How Does a Rear Differential Work—and Why That Changes How You Inspect Every Metal Part

Ask three engineers "how does a rear differential work" and you'll get three versions of the same answer. Ask them how to inspect a differential housing and you'll get three different checklists—and at least two of them will be wrong for the part sitting on the bench.

I've been on both sides of that bench. I'm the quality and brand compliance manager at beru, an automotive metal parts supplier that runs stamping, die making, CNC machining, forging, and extrusion under one roof. I sign off on roughly 40 to 60 production lots a month before they ship, and I handle the customer complaints when something slips through. In our Q1 2024 audit, about 11% of first-article submissions from new suppliers came back with dimensional or documentation issues. Not because the suppliers were bad. Because the inspection plan didn't match the part.

Here's what I've learned: there's no universal QC recipe. What you check—and how hard you check it—depends on how the part fails. I sort most automotive metal parts into three buckets. Find yours, and the checklist writes itself.

The three failure modes (and why they matter more than part price)

Every part fails in one of three ways: it breaks and someone gets hurt, it leaks and someone eats the warranty cost, or it looks wrong and nobody notices until the customer does. Those are very different problems. Price tells you almost nothing about which one you're buying.

• Load-bearing / safety-relevant: differential cases, ring and pinion gears, axle flanges, suspension brackets, brake caliper mounts.
• Fluid-containment / sealing: flat radiator pipe collars, fuel filter hose fittings, coolant outlets, oil pan flanges, thermostat housings.
• Fit-and-finish / aftermarket: brackets, covers, trim mounts, non-structural spacers, cosmetic caps.

I'm not going to pretend these lines are always clean. A radiator collar on a track car is closer to safety-relevant than a bracket on a garden trailer. But the sorting still works, because it forces you to ask the only question that matters: what happens when this part fails?

Scenario A: The part carries load or protects people

Start with the rear differential, because it's the cleanest example of why "visual inspection plus a few calipers" isn't enough.

How a rear differential works, short version: the driveshaft spins a pinion gear, the pinion turns the ring gear, and the ring gear is bolted to the differential case. Inside the case, spider gears on a cross pin drive the two side gears, which connect to the axle shafts. Going straight, the spider gears don't rotate on the pin—everything turns as one unit. In a turn, the spider gears rotate and let the outside wheel spin faster than the inside wheel. That's the whole trick. (Limited-slip versions add clutches or gear sets to stop one wheel from spinning uselessly, but the basic geometry is the same.)

Now think about what that means for the parts. The differential case has to hold the ring gear flat and keep the bearing bores coaxial. The ring and pinion have to mesh at the right depth and backlash. If the case face is out of flat by a few thousandths, the gear pattern shifts, and you get whine at 60 mph that the dealer "can't reproduce." If the bearing bores are off-axis, the bearings wear unevenly and the unit fails at 40,000 miles instead of 150,000.

So for this bucket, my checklist is expensive on purpose:

• Full dimensional first article on a CMM, not a caliper and a prayer
• Material certs with heat numbers, traceable to the lot
• Hardness and, where specified, case depth
• Magnetic particle or ultrasonic inspection on load paths
• Gear pattern check on a rolling tester (for ring and pinion)
• PPAP Level 3 or higher (AIAG), backed by an IATF 16949 quality system, plus process capability data—Cpk 1.33 minimum on critical characteristics

Here's the counterintuitive part: a perfect sample proves almost nothing. I've seen beautiful first articles from a forging die that was already washing out. By unit 5,000, the flash line had moved and the machined face was out of tolerance. That's why I ask for PFMEA and control plans before I ask for another sample. Show me how you'll catch drift, not how good you are on a good day.

Anyone can make ten good parts. The question is whether part 8,000 looks like part one.

Scenario B: The part holds fluid—and the leak shows up six months later

Flat radiator pipe collars and fuel filter hose fittings live in a different world. They rarely snap. They seep. And a seep doesn't announce itself on the loading dock. It shows up as a coolant smell in a customer's garage, or a fuel odor that turns into a safety campaign.

The failure modes are boring and specific: sealing face flatness, roundness, surface finish, plating adhesion, and material compatibility with whatever's flowing through it. A collar that passes a dimensional check can still leak because the sealing surface has a tool mark deep enough to break the gasket line. A fuel filter hose fitting can pass a pressure test cold and fail hot, because the elastomer or the coating wasn't rated for the actual fluid and temperature.

For this bucket I lean on system-level standards and real-world cycling, not just print dimensions:

• Sealing face flatness and surface roughness (Ra) called out separately from general tolerance
• Salt spray per ASTM B117 where coatings are specified
• Pressure and thermal cycling, not just a single static burst test
• Reference the relevant SAE hose and fitting standards—SAE J20 for coolant systems, SAE J30 for fuel and oil—and confirm the current revision with your customer's engineering group
• Plating adhesion and coating thickness coupons from the same lot

Now the part that gets me in trouble with sales: cheap collars are usually the most expensive parts we make. I've watched a $0.40 stamped collar with a marginal sealing face turn into a $200 cooling-system rework, plus a line stoppage, plus a customer who now audits us twice a year. The unit price was a rounding error. The failure cost wasn't.

So when a buyer sends me three quotes and picks the lowest, I ask what's missing. Does the quote include first-article inspection? Plating thickness reports? Pressure cycling? Packaging that won't dent the sealing face in transit? A quote that excludes all of that isn't cheaper. It's just unfinished. I'd rather see a higher number with everything listed than a low number with a surprise invoice later. (This is the same reason I ask "what's not included" before I ask "what's the price." Learned that one the hard way.)

Scenario C: The part is cosmetic or aftermarket—and over-inspection is a real cost

Not every bracket needs a CMM report. For non-structural, cosmetic, or aftermarket parts, the failure mode is usually fit or appearance, and the right move is often less inspection, not more.

My default here:

• First article against the drawing, with a go/no-go gauge for the interface dimensions that actually matter
• AQL sampling on the rest (usually 1.0 or 2.5 depending on the customer)
• Color and gloss check against a physical standard if it's coated
• Packaging drop test if the finish is visible

The counterintuitive point: buyers routinely over-test this bucket and under-test the two above. A full PPAP on a cosmetic trim bracket burns weeks and adds cost the customer will never feel. Save that rigor for the parts where failure has consequences. Inspection budget is finite—spend it where the risk lives.

How to figure out which scenario you're in

Three questions, in this order:

1. What happens if it fails? Injury or stranded vehicle → Scenario A. Leak, smell, or warranty claim → Scenario B. Rattle, gap, or "meh" → Scenario C.
2. Who's the end customer, and what do they require? OEM programs usually mean PPAP, IMDS, and a documented control plan. Aftermarket usually means interchangeability and packaging. Mixing the two sets of rules is how you get either a rejected submission or a wildly overpriced bracket.
3. Will the tooling wear out during my order? If annual volume is 50,000 and the die was built for 200,000 hits, you're fine. If the die is near end of life, you need wear allowance data and a maintenance schedule—not another sample.

If you're still unsure, pick the higher scenario. The delta in inspection cost is almost always smaller than the delta in failure cost. That's not a sales pitch. It's the pattern I've seen across hundreds of part numbers since I started tracking rejection reasons in 2022. On our beru auto parts programs, that conversation now happens before the PO is issued, not after the first leak.

One more thing, and then I'll get off the soapbox. None of this works if the drawing and the purchase order disagree. I still kick myself for a 2023 lot where the print called out a flatness tolerance and the PO said "per standard." The parts measured fine against the PO. They leaked against the print. We ate the rework. Now every PO we issue references the drawing revision and the inspection standard in the same line. Boring? Yes. Effective? Also yes.

Match the checklist to the failure mode. Ask what the quote doesn't cover. And don't let a good sample talk you out of a process audit. That's the standard I hold every beru brand shipment to—and it's the reason I sleep through most nights. Most of them.