Maintenance Bad Actor Analysis Tool

Identify recurring losses. Correct the failure mechanism. Verify the result.

A maintenance bad actor is an asset, subsystem or failure mode that repeatedly consumes disproportionate maintenance effort or disrupts the compressed-air service that production needs. Identify it by combining verified failure events, operating exposure, downtime, repair cost, production consequence and recurrence after corrective action—not by counting work orders alone.

This page explains how to turn maintenance history into an actionable bad-actor register, with calculation rules, reusable worksheet fields and worked examples. It covers compressors, motors, drives, cooling systems, dryers, filters, drains, valves and their shared utilities. The starting point is the required service: adequate air flow, pressure, quality and availability at the users that depend on it. [1, 2]

1. Purpose and engineering decision

The practical question is not simply “Which compressor failed most often?” It is “Which recurring defect should we address next, why does it matter, and what evidence will show that the action worked?” A frequently used machine can accumulate more failures than a lightly used machine while having a lower occurrence rate. Conversely, a cheap drain can create a serious treatment-system problem without appearing near the top of a maintenance-spend report.

A serious loss of a critical function requires the site's incident response regardless of its position in a monthly ranking. Among non-emergency defects, compare the burden and evidence to select the next investigation. Only then decide whether the appropriate remedy is maintenance, operating practice, spares, system modification or replacement.

The output is a justified action, not a label attached permanently to a machine. Keep people, suppliers and equipment brands out of blame-based rankings; record what failed and what the evidence establishes.

2. The dashboard and its underlying records

A credible dashboard must allow the user to move from an aggregate result back to the event records that produced it. Retain the original work-order references, maintenance notes and measured values. Do not replace uncertain source entries with guessed data merely to fill a chart.

Record cost, exposure and consequence completeness separately. A dashboard can have excellent failure timestamps and poor cost records; a single “data quality: good” badge would conceal that difference.

3. Define the analysis unit and failure event

Choose asset, subsystem, component or failure-mode analysis before calculating. Compressor C-03, its cooling subsystem and its cooling-fan motor are different levels of the same hierarchy. Show roll-ups, but do not add parent and child costs together as though they were independent expenditure.

For this worksheet, a functional failure is a verified inability to meet the defined duty, or a verified loss of a required standby or protective function. An alarm, inspection observation or planned service is not automatically a failure. A defect found during preventive maintenance may reveal a genuine hidden failure; retain that failure event while keeping the planned work order in its own maintenance category. [2]

Several work orders may concern one breakdown: diagnosis, parts procurement and final repair. Link them to one event ID. Conversely, a new loss after documented restoration is a new event even when the symptom repeats. Do not merge events using an arbitrary time gap alone.

When a shared electrical or cooling disturbance affects several assets, retain each asset's history and a common incident ID. Calculate plant-level service loss from the common timeline, not by summing every affected machine's outage.

4. Core metrics and their exact meaning

Use a fixed observation window with a stated time zone. Let H be verified operating hours and N the number of qualifying failures whose onset falls within that window. For N greater than zero, the descriptive operating-time MTBF estimate is H/N. It is not the predicted date of the next failure. Interpreting it as a constant-rate reliability parameter requires an appropriate model and comparable operating conditions. [5]

The familiar simplified expression MTBF/(MTBF + MTTR) represents an alternating operating/repair model under consistent assumptions. It must not be presented as observed operational availability when the repair term excludes parts waiting, access delay, planned unavailability or other relevant losses. Report the directly measured required-service ratio separately.

Include failures beginning in the window in N even when still open at the cutoff. Include their measured downtime up to the cutoff, but exclude incomplete durations from completed-repair means and display the open-event count. Events already open at the window start contribute in-window downtime, not a new onset. This prevents unresolved outages disappearing from the analysis.

5. Operating exposure, standby demands and uncertainty

For repairable equipment, “failure occurrence rate” or “rate of occurrence of failures” is more precise than treating the result as a non-repairable component's hazard rate. The selected denominator must be visible beside the result. [4]

Running hours suit some comparisons; starts, load cycles or valid standby demands suit others. Show failed valid demands divided by total valid demands as a separate demand-based measure. Do not blend test demands and actual production demands without identifying them. A failed standby start can be consequential despite almost no accumulated running time.

Compare like duties: base-load versus base-load, similar pressure conditions, similar environment and comparable event definitions. Pool event counts and exposure before calculating a group rate; do not take an unweighted average of machine rates. Flag missing counter history, meter resets, major configuration changes and partial monitoring coverage.

With zero failures and positive exposure, display “0 observed events in H hours.” Do not display a proven infinite MTBF or a zero-risk conclusion. With zero or unknown exposure, suppress exposure-based metrics rather than dividing by zero or quietly assigning a favourable score.

6. Downtime burden: separate the machine from the air service

Asset downtime measures when the asset could not meet its defined duty. Air-service loss measures when the required pressure, flow, quality or backup function was not available at the agreed boundary. They are not interchangeable. Standby transfer can preserve production during a compressor outage while temporarily reducing resilience.

Calculate each asset's downtime from the union of its unavailable intervals. For example, two records covering 08:00–12:00 and 10:00–14:00 represent six hours of unavailability, not eight. Retain both source records and explain the overlap resolution.

Use mutually exclusive timeline categories: diagnosis, active repair, parts waiting, specialist waiting, access waiting, and restart/verification. When tasks occur in parallel, assign the elapsed interval to the documented dominant constraint or a declared allocation; do not count it twice. Keep labour-person-hours separately because several technicians can work during one hour of elapsed downtime.

7. Maintenance cost and operational consequence

Direct corrective cost comprises attributable parts, labour and external service, plus separately described direct costs. Assign each transaction a unique reference. Check whether an external invoice already includes parts or labour before adding internal entries. Record currency, treatment of credits and the analysis boundary consistently.

Keep planned maintenance expenditure visible but separate from corrective-failure cost. Open-event accrued cost, final invoiced cost and estimates should remain distinguishable. Missing invoices mean “cost incomplete,” not “zero cost.”

Show lost output, rejected product, recovery effort and verified energy penalties in separate fields. Any conversion to money needs an explicit basis and an approved plant estimate. Lost sales revenue is not automatically the economic loss, and restart hours may overlap already-recorded service loss.

8. Priority selection and optional weighted scores

The recommended default is separate rankings for occurrence rate, downtime, corrective cost and consequence. Show critical-function exceptions above the ranked list. A rare but serious event must not be hidden by numerous inexpensive routine defects.

A composite score is optional, not an industry-standard risk calculation. Where the plant elects to use one, document each input's meaning, scale, direction and weight. Do not directly add hours, rupees, counts and consequence classes. Avoid counting the same effect twice, such as including production-loss cost and an identical consequence score as independent dimensions.

One transparent design for a non-negative quantitative input x is s = 100 × min(1, x/T), where T is a positive, plant-approved reference value. An overall score can then be the sum of w × s, with non-negative weights summing to one. Consequence needs its own approved anchored rubric; an ordinal category is not automatically a numerical interval scale.

Show thresholds and uncapped input values. Missing inputs suppress the composite result rather than silently becoming zero. A score expresses a chosen prioritisation policy, not failure probability, remaining life or engineering approval. Changing the weights must not remove immediate consequence-based escalation.

9. Pareto analysis that can be reproduced

A Pareto view orders category contributions from largest to smallest and may show their cumulative percentage. Decide the metric, period and grouping before comparing results. [6]

For non-negative contributions x, share equals 100 × x divided by the total; cumulative share is the running sum after sorting. Use a stable secondary sort, such as asset ID, for ties. When the total is zero, show “no recorded burden in this view” rather than invalid percentages. Investigate negative net amounts before using them in a cost Pareto.

Make the leading bars traceable to their events. Separate charts can reveal a dryer dominating unavailable hours, a compressor dominating repair cost and a drain dominating quality interruptions. The data need not follow an exact 80/20 split. Concentration is an observation, not a rule that forces a predetermined number of machines into the bad-actor list.

10. Failure-mode recurrence without double counting

Standardise failure-mode descriptions while retaining original technician notes. “High temperature,” “hot running” and a controller overtemperature code may describe one symptom family; none alone proves the same underlying cause. Keep symptom, mode and cause in separate fields.

Use an explicit repeat rule. For a simple in-window measure, the first qualifying event in each asset-and-mode group is the reference event; later events after restoration are repeats. Repeat-event share equals repeat events divided by all qualifying events, multiplied by 100. Three restored-and-recurred events of one mode plus one unrelated event produce two repeats in four events: 50%.

This is different from the selected mode's share of all failures, which would be 75% in that example. Neither measure proves poor repair quality. Link known pre-window history separately, and label whether the calculation uses in-window recurrence or an expanded lookback.

11. Root-cause confidence and action status

Keep two fields rather than one misleading progress bar. Cause confidence records unknown, suspected or confirmed. Action status records not assigned, planned, implemented, under verification, effective within the agreed scope, or reopened. A corrective action may be implemented before the cause is fully established; the dashboard must not convert implementation into confirmation.

Confirmation requires evidence explaining the failure mechanism and considering plausible alternatives. Effectiveness requires post-action evidence, appropriate duty exposure and satisfaction of the stated acceptance criteria. This distinction follows the investigation-to-corrective-action-to-effectiveness sequence in the supporting guidance. [7]

Record who made each finding, when, and what evidence supports it. A new validated occurrence of the same mode should flag the linked action for review and reopen it where the event falls within its defined scope. Do not silently erase the previous closure or presume the cause is identical without review.

12. Symptoms are investigation starting points

These are hypotheses, not remote diagnoses. Preserve controller history, as-found photographs, failed parts where appropriate and recent maintenance records. Intrusive tests, electrical work and pressure-system interventions require the applicable site controls and competent personnel. Do not bypass protective settings to keep a ranked asset operating. [1, 2]

13. Condition trends and the missing operating context

Trend pressure, temperature, differential pressure, dew point, flow, power, operating state, starts and alarms where suitable instrumentation exists. Record location, units, timestamps and instrument validity. Compare like-for-like states: a lower temperature after repair during a cooler, lightly loaded shift does not by itself demonstrate improved cooling.

Link the failure timeline to recent changes: element replacement, lubricant change, ventilation alteration, controller adjustment, production expansion or upstream utility interruption. Ask which measurement would distinguish competing explanations instead of collecting every available signal without a question.

Use existing service and installation records first. Compare the current arrangement with the documented ventilation, access, pipework and equipment configuration. A change in the surrounding installation can affect the same machine without an internal component defect. Preserve that context alongside the repair history. [3]

14. Worked case A — raw failure count can mislead

Illustrative data, not a Reddy Compressors customer result. Assume both machines use the same qualifying failure definition and their operating exposures are complete.

Compressor A leads the raw count, but B has the higher observed occurrence rate: 2.00/0.75 = 2.67 times A's rate. The pooled descriptive fleet rate is 1,000 × 10/10,000 = 1.00 event per 1,000 hours, not the unweighted average of 0.75 and 2.00.

B deserves investigation on this metric, not automatic condemnation. Check duty, failure modes and the small event counts before inferring a persistent reliability difference. A's total operational consequence could still be greater. The table establishes arithmetic, not statistical superiority or a replacement recommendation.

15. Worked case B — downtime, repair time and spares differ

Illustrative data. Each asset is required for 2,000 hours. The stated downtime occurs within that required window; all events are restored, non-overlapping and fully recorded.

D-01 dominates this asset-downtime Pareto despite fewer failures. Its parts-delay share is 100 × 20/30 = 66.67%. Reducing two hours of hands-on repair is not necessarily the most valuable intervention when parts availability dominates the outage.

The 40-hour total is suitable for an asset-burden chart. It is not automatically 40 hours of production stoppage; confirm whether the outages overlapped or whether redundancy protected the process.

16. Worked case C — a low-cost drain can have high consequence

Illustrative scenario. A drain experiences three separate, restored-and-recurred failures of the same mode. Parts expenditure totals INR 5,000, but the plant records eight hours of treatment-related service loss. At an assumed production rate of 60 units per hour, the interrupted-output estimate is 480 units.

That output estimate does not establish monetary loss or prove the drain alone caused every interruption. The investigation must substantiate the treatment and process timeline. It must also determine whether the repeated events relate to contamination, unsuitable application, installation, control or another cause.

With no other failures in the chosen window, two of the three events are repeats: 66.67%. Show the quality consequence independently of the inexpensive part. Replacing the drain again without checking the mechanism and downstream recovery may close a work order while leaving the bad actor unresolved.

17. Worked case D — verify the action at representative exposure

Illustrative scenario. C-04 had four overtemperature failures during 2,000 operating hours. Investigation documents cooler contamination and hot-air recirculation. After appropriate cleaning and air-path correction, matched-load observations meet the agreed thermal criteria and no qualifying trip is recorded over 1,000 further operating hours.

Record the changed configuration, acceptance limits, ambient and load conditions, post-action exposure and monitoring coverage. The operational statement is “acceptance criteria met during the recorded verification window,” not “the defect can never recur.”

Under a constant-rate Poisson model, with complete observation over a fixed post-action window, zero failures give a one-sided 95% upper occurrence-rate bound of −ln(0.05)/H. At H = 1,000 hours this is 0.002996 per hour, or 2.996 events per 1,000 hours. The equivalent lower MTBF bound is 333.81 hours. [5]

The pre-action observed rate was 2.00 events per 1,000 hours. This limited failure-free follow-up alone therefore does not establish, at that confidence level, a post-action rate below the historical point estimate. Engineering acceptance and statistical evidence answer different questions; both must state their limitations.

18. Bad actor or unsuitable maintenance strategy?

Do not assume the machine is inherently defective. A pattern can arise from ineffective cleaning, an unsuitable task interval, incorrect component identification, repeated control changes, inaccessible equipment or a shared environmental problem. Compare the failure mechanism with the work actually being performed—not merely the percentage of scheduled jobs marked complete. [2]

If several similar machines fail after the same maintenance intervention, review common parts, methods and configuration changes. If failures cluster during peak ambient conditions, compare cooling and duty evidence. If one compressor looks poor only because it carries almost all production, examine load allocation before recommending a rebuild.

The remedy should follow the demonstrated mechanism. More maintenance is not automatically better maintenance; unnecessary intrusive work can introduce new defects. Equipment-specific procedures and service requirements remain the governing inputs.

19. Repair quality and a fair repeat-repair measure

For a repeat-repair indicator, define an index repair, the same-mode rule and a follow-up horizon in operating hours, starts or calendar time. Count each eligible index repair once, even if it produces several later work orders.

For a fully observed cohort, repeat-repair percentage equals 100 × eligible index repairs followed by a qualifying recurrence divided by all eligible index repairs. Use only index repairs whose entire prescribed follow-up can be observed with adequate records; show pending follow-ups separately. Early recurrences in pending cases still trigger investigation but must not be mixed into a supposedly complete cohort rate.

For example, three recurring cases among ten fully followed index repairs produce 30%. Two other repairs awaiting sufficient follow-up are shown as pending, not counted as successful. The indicator signals where to investigate diagnosis, installation, parts or operating conditions; it is not proof of workmanship failure or supplier fault.

21. Reusable maintenance bad-actor worksheet

Complete one assessment per asset or asset-and-mode combination. Maintain the event ledger beneath it; this summary is not a substitute for source records. Empty entries mean “not supplied,” not zero.

Close the assessment with a short decision statement: “Investigate [asset/mode] because [metric and consequence], using [evidence]. Complete [action] by [owner/date]. Accept effectiveness only when [criteria and exposure] are satisfied.”

22. Calculation and data-quality test cases

These are explicit expected behaviours for the worksheet and future application, not a claim that a website implementation has passed testing.

23. Decision output and practical next step

The result should identify the asset and failure mode, the observed burden, the comparison basis, the consequence and the proposed response. Include the exact observation window, raw numerator and denominator, data gaps, owner and verification criteria.

Where the cause remains uncertain, the next action may be an evidence-gathering task rather than a repair order.

Primary enquiry: Review My Recurring Compressor Failures. Supply asset identification, failure history, running hours or demand counts, outage timelines, work orders and available condition trends. State any current critical service loss first. The review scope and response arrangements depend on the actual equipment, site information and agreed service; this page makes no guaranteed response-time or remote-diagnosis promise.