicates bearing wear |
| Prescriptive | Analytics recommends both timing and action | Model predicts bearing failure in 40 days; schedule now |
The four categories exist on a maturity axis. Reactive is the least mature and most expensive per event. Prescriptive is the most mature but requires substantial data infrastructure and analytics capability.
The PF curve: why the two approaches are complementary
The classic reliability engineering PF curve shows that potential failure (P) is detectable well before functional failure (F). Between P and F is the window in which condition monitoring can catch degradation. For assets with a long PF interval (weeks to months), condition monitoring works. For assets with a short PF interval (hours to days), time based PM is the better tool because the sensor data does not give useful warning.
The Society for Maintenance and Reliability Professionals published the canonical PF curve reference, and the ISO 55000 asset management framework incorporates it in condition monitoring guidance.
Which assets belong on which strategy
| Asset class | Best strategy | Rationale |
|---|---|---|
| Critical duty pumps | Predictive (vibration, amperage) | Long PF interval, high failure cost |
| Standby pumps | Preventive (runtime + exercise) | Rarely running, condition data limited |
| Blowers | Predictive (vibration, temperature) | Long PF interval, high energy cost |
| Motors | Predictive (insulation, temperature) | Slow degradation curve |
| Valves | Preventive (time based) | Short PF interval, no cost effective sensor |
| Instruments | Preventive (time based calibration) | Calibration drift is time based |
| UV disinfection lamps | Preventive (runtime based) | Well characterised runtime to end of life |
| Membrane systems | Predictive (pressure, permeability) | Long PF interval, direct sensors |
| Aeration diffusers | Predictive (dissolved oxygen, air flow) | Slow degradation, cost effective monitoring |
| Chemical dosing pumps | Preventive (time based) | Small size, limited sensor value |
Starting from a mature preventive programme
Most water utilities are still building preventive maturity. The path is well defined:
- Complete asset register in CMMS.
- PM templates for each asset class with defined intervals and evidence requirements.
- PM schedules generating work orders at the right cadence.
- Field execution on mobile with evidence capture.
- Compliance measurement (PM completion percent) reported monthly.
- PM effectiveness review every 12 months.
Once the preventive programme is running above 95 percent compliance for 12 consecutive months, the utility has the foundation to start the transition to predictive.
Transitioning to predictive
Step 1: identify high value candidates
The ISO 55000 framework suggests focusing on assets where reliability improvement has high consequence value. In wastewater, this is usually critical duty pumps, blowers, and process instrumentation.
Step 2: instrument the candidates
Vibration sensors, motor current monitoring, temperature sensors, or process sensors specific to the asset class. Cost per point ranges from USD 300 for basic runtime hour tracking to USD 3,000 for high resolution vibration monitoring.
Step 3: baseline the healthy signature
Every asset has a "healthy" sensor signature. Baseline it during a period of known good operation. Anomaly detection is only as good as the baseline.
Step 4: define alert thresholds
Amber and red thresholds should be set based on manufacturer guidance, industry standards (like ISO 10816 for vibration), or historical data. Start conservative and tighten as the programme matures.
Step 5: connect thresholds to CMMS work orders
When a threshold is crossed, the CMMS should auto generate a work order. The technician response is the same as any other work order, but the trigger is condition data rather than time.
Step 6: review and refine
False positive rate matters. Too many false alerts and crews stop responding. Too few and real failures slip through. Quarterly review of alert to failure correlation.
Predictive analytics: the next layer
Threshold based predictive maintenance is the entry point. Predictive analytics adds statistical models that forecast failure probability from time series data. Modern analytics platforms use survival analysis, gradient boosted trees, or recurrent neural networks to predict remaining useful life.
The data set required is significant. Typically 18 to 36 months of high resolution sensor data plus failure event labels. Utilities that started building the data set in 2020 are now positioned to deploy analytics; utilities starting today should not expect predictive analytics before 2027.
Cost comparison
| Approach | Setup cost per asset | Ongoing cost per asset per year |
|---|---|---|
| Preventive (time based) | Nil to USD 100 | Labour cost of PM cycles |
| Predictive (threshold based) | USD 500 to 3,000 | USD 100 to 300 (data + review) |
| Predictive (analytics) | USD 3,000 to 8,000 | USD 500 to 1500 |
| Prescriptive | USD 5,000 to 15,000 | USD 1000 to 3000 |
Measuring PM effectiveness
A preventive programme that hits 95 percent compliance is not necessarily effective. Effectiveness measures whether the PM actually prevented the failure. Metrics:
- PM compliance: percent scheduled PMs completed on time.
- PM to failure ratio: unplanned failures divided by scheduled PMs.
- Finding rate: percent of PMs that identified something requiring action.
- MTBF trend: mean time between failures improving over time.
- PM cost effectiveness: cost per failure prevented.
A well tuned preventive programme has a finding rate around 15 to 25 percent. Below 10 percent, the PMs are over frequent and should be relaxed. Above 40 percent, the PMs are under frequent and should be tightened.
Condition monitoring in practice
| Signal | Asset | What it detects |
|---|---|---|
| Vibration (velocity, acceleration) | Pumps, blowers, motors | Bearing wear, imbalance, misalignment |
| Motor current signature | Any electric motor | Impeller wear, bearing wear, electrical issues |
| Temperature | Bearings, motors, VFDs | Overheating from wear or load |
| Oil analysis | Gearboxes, hydraulic systems | Wear metal accumulation, contamination |
| Ultrasonic | Steam traps, compressed air | Leaks, cavitation |
| Insulation resistance | Motors, transformers | Winding degradation |
| Thermography | Electrical panels, motors | Loose connections, hot spots |
Return on investment
Predictive maintenance on top of a mature preventive programme typically delivers 20 to 40 percent additional downtime reduction, 15 to 25 percent equipment life extension, and payback of the sensor investment within 2 to 4 years. Utilities that skip preventive and go straight to predictive rarely capture these benefits because the underlying execution discipline is not in place.
Data infrastructure for predictive
Predictive maintenance requires a data pipeline that many utilities have not built yet. The pipeline needs to move sensor data at 1 second to 1 minute resolution from field devices through the SCADA layer, into an historian or time series database, then out to the CMMS or analytics platform. Common architectures use OPC UA at the field, an on premises or cloud historian (OSIsoft PI, InfluxDB, TimescaleDB) for storage, and REST APIs for downstream consumption. Storage volume is significant: a wastewater plant with 200 sensors sampled every minute generates roughly 100 gigabytes per year, and analytics grade data at 1 second resolution multiplies by 60. Retention policies typically keep raw data for 12 to 24 months and downsampled data for 5 to 10 years.
Team roles for predictive
A predictive maintenance programme typically requires roles that time based PM programmes did not: a data engineer to build and maintain the pipeline, a reliability engineer to define thresholds and models, and an analytics developer to build dashboards and alerts. Smaller utilities can outsource some of these to platform vendors or consultants. The pattern that consistently works: keep the reliability engineer role internal (they understand the assets), consider outsourcing the data engineering (specialist skill, spiky demand), and be pragmatic about the analytics developer (often part time or shared with other utility functions).
Common failure modes
- Sensor calibration drift: false alerts erode trust.
- Alert fatigue: too many alerts, crews stop responding.
- Insufficient baseline: models trained on unusual operation give bad predictions.
- Data pipeline gaps: sensor to CMMS integration breaks and no one notices.
- Missing failure event labels: models cannot learn without labelled outcomes.
- Vendor lock in: proprietary sensor data formats make platform changes expensive.
- Overcomplication: analytics on assets that would be fine on time based PM.
Vendor selection for predictive
The vendor market for predictive maintenance splits into three tiers. Tier one: full analytics platforms with integrated sensors, data pipeline, and predictive models (typically enterprise pricing). Tier two: analytics platforms that connect to existing sensors and historian systems. Tier three: point solutions for specific asset classes (a vibration analytics tool for pumps, for example). Selection depends on utility scale, existing data infrastructure, and analytics maturity. Smaller utilities benefit from tier three tools with focused scope; larger utilities with existing historian and SCADA investments benefit from tier two solutions. Tier one platforms are appropriate when the utility is building the data infrastructure from scratch.
Where this is going
Predictive maintenance is becoming mainstream in wastewater over 2025 to 2030. The two major trends: standardisation of sensor interfaces (OPC UA, MQTT) reducing integration cost, and continued cost reduction in sensors (vibration sensors under USD 200 are now viable). The International Energy Agency flags predictive maintenance as one of the top energy efficiency technologies in industrial and utility sectors.
Frequently asked questions
Should we skip preventive and go straight to predictive?
No. Predictive requires the disciplined execution that comes from a mature preventive programme. Skip the foundation and the analytics never work.
How much of our maintenance should be preventive versus predictive?
Well maintained plants often see roughly 40 percent preventive, 40 percent predictive, 20 percent reactive after 3 to 5 years of investment. There is no single right answer.
What is the minimum data set for predictive analytics?
18 to 24 months of continuous sensor data on target assets, with at least 5 to 10 labelled failure events for model training.
Do we need data scientists?
Not necessarily. Threshold based predictive can be run without data science. Analytics driven predictive typically needs at least one internal or contracted data science role.
Which sensor gives the highest value first?
Vibration on critical duty pumps. Highest failure cost, best characterised PF curve.
How reliable are false positive rates?
Mature threshold based systems run 5 to 15 percent false positive. Analytics driven systems can achieve 2 to 5 percent with mature models.
Should we buy sensors from the pump vendor or independent?
Depends on lock in tolerance. Independent gives flexibility, vendor integrated gives easier installation.
Can we do condition monitoring without full time online sensors?
Yes. Portable vibration analysers on a monthly walk down are much cheaper and cover many assets acceptably.
What about acoustic emissions?
Emerging technology, useful for specific failure modes (early bearing wear, valve leakage). Cost per point still relatively high.
Do we need to update the FMEA when moving to predictive?
Yes. Predictive changes the detection score materially. Rerun the FMEA to reflect the new detection capability. See FMEA for wastewater plants.
Summary
Preventive and predictive maintenance are complementary, not competing. Preventive fits assets with short PF intervals and time driven degradation. Predictive fits assets with long PF intervals, high failure cost, and characterisable sensor signatures. A mature wastewater utility runs both on the same platform, chooses the right strategy per asset class, and reviews effectiveness quarterly. The transition path starts with a disciplined preventive programme and layers predictive on top over 2 to 5 years.
Next reading
- What is a CMMS for water utilities?
- FMEA for wastewater plants
- How a CMMS reduces unplanned downtime in pumping stations
- CMMS ROI at real wastewater plants
- Browse the wastewater plants directory
See the assets in this article
Explore 177,000+ utility infrastructure sites
Locations, capacity, operators, and permits across 24 sectors: the same records our writers pull from.
Start browsingMaintenance guides from the UtilityRadar team.