Smart Manufacturing Systems Case Studies
Industrial5 Case StudiesUpdated Sep 2026
Smart manufacturing work tends to break down where software, sensors, and equipment meet. The five cases here come from fictional plants and integrators recovering from problems already in front of them. Each turns on one decision: a crack pattern pointing two ways, a bottleneck hidden behind a quality alarm, a safety function missing its paperwork, a controller shortage threatening a handover, and an energy retrofit that could not cost output. The figures belong to each organization alone. Read for the order of questions the practitioners asked, since that order often matters more than the answer they reached.
Case Study 1
Repeated spindle housing cracks on an automated pouring line
Harrow Lane Castings, a mid-sized foundry supplying machined transmission housings to tractor drivetrain makers.
The situation
Harrow Lane began seeing cracked spindle housings in the second month after its third pouring cell went live. The part is a cast iron housing for a tractor transmission, and the customer accepts it only after a hardness check and a pressure test on every piece. Field returns were rare at first. Then one shift produced nine failed pressure tests out of 2,300 castings from the same cell. The cell ran a gating layout that process engineering had modeled but never validated at full volume. Thermocouple logs from the mold showed the boss section cooling about 18 percent faster than the model predicted during the first weeks of production.
The challenge
The quality group wanted a fast answer: tighten pour temperature and move on. The design group held that the gating was sound and sand was the variable. Maintenance records showed the sand reclaimer had two unscheduled stops in six weeks, and the mold coating supplier had changed its formulation in the third week of the run. No one could say which change produced the cracks. The customer expected a corrective action report within ten working days.
Analysis
Teams grounded in Smart Manufacturing Systems usually begin by separating what the design assumed from what the process actually delivered. In this case the gating model assumed a uniform cooling rate across the boss and the flange, and the thermocouple data contradicted that assumption within the first few hundred pours. The tolerance on wall thickness was tight, but it was the solidification window that set the real margin. A section cooling at a different rate than modeled creates a stress gradient that the hardness test does not catch. Harrow Lane had sensors on pour and mold temperature but none on the boss, so the team reasoned from flange data alone.
The sand reclaimer had two unscheduled stops, and the sand moisture reading stayed within specification at each stop but drifted high afterward. When the team overlaid the reclaimer history on the crack lot, the timing lined up with the coating change rather than the gating itself. That overlap did not prove causation. It did mean that a design change alone would have been expensive and possibly wrong. The team ranked three failure modes: gating-driven cooling gradient, coating-driven surface defect acting as an initiation site, and sand moisture variation raising mold gas. Each mode predicted a different crack location, which made the next step testable.
Actions taken
- Install a thermocouple in the boss cavity of two molds and log cooling rate for every pour during the next three production weeks.
- Section twelve retained castings from the cracked lot and compare crack origin locations against each of the three candidate failure modes.
- Hold the coating formulation at the previous supplier lot until the sectioning results return, and record every lot change in the cell log.
- Revise the gating model with measured cooling curves, then run a short validation batch before changing any drawing or tolerance callout.
Results
Sectioning showed crack initiation at the coating-affected surface in nine of twelve parts, while the other three showed porosity near the boss that matched sand moisture. The gating model was not the primary cause, though its cooling prediction missed consistently enough to justify revision. Holding the previous coating lot cut failures to two in the next 1,400 castings, inside the customer's acceptance level. The corrective action report went out on the ninth working day with a change to supplier qualification and a shorter reclaimer service interval. The gating revision stayed on the shelf until the validation batch confirmed it.
Questions for discussion
- Which of the three candidate failure modes would you have tested first, and what evidence would have changed that order?
- How should a team decide whether a sensor gap on the boss justifies the cost of instrumenting production molds?
- When a design model and field data disagree, what sets the threshold for revising the model rather than the process?
Case Study 2
Throughput loss and leak rejects on a valve body line
Calder Ridge Components, a contract machinist producing hydraulic valve bodies for agricultural equipment makers.
The situation
Calder Ridge runs a valve body line with five stations: rough mill, bore finish, deburr, leak test, and pack. Over the quarter, output dropped about 11 percent against plan even though the mill ran more hours. The leak test reject rate rose from 0.6 percent to 2.1 percent in the same window. Bore finishing is the constraint, and its single boring head loses time at every tool change. Operating data showed cycle time at that station ranging from 74 to 118 seconds, with the longest cycles clustered after the second tool change of each shift. Spindle maintenance records showed one bearing replacement and two coolant filter stops.
The challenge
The plant manager wanted throughput recovered before quarter end and asked for a second boring head. The quality lead pointed to the leak test and asked for a tighter bore tolerance across the board. The capital request had a ceiling, so the team could not buy both. Bore diameter tolerance was plus or minus 0.01 millimeter, and the leak test passed a seal only when bore roundness stayed within 0.006 millimeter. Nobody had measured roundness on the line since the spindle bearing was replaced.
Analysis
Mapping the line tends to show that the station with the alarm is not the station causing the trouble. In this case the bore finish station generated most of the leak rejects, but the alarm appeared downstream at the leak test, which is why the first explanations pointed at deburr. Overlaying the five stations on time stamps revealed three sources of variation: tool life at the boring head, clamp force on the fixture that shifted with operator loading, and manual deburr time that stretched whenever batches arrived out of sequence. Each source touched cycle time, but only one touched roundness. The map made the handoffs visible for the first time.
Prioritizing the fixes required putting numbers on each source. Tool life data suggested the boring head was changed on a fixed schedule rather than on wear, which left the end of each tool run out of tolerance. Fixture clamp readings varied by roughly 15 percent between operators, which affected roundness mainly on parts handled in the first hour after a changeover. The deburr delay was the cheapest to fix, though it limited throughput more than quality. The team set acceptance criteria before any change: bore finish cycle time under 90 seconds for 95 percent of parts, and a leak test reject rate under 0.8 percent.
- 1Rough millThree spindles with stable cycle time and minimal reject contribution.
- 2Bore finishConstraint station where tool life sets the rate limit.
- 3DeburrManual wipe step adds variable time between parts.
- 4Leak testRejects trace back to bore roundness, not deburr.
- 5Pack and shipBatch release waits on each hour of leak test results.
Actions taken
- Replace the fixed tool change schedule with a wear-based trigger, using the existing tool life log to set a limit at a measured bore size drift.
- Add a clamp pressure check at each changeover, log the reading, and stop the line if clamp force falls outside the setup window.
- Move manual deburr into a fixed sequence with a buffer of twelve parts, so batches arrive in order and the wiping step keeps a steady rhythm.
- Measure bore roundness on every fifth part for two weeks and compare the results against leak test outcomes before approving the second boring head request.
Results
Within the first month after the tool change schedule moved to wear-based triggers, cycle time at bore finish fell to a median of 83 seconds, and 96 percent of parts met the 90-second target. The leak test reject rate dropped to 0.9 percent, short of the 0.8 percent goal but well below the starting point. Roundness data showed the clamp check caught most out-of-window setups before parts reached the leak test. The second boring head request was withdrawn once the measured recovery covered most of the lost output. Calder Ridge kept the buffer rule as a permanent work instruction.
Questions for discussion
- Which measurement would you have taken first to separate tool wear from clamp variation, and why?
- When a downstream alarm points away from the true source, how much mapping is enough before changing equipment?
- How should a plant weigh a capital request against process changes that cost little but need discipline to hold?
Case Study 3
Missing safety evidence on a palletizing cell retrofit
Northgate Robotics Integration, a systems integrator that retrofits palletizing cells on beverage packaging lines.
The situation
Northgate was retrofitting a palletizing cell on a bottling line, replacing a manual stacking station with a six-axis robot and a two-lane infeed. The old cell had fixed guarding and a single interlocked gate. The new layout placed the robot's reach across the path forklifts used to restage pallets, and the light curtain was moved to cover a wider opening. Operating data from the old cell showed 3,400 pallets per shift with two jam stops each day. Maintenance records showed the light curtain muting sensor had been replaced once, and its alignment was never rechecked afterward. The safety PLC program had been revised twice without a matching validation report.
The challenge
The bottler wanted acceptance testing finished before its seasonal volume increase, which left about three weeks of schedule. Ines, the project lead, was ready to sign the commissioning checklist, but the risk assessment still described the old layout. The stop function had no recorded performance calculation. Marcus, the customer's safety officer, asked for evidence before any operator would work in the cell.
Analysis
A review like this starts with the hazard the new layout created, not the components it added. The robot's reach now overlaps the restaging path, so the hazard is a person entering that path while the robot is still moving, and the light curtain was never tested against that entry point. The applicable robot safety standard calls for a fresh risk assessment after any layout change. The performance level of the safety function depends on component reliability and diagnostic coverage, and the team had not calculated either. Without those numbers, the existing guarding cannot be credited. The fix may be physical, such as a fixed barrier and a second interlock, or it may be a revised function with a documented calculation behind it.
Documentation gaps were the more serious problem. The risk assessment, the safety requirements specification, and the validation report each described a different version of the cell. The approval gates were defined as design freeze, factory acceptance, site installation check, and handover, but the checklist allowed handover to be signed before the validation report existed. Maintenance records added a second concern, since the muting sensor replacement had no verification record. A reviewer can accept a risk control only when its evidence chain runs from the hazard to the test result, and here the chain broke at two links.
- First weekLayout change puts forklift path under robot reach; assessment not updated.
- Early second weekSafety PLC revised twice with no matching validation report filed.
- Mid second weekLight curtain muting found misaligned during site walkdown.
- Third weekCommissioning checklist drafted against the old layout drawing.
- Before handoverReview holds sign-off until the evidence chain is complete.
Actions taken
- Repeat the hazard analysis for the full restaging path, including forklift movement and the robot's maximum reach, and sign the revised assessment before any further testing.
- Calculate the performance level of the stop function using component data and diagnostic coverage, and file the calculation with the safety requirements specification.
- Install a fixed barrier across the restaging path and add a second interlock, so the light curtain is no longer the only protection for that entry point.
- Revise the approval gate so handover requires a matching validation report, and retest muting alignment with the sensor replacement record attached.
Results
The barrier across the restaging path removed the cell's exposure to forklift movement at the entry point. The revised assessment also identified a fourth hazard, a pinch point at the pallet transfer that the original design had not shown. The performance calculation required a second interlock on the gate, which added about two weeks of hardware lead time. Acceptance testing moved past the seasonal increase, and the bottler ran the cell at reduced rate for the first eleven days after handover. Marcus signed the handover once the evidence chain was complete. Northgate now treats a revised hazard assessment at every layout change as a gate, not an optional step.
Questions for discussion
- Which piece of missing evidence would you have requested first, and what would it have told you about the others?
- How should a retrofit team decide when a layout change requires a full reassessment rather than an addendum?
- When a schedule deadline and a sign-off requirement conflict, who should carry the decision and on what written basis?
Case Study 4
Controller shortage threatens a mixing line handover
Varnell Process Equipment, a builder of custom batch mixing skids for specialty chemical and food plants.
The situation
Varnell was building two automated mixing skids under a contract valued at 1,850,000 dollars, with a customer factory acceptance test set for the end of the quarter. The skid controller was a motion and recipe platform, and the supplier had moved it to a longer lead time after a component allocation change. The original units were due in the sixth week of the build, but the revised date landed nine weeks later. The customer then requested a second dosing valve on the second skid, which required a panel redesign. Coordination between the electrical subcontractor and the controls integrator had run through email, and the panel drawings went out with two revision letters.
The challenge
Every week past the acceptance date carried a penalty of 42,000 dollars under the contract, and the customer had made clear that a late skid would cost them a production window. Dara, the project manager, had three options, and each one moved cost somewhere else. Expediting controllers through a distributor was expensive and carried a risk of non-conforming firmware. Redesigning the panel for a different controller family meant re-certifying the recipe logic. Holding the original order meant accepting a full slip.
Analysis
The trade-off has to be stated in money and weeks, not in preferences. Distributor controllers were quoted at a premium of about 61,000 dollars for the two skids, and their firmware was a different revision from the one the recipe logic had been tested against. Holding the original order carried a penalty of 42,000 dollars per week after the acceptance date, so a nine-week slip would have cost far more than the premium. The panel redesign for the second dosing valve added roughly three weeks of drawing work whether or not the controller changed. Separating those costs showed the partial swap was cheaper only if the firmware risk could be closed within a week.
Coordination was the hidden constraint. The electrical subcontractor and the controls integrator were working from different panel revisions, and the dosing valve change had reached only one of them. The partial swap would therefore need a single controlled drawing set before any wiring began. Design constraints also narrowed the options: the recipe logic had been validated against a specific function block library, so distributor units had to run the same library version or the validation would have to be repeated. Measuring that repeat effort in hours, rather than assuming it, changed the answer.
Actions taken
- Freeze one panel drawing set with a single revision letter and send it to both the subcontractor and the integrator on the same day, with a read-back confirmation.
- Run the distributor firmware against the existing recipe test set in a bench rig for five working days before any unit is installed on skid one.
- Ask the customer in writing to accept a revised acceptance date for skid two, with a fixed penalty cap tied to the controller delivery date.
- Log each recovery cost, including bench hours and expedite fees, in the change ledger so the final margin can be compared with the original estimate.
Results
The bench run took six working days and found one firmware fault in alarm handling, which the distributor patched before installation. Skid one reached factory acceptance on the original date. Skid two slipped four weeks under a written revision the customer accepted, which capped the penalty at a small fraction of the original exposure. Total recovery cost came to about 74,000 dollars, above the premium estimate because of bench hours and a second panel revision. The contract margin was reduced but held above the level set at bid. Varnell now requires a single drawing set for any change order that touches a controller.
Questions for discussion
- What would have to be true about the firmware for the partial swap to be the wrong choice?
- How should a project team price the cost of rework on drawings that went out with two revision letters?
- When a customer change request and a supplier delay arrive together, which should be negotiated first, and why?
Case Study 5
Cutting energy and regrind in an injection molding cell
Arden Fold Plastics, an injection molding plant making closures for beverage bottlers, running twelve presses on two shifts.
The situation
Arden Fold set a target to cut energy per 1,000 closures for the current quarter, and one cell was the largest consumer on the floor. The cell ran a 48-cavity mold on a hybrid press at a cycle time of about 6.8 seconds, with a barrel heater setpoint that had been raised during an earlier resin change and never lowered. Regrind was being added at 25 percent of shot weight, and scrap on that mold had climbed. Submetering showed the cell using 14 percent more kilowatt-hours per 1,000 parts than its sister cell, which ran the same mold on a different resin lot. Maintenance records showed the hydraulic pump had been serviced for a leak but never recalibrated.
The challenge
Pilar, the process engineer, wanted to lower the heater setpoint and raise regrind to 35 percent, which would save energy and material together. Tomas, the quality lead, warned that wider regrind had already produced gate vestige variation and that the seal test margin was thin. Any change had to keep closure torque and leak rate inside specification, and the customer had an audit scheduled on seal test results. Each option claimed a saving, but none had been measured against energy and scrap in the same trial.
Analysis
Comparing the options required two kinds of measure that usually live in different reports. Energy per 1,000 closures is an operational figure the plant already tracks, but a lifecycle view also includes the embodied energy of virgin resin displaced by regrind and the cost of disposing of scrap that could not be reground. A regrind increase looks attractive on material alone. Once the seal test margin and the extra rework hours are counted, the net benefit shrinks. The team wrote the comparison as a single sheet with energy, scrap, seal test margin, and cycle time in the same row for each option.
The constraints set the order of work. The seal test had tolerances on closure torque and leak rate that left little room for resin variation, so any change needed validation across at least three lots before scale-up. The hydraulic pump was the only item that cut energy without touching the part, which made it the first move. Heater setpoint and regrind share both affected melt viscosity and therefore fill and gate vestige, so they were tested together in a designed trial rather than one at a time. Maintenance history explained why the pump gain had been hidden: the leak repair had restored flow but not the calibrated pressure setting.
- Heater setpoint Lowering it cut energy but risked underfilled gate areas in early trials.
- Regrind share Raising it to 35 percent saved material, but seal test margin narrowed.
- Hydraulic recalibration Restored pump efficiency; the largest single energy gain on the cell.
- Cycle time review Shorter cooling time reduced cycle length without changing part weight.
Actions taken
- Recalibrate the hydraulic pressure setting on the cell pump, then rerun the energy submeter for one week to confirm the baseline before any process change.
- Run a designed trial of heater setpoint and regrind share at three levels each across three resin lots, measuring energy, scrap, and seal torque.
- Set regrind share at the highest level that keeps seal torque within the lower control limit in every trial lot, rather than at a fixed target.
- Shorten cooling time only after the trial confirms gate vestige stays inside the drawing limit, and record the cycle change in the cell standard work.
Results
Pump recalibration alone cut energy per 1,000 closures by about 9 percent in the baseline week, the largest single gain. The designed trial showed regrind could rise to 30 percent without moving seal torque below the lower control limit, but 35 percent pushed two of three lots under it. The heater setpoint came down modestly, and the cooling time reduction was approved only after gate vestige stayed inside drawing limits. Combined energy savings reached roughly 17 percent against the starting point, with scrap at or below the earlier rate. Arden Fold kept the 30 percent regrind ceiling as a documented standard for this mold.
Questions for discussion
- Which of the four levers would you have tested first, and what measurement would have changed your order?
- How should a lifecycle comparison treat regrind that cannot be reground a second time?
- When a sustainability target and a seal specification pull in different directions, what should decide the limit?
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.