Most organisations planning a predictive maintenance programme spend months debating which data to collect. The real question is simpler: what does your equipment actually tell you, and are you listening? This guide cuts through the noise.
If you have searched for “predictive maintenance data requirements,” you have probably landed on articles listing dozens of sensor types, petabytes of historian data, and enterprise AI platforms that cost more than the machines they are supposed to protect. That advice is not wrong exactly, but it is often wildly disproportionate to the actual problem.
The truth is that a well-scoped predictive maintenance (PdM) programme can start with remarkably little. What matters is collecting the right data for the failure modes that actually threaten your plant, not the most data. Here is what that looks like in practice.
First: why data requirements depend on your maintenance strategy
Before listing data types, it is worth understanding where predictive maintenance sits within the broader maintenance strategy landscape. As Faraday Predictive’s own education resources explain, there are essentially three generic approaches: breakdown maintenance (reactive), planned preventive maintenance (time-based), and condition-based/predictive maintenance (CBM/PdM).
CBM is the optimum strategy because it means work is only done when condition data indicates it is genuinely needed. That reduces unnecessary interventions, cuts costs, and avoids the well-documented problem of “infant mortality” failures caused by the act of maintenance itself.
Key insight: Most equipment does not fail on a predictable schedule. Repeated studies show that the majority of machines exhibit random failure patterns rather than age-related deterioration. Time-based maintenance intervals are therefore almost always either too frequent or too infrequent. Condition data is the only reliable guide.
The myth vs. reality of PdM data
Common myth
You need continuous, high-frequency sensor data from every bearing, seal, and winding before you can begin.
Reality
Periodic data collection from the right measurement points, matched to the P-F interval of your critical failure modes, is often sufficient to catch problems in time.
Common myth
Vibration sensors on every machine are the foundation of any serious programme.
Reality
Vibration is a leading technique, but it cannot detect electrical faults, operational issues like cavitation, or rotor asymmetry. Selecting techniques to match your actual failure modes matters far more than standardising on one technology.
Common myth
You need a large data science team to interpret PdM data.
Reality
Modern condition monitoring systems, including MBVI-based approaches, are designed so that plant engineers, not data scientists, can interpret outputs and act on them.
The core data categories you actually need
Effective predictive maintenance data falls into three domains. For rotating equipment driven by electric motors, which covers a very large proportion of industrial assets, all three can be captured from a single measurement point.
| Domain | What it detects | Typical source | Priority |
| Mechanical | Bearing wear, imbalance, misalignment, looseness, gear mesh faults | Vibration, motor current/voltage (MBVI) | Essential |
| Electrical | Supply imbalance, harmonic distortion, winding faults, rotor bar defects, power factor issues | Motor voltage & current waveforms | Essential |
| Operational / Process | Cavitation, blockages, throttling, belt slip, process load variation | Motor current signatures, flow/pressure where available | Essential |
| Thermal | Hot spots, insulation degradation, cooling system issues | Thermography cameras | Supplementary |
| Acoustic / Ultrasound | Leaks, early bearing defects, electrical discharge | Ultrasonic probes | Supplementary |
| Oil / Fluid analysis | Contamination, wear particles, degradation | Lab sample or inline sensor | Supplementary |
The key observation in this table is that the three essential domains, mechanical, electrical, and operational, can all be interrogated using a single technique: measuring the voltage and current drawn by the motor. This is the basis of Model-Based Voltage and Current (MBVI) analysis, and it is why MBVI systems are increasingly replacing multi-sensor configurations for motor-driven equipment.
Learn more: What is MBVI technology?
What MBVI data actually looks like in practice
When you install a system that captures voltage and current waveforms, you gain visibility that most sites have never had. You can see three-phase voltage and current traces, detect harmonic distortion, observe frequency and phase balance, and identify waveform shapes under different operating conditions.
From those signals, a model-based approach can simultaneously extract information about bearing condition, rotor asymmetry, air-gap variation, belt drive integrity, cavitation events, blockage, throttling, and electrical supply quality from a measurement made safely in the motor starter cabinet without any sensors on the rotating equipment itself.
Real-world example: A seawater lift pump was monitored using MBVI. The system detected intermittent cavitation caused by sea swell, varying the inlet pressure, catching events that lasted only seconds and occurred unpredictably. Conventional vibration analysis, applied periodically, would almost certainly have missed them entirely. The pump was corrected by adjusting the antifouling system, with no maintenance intervention on the pump at all.
Read: Common Predictive Maintenance Failures (and How to Avoid Them)
A practical checklist: data requirements for getting started
If you are scoping a predictive maintenance programme for the first time, work through these steps before specifying data infrastructure.
Identify your critical assets
Score each asset by production impact, current maintenance cost, and failure frequency. Rank them. Apply your data investment to the top of that list first, not uniformly across the site.
Define the likely failure modes for each critical asset
For each machine, identify the dominant failure modes. Are they mechanical (bearing, shaft, coupling)? Electrical (winding, rotor, supply)? Operational (cavitation, surge, throttling)? The failure modes determine which condition monitoring technique is appropriate, and therefore which data you need.
Understand the P-F interval
The P-F interval is the time between a detectable potential failure and a functional failure. If the P-F interval for your dominant failure mode is measured in days, continuous monitoring is needed. If it is weeks or months, periodic data collection may be entirely sufficient.
Match your technique to the failure mode
Vibration analysis is excellent for mechanical faults in accessible machines. MBVI is better suited to inaccessible or hazardous locations and is essential when electrical or operational failure modes are in scope. Thermography is valuable for electrical panels and equipment with significant thermal signatures. No single technique covers every situation.
Establish a baseline
Condition monitoring produces its most actionable results when you have a reference point. A baseline spectrum or waveform captured from a healthy machine provides the comparison against which deterioration is measured. Many sites skip this step; those that do it correctly see faster, more confident diagnosis.
Define alert thresholds and escalation paths before you start collecting data
Data without a process for acting on it is just storage cost. Before installation, agree on who receives alerts, what diagnostic process follows an alert, and what the escalation path is if a machine needs to come out of service.
Not sure which technique fits your equipment?
Faraday Predictive offers on-site condition assessments that give you a snapshot of your equipment’s health and a clear recommendation on the right monitoring approach, before you commit to any system.
Data you probably do not need (at least not yet)
A common mistake is to over-specify the data infrastructure at the start of a programme. This creates cost, complexity, and often a mountain of data that nobody has the bandwidth to interpret. The following are frequently cited as requirements, but are rarely necessary for a first deployment.
- High-frequency continuous streaming from every asset. Most failure modes have P-F intervals long enough that periodic measurements, whether weekly snapshots with a portable instrument or measurements every few minutes from a fixed system, are more than adequate. Storing raw high-frequency waveforms from every motor on a large site is expensive and usually unnecessary.
- Process historian integration from day one. Correlating condition data with process variables is genuinely useful, but it is an optimisation step, not a prerequisite. Start with condition data alone and add process context once you have established the monitoring workflow.
- Enterprise IoT platforms and data lakes. Many sites have successfully deployed predictive maintenance using purpose-built condition monitoring hardware with its own software, entirely independently of broader IT infrastructure. Integrating with enterprise systems can come later, once the programme is generating clear value.
The role of electrical supply data
One data category that is systematically underrated is the quality of the electrical supply to your motors. Most sites have no visibility of frequency, phase balance, harmonic distortion, or power factor at the individual machine level, let alone the shape of voltage and current waveforms.
Electrical supply issues are a significant cause of both motor and driven-equipment failures, and they are often misdiagnosed as mechanical problems. A site that monitors voltage and current at the machine level can identify supply-related deterioration before it causes damage, attribute failures correctly, and often resolve problems at the supply level rather than replacing equipment unnecessarily.
Read: Electrical Supply Issues and What They Reveal
Putting it together: minimum viable PdM data set
For a plant running motor-driven rotating equipment (pumps, fans, compressors, conveyors), the minimum viable data set for a credible predictive maintenance programme is:
- Voltage and current waveforms at the machine level, captured at sufficient frequency to resolve the spectral features associated with your failure modes, and with sufficient baseline history to detect trend changes.
- Operational context, even basic information such as whether the machine was running, at what load, and under what process conditions, to avoid false alerts from legitimate process variation.
- A history of past failures for each asset, to calibrate alert thresholds and understand which failure modes are historically dominant.
That is it. A system that captures this data reliably, stores it accessibly, and presents it to engineers in an interpretable form is sufficient to run an effective CBM programme on the majority of motor-driven assets.
Further reading from Faraday Predictive
Summary
Predictive maintenance does not require a vast data infrastructure. It requires the right data for the failure modes that threaten your most critical assets, collected at an appropriate frequency, and reviewed by engineers who have a clear process for acting on what they find.
For motor-driven rotating equipment, measuring voltage and current at the machine level gives you simultaneous visibility of mechanical, electrical, and operational conditions. That single data source, properly analysed, covers the failure modes that account for the vast majority of unplanned downtime on most industrial sites.
Start with your most critical assets. Establish baselines. Define your escalation process. Then expand the programme as you build confidence and demonstrate value.
Ready to see what your equipment is telling you?
Start with a site assessment, not a system
Faraday Predictive can carry out an on-site condition assessment of a selection of your motor-driven equipment using our portable P100 system. You get a clear report on asset health and actionable maintenance recommendations before committing to any installed solution.