Choose your reading route
eduKateSG · Why Mathematics? · Preventive Maintenance Optimisation: Age Replacement, Inspection Intervals and Risk
Start with the question, audit the worked example, or jump to the assumptions and practice. Every chapter stays open for uninterrupted reading.
Understand the question
Define the quantity and follow the mechanism.Route 2 · Chapters 8–14
Reproduce the example
Audit the numbers one step at a time.Route 3 · Chapters 15–21
Check assumptions
See where design and calibration matter.Route 4 · Chapters 22–28
Explore applications
Connect the method to a concrete investigation.Route 5 · Chapters 29–45
Study and verify
Practise, inspect failure cases and read further.
Full chapter index · Practice and checking · Mathematics Learning Hub
Why is mathematics important in preventive maintenance? It helps teams choose an age, inspection interval or condition trigger from evidence about failure behaviour, detection, cost, downtime and risk. The calculation supports a decision only after the task’s physical mechanism and safety constraints are clear.
Did You Know? Replacing a healthy item more often can increase faults when each intervention carries a small chance of damage or incorrect reassembly.
Full chapter index
Chapters 1–7 · Understand the question
Chapters 8–14 · Reproduce the example
Chapters 15–21 · Check assumptions
Chapters 22–28 · Explore applications
Chapters 29–45 · Study and verify
- Common misconceptions
- Connection to recent mathematics articles
- Practice questions
- Frequently asked questions
- A reporting checklist
- A final audit
- Useful next reading
- Discrete versus continuous choices
- Bundled maintenance
- Opportunistic replacement
- Imperfect inspections
- Degradation thresholds
- Seasonal access
- Learning after interventions
- Multi-objective decisions
- Verification after maintenance
- An end-to-end policy review
CHAPTER 1 OF 45 · Understand the question
1. Why preventive maintenance needs optimisation
Preventive maintenance uses inspections, servicing or planned replacement before an unwanted failure. The useful question is not whether maintenance sounds good, but which task, interval and trigger reduce a defined risk or cost. Mathematics connects lifetime distributions, renewal cycles, detection delay, task cost and consequence. It also reveals when a calendar task has no evidence of age-related benefit. A sound policy begins with failure modes and constraints, then compares feasible choices rather than declaring that more intervention must be better.
A maintenance task should prevent, detect or control a specific failure mode. Describe the physical degradation, how it becomes observable and what consequence follows if it is missed. Lubrication may address wear, an inspection may reveal a crack, and a proof test may expose a hidden protective-device failure. These mechanisms need different mathematics. Grouping them under one generic maintenance interval can hide both benefit and harm.
Under an age-replacement policy, replace at failure or planned age T, whichever occurs first. If lifetime is X, the cycle length is min(X,T). A failed cycle incurs the failure-replacement consequence; a surviving cycle incurs the planned-replacement consequence. After a genuine as-good-as-new replacement, cycles can support renewal-reward analysis. If overhaul is imperfect or the surrounding system continues ageing, the reset assumption needs a richer model.
For nonnegative lifetime X with survival function S(x), expected truncated life is E[min(X,T)], equal to the integral of S(x) from zero to T under standard conditions. This denominator is easy to overlook. Dividing expected cycle cost by T assumes every cycle reaches the planned age, which is false when failures occur earlier. The survival curve therefore affects both the expected cost and the time over which that cost is earned.
Let Cf be the consequence assigned to failure replacement and Cp the cost of planned replacement. A simple expected cycle cost is Cf P(X≤T)+Cp P(X>T), with conventions at the boundary stated. Real analyses may add downtime, inspection, environmental or safety outcomes separately. Monetary values should be dated and sourced. Consequences that cannot ethically or legally be traded for money should remain constraints rather than hidden inside an invented price.
The renewal-reward cost rate is expected cycle cost divided by expected cycle length. Suppose an illustrative policy has expected cost $300 and expected cycle length 60 days. The rate is $5 per day. This is a long-run average under regenerative assumptions, not a bill that arrives every day. Compare candidate ages with the same cost boundary, exposure unit and reset model. Keep safety and service constraints outside the cost race.
| Policy age | Expected cycle cost | Expected cycle length | Cost rate |
|---|---|---|---|
| 50 days | $180 | 42 days | $4.29/day |
| 100 days | $300 | 60 days | $5.00/day |
| 150 days | $390 | 70 days | $5.57/day |
The illustrative ages of 50, 100 and 150 days have modelled rates of $4.29, $5.00 and $5.57 per day. The smallest displayed rate is age 50 in that fictional table, but it is not automatically the recommendation. The table omits risk limits, maintenance-induced damage, scheduling capacity and parameter uncertainty. A policy is acceptable only after those checks. Numbers in a worked example teach the calculation; they are not a maintenance instruction for any real asset.
Calendar replacement is most plausible when the conditional failure tendency rises with age or when a known life limit applies. With a constant hazard, replacing a functioning item solely because it reached an arbitrary age does not make an exponential lifetime younger in the memoryless model. With a decreasing hazard, routine replacement can even remove robust survivors. Estimate age behaviour with censored data, diagnostics and engineering knowledge rather than inferring it from a bathtub-curve cliché.
An inspection policy searches for a condition or hidden failure at interval I. Shorter intervals can reduce detection delay, but add labour, interruption, false alarms and possible damage. If failure occurrence is approximately uniform between perfect inspections, expected detection delay is about I/2. That result changes when occurrence is not uniform, detection is imperfect or demand reveals the failure. State the detection model before using a half-interval rule.
Protective devices may fail silently while appearing available. A proof test can reveal the hidden state, so unavailability depends on failure rate, test interval, test coverage, repair and common causes. The interval is often constrained by safety analysis and regulation. A proof test that misses important modes does not reset the device to certainty. Record coverage and test-induced risk; do not optimise an interval from one simplified formula alone.
Condition-based policies act when a validated indicator crosses a threshold or shows a concerning trend. Vibration, temperature or wear debris can be informative only if measurement error, normal operating variation and warning lead time are understood. A threshold trades missed detections against false alarms. Historical labels can be biased because only failed or inspected units receive confirmation. Use held-out validation and preserve the engineering mechanism.
Opening, adjusting or replacing a healthy component can introduce contamination, loose connections, software errors or infant mortality. Include a probability or scenario for intervention-induced harm when evidence supports it. More frequent maintenance then has both a benefit and a cost beyond labour. The point is not to avoid necessary work; it is to choose tasks that demonstrably control the target mode and to improve procedures, tools and verification.
NASA’s Reliability-Centered Maintenance guide frames maintenance as an integrated choice among proactive and reactive approaches based on function, failure and consequence. It cautions that time-directed work needs evidence of an age-reliability relationship. Read the NASA RCM Guide. The source supports a disciplined method; a local policy still requires asset data, qualified engineering review and applicable rules.
| Policy | Evidence needed | Main limit |
|---|---|---|
| Fixed age | age-related failure pattern | can cause unnecessary intervention |
| Inspection | detectable degradation and detection performance | hidden faults between inspections |
| Condition based | validated indicator and warning lead time | sensor and threshold error |
| Run to failure | acceptable consequence and restoration plan | unplanned downtime |
The same component may need different policies under different duty cycles, environments, redundancy and consequences. Calendar time, operating hours, starts and load cycles are different exposure clocks. Translate between them only with evidence. A policy copied from a hotter, busier or safety-critical setting may waste resources or under-control risk. Record the exposure that actually drives degradation and the service calendar that determines access.
Build histories with installation, operating exposure, inspection findings, failure mode, intervention, restored configuration and observation end. Keep censored units that have not failed. Separate planned discoveries from functional failures and distinguish active work from logistics delay. Missing inspections and changed thresholds can bias apparent performance. A decision log should explain why a task was added, changed or retired.
Units still operating at the cutoff are right-censored, not proven immortal. Analysing only failed components shortens estimated life and can favour over-maintenance. Conversely, examining only survivors can hide early problems. Fit lifetime or degradation models that use both event and censored exposure. Record retirement for unrelated reasons separately because removal may be informative. A clean censoring flag is one of the most valuable columns in the data.
A component can fail through wear, corrosion, random shock or an external system problem. One replacement interval cannot be justified by pooling modes with different mechanisms. A task that prevents wear may do nothing for shocks. Use cause-specific evidence and consider competing-risk methods when the first observed mode removes the item from observation. Preserve mode uncertainty instead of forcing every event into a convenient category.
Preventive work occupies time even when it avoids future breakdowns. Compare planned and unplanned downtime, staffing windows and user demand. A policy with slightly higher direct cost may meet a service constraint better, while a cheap policy may create unacceptable outage clustering. Availability analysis links failure and restoration; renewal-reward analysis provides long-run rates. Choose the objective before calculating an optimum.
For safety, environmental or legal requirements, define the feasible region first. A cost minimum outside that region is irrelevant. Risk estimates need consequence, exposure and control effectiveness, not only failure frequency. Document mandatory inspections, qualified-person requirements and escalation rules. Optimisation can then compare feasible alternatives. Mathematics clarifies trade-offs; it does not authorise relaxing a safety barrier.
Lifetime parameters, costs, detection probabilities and repair times are estimated. Propagate uncertainty through each candidate policy with bootstrap, Bayesian simulation or scenario analysis suited to the evidence. Show whether the preferred interval changes across plausible inputs. A flat cost curve suggests a robust range may be wiser than a falsely precise day. Highlight parameters that drive the decision so future data collection has a purpose.
Simulation can combine random failures, inspections, imperfect detection, repair queues, spares and seasonal access. Validate it against simple age-replacement results before using richer logic. Use documented seeds and enough replications to estimate Monte Carlo error. Simulation explores consequences under a model; it does not validate the degradation mechanism. Keep event traces so reviewers can see why downtime and cost arose.
A mathematically short interval can overload technicians, compress planning and increase procedural errors. Access may require shutdowns, permits or specialist equipment. Include resource constraints and task duration. Design improvements such as better access, modular replacement or built-in tests can dominate interval tuning. Maintenance is performed by people in a system, so a policy should support clear instructions, verification and stop-work authority.
Premature replacement consumes materials and creates waste, while delayed work can cause inefficient operation or damaging failure. Measure environmental outcomes separately when monetary conversion would obscure them. Repairability, refurbishment and condition monitoring can change the feasible policy set. A lifecycle view prevents optimisation from shifting cost or harm to another stage. State which outcomes the model includes and which require stakeholder judgement.
Use simulated lifetimes rather than real safety equipment. Draw lifetimes from two distributions with the same mean but different ageing, then compare replacement ages. Calculate failure probability, expected cycle length and illustrative cost rate. Add a small intervention-damage probability and observe whether the preferred age changes. The exercise shows why the shape of uncertainty matters, and why a modelled optimum depends on assumptions rather than appearing from arithmetic alone.
Place candidate ages in rows. For each, compute survival probability, expected truncated life, expected cost and long-run rate. Add columns for downtime, detection delay and constraint status. Keep parameter inputs and sources on a separate sheet. Chart the rate across ages and show a sensitivity band. Label illustrative costs and never hide mandatory limits inside a formula. Recalculate with at least one alternative lifetime model.
Represent policies as explicit event rules and write unit tests for failure before age, survival to age and boundary cases. Fit censored lifetimes, simulate cycles and compare Monte Carlo rates with analytic results. Log seeds, data versions and parameter draws. Return a feasible set and uncertainty distribution rather than one unsupported interval. Review code with a maintenance engineer who can spot a physically impossible reset or task sequence.
Maintenance optimisation supports transport, manufacturing, buildings, utilities, computing infrastructure and healthcare technology. Roles combine statistics, operations research, engineering, data systems and field knowledge. Mathematics supports informed career pathways but does not guarantee salary or access. Students can explore probability and simulation now, then consider engineering, analytics, polytechnic or university routes according to interests and current admissions information.
Start with one question: what failure mode does this task control? Students should draw the cycle, label the exposure clock and reproduce a cost-rate calculation before using software. Parents can encourage them to challenge a universal interval and ask about censored survivors. Use simulated or non-critical examples. The transferable benefit is learning to combine mathematical evidence with mechanism, constraints and uncertainty.
More maintenance is not always safer. An average lifetime is not a replacement date. A low cost rate is not automatically acceptable. The shortest inspection interval does not guarantee perfect detection. A bathtub curve should not be assumed without data. Censored survivors are not failures or immortal units. Condition monitoring requires validated indicators. A modelled optimum is not universal across environments and consequences.
The four most recent companions are Renewal Theory, Maintainability Engineering, Availability Engineering and Reliability Allocation. Each answers a neighbouring question, so use the one whose mechanism matches the decision. The Mathematics Learning Hub connects these ideas to probability, statistics, algebra, computing and responsible modelling.
1. Define an age-replacement cycle. 2. Explain E[min(X,T)]. 3. Calculate $300/60 days. 4. State when I/2 is a detection-delay approximation. 5. Give one maintenance-induced failure. 6. Separate a cost objective from a risk constraint. 7. Explain censoring. 8. Name two exposure clocks. 9. Design a sensitivity analysis. 10. Explain why condition thresholds need validation.
### Is the cheapest interval always best? No. It must satisfy safety, service and regulatory constraints.
### Does preventive maintenance always improve reliability? No. The task must address a real failure mechanism and can introduce harm.
### Why use renewal reward? It compares expected outcome per expected cycle time.
### Can a schedule be copied between assets? Only after checking mechanism, exposure, environment and consequence.
Define asset function, failure mode and policy. State exposure clock, lifetime evidence, censoring and reset assumption. Give cost and downtime boundaries, safety constraints, detection performance and intervention risk. Show candidate intervals, uncertainty and sensitivity. Record resource feasibility, applicable rules and decision ownership. Link NASA guidance accurately. Label examples and avoid converting a model into an unauthorised maintenance instruction.
Recalculate expected costs and lengths independently. Confirm every probability lies between zero and one and every rate has units. Check boundary ages and missing survivors. Inspect whether a task can physically prevent or detect the named mode. Verify that all safety requirements remain constraints. Test the simulation against an analytic case. Remove universal claims and ensure the illustrative table cannot be mistaken for a real schedule.
Read the NASA Reliability-Centered Maintenance Guide and the NIST reliability handbook sections on lifetime distributions and censored data. Continue with renewal theory for repeated cycles, maintainability for repair time and availability for user-facing downtime. The Mathematics Learning Hub connects survival functions, expected values, optimisation and simulation.
Some intervals are chosen from permitted service windows rather than every possible day. Optimise over feasible discrete choices after incorporating access, staffing and bundled work. A continuous calculus optimum may fall between shutdown opportunities and be unusable. Compare neighbouring feasible intervals and show how much the objective changes. A broad near-optimal region can support a practical schedule with less disruption than an exact but fragile mathematical minimum.
Several tasks may share one access window, reducing setup downtime but causing some items to be serviced early. Model common setup cost and task-specific ageing. Bundling can also increase the duration and complexity of a shutdown. Evaluate groups rather than adding independent optimum intervals that cannot coexist. Record which task drives the window and whether deferment rules preserve safety. Coordination is an optimisation variable, not an afterthought.
When a system is already open for another repair, replacing a nearby ageing item can save future access cost. The choice depends on conditional remaining life, added intervention risk and inventory. A simple age threshold may be used, but it should be derived from the opportunity state. Do not replace every accessible component automatically. The opportunity changes cost and downtime, not the underlying evidence about the component’s failure mechanism.
Inspection sensitivity and specificity can depend on defect size, operator, tool and environment. Repeated inspections are not independent when the same blind spot persists. Include probability of detection as a curve where evidence allows, rather than one perfect-detection assumption. False positives create unnecessary intervention; false negatives extend hidden exposure. Validation samples should represent the field condition. Calibration alone does not prove inspectors can detect the target defect.
A degradation path can be modelled as a stochastic process with measurement error. A maintenance threshold must leave enough lead time for planning before functional failure. Too conservative a threshold creates early work; too late a threshold raises risk. Estimate crossing distributions with censored paths and account for irregular measurement times. Review whether the indicator remains meaningful after software, sensor or operating changes.
Weather, school calendars, production peaks or regulatory windows can restrict maintenance. A nominal six-month interval may actually produce variable exposure between visits. Simulate calendar and operating clocks together and define rules for early or late windows. Seasonal demand can make the consequence of downtime asymmetric. A policy that is acceptable in a quiet period may be unacceptable during peak service, even with the same average failure probability.
Maintenance records can reveal which tasks found actionable conditions and which created repeat work. Track findings per inspection, post-maintenance failures and time to recurrence. Declining findings can mean improvement or an ineffective task applied to a stable population. Use a control group or phased change where feasible. Retire tasks only through authorised review, with monitoring for unintended consequences. Evidence should update the policy instead of freezing the first schedule forever.
Cost, downtime, risk, waste and workforce burden may conflict. Show a Pareto set of policies that are not dominated across the chosen outcomes. Stakeholders can then select among transparent trade-offs. Converting every outcome to dollars can obscure values and uncertainty. A multi-objective view also reveals when one policy is worse on every measure and can be removed without debate. Document the final decision rule separately from the mathematical frontier.
A task is incomplete until the required function and configuration are verified. Add verification time and the chance of detecting installation error to the cycle. Functional tests may themselves impose load or unavailable time. Record failed verification and rework rather than counting only successful closures. This feedback is essential for estimating maintenance-induced risk and improving procedures. A short active task with frequent rework is not truly efficient.
Define the function and target mode, fit a censored lifetime or degradation model, and specify candidate age, inspection or condition policies. Calculate expected cycle length and outcomes, enforce safety and service constraints, and propagate uncertainty. Add resources, induced faults and seasonal windows. Compare the feasible set and document why one policy was chosen. Establish monitoring and a review date. This complete chain prevents an elegant formula from becoming an unsupported work instruction.
Extended practice laboratory
Practice case 1: Map a cycle
Setup: Replace at failure or age 100 days.
Task: Define cycle length as min(X,100).
Checking target: Keep failure and planned-replacement costs separate.
Practice case 2: Calculate a simple rate
Setup: Use expected cycle cost $300 and expected cycle length 60 days.
Task: Divide cost by time.
Checking target: Obtain $5 per day as an illustrative long-run rate.
Practice case 3: Compare intervals
Setup: Evaluate planned ages 50, 100 and 150 days.
Task: Calculate cost and downtime rates.
Checking target: Do not choose from cost alone when risk constraints apply.
Practice case 4: Check ageing
Setup: Plot hazard or conditional failure probability by age.
Task: Look for an increasing pattern.
Checking target: Calendar PM needs evidence of age-related failure.
Practice case 5: Model inspection detection
Setup: Use a fault that can remain hidden between inspections.
Task: Relate interval length to expected detection delay.
Checking target: Shorter intervals reduce delay but add inspection burden.
Practice case 6: Include maintenance-induced failure
Setup: Assign a small probability of damage per intervention.
Task: Add it to scenario analysis.
Checking target: More maintenance is not automatically safer.
Practice case 7: Separate safety constraints
Setup: Set a maximum acceptable risk before optimisation.
Task: Minimise cost only inside the feasible set.
Checking target: Do not trade away mandatory controls.
Practice case 8: Use condition data
Setup: Define a measurable degradation indicator.
Task: Compare threshold and fixed-interval policies.
Checking target: Validate the indicator and lead time.
Practice case 9: Audit task effectiveness
Setup: Link each PM action to a failure mode it can prevent or detect.
Task: Remove unsupported ritual tasks carefully.
Checking target: Preserve regulatory and safety requirements.
Practice case 10: Write the conclusion
Setup: State failure behaviour, policy, costs and constraints.
Task: Report sensitivity and evidence gaps.
Checking target: Avoid presenting a modelled optimum as a universal schedule.