Predictive Maintenance Technology Case Studies
Industrial5 Case StudiesUpdated Sep 2026
Predictive maintenance sits where sensor data, plant economics, and technician judgment meet. The five cases below come from different corners of industrial operations: a pump fleet with repeated bearing failures, a filling line that lost throughput, a review of monitoring hardware on pressurized equipment, a late delivery that threatened a seasonal commitment, and an energy program at a packaging plant. Each case is written around the same four questions. What were the business goals, what evidence existed, which constraints bound the response, and who carried the consequences. The organizations and people are fictional, but the decisions are the kind that come up repeatedly on real floors. Read them as working notes rather than templates, since the right call often depended on details that do not transfer cleanly.
Case Study 1
Bearing failures on a cooling water pump fleet
Calder Vale Water Treatment, a regional utility operating twelve high-lift pumps across two intake stations
The situation
Calder Vale runs twelve high-lift pumps that move raw water from an intake station to the treatment plant. Over the past several months, the drive-end bearings on four units failed well before the replacement interval printed in the maintenance manual. Two failures showed little warning in the vibration trend, and one pump seized during a peak demand window in the middle of the summer. The operations group had fitted accelerometers and temperature sensors across the fleet a few seasons earlier, so there was data, but it had been collected for general health checks rather than for bearing defect frequencies.
The challenge
Priya, the reliability engineer, had to decide whether the failures pointed to a bad bearing batch, a design flaw in the housing, or a control problem that was loading the shaft during starts. The plant manager wanted a fix that did not require the intake to be offline for longer than one shift each time. Procurement had already been asked to source a higher rated bearing, and the budget for a full fleet retrofit covered only half the pumps.
Analysis
The business objective was narrow: keep the intake available through the high demand months without spending the whole capital allowance on a single pump design. The evidence was uneven. Vibration logs existed for all twelve units, but the sampling rate was too slow to separate outer race wear from misalignment. Lubrication records were paper based for the first year. Failed bearings had been retained, which helped considerably. Constraints included a fixed outage window and a supplier lead time of several weeks on the upgraded part. Stakeholders included the treatment operators, who carried the risk of low pressure, and the finance office, which needed a case that could withstand review.
Corrective options were not equal. Replacing the bearing with a higher rated part addressed the symptom quickly but left the shaft loading untouched. Rebuilding the housing with a different fit would have meant pulling every pump, with the highest outage cost. Adjusting the control logic to soften start transients was inexpensive and could be tested on one unit first. Priya weighted the options by what each would teach her. The retained bearings showed brinelling consistent with repeated shock loading during starts, which pointed toward control rather than batch quality. That reading stayed provisional until sampling was reconfigured to capture high frequency content during starts.
- 1Retained bearing inspectionBrinelling pattern suggests repeated shock loading during starts.
- 2Start transient loggingSampling reconfigured to capture high frequency content at start.
- 3Control logic adjustmentSoft start ramp tested on one pump before any fleetwide change.
- 4Housing redesign reviewSet aside for now due to full fleet outage and cost.
- 5Upgraded bearing trialHigher rated part fitted to the worst unit as a stopgap.
Actions taken
- Reconfigure vibration sampling on all twelve pumps so start events are captured at higher frequency. Keep baseline logs for comparison across two full operating cycles before drawing firm conclusions.
- Pilot a soft start ramp on the most affected pump during a planned low demand window. Record bearing temperature and vibration against the unmodified units over the same period.
- Fit the higher rated bearing to the single worst unit only, so the stopgap does not consume the capital allowance meant for the control change.
- Document lubrication intervals digitally across the fleet, with each entry tied to a pump identifier and a time stamp, so future trend analysis starts from clean inputs.
Results
Within the first month of revised monitoring, the start transient logs showed peak shock events on the four failed units roughly three times the level seen on the untouched pumps. The soft start pilot cut that peak by about half on the test unit and kept bearing temperature within normal range through two weeks of demand peaks. No further bearing failures occurred on the pilot unit. The control change was approved for the remaining pumps and staged over two low demand periods. Total spend stayed within the capital allowance, and the housing redesign was deferred with a documented trigger: a second failure pattern after the control change.
Questions for discussion
- Which evidence would have changed the root-cause conclusion before any hardware was replaced?
- How should a utility weigh a fast stopgap against a slower control change when both affect the same asset?
- What monitoring gaps did this failure history expose that a routine maintenance audit would likely have missed?
Case Study 2
Filling line throughput and reject rate decline
Harbor Ridge Bottling, a beverage plant with three filling lines feeding one shared palletizing cell
The situation
Harbor Ridge runs three filling lines that feed a shared palletizing cell. Over about six weeks, hourly output on line two fell to roughly four fifths of its usual rate, while the reject rate at capper inspection climbed from under one percent to a level that consumed a day of rework each week. The recipe had not changed. A new closure supplier had been qualified in the same period, and a second shift had been added to cover a staffing gap that started at about the same time.
The challenge
Olumide, the operations manager, had until the end of the quarter to restore committed volume to a major retail account, and he could not add a line. Quality wanted the reject rate back under the original threshold before any speed increase. Maintenance wanted to replace the capper drive, which had been flagged in a recent inspection, but the production planner warned that every hour of planned downtime came out of volume that was already short.
Analysis
The commercial target was volume with a defect ceiling, not volume alone. Speed gains that pushed rejects back up would only move the loss downstream. The evidence was better than usual: the line controller logged cycle times, and the inspection system recorded reject reasons by code. Those codes showed loose closures and fill level variation as the dominant causes, far more than label faults. The weakness was that shift handovers were recorded on paper, and the second shift began when the closure supplier changed. Separating supplier effects from crew effects meant pulling torque readings and comparing them by shift. The retail account team needed a dated recovery plan more than a precise root cause, and second shift operators were being judged on a rate they had not been trained to hold.
Teams grounded in Predictive Maintenance Technology usually start by mapping where variation enters the process, not where the output is measured. On this line the capper torque spread was wider on the new closure lot, and fill level drifted most in the hour after each changeover. Those two sources explained most of the rejects and some of the slowdown, since operators stopped the line to re-check. Ranking improvements meant ordering them by reject reduction per hour of effort. The torque setpoint and a changeover checklist were cheap. The closure specification was slower to change but had to be fixed in writing. Acceptance criteria were set in measurable terms: reject rate below the prior threshold for ten consecutive production days, and hourly output within five percent of rated speed.
- Week oneHourly output starts drifting on line two after the closure lot change.
- Week threeReject codes show loose closures and fill level variation dominating.
- Week fourHandover records reveal inconsistent torque checks across shifts.
- Week sixTorque setpoint and changeover checklist introduced on the line.
- End of quarterReject rate back under threshold and volume restored to commitment.
Actions taken
- Narrow the capper torque window on the new closure lot and log every reading for each bottle sample. Target a spread that keeps seal failures under the prior threshold across two full shifts.
- Add a fill level check at the start of every changeover and at the first hour after restart. Operators record the reading on a shared sheet so drift can be seen by shift rather than inferred.
- Pilot the revised closure specification with the supplier on a single lot, with incoming inspection criteria written into the purchase order before any wider use.
- Set acceptance criteria before any speed increase: reject rate below threshold for ten consecutive production days, and hourly output within five percent of rated speed.
Results
Within the third week of the changes, reject codes for loose closures had fallen sharply and fill level variation after changeovers settled into a narrow band. Hourly output recovered most of the gap by the sixth week, with the line running near rated speed during the day shift. The reject rate crossed back under the prior threshold and held through the full acceptance window. The food safety audit found the torque records complete, which had not been true before. Volume to the retail account was restored by the end of the quarter, though the plant accepted a higher overtime cost during the recovery weeks.
Questions for discussion
- Which signals in the reject data would have pointed to the closure supplier sooner than the shift records did?
- When a commercial deadline and a validated process conflict, who should own the decision to change line speed?
- How would you decide whether to keep the tighter torque window once the recovery period ends?
Case Study 3
Gate review for condition monitoring on pressurized compressors
Tamsin Gorge Chemical Works, a specialty chemicals site with four reciprocating compressors on a nitrogen service loop
The situation
Tamsin Gorge runs four reciprocating compressors that feed a nitrogen loop serving two reactors. The maintenance group proposed adding wireless vibration sensors and a gateway to all four machines, with data flowing to a cloud analytics service run by an outside vendor. The aim was to move from fixed-hour overhauls to condition-based intervals. The project sponsor, Hannah, wanted the pilot installed before the next planned shutdown, which was about five weeks away when the review took place.
The challenge
Hannah and the site safety lead, Mateus, asked for a gate review before any hardware was ordered. The review had to establish whether the installation touched pressure boundaries, whether the wireless devices fell inside a classified area, whether the vendor's data handling met the site's network policy, and what documentation existed to support a change request. The proposal described the sensors well but said little about any of those questions.
Analysis
The business case held up. Overhauls ran on fixed operating hours, and one compressor had gone past its interval with little wear while another had been opened early for nothing. Condition data could justify longer intervals where the signal was real. The evidence was thinner than the proposal suggested. The vendor supplied ingress and temperature ratings but no certificate tied to the site's area classification, and the drawings placed mounting points on flanged connections inside the nitrogen boundary. Constraints included the shutdown window, a network policy that blocked outbound traffic from control segments, and a change process requiring sign off from engineering, operations, and safety. The reactor operators would inherit a new alarm source, so their concern about nuisance alarms carried real weight.
The review separated what could be approved from what needed more work. Installations on pressure boundaries required the pressure equipment documentation to be updated, including the boundary drawings. Any device in a classified area needed a certificate matching the zone and gas group. Those items were gates, not preferences. The vendor's data handling was assessed against the site's own records and network policy, which governs retention and access, not only against general security guidance. The reviewers asked for a plan that kept sensors off pressure boundaries in the first phase, used magnetic or stand-off mounts where access was limited, and held the cloud feed behind an approved gateway. That kept the pilot moving while leaving the boundary question for a separate change.
Actions taken
- Pull the area classification drawing for each compressor and confirm the zone and gas group for every candidate mounting point before any sensor is selected or ordered.
- Request a certificate matched to the zone and gas group from the vendor, and hold procurement until that certificate is placed in the change file.
- Limit phase one to stand-off or magnetic mounts on non-pressure supports, with the cloud feed routed through an approved gateway that meets the site records policy.
- Update the pressure equipment documentation and boundary drawings as a separate change, approved before any mounting point on a flanged connection is used.
Results
The pilot went ahead with sensors on four non-pressure supports, communicating through a gateway that the network team approved after a short review. The vendor supplied the missing certificates two weeks before the shutdown, which allowed installation during the window without a waiver. Two nuisance alarms in the first month were traced to a loose bracket and corrected. The boundary change was deferred to the next outage and handled as its own change with full documentation. The corporate risk function accepted the phased plan, and the monitoring data is now informing a review of interval lengths on one compressor.
Questions for discussion
- What evidence would have been enough to approve sensor mounting on a pressure boundary in this case?
- How should a review weigh a vendor's general ratings against a site-specific certificate requirement?
- Where should the line sit between a phased pilot and a change that the approval process was never designed to handle in pieces?
Case Study 4
Sensor analytics project slips after a supply shortfall
Ashcombe Power Cooperative, a generation operator with two gas turbines and a shared spares pool
The situation
Ashcombe runs two gas turbines and planned to add a condition monitoring package to both, with vibration and oil debris sensors feeding a shared diagnostics platform. The package was scheduled to be live before the summer peak, when the turbines carry their heaviest load. In the fourth month of the project, the vendor reported that the vibration modules for the first turbine would arrive about six weeks late because of a component shortage upstream of its own supplier. The shared spares pool held no compatible substitute.
The challenge
Ruth, the project manager, had to present a recovery plan to the asset committee within the week. The finance director wanted the overrun held under the contingency already approved. The operations head wanted no commissioning activity during the peak period. The engineering lead warned that any substitute module would need validation against the existing signal chain, which could consume time the team did not have.
Analysis
The business objective was readiness for the peak period, not monitoring for its own sake. The committee had approved the package on the basis of fewer forced outages and better planning for overhauls, and a late go live would delay those benefits by a full operating season. The evidence was a vendor schedule with a revised date, a parts tracking log from the upstream supplier, and the fault history from both turbines. The schedule evidence was the weakest. The vendor gave a range rather than a firm date, and the upstream supplier had not confirmed allocation in writing. Constraints included a contractual cap on expedite spend, the peak period itself, and a signal validation for any substitute that took several weeks. The operations crew would have to take the systems on shift, and the finance office had already committed contingency and had no appetite for a second request.
Teams grounded in Predictive Maintenance Technology tend to compare recovery plans by what each one protects and what each one spends, not by how quickly the schedule moves on paper. Expediting freight was the most expensive option per week gained, and it recovered the original date only if the upstream allocation held. Phasing the rollout kept the first turbine on time and moved the second to the following outage, which avoided the premium but deferred benefits for one asset. Substituting hardware looked attractive until validation was costed, which roughly doubled the engineering hours. Accepting the delay cost nothing up front and produced the cleanest commissioning data, but it gave up the peak period entirely. Laying the options against one another in a single table made the decision easier than the schedule chart had.
- Expedite supplier Pay premium freight for the vibration module shipment and accept the cost.
- Phase deployment Commission the first turbine now, defer the second to the next outage.
- Substitute hardware Use a qualified alternative module after engineering validation of signal output.
- Accept delay Hold the original scope and absorb the slip into the next budget cycle.
Actions taken
- Ask the vendor for written allocation from the upstream supplier, and tie the expedite authorization to that confirmation so premium freight is not paid against an unverified date.
- Commission the first turbine's monitoring package on the original date with the vibration modules that arrive, and defer the second turbine's installation to the next planned outage.
- Start signal validation for the alternative module now as a parallel track, so it is ready if the expedited shipment slips again.
- Log the cost of each option against the contingency line in the same format the finance office uses, so the committee sees the trade-offs in its own terms.
Results
The vendor confirmed allocation in writing in the second week, and the expedite premium was authorized for one shipment only. The first turbine's modules arrived with six days of slack before the commissioning window, and the package went live ahead of the peak period. The second turbine was deferred to the next outage as planned. Validation of the alternative module finished without being needed, which kept the parallel effort to a modest engineering cost. Total overrun stayed within the approved contingency, though the second turbine's benefits arrived a full cycle later than first proposed.
Questions for discussion
- When is paying an expedite premium justified, and what evidence should exist before authorizing it?
- How should a committee weigh a partial delivery against a full scope delayed into another budget cycle?
- What would make starting a parallel substitute track too expensive before the primary supplier confirms allocation?
Case Study 5
Cutting compressed air waste in a food packaging plant
Brightwater Foods, a packaging plant producing sealed snack trays across six lines and a central utilities building
The situation
Brightwater Foods runs six packaging lines off a shared compressed air system. The utilities team estimated from a flow meter survey that roughly a third of the air generated was lost to leaks, unnecessary header pressure, and open drains. That figure described total demand, not the source of each loss. Compressed air is one of the more expensive utilities per unit of energy, and the plant's electricity cost had risen faster than output. The site manager, Dorota, wanted a program that reduced energy use and also cut packaging scrap, with no drop in line performance.
The challenge
Dorota had to choose which measures to fund from a limited energy budget, and she needed visible results within the current budget cycle. The packaging lines could not tolerate pressure drops during the sealing step, since seal quality failed quickly below a set point. Kenji, the sustainability lead, wanted lifecycle figures for each option, not only payback, and the maintenance team warned that several candidate upgrades required outages the production schedule could not absorb.
Analysis
The business objectives were energy cost, a measurable reduction in waste, and no loss of sealing performance. Lifecycle measures mattered because the proposals differed in embodied impact: a new compressor carried manufacturing emissions and disposal obligations that a leak repair did not. The evidence was partial. The flow meter survey gave a total demand figure but could not assign losses to specific lines or drains. A short ultrasonic survey covered most headers and produced a tagged list, yet it missed a segment routed behind a wall panel. Operational constraints included the sealing pressure floor, the outage calendar, and a maintenance contract that limited changes to compressor controls. The line operators would notice any pressure swing, and the finance group tracked energy and waste as separate budget lines, so claims that crossed them drew scrutiny.
Comparing options meant putting energy and lifecycle measures on the same page. Kilowatt hours saved were easy to compare, but so were the carbon content of the electricity and the disposal burden of replaced parts. The team compared four groups. Leak repair and a pressure setpoint trim scored highest on energy saved per unit of disruption, with almost no embodied cost. A central compressor replacement cut losses most but added manufacturing impact and a long shutdown. Housekeeping, such as tagging unused drops and cleaning filters, was small but cheap and kept gains from eroding. Replacing working dryers for marginal gains did not hold up once lifecycle impact was included. The ranking was sensitive to survey coverage, so the team treated that as a reason to measure before committing.
Actions taken
- Complete a full ultrasonic leak survey of all headers and drops, including the segment behind the wall panel, and tag each leak by line and estimated flow rate.
- Lower the header pressure setpoint in steps, holding each step for a full production cycle while sealing quality is checked on every line.
- Schedule the compressor controls update for the next planned outage, and record the disposal plan for any replaced filter or valve before the work order is raised.
- Report energy saved and scrap reduced in the same monthly pack, with a lifecycle note on each change, so finance and sustainability review one set of figures.
Results
The leak survey tagged about forty losses, and repairs on the largest ones cut total compressor demand noticeably within two months. The pressure trim was taken in three steps, and seal quality held at each one. The final step was stopped one notch short after a single line showed borderline results. Scrap from one tray line fell once a nozzle fault was found during a drop inspection. The compressor controls change was deferred to the next outage, and the lifecycle note recorded that decision along with its reasoning.
Questions for discussion
- How would you decide whether a gap in leak survey coverage is large enough to delay a pressure change?
- Should the lifecycle impact of replacement equipment carry the same weight as operating energy savings in a plant budget?
- What signals on a packaging line would tell you a pressure reduction has gone too far before defects appear?
Learning Resources
Industrial CASE STUDIES
China, India, United States, Indonesia, Pakistan, Nigeria, Brazil, Bangladesh, Russia, Japan, Mexico, Philippines, Vietnam, Egypt, Turkey, Germany, United Kingdom, France, Italy, Canada, South Korea, Australia, Spain.