// ref: MAINT-CASEBOOK-09 // classification: practitioner-knowledge // status: not in any manual
Failure Patterns That
Don't Appear in Textbooks
// the ones you learn from the machine, not the curriculum
The textbooks cover the classic failure modes. Fatigue. Corrosion. Overload. Wear. They are correct as far as they go. They don't go far enough. This is a collection of the patterns that only appear in a plant — compound, contextual, intermittent, and quietly repeating until someone finally notices the shape.
Photo: Unsplash — Maintenance practice
The textbooks are written by people who know failure modes. They know them as categories — fatigue fracture looks like this, fretting corrosion leaves this signature, galvanic attack progresses in this sequence. The categories are real and useful. They give you a language for what you're looking at. But in twenty-plus years on the floor, the failures that were hardest to diagnose, most expensive to ignore, and most instructive after the fact were almost never the clean, single-mode failures the textbook describes.
They were compound. They were context-dependent. They were intermittent in a way that made them nearly impossible to catch with scheduled inspection. They happened because of something that changed — in the load cycle, in the environment, in the maintenance sequence, in the operating practices — and they appeared as classical failure signatures that turned out to be masking something entirely different underneath.
These are field notes, not a literature review. Each pattern described here is representative of something encountered across steel plant and crane operations — observed, investigated, and eventually understood. Some were found early. Some were not. The names are general; the patterns are genuine.
// Pattern Index — Field Cases Documented Below
The Thermally-Triggered Intermittent
// fault present, untraceable, then gone — until afternoon
The Corrected-Into-Failure
// the maintenance that caused the breakdown
Cascade Through an Unexpected Path
// secondary failure in a component that looked unrelated
The Load-Cycle Mismatch
// equipment designed for one duty, running another
Normalised Abnormal
// the fault everyone knows about and nobody has fixed
The Survivor Failure
// correct diagnosis, wrong equipment — the one that kept running
Sequential Competency Erosion
// each technician did it correctly — the sequence did not
The Slow Drift
// parameter moving 0.3% per month — invisible until it isn't
The crane had a fault that appeared at roughly 14:00 every afternoon and cleared by 06:00 the following morning. It had been recurring for three weeks. The electrical team had changed the drive card. Changed the encoder. Changed the feedback potentiometer. Cleared the fault codes. Run diagnostics. Every time: no fault found. Every afternoon: fault present.
What the textbook describes as "intermittent electrical fault" is a category that covers an enormous range of physical causes. The diagnostic approach it suggests — measure the circuit parameters, check insulation resistance, inspect connections — is correct but incomplete. It doesn't account for the fact that the circuit's physical characteristics change with temperature, and that the ambient temperature in a steel plant bay at 14:00 in the summer is not the same as at 06:00.
The actual fault was a hairline crack in a PCB trace on the control card — invisible to visual inspection, passing continuity and resistance tests at room temperature. When the card reached operating temperature after four to five hours of running in a bay ambient of 47°C, the trace expanded differentially and opened. When it cooled overnight, it closed again. The fault was present only in the thermally-expanded state.
Intermittent faults — check connections, check insulation resistance, replace suspect components, run diagnostic cycle.
If the fault is time-of-day dependent, check the ambient temperature profile. Thermally-triggered failures cannot be found at room temperature.
// Signal Patterns — Normal vs Thermally-Triggered vs Intermittent
// illustrative signal pattern representations — not scope captures
This is the most uncomfortable pattern to document because it implicates maintenance action as the proximate cause of the failure. It is also the most underreported, because nobody wants to write up a breakdown caused by their own team's work. But it is real, it is repeatable, and understanding it is essential for anyone responsible for maintenance quality.
A hoist brake was due for adjustment. The brake clearance had increased beyond the upper limit — measured at 0.6 mm against a specification of 0.2–0.4 mm. The technician adjusted it correctly, torqued the fasteners correctly, and tested the brake. Brake tested satisfactory. Job closed. Four days later, the brake ran hot and the friction material overheated.
What had happened: the brake drum surface had developed a slight taper over time — barely perceptible to touch, not caught by the visual inspection during the job. The adjustment brought the brake pads into contact with the tapered surface, creating uneven pressure distribution across the pad face. Under light use, it was adequate. Under the load profile of a full production shift, the high-pressure zone on the pad overheated, the friction material began to glaze, and braking performance degraded. The brake was technically "correctly adjusted" and it failed in service within a week.
Adjust brake clearance to specification. Test brake function. Close job.
Check drum surface condition and geometry before adjusting clearance. A correct adjustment on a worn surface produces incorrect contact.
// Field Pattern Recognition
"If the same maintenance action keeps having to be repeated at shortening intervals, the action is treating a symptom. The underlying cause is not in the component — it's in the conditions the component is operating in."
// maintenance casebook — observed pattern in bearing and brake maintenance cycles
Cascade failures are in the textbooks. The textbook version is orderly: Component A fails, which loads Component B beyond its rating, which then fails in a predictable sequence. What the textbook doesn't capture is the cascade that travels through an unexpected pathway — through a connection that wasn't part of the original design intent, through a mounting arrangement that shared structure between two supposedly independent systems, through an earthing path that wasn't supposed to carry load current.
We had a crane long-travel drive that developed an insulation breakdown to chassis on the drive's input stage. The fault current found an earth path — not through the designed protective earth, but through the crane runway rail, via the rail clips, to the building structural steel, and back through the building electrical earthing system. The drive tripped on overcurrent. The investigation focused on the drive. Nobody initially followed the fault current path.
What the unexpected path had done in transit: the rail earth fault current had flowed through a rail joint that was not welded — it was bolted, with a fishplate. The current flow through the bolted joint caused arc erosion of the contact surfaces over several weeks, which progressively increased the joint resistance, which produced local heating at the joint, which eventually caused differential thermal expansion of the rail sections at the joint, which affected crane travel — the long-travel motor began showing intermittent overcurrent faults that were misattributed to the motor itself.
Fault current follows the path of least resistance. Protective earth provides the designed fault return path.
Fault current also follows unintended paths through structural members, rail systems, and shared metalwork. These paths can cause secondary damage remote from the original fault.
A Real Compound Failure — Traced Backwards
This is the sequence of a compound failure reconstructed through a thorough investigation. At first presentation, it appeared as a hoist motor overcurrent failure. It was not. Each step below represents a real link in the chain — traced backwards from the consequence to the origin.
// origin
Bay ambient temperature rose significantly over two months as summer arrived and the arc furnace campaign intensified. Hoist gearbox sump temperature increased from typical 58°C to sustained 74°C.
// developing condition
Gearbox lubricant viscosity at 74°C was below specification for the ISO VG 320 oil grade — viscosity reduces with temperature. The lubricant film on gear tooth flanks became marginal under peak load conditions.
// damage initiation
Micropitting began on the gear tooth flanks — a form of surface fatigue that occurs under boundary lubrication conditions. The micropitting produced fine metallic debris in the oil, detectable only by oil analysis — which was not performed at this facility.
// secondary effect
Metallic debris from micropitting contaminated the gearbox bearing lubricant. The output shaft bearing began to run with abrasive contamination in the grease film. Bearing noise increased marginally — noted by the crane operator as "slightly different" but not escalated.
// failure event
Output shaft bearing failed — spalled race, seized bearing. This increased the load on the hoist motor, which tripped on overcurrent. The breakdown was recorded as "hoist motor overcurrent trip — cause unknown."
// investigation
Motor replaced. Bearing replaced. Gearbox oil changed (incidentally, during the bearing replacement). Root cause recorded as "bearing failure." The micropitting on the gear flanks was not identified. Six months later: same failure sequence, different bearing.
Equipment is designed and rated for a specific duty. The nameplate says M5 duty — moderate duty, typically 40–50% load spectrum, 150–300 starts per hour. The crane it was installed on was an M5 machine when it was commissioned. The process it served changed. The frequency of lifts increased. The loads became heavier and more consistent. The duty, in practice, became M6 or M7. Nobody updated the classification. Nobody re-specified the equipment. Nobody reviewed the original design assumptions against the new operating reality.
This pattern is common in production environments that evolve gradually. Individual changes in operating practice are each too small to trigger a formal design review — but their cumulative effect substantially changes the loading environment. The equipment continues to be maintained against its original duty classification — lubrication intervals, inspection frequencies, component life calculations — all based on M5 assumptions applied to an M7 reality.
The symptoms of load-cycle mismatch are characteristically time-compressed: component lives achieved in practice are shorter than specified, but the shortening is attributed to poor maintenance execution or component quality rather than to the fundamental mismatch between design duty and actual duty. The real diagnosis requires measuring what the crane is actually doing — load spectrum measurement, start-stop frequency monitoring, thermal duty cycle assessment — and comparing it to the original design assumptions.
// Field Pattern Recognition
"The component that fails at 60% of rated life is not always a poor-quality component. It is sometimes a correctly-rated component operating in conditions it was not rated for."
// load-cycle analysis — steel plant crane fleet audit observations
Every plant has a list of things that are wrong and have been wrong for so long that they are no longer treated as wrong. A crane that always sounds slightly rougher on the East runway than on the West runway. A panel that always runs hotter than the adjacent panel. A motor that always draws 5–8% more current than its nameplate rating. These are not acceptable; they have simply been accepted — because they have always been like this, because nothing has catastrophically failed yet, and because investigating them would require stopping production.
The normalised abnormal is dangerous not primarily because of what it is, but because of what it does to the diagnostic reference. When the rough sound on the East runway is normal, the technician calibrates their baseline against it. If the sound worsens — genuinely worsens, as a developing fault — the reference point against which worsening is measured is already elevated. The fault signature is measured against abnormal-normal, not against true normal. The margin for detection narrows substantially.
The rougher runway had caused progressive wear to the crane bridge buffer stops, which was creating slight lateral oscillation of the bridge on acceleration and deceleration. Over time, this was producing fatigue loading on the bridge end carriage welds that had not been designed for lateral cyclic loading. The visual inspection of the welds had been routine for three years. Nobody had connected the sound to the weld condition.
Inspect welds on a scheduled basis. Replace end-of-life components based on condition assessment.
The inspection baseline is corrupted when abnormal conditions are normalised. Return abnormal conditions to normal before accepting them as the baseline. Otherwise you are measuring change against the wrong zero.
This pattern is counterintuitive and requires a brief explanation of how it works. A facility has a fleet of identical cranes — eight units of the same specification, installed at the same time, running similar duty cycles. One of them fails at an earlier-than-expected point in its service life. The failure is correctly diagnosed. The component is replaced. The remaining seven are inspected for the same failure mode. Three of them show early signs of the same condition. They are corrected. The other four show nothing. They continue in service.
Five years later, one of the "clean" four fails — unexpectedly, at a point significantly earlier than the three that had shown early signs and been corrected. The investigation is puzzled. The inspection five years ago was clean. What happened?
What happened is survivorship bias applied to maintenance. The three that showed early signs were found because the inspection was looking for them. The four that showed nothing may have had sub-surface defects that were not detectable by the inspection method used — surface visual or basic NDE — and had been silently growing while the attention was focused on the identified cases. The "survivor" that failed was not clean five years ago. It was undetected.
Each individual maintenance action was executed correctly by a competent technician. The sequence of actions, applied by successive technicians over time, produced an outcome that none of them individually intended or would have predicted. This is the failure pattern that makes individual blame-attribution useless — because nobody did anything wrong, and the result was still a failure.
A VFD drive unit on a hoist motor had been modified three times over eight years. The first modification changed the motor feedback arrangement when the original encoder type became obsolete. The second modification extended the motor cable run when the control panel was relocated for electrical room reconfiguration. The third modification updated the braking resistor specification when the original units were discontinued. Each modification was individually correct, implemented by different technicians, documented in isolation.
The combination of the extended cable run (which changed the cable capacitance affecting VFD switching characteristics), the updated feedback arrangement (which had slightly different response timing), and the new braking resistor (which had slightly different impedance characteristics) produced an interaction that caused the drive to exhibit regenerative energy surges that the original installation had been immune to. The result was intermittent DC bus overvoltage trips that no individual change — taken in isolation — would have caused.
The slow drift is perhaps the hardest failure pattern to catch, because at any individual inspection, everything appears within specification. The parameter drifts slowly enough that periodic inspection captures it within limits — until a measurement that is still technically within limits conceals a trend line that, extended forward three months, terminates in failure.
A crane runway wheel had been wearing at approximately 0.3 mm of flange wear per month. The inspection interval was six weeks — nominally long enough to catch significant change, but short enough that each six-week increment of 0.45 mm looked unremarkable against the wear limit of 3 mm. At month eight, the remaining flange thickness was 1.2 mm — still within the limit. At month eleven, the flange thickness was 0.3 mm. The crane derailed. The last inspection, six weeks before derailment, had recorded "acceptable — within limits." It was within limits. The trend was not.
The fix is not a shorter inspection interval. The fix is trend-based assessment rather than point-in-time assessment. Every measurement is data. The rate of change is the diagnostic variable — more important than the absolute value in most slow-drift failure modes. A parameter at 40% of limit and drifting at 8% per month is more urgent than a parameter at 65% of limit and stable.
Inspect at scheduled intervals. Replace when limit is reached.
Record every measurement. Plot the trend. Calculate the date the parameter will reach its limit at current rate. Schedule intervention before that date — not at it.
// Field Pattern Recognition
"A measurement within limits is not the same as a measurement that is safe. The trend is the diagnosis. The number is just the evidence."
// wear monitoring — crane runway wheel and rope inspection practice
What These Patterns Have in Common
Looking across the eight cases documented above, some common threads emerge that don't appear in any individual textbook chapter but that experienced practitioners recognise as the real terrain of industrial maintenance diagnosis.
First: context dependency. Most of these failures cannot be understood without understanding the operating context — the time of day, the season, the history of modifications, the evolution of the production load, the accumulated effect of previous maintenance actions. The textbook describes the failure mode in isolation. The failure on the floor happens in a specific context that modifies its behaviour, timing, and detection characteristics.
Second: temporal depth. Many of the most instructive failures had origins that were weeks, months, or years before the failure event. The slow drift began accumulating long before it crossed any threshold. The load-cycle mismatch developed as process demands changed gradually. The compound gearbox failure started with a seasonal temperature change. Inspection systems that capture point-in-time snapshots miss the temporal dimension — which is where many of these patterns live.
Third: the mismatch between designed and actual. The textbook describes the failure mode of a nominal component in a nominal environment. Real failures often occur at the boundary between the designed and the actual — where the specification meets the real operating condition, where the designed current path meets the unintended structural path, where the rated duty meets the actual duty. These boundaries are exactly where the textbook offers the least guidance and where field experience offers the most.
Fourth, and most important: these patterns are recognisable once seen. The thermally-triggered intermittent, the normalised abnormal, the slow drift — once you have seen each of these once and understood what produced it, you recognise the pattern signature when you encounter it again. This is why field experience compounds in a way that textbook knowledge doesn't. The textbook gives you the vocabulary. The floor gives you the grammar that combines vocabulary into sentences the machine is actually speaking.
Write down what you see. Not just the failure — the conditions, the history, the context. The failures that are hardest to diagnose the first time are the most valuable to document, because someone on a different plant, in a different year, will encounter the same pattern and will search for anything that looks familiar. Field notes become someone else's early warning.
Sources & References
- Moubray, J. (1997). Reliability-Centred Maintenance. 2nd ed. Butterworth-Heinemann. [Failure mode taxonomy and P-F interval concepts]
- Smith, A.M. & Hinchcliffe, G.R. (2004). RCM — Gateway to World Class Maintenance. Elsevier. [Compound failure mode analysis]
- Neville W. Sackett & Associates. (2005). Failure Analysis of Engineering Structures. ASM International. [Case-based failure analysis methodology]
- Wulpi, D.J. (1999). Understanding How Components Fail. 2nd ed. ASM International. [Surface fatigue, micropitting, fretting corrosion failure modes]
- ISO 10816-3:2009. Mechanical Vibration — Evaluation of Machine Vibration. ISO. [Vibration baseline and trend assessment]
- ISO 13379-1:2012. Condition Monitoring and Diagnostics — Data Interpretation. ISO. [Trend-based vs threshold-based diagnosis]
- FEM (Fédération Européenne de la Manutention). FEM 1.001 — Rules for the Design of Hoisting Appliances. FEM. [Crane duty classification and load spectra — M-class system]
- Bureau of Indian Standards. IS 3177:1999 — Code of Practice for Electric Overhead Travelling Cranes. BIS, New Delhi.
- Reason, J. (1990). Human Error. Cambridge University Press. [Latent failures — sequential competency erosion model context]
- Dekker, S. (2006). The Field Guide to Understanding Human Error. Ashgate. [Hindsight bias in failure investigation]
- Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. [Survivorship bias and normalisation of abnormal — cognitive basis]
- Health and Safety Executive (UK). HSG176: The Storage of Flammable Liquids in Tanks & HSG85: Electricity at Work. HSE. [Fault current path — unintended current paths in structural metalwork]
No comments:
Post a Comment