Predictive Maintenance: Signals Your PLC and VFD Give Before Failure

Unplanned downtime on a production line rarely stays silent. Long before a PLC trips or a VFD alerts a fault, both leave tattle-tale bits of information behind — scan-time drift, sagging capacitor voltage, creeping drive heatsink temperature, fault-counter ticks. Predictive maintenance for PLC and drive hardware is mainly about learning how to read this information early enough to schedule a part swap on your schedule, not the production line's. This guide covers the nine warning signals PLCs and VFDs show before they fail, how to set thresholds that don't cry wolf, and how to assemble a monitoring stack without over-investing on instrumentation you do not need.

Quick Specs: PLC & VFD Predictive Maintenance

• PLC core info signals: scan time, battery voltage, I/O fault counters, comm packet retries
• VFD core info signals: DC bus voltage, IGBT/heatsink temps, output current imbalance, fault history
• Threshold levels: Info / Warn / Alarm / Trip (4-level classification)
• Common protocols: Modbus TCP, EtherNet/IP, PROFINET, OPC UA
• Monitoring brands: Allen-Bradley, Siemens, Mitsubishi, Schneider, ABB, Omron, Yaskawa
• Investment range: under $5,000 (basic retrofit) → $50,000+ (cloud + AI analytics)

What Predictive Maintenance Means for PLCs and VFDs

Predictive maintenance (PdM) is the proactive tech practice of acting on equipment telemetry to flow maintenance schedules pre-failure, rather than post-breakdown (reactive) or pre- calendar (calendar-based maintenance). For PLCs and drive hardware, the data collection step is most of the way already done within the controller- diagnostic registers, fault counters, drive parameter history. Reading this variable info consistently is the real challenge, normalizing it to historical norms, then turning minor deviations into work orders. A well-planned predictive maintenance programme treats every PLC and VFD as its own self-instrumenting asset and pulls its existing telemetry into an analytics layer to run parallel with, and not interfere with, the control system.

Cost benefit analysis now recommends predictive over reactive maintenance for most asset-heavy operations. Industry data put together by Aberdeen Research and cited in the Siemens' True Cost of Downtime 2024 report puts the average cost of unplanned downtime at $260,000 per hour while in operation across manufacturing operations; automotive plants average $2.3M/hr. Plants with mature condition monitoring reduce unplanned downtime breakdowns by about 70-75%. Worldwide, the predictive maintenance market expanded from $11.82B in 2025 to $15.29B in 2026 according to The Business Research Co company's 2026 industry report, a CAGR about 28.6% - rapid enough to cause vendor solutions stacks to change every 12-18 months.

Few manufacturing plants choose between strategies they combine them by asset need: run-to-failure for cheap redundant pumps, clockwork-based preventive maintenance for common parts, and predictive scans for any controller or VFD whose failure halts a line.

💡 Pro Tip: If a single hour of downtime on a given asset costs more than $5,000, predictive monitoring usually pays for itself within the first prevented failure.

The 9 Early Warning Signals PLCs and VFDs Give Before Failure

Much of the PLC and drive predictive maintenance coverage takes the sensors as the beginning - bolt on a vibration sensors, watch the trend. That perspective is helpful but not complete. Both the PLC and the VFD surface internal data that accurately predicts their failure long before any external sensors shows them as a symptom. Here are nine signals organized by source: four from the PLC, four from the VFD, one from the environment they occupy.

What are the most overlooked early warning signs of PLC failure?

Operators typically watch for I/O alarms and outright faults, but the most useful PLC signals are subtle trend changes: scan time rising over weeks, communication retry rates climbing slowly, or battery voltage drifting toward the brand-specific replacement threshold. None trip a fault on their own — taken together, they predict a service call. A peer-reviewed analysis in MDPI Actuators (2025) on time drift in programmable logic controllers documents how cycle-time anomalies correlate with deeper system stress, validating scan-time monitoring as a real PdM signal.

PLC-side signals (4):

1. Scan-time drift. For a modern PLC the nominal average scan time should be fairly fixed to within a few percent. A creeping trend of 15% above the last 30 day average, could be a symptom of logic bloat, memory fragmentation, or general communications overhead, all of which precede a CPU fault.
2. Battery voltage decline. plcs memory backed-up have volatile retention when battery level gets down near the OEM-specific threshold. Siemens fault code F0645 triggers when the battery's down around 2.8 VDC and many other PLCs' warnings occur around 2.5V. Industry vendor data sets report battery-related culprits for roughly 90% of memory back-up failures, in which case it is the most profitable single check in any monitoring routine.
3. I/O module fault counter creep. Most models of plcs expose individual error counters by circuit board. A counter that creeps slowly week over week predicts module failure far enough in advance to swap during a missed outage rather than a mid-shift.
4. Communication packet loss / retry rate. TCP and EtherNet/IP packet retry counter trending higher indicate a performance drag from cabling, switch, or device aging - frequently the least expensive issue to address and the most costly to ignore.

VFD-side signals (4):

5. DC bus voltage integrity. Diminishing electrolytic life appears as rising ripple-voltage and bus voltage dips during deceleration. Overall difference matters more than any specific instant - a functioning VFD maintains a stable bus voltage; a failing one loses that stability over the course of months.
6. Heatsink temperature creep. An increase of 5-8°C above nominal at the same load point over time indicates fan failure, dust buildup, or IGBT degradation. Absolute value matters less than the trend, since identical drives can run as much as a hundred degrees apart in different enclosure.
7. Output current asymmetry. Phase-to-phase imbalance over approximately 5% suggests motor bearing degradation, winding insulation breakdown, or VFD output-stage stress. Drives already measure this — most operators simply never look at the parameter.
8. Switching-frequency variation. Sometimes IGBT gate driver problems manifest initially as audible tone variation or a drifting in the drive reported carrier frequency, several days before a fault.

Cross-system signal (1):

9. Cabinet enclosure temperature trend. HVAC degradation and power-supply heat builds up, and all the components below them are forced to conduct. A cabinet that gets a little warmer over a quarter can indicate issues worth a service call before the controller or drive exhibits its own symptoms.

📐 Engineering Note: Polling rate matters. Sub-second polling for scan time and bus voltage; minute-rate polling for battery voltage and fault counters; per-shift sampling for cabinet temperature. Polling everything at one-second granularity will itself drift PLC scan time — a self-defeating monitor.

⚠️ Important: Do not chase nine signals on every asset. Pick three based on what fails most often in your inventory. Repair-shop refurbishment patterns for VFDs typically focus on fans, electrolytic capacitors, and thermal paste — three components tied directly to the heatsink-temperature signal. That tells you where the highest yield is for monitoring effort.

How PLCs Capture Predictive Maintenance Data

How does a PLC work for predictive maintenance?

A programmable logic controller continuously runs an input scan, program scan, output scan, and housekeeping cycle, and on every cycle it has the chance to log diagnostic data alongside its control logic. Modern PLCs expose dozens of diagnostic registers (scan time min/avg/max, battery state, I/O health, communication statistics) without any additional sensors. Controllers don't become smarter for predictive maintenance — operators become more attentive to what the controller already publishes.

Not every PLC is equal in what it offers or how easily it can be read. Three capability tiers cover most installed control systems today.

Data flow for any tier looks similar: sensor or internal register → PLC tag → SCADA or OPC UA broker → time-series database → dashboard or alert engine. Interfaces between the controller and the analytics layer (HMIs, SCADA platforms) determine how quickly maintenance teams see deviations. A common rookie mistake is buying new sensors before exhausting what the PLC already publishes — start with a list of every available diagnostic tag, then decide what to add.

For factories running older controllers, a deeper grounding in how to read these registers helps: our PLC troubleshooting guide walks through I/O module faults, watchdog timeouts, and other patterns that show up before catastrophic failure.

VFD-Specific Diagnostics: Reading the Drive's Built-In Telemetry

VFDs live in a bit of an odd space within most monitoring cases: they are critical assets, subject to failure in predictable ways, and they already publish much of the data needed to predict their own failure, yet most plants only read VFD parameters during fault-clear, not continuous telemetry. End result: every drive failure becomes a surprise, no matter how much warning the drive is sending for months.

Six categories of diagnostic data are available on just about every modern variable frequency drive without adding any external instrumentation:

• DC bus voltage (real-time value plus min/max history) — the single best indicator of capacitor health and supply quality.
• Heatsink and IGBT temperature (current value plus lifetime peaks) — the earliest signal for cooling-system degradation.
• Output current per phase (RMS plus peaks) — reveals motor and drive stress before either trips an overload.
• Fault history register (last 5-20 faults with timestamps) — the cheapest predictive analytic in industrial automation, because the drive already keeps the log.
• Run-time meter and start counter — usage tracking that lets you compare actual run time against the manufacturer's expected life span.
• Capacitor health estimate (newer drives only, vendor-specific) — some manufacturers publish a derived health percentage based on bus ripple analysis.

Economics here are bizarre: without doing an additional thing, a standard VFD is already aggregating six different streams, and its firmware's fault-history register persists through power cycles. Simply exporting these as a SCADA tag every couple minutes costs nothing in hardware, effectively nothing in PLC scan time overhead. According to industry-repair-shops monitoring cross-brand inventory, the failure-relevant elements - fans, electrolytic capacitors, thermal paste- all map to one or two of these parameters. Drives tell you which; most plants just haven't been paying attention.

For firms purchasing across brands, the hard reality is that parameter name and P-code conventions differ radically for Allen-Bradley, Siemens, Yaskawa, Mitsubishi, ABB etc.- but the six fundamental categories are the same across the board. Cross-vendor VFD inventories take a lookup table to normalize parameter names, then the analytics layer sees them all as common signals.

Setting Failure Thresholds Without Causing False Alarms

The primary reason that predictive maintenance programs fall flat on their face within one year is threshold settings. Alarms may trigger too often (gotta be careful with alarm fatigue-if maintenance views them as a nuisance they're all ignored), or too infrequently (errors will happen anyway and we get no value). Better practice: a tiered framework that isolates "log for background" from "notify an operator right now".

The 2025 NIST systematic review of condition monitoring-based technologies in industrial maintenance by Dadfarnia, Sharp, and Herrmann (Journal of Manufacturing Systems) explicitly identified a lack of standardization in evaluation methods and threshold-setting by industry. In other words: there is no industry-mandated threshold table you can copy. A four-tier framework based on statistical deviation from a rolling baseline fills that gap practically:

Why the baseline beats the value: two identical VFDs in two different cabinets in the identical facility can operate 8C apart simple by going at different locations, weather, and panel arrangements. An OEM-suggested 70C heatsink threshold could be a full 50C of margin on one drive or maybe a single digit percent margin on the other. Having a 30 day moving baseline makes that fluctuation into relevant local information-each motor learns its own family deviation, and only its own deviation is significant.

📐 Engineering Note: Use sigma multipliers as a starting framework, then tune within the first quarter. Sites running heavy seasonal loads (HVAC plants, food processing) often need a seasonal baseline rather than a single rolling 30-day window. This threshold framework is a tool, not a recipe.

Building Your PdM Stack: Sensors, PLC, and Analytics Integration

What hardware and software does a PdM program actually need? It depends on budget, criticality, and how much of the work the existing PLCs and VFDs can do unassisted. Three investment tiers cover most factory installations, and the right choice is rarely the most expensive one.

The Standard tier is where most factories should land for the first 12 months. Starter is fine if the existing PLCs are modern and SCADA is already in place — the data is being collected; nothing new is needed except the analytics. Advanced is appropriate when downtime cost crosses roughly $100,000 per hour, or when a regulatory or quality regime demands automated audit trails. For motors specifically, vibration analysis remains the highest-yield external signal once the drive's built-in telemetry is being captured. Cross-brand sourcing for sensors and gateways is straightforward; selecting motor and driver hardware is well covered in our guide to servo motor and driver brands.

One related purchasing variable is power supply specification. A predictive maintenance stack drawing telemetry from dozens of devices needs reliable 24 VDC for sensors, gateways, and panel-mount analytics — a topic we cover in detail in our 24VDC power supply sizing guide. The wider context of SCADA, PLC, and HMI integration is covered in our industrial automation and control systems overview.

Common Pitfalls That Make Predictive Maintenance Fail

Predictive maintenance programs fail more often than they succeed in their first year. The reasons are not technological — the data is there and the analytics work — but organizational and methodological. These five patterns account for the majority of industrial PdM failures, compiled from field reports and post-mortems.

1. Drowning in data, starving for action. Plants stand up dashboards full of trends and alerts but never define which alerts trigger which work orders. Without a closed loop between alarm and action, the data becomes wallpaper.
2. Skipping baseline measurement. Generic manufacturer-suggested thresholds applied to specific assets produce false positives. Within weeks, maintenance teams stop trusting the alerts. Spend the first 30 days measuring, not alerting.
3. Buying sensors before reading existing data. Modern PLCs and VFDs already publish 60% of the useful signals. New vibration sensors are valuable, but they are an addition to PLC and drive telemetry, not a substitute for it.
4. No spare parts plan tied to alerts. A "warning" alert is useless if the spare board has a 12-week lead time. Predictive maintenance only pays off when it shortens repair time — which means a parts strategy in lockstep with the alerting strategy. For hard-to-find or discontinued PLC and drive models, having a sourcing partner ready to handle a quote request often determines whether a "warning" tier alert turns into a planned swap or an unplanned outage.
5. Ignoring the cabinet itself. Plants monitor motors and drives but not the panel that holds them. HVAC degradation, dust ingress, and thermal stratification together accelerate every component below them. The cabinet is the ninth signal for a reason.

💡 Pro Tip: Field practitioners commonly report that the maintenance teams who succeed with PdM are the ones who treat the first six months as a learning period — gathering data, tuning thresholds, building trust — before promising specific downtime reduction numbers to plant management.

2026 Outlook: AI, Edge Computing, and Where Predictive Maintenance Is Heading

What are the benefits of AI-driven predictive maintenance?

AI-driven predictive maintenance pairs traditional threshold logic with machine learning models that detect cross-signal anomalies — patterns no single sensor would flag on its own. The 2025 NCBI peer-reviewed paper on Artificial Intelligence of Things for next-generation predictive maintenance describes how this combination raises detection accuracy beyond what threshold-only systems achieve, particularly for slow-developing failures across coupled equipment. Trade-off: AI-driven systems require more historical data to train and more discipline to interpret — they do not replace the threshold-tier framework above; they augment it.

Three specific shifts are reshaping predictive maintenance entering 2026.

Cloud-to-edge migration is now the default architecture. Cloud-based analytics platforms dominated 2020-2023 PdM rollouts. By 2026, edge-computed analytics — running on the controller itself or on a panel-mounted edge box — are becoming the default for new deployments. Latency, data sovereignty, and bandwidth cost all push the same direction. Rockwell's GuardianAI, Siemens Senseye, AutomationDirect's BRX cloud-enabled PLCs, and Banner Engineering's wireless platforms all reflect this architectural shift, and they market themselves explicitly as IIoT-ready stacks rather than pure cloud subscriptions.

Condition-based maintenance is overtaking interval-based PM. Multiple market reports (The Business Research Company 2026, Fortune Business Insights, IMARC Group) project the predictive maintenance segment growing at 21-29% CAGR through 2030. Growth is being driven primarily by a shift away from calendar-based preventive maintenance, not from reactive maintenance — meaning displacement is happening among already-mature maintenance programs.

Standardization is finally on the horizon. The 2025 NIST review explicitly called out the lack of standardized evaluation methods for condition monitoring as a research priority. Expect the next 24-36 months to produce reference frameworks from NIST and IEEE that look much like the threshold-tier table above — at which point early adopters will already be aligned by accident.

"The introduction of predictive maintenance does not phase out human expertise from the maintenance function — it places the maintenance technician as a data-informed decision maker rather than a reactive responder."

— Adapted from Siemens, "5 Common Misconceptions Surrounding Predictive Maintenance" (2023, public Siemens blog)

What to do in the next 12 months: If you are running modern PLCs and modern drives but still scheduling maintenance on a calendar, start by reading what those controllers already publish before you spend a dollar on new sensors. If you are running legacy controllers, the cost-effective move is a Standard-tier retrofit (gateway plus a small set of wireless sensors) rather than a controller upgrade.

Frequently Asked Questions

Q: What is the difference between predictive and preventive maintenance for PLCs?

Preventive maintenance for PLCs runs on a fixed calendar — replace the backup battery every 24 months, swap the cooling fan every 5 years, regardless of condition. Predictive maintenance reads the controller's actual data — battery voltage, scan time, fault counters — and acts when the data deviates from baseline. Predictive avoids replacing parts that are still healthy and catches anomalies that the calendar would miss.

Q: Can older PLCs do predictive maintenance, or do I need new hardware?

Most legacy PLCs (PLC-5, SLC 500, Mitsubishi A-series, Siemens S5) can support a basic PdM program through external SCADA polling — you read what they expose, and trend it externally. The richer telemetry on modern controllers makes things easier and cheaper, but it is not a prerequisite. A retrofit gateway and a time-series database are usually a better first investment than a controller upgrade, unless the legacy hardware is approaching end-of-support for other reasons.

Q: What's the most common early warning sign before a VFD fails?

Heatsink and IGBT temperature creep tied to fan or cooling-system degradation is the highest-yield single signal. It precedes capacitor degradation by months in most cases, and it shows up clearly in the drive's existing temperature parameter without any external sensor. Repair-shop refurbishment patterns for failed VFDs almost always begin with cooling-related parts: fans, thermal paste, and electrolytic capacitors.

Q: How much does it cost to add condition monitoring to a VFD?

If the drive is modern and a SCADA or OPC UA layer already exists, the marginal cost is near zero — the drive parameters are already there, and pulling them into a dashboard is a configuration task. Adding wireless vibration sensors on the connected motor for cross-validation typically runs $300-$1,500 per asset depending on the sensor and the gateway. Cloud analytics subscriptions add $50-$300 per asset per year for commercial platforms.

Q: Do I need AI to do predictive maintenance, or are simple thresholds enough?

For most factories, threshold-based monitoring against a rolling baseline catches 70-80% of the value. Machine learning adds value primarily for cross-signal anomaly detection — failures that show up only as a coordinated drift across several signals at once. Start with thresholds, get the data discipline right, and add ML on top after 12-18 months of clean baseline data.

Q: Can I retrofit predictive maintenance onto an existing Allen-Bradley, Siemens, or Mitsubishi PLC?

Yes. Allen-Bradley CompactLogix and ControlLogix expose extensive diagnostic registers via EtherNet/IP. Siemens S7-1200 and S7-1500 publish similar data via PROFINET and the TIA Portal diagnostic buffers. Mitsubishi FX5U and Q-series PLCs offer comparable telemetry. The retrofit work is mostly on the analytics side — gateway, broker, dashboard — not the controller itself. For ladder-only legacy units, a minor ladder addition can copy diagnostic words to readable tags before the gateway picks them up.

Source Replacement Drives, PLCs, and HMI Components

A predictive maintenance program eventually generates a "warning" or "alarm" tier alert that requires a part. Sourcing replacement PLCs, drives, sensors, and HMIs across legacy and modern brands — including discontinued and refurbished units — is where most plants discover their parts strategy was the weakest link. Browse PLC and VFD parts inventory for in-stock units, or contact us for industrial electronics repair service when the cheapest path forward is restoring an existing unit rather than buying a replacement.

Explore PLC and VFD inventory →

Our Perspective

This guide reflects what cross-brand industrial parts distributors and repair shops see every week: PLCs and VFDs that arrive for refurbishment with the same handful of failure patterns, repeating across Allen-Bradley, Siemens, Mitsubishi, Schneider, ABB, and Omron inventory. The nine-signal taxonomy and the four-tier threshold framework are how we organize what those patterns are telling us. The market and downtime data come from third-party industry reports cited in the references; the practitioner observations come from years of handling the same parts on the bench.

Related Articles

• Introduction to PLC Troubleshooting — the reactive companion to this predictive guide, covering how to diagnose I/O module faults and watchdog timeouts when an alert escalates.
• Top 8 Servo Motor and Driver Brands Worldwide — when the predictive maintenance program flags a motor and the spare needs sourcing.
• 24VDC Power Supply Sizing: Load Calculation, Derating, and Safety Margins — a healthy PdM stack starts with reliable panel power; this is the load-budget primer.
• Industrial Automation and Control Systems — context on PLC, SCADA, and HMI integration for plants designing their first end-to-end PdM architecture.

References & Sources

1. Systematic Review of Condition Monitoring-Based Technologies in Industrial Maintenance — Dadfarnia M., Sharp M., Herrmann J., Journal of Manufacturing Systems, July 2025 (National Institute of Standards and Technology)
2. Analysis of Time Drift and Real-Time Challenges in Programmable Logic Controllers — Actuators, MDPI 2025 (peer-reviewed)
3. Artificial Intelligence of Things for Next-Generation Predictive Maintenance — PMC NCBI (U.S. National Library of Medicine), 2025
4. Predictive Maintenance Global Market Report 2026 — The Business Research Company
5. The Real Cost of Unplanned Downtime in Manufacturing (2026 Data) — Reliamag, citing Aberdeen Research and Siemens True Cost of Downtime 2024
6. 5 Common Misconceptions Surrounding Predictive Maintenance — Siemens Blog, 2023