Die Druckversion wird nicht mehr unterstützt und kann Darstellungsfehler aufweisen. Bitte aktualisiere deine Browser-Lesezeichen und verwende stattdessen die Standard-Druckfunktion des Browsers.

Predictive Maintenance

Introduction

Predictive maintenance is a maintenance strategy that uses the actual condition of equipment, operating data, and analysis to estimate when maintenance should be carried out. Instead of waiting for a failure or replacing parts only because a fixed calendar date has arrived, you look for evidence that the condition of an asset is changing. The aim is to plan the right maintenance work at a useful time before a functional failure occurs.

This aiMOOC is designed for apprentices, trainees, and vocational students in maintenance, mechatronics, industrial mechanics, electrical engineering, automation, and related occupations. You will learn how technicians combine practical inspection skills with condition monitoring, sensors, data analysis, work orders, and safe maintenance procedures.

Predictive maintenance is not a magic prediction system. A useful program depends on good measurements, correct interpretation, knowledge of the machine, and clear action rules. A sensor can indicate an abnormal trend, but a trained person must still decide what the signal means and what work is justified.

Learning Goals

After completing this course, you should be able to explain the difference between reactive, preventive, condition-based, and predictive maintenance; identify useful machine-condition signals; describe a basic sensor-to-work-order workflow; recognize common faults from trends and inspection evidence; discuss data quality and false alarms; and propose a simple predictive-maintenance plan for a vocational workplace.

You should also be able to communicate findings clearly to operators, maintenance planners, electricians, mechanics, and supervisors. Good predictive maintenance is a team activity because the data, the machine, and the maintenance process are connected.

Why Predictive Maintenance Matters

Industrial equipment rarely fails at a convenient time. An unexpected failure can stop production, damage other components, create scrap, delay deliveries, and require urgent repairs. In some systems it can also create a safety or environmental hazard. Predictive maintenance tries to identify deterioration early enough that the work can be planned.

The main practical benefits can include fewer unplanned stops, better use of spare parts, longer useful life of components, more focused inspections, and better planning of labor. These benefits are not automatic. A poorly selected sensor, an incorrect alarm limit, or unreliable data can waste time and create false confidence.

For vocational practice, the central question is simple: What evidence tells you that this asset needs attention, and what safe action should follow?

Four Maintenance Strategies

Strategy	Typical trigger	Typical advantage	Typical limitation
Reactive maintenance	A failure has already occurred	Simple to organize for low-criticality items	Can create unplanned downtime and secondary damage
Preventive maintenance	Time, calendar date, operating hours, cycles, or planned interval	Easy to schedule and standardize	Work may be done too early or too late
Condition-based maintenance	A measured condition crosses an action limit or shows deterioration	Work is linked to actual equipment condition	Requires suitable monitoring and interpretation
Predictive maintenance	Condition data and analysis indicate a developing fault or future maintenance need	Can support earlier planning and better timing	Requires reliable data, models, expertise, and follow-up

Predictive maintenance is closely related to condition-based maintenance. The important difference is the predictive step: the system or analyst uses condition information to estimate how the asset is likely to develop, not merely whether it is currently inside or outside a limit.

From Machine to Maintenance Decision

A practical predictive-maintenance workflow connects the physical machine to a maintenance action. You can remember the chain as asset → signal → data → interpretation → decision → work order → verification.

Step One: Choose the Right Asset

Not every component needs continuous monitoring. Start with an asset where failure has meaningful consequences, where a measurable failure pattern exists, and where maintenance can be planned if you receive useful warning.

Useful selection questions include: Does the machine stop a whole process when it fails? Is the failure expensive or dangerous? Is the component difficult to replace quickly? Does the machine show a measurable change before failure? Can your team act on the warning?

A cheap, non-critical lamp may be suitable for run-to-failure maintenance. A production motor, pump, compressor, gearbox, conveyor drive, or critical fan may justify more active monitoring.

Step Two: Measure Condition Signals

Different faults create different physical effects. The job is to choose a signal that can reveal the degradation mechanism you care about.

Vibration

Vibration analysis is widely used on rotating equipment. Imbalance, misalignment, looseness, damaged gears, and bearing defects can change vibration amplitude or frequency content. An accelerometer converts mechanical motion into an electrical signal that can be stored and analyzed.

A single high reading is not always a fault. Load, speed, mounting position, nearby machines, and the measurement method can change the signal. Trends and comparable operating conditions are therefore important.

The balancing-machine diagram includes a vibration sensor and shows why vibration is directly connected to the behavior of rotating components.

Temperature and Thermography

Temperature rises can indicate friction, overload, poor electrical contact, blocked cooling, insufficient lubrication, or other abnormal conditions. Infrared thermography helps technicians compare heat patterns without touching live or moving components.

The thermogram above shows an abnormal hot area in an electrical fuse block. A thermal image can be an important warning sign, but it must be interpreted with load, ambient conditions, component type, and electrical safety requirements in mind.

Lubricating Oil

Oil analysis can reveal contamination, lubricant degradation, and wear particles. For engines, gearboxes, turbines, and hydraulic systems, a sample can provide information about what is happening inside the machine without complete disassembly.

Good sampling matters. A sample from the wrong point, a dirty container, or inconsistent timing can make the result misleading. Trend samples taken under comparable conditions are usually more useful than isolated results.

Acoustic and Ultrasonic Signals

Leaks, electrical discharge, friction, and some bearing defects can create characteristic sound patterns. Ultrasonic condition monitoring can detect frequencies above normal human hearing and can support inspection of compressed-air leaks, steam traps, bearings, and electrical systems where the method is appropriate.

Electrical and Process Signals

Motor current, voltage, power, pressure, flow, speed, torque, humidity, and production-cycle data can add context. A temperature rise during a heavy-load period may be normal, while the same temperature rise at light load may be abnormal. Predictive maintenance becomes stronger when machine-condition data is interpreted together with operating conditions.

Step Three: Establish a Healthy Baseline

A baseline describes how the asset behaves when it is known to be healthy under defined operating conditions. Without a baseline, it is difficult to decide whether a new measurement is unusual.

The baseline may include normal vibration levels, temperature ranges, motor current, pressure, cycle time, sound characteristics, or other variables. If operating conditions change, the baseline may also need to change.

An apprentice should learn to ask: Was the machine at the same speed? Was the load comparable? Was the sensor mounted in the same place? Was the measurement taken with the same method? Was maintenance recently performed? These questions often matter as much as the number on the screen.

Step Four: Clean and Prepare the Data

Real industrial data is rarely perfect. Sensors can fail, cables can be loose, wireless packets can be lost, timestamps can be wrong, and operating states can change. Before analysis, data may need to be checked for missing values, impossible values, noise, duplicated records, incorrect units, and inconsistent sampling rates.

Data preparation should never hide inconvenient evidence. If a measurement is removed, the team should know why. Traceability is important because maintenance decisions can affect safety, cost, and production.

Step Five: Detect Change and Diagnose Faults

An anomaly is a pattern that differs from expected behavior. An anomaly can be caused by a developing fault, but it can also come from a different operating condition, a sensor problem, or a process change.

Technicians and analysts may use alarm limits, trend rules, frequency analysis, statistical models, or machine-learning methods to identify unusual behavior. The next step is diagnosis: deciding which fault mechanism best explains the evidence.

For example, increasing vibration at a characteristic rotational frequency may suggest imbalance, but a diagnosis should also consider alignment, looseness, load, foundation condition, and sensor quality.

Step Six: Estimate Future Condition

Prognostics means estimating how the condition may develop in the future. Some systems estimate remaining useful life, often abbreviated as RUL. RUL is an estimate, not a guaranteed countdown.

A prediction should ideally include uncertainty. Saying “this bearing has exactly 63 hours left” creates false precision unless the evidence truly supports that accuracy. A more useful maintenance message may be: “The degradation trend is accelerating; inspect within the next planned stop and prepare a replacement bearing.”

A damaged bearing can produce vibration, sound, temperature, and lubricant evidence. Predictive maintenance is strongest when several independent signals point toward the same developing problem.

Step Seven: Turn Information into Work

A prediction only creates value when it leads to a useful maintenance action. Results should connect to a computerized maintenance management system, maintenance board, or another controlled workflow.

A good work request describes the asset, the observed condition, the trend, the suspected fault, the recommended action, the urgency, relevant safety requirements, and the evidence used. The technician who performs the work should then record what was actually found.

This feedback is essential. If the model predicted a bearing fault but the bearing was healthy, that result should be investigated. If the model gave no warning before a failure, that missed detection should also be investigated. Predictive-maintenance systems improve when predictions are compared with real inspection findings.

Data, Models, and Human Judgment

Predictive maintenance can use simple rules or advanced algorithms. A simple rule might alert when vibration increases by a defined amount from a stable baseline. A more advanced model may combine vibration, temperature, speed, and load to estimate failure risk.

Machine learning can be useful when relationships are complex and enough representative data exists. However, a complex model is not automatically better. A transparent threshold with good process knowledge may be more reliable than a sophisticated model trained on poor data.

Common Data Problems

Sensor drift means that a sensor output slowly changes even though the true condition may not have changed. Missing data can hide events. Label errors occur when historical records say a component was healthy or faulty when the real condition was different. Class imbalance occurs when normal operation is common but true failures are rare.

Failures are often rare because good plants prevent them. This creates a difficult learning problem for data-driven models. Teams may need long-term historical data, test-rig data, simulated faults, or expert rules to support model development.

False Alarms and Missed Detections

A false alarm tells the team there is a problem when the machine is actually acceptable. Too many false alarms waste labor and can make people ignore future warnings.

A missed detection is more serious in many applications: the system fails to identify a real developing fault. Alarm limits must therefore balance sensitivity with practical consequences.

The correct balance depends on asset criticality. A false alarm on a non-critical fan may be annoying. A missed detection on a safety-critical asset can have much greater consequences.

Safety in Predictive Maintenance

Condition monitoring can reduce unnecessary intrusive inspections, but it does not remove normal workplace hazards. Rotating shafts, electrical equipment, hot surfaces, pressure systems, chemicals, and moving production lines remain dangerous.

Never use predictive data as permission to bypass guarding, isolation, lockout or tagout procedures, electrical safety rules, permits, or manufacturer instructions. If a measurement requires access to a hazardous area, use the approved procedure and the correct personal protective equipment.

Remote and non-contact measurements can reduce exposure, but even a thermal camera or wireless sensor must be used within the site safety system. When a sensor reports a severe anomaly, the response should follow the defined escalation process rather than improvisation.

Vocational Case Study: Motor and Bearing

Imagine a conveyor driven by an electric motor. The motor has run reliably for months. A vibration sensor on the bearing housing now shows a slowly increasing trend. Temperature also rises slightly, while motor current remains close to its usual level.

A useful first response is not to declare that the bearing will fail immediately. Instead, check measurement quality, operating load, sensor mounting, and recent maintenance. Compare the vibration spectrum with earlier healthy data. Listen for unusual sound if safe. Review lubrication history and inspect for looseness or alignment problems.

If the evidence supports bearing deterioration, the maintenance planner can prepare the correct bearing, tools, and labor for a planned stop. After replacement, the team should verify that vibration and temperature return toward the healthy baseline. The removed bearing can be inspected to confirm the failure mode.

This final verification closes the learning loop: prediction → maintenance → physical evidence → improved future decision.

Example Application: Railway Infrastructure

Predictive maintenance is also used beyond factory machines. Rail infrastructure can be monitored using sensor data to detect changes in switches, tracks, vehicles, and other assets. The same logic applies: collect relevant condition data, detect degradation, estimate risk, plan work, and verify the result.

The image above shows a power-operated railway switch machine. The specific sensors and failure modes differ between a railway switch and a factory motor, but the maintenance reasoning is similar. This is an important transfer idea: predictive maintenance is a method, not one single device.

Measuring Whether the Program Works

A predictive-maintenance project should be judged by maintenance and operational results, not by the number of sensors installed.

Useful indicators include unplanned downtime, planned versus unplanned work, failure frequency, repeat failures, warning lead time, maintenance cost, spare-part usage, false-alarm rate, missed detections, and time from alert to completed action.

Mean time between failures can describe reliability trends for repairable assets. Mean time to repair can describe how long restoration takes. These indicators are useful only when definitions and data collection are consistent.

A successful pilot should show that the monitoring information changes decisions in a useful way. If alerts are generated but nobody acts on them, the technical system has not created an effective maintenance process.

Planning a Small Predictive-Maintenance Pilot

A vocational training project can begin with one suitable machine. Choose a clear failure mode, identify a measurable condition signal, collect healthy baseline data, define an inspection or alert rule, and connect the result to an actual maintenance response.

Keep the first pilot understandable. A small team should be able to explain why the sensor was selected, what the data means, what triggers action, who receives the alert, what safety rules apply, and how the team will confirm whether the warning was correct.

Documentation should include the asset name, sensor location, units, sampling method, normal operating conditions, baseline period, alarm rule, maintenance findings, and lessons learned. This creates a repeatable process instead of a one-time experiment.

Interactive Tasks

Quiz: Test Your Knowledge

What is the main purpose of predictive maintenance? (Plan maintenance using evidence of developing equipment condition) (!Repair every asset only after it fails) (!Replace every component at a fixed calendar interval) (!Eliminate the need for technicians)

Which maintenance strategy is normally triggered by a failure that has already happened? (Reactive maintenance) (!Predictive maintenance) (!Condition monitoring) (!Prognostics)

Which signal is especially useful for many rotating machine faults? (Vibration) (!Paint color) (!Workshop lighting) (!Inventory label)

What is a healthy baseline used for? (Comparing new measurements with known normal behavior) (!Guaranteeing that a machine will never fail) (!Replacing all future inspections) (!Setting every sensor to the same value)

What does a false alarm mean? (The system indicates a problem when the asset is acceptable) (!The system correctly detects a developing fault) (!The technician replaces the correct component) (!The machine operates at its normal baseline)

What can oil analysis help reveal? (Contamination and internal wear) (!Only the color of external paint) (!Only the age of the machine) (!Only the name of the operator)

What should happen after a predicted fault is repaired? (Verify the machine condition and record what was found) (!Delete the earlier measurement history) (!Ignore the removed component) (!Disable all future alarms)

What is remaining useful life? (An estimate of how long an asset may continue to perform acceptably) (!A guaranteed exact countdown to failure) (!The fixed age of every component) (!The time needed to purchase a sensor)

Why should operating conditions be recorded with sensor data? (They can change the meaning of a measurement) (!They make safety procedures unnecessary) (!They remove the need for units) (!They guarantee perfect model accuracy)

Which statement about machine learning is correct? (It can support predictive maintenance when suitable data and validation are available) (!It always performs better than simple engineering rules) (!It removes the need for maintenance feedback) (!It makes sensor quality unimportant)

Memory Game

Baseline	Known healthy behavior used for comparison
Accelerometer	Sensor used to measure vibration or acceleration
Thermography	Technique that visualizes surface temperature patterns
Prognostics	Estimation of how equipment condition may develop
Work Order	Authorized maintenance instruction linked to a task
Anomaly	Pattern that differs from expected behavior

Drag and Drop

Match the correct terms.	Topic
Healthy reference condition	Baseline
Unusual pattern in data	Anomaly
Estimate of future asset condition	Prognostics
Thermal pattern inspection	Thermography
Instruction for maintenance execution	Work order

...

Crossword Puzzle

Sensor	What device converts a physical condition into usable measurement data?
Vibration	What machine signal is commonly monitored on rotating equipment?
Baseline	What word describes known healthy behavior used for comparison?
Anomaly	What word means an unusual pattern that differs from expected behavior?
Prognostics	What discipline estimates how equipment condition may develop?
Thermography	What technique uses infrared temperature patterns for inspection?

LearningApps

Cloze Text

Predictive maintenance uses equipment

to support better maintenance timing. A healthy

helps you compare new measurements with normal behavior. Rotating machines are often monitored using

data. Infrared inspection can reveal abnormal

patterns. Lubricant sampling can provide evidence of contamination and internal

. An unusual data pattern is called an

. Estimating how an asset may develop in the future is called

. A remaining useful life value is an

rather than a guaranteed countdown. An alert creates value only when it leads to an appropriate maintenance

. After the repair, the team should verify the result and record the physical

.

Open-Ended Tasks

Easy

Maintenance Strategy Comparison: Choose one machine you know and explain whether reactive, preventive, condition-based, or predictive maintenance would be most suitable and why.
Sensor Walkaround: During an approved workshop walkaround, identify three machines and suggest one condition signal that could be monitored on each without entering a hazardous area.
Trend Sketch: Draw a simple graph showing a healthy baseline, a gradual degradation trend, an alert point, and a planned maintenance event, then explain the graph in your own words.
Technician Interview: Interview a technician, trainer, or supervisor about one recurring equipment fault and summarize the warning signs that usually appear before failure.

Standard

Vibration Mini Project: Use a safe training rig or provided dataset to compare vibration under two operating conditions and write a short interpretation of the differences.
Thermography Report: With approved equipment and supervision, produce a one-page thermal inspection report that records load, ambient conditions, observed hot spots, and a recommended next step.
Oil Sampling Procedure: Create an illustrated or video-based procedure that explains how to take a representative lubricant sample while avoiding contamination and following workplace safety rules.
Maintenance Work Order: Convert a fictional predictive-maintenance alert into a professional work order containing asset identification, evidence, suspected fault, urgency, safety requirements, and verification steps.

Advanced

Predictive Maintenance Pilot: Design a small pilot for one critical training asset, including failure mode, sensor, baseline, alert logic, response workflow, and success indicators.
False Alarm Investigation: Analyze a case in which an alert was triggered but no fault was found, then identify possible causes in the sensor, data, operating condition, threshold, and model.
Remaining Useful Life Study: Use a supplied degradation dataset to estimate a maintenance window, communicate uncertainty, and explain why an exact failure time cannot be guaranteed.
Cross-Trade Maintenance Project: Form a team with mechanical, electrical, automation, or IT roles and produce a short presentation or video showing how data moves from a machine sensor to a verified maintenance action.

Learning Assessment

Failure Mode Reasoning: Given a motor with rising vibration and stable current, propose at least three possible causes, identify additional evidence you would collect, and justify the safest next action.
Strategy Selection: Compare preventive and predictive maintenance for a critical gearbox and decide which combination of methods is most appropriate for a plant with limited sensor budget.
Data Quality Diagnosis: Review a dataset with missing readings, a sudden sensor offset, and changing machine load, then explain which records can be trusted and what checks are required.
Maintenance Decision Transfer: Apply the predictive-maintenance workflow to a non-factory asset such as a railway switch, HVAC fan, pump station, or vehicle and explain what must change.
Alarm Design: Propose an alarm strategy that balances false alarms and missed detections for one safety-relevant or production-critical asset and justify the trade-off.
Verification Loop: Explain how inspection of a removed component and post-repair measurements can improve future predictions, maintenance planning, and technician confidence.

Evidence of Learning

Evidence of learning should show more than vocabulary recall. Important knowledge includes maintenance strategies, condition signals, baselines, anomalies, prognostics, remaining useful life, data quality, and maintenance workflow.

Important skills include selecting a meaningful sensor, comparing measurements with operating conditions, interpreting trends, documenting uncertainty, writing a clear work request, following safety procedures, and verifying the result after maintenance.

Useful products include a condition-monitoring report, sensor-location plan, trend graph, maintenance work order, thermography or vibration study, oil-sampling procedure, pilot proposal, or short explanatory video.

Strong transfer achievement means you can take the same reasoning process to a different asset, recognize which signals and risks change, and design a new evidence-based maintenance approach instead of copying a fixed solution.

OERs on the Topic

Linked Learning Areas

Predictive Maintenance

Predictive maintenance connects mechanical engineering, electrical engineering, automation, data analysis, reliability, safety, and maintenance planning. In vocational education, the most important connection is between physical evidence from the machine and a safe, documented maintenance action.