Three hundred A356-T6 aluminum housings landed in our yard in February. Gravity cast, about 1.4 kg each, a machined face on the front and a bored bearing boss behind it. The drawing called for a 0.8 mm cleanup cut on that face and a 4 mm nominal wall under it. We had run two smaller jobs for the same foundry before and neither one had given us trouble.
We cut the first face on part one and stopped. Three dark clusters sat on the freshly opened surface, each about the size of a pin head, arranged in a ring where the boss transitions into the wall. Dye penetrant made it worse. The clusters bled along the surface for two or three millimetres. Part one went into the scrap bin.
By part 300 we had scrapped 71 housings, 23.7 percent of the order, and re-machined 22 more after the foundry agreed to weld them. One part failed a 0.6 bar air test on the boss. That is the one that worried me. A porous bearing boss on a housing that holds oil is not a cosmetic problem.
The radiographs said the castings were fine
The foundry had X-rays, and they sent them with the first article. Three pieces from the heat lot, all rated clean against their internal standard. That is the part that took me a week to understand.
Their radiography was taken on the as-cast part, at the as-cast surface. The porosity we were hitting sat 1.6 to 2.2 mm below the skin, in the thermal centre of the boss, exactly where the wall is thickest and the metal is last to freeze. An X-ray of the outside of a casting does not see the middle of it. They were not lying to us. They were measuring a different thing than the one that mattered to us.
Sampling was the second problem. Three pieces out of 300 tells you the heat lot was poured. It does not tell you every part in that lot is sound. Casting defects are not evenly distributed, and the failures were not spread evenly through the pallet either. Once we started numbering the housings we found 41 of the 71 came from the second half of the pour.
Why the 0.8 mm cleanup cut was the trigger
Cast aluminium forms a chill zone at the surface. That thin skin freezes fast, traps almost no gas, and is the soundest material in the whole part. Underneath it sits the thermal centre, where shrinkage voids and gas porosity collect. A 0.8 mm face cut removes most of the good skin and exposes the layer underneath it.
The wall was 4 mm nominal. After a 0.8 mm cleanup outside and a 0.5 mm bore inside we were down to a 2.7 mm wall with a 1.5 mm porous band running through the middle of it. The drawing was satisfiable. The casting we were handed was not.
What we put on the purchase order
Three clauses, and they now fit in one paragraph of the PO.
First, radiography on the machined exposure zone rather than the as-cast surface, referenced to a severity level we name in the clause. Second, destructive sectioning of three pieces per heat lot at the boss, cut and photographed, travelling with the material certification. Third, a 1.2 mm casting stock allowance on the machined faces instead of 0.8 mm, so the cleanup cut gets below the porous band rather than landing in it.
The foundry also changed the gating on their tool, which was their fix and their cost, and scrap in the next lot dropped to nine pieces. We have not seen a repeat in four runs since.
None of that is exotic. It is the same discipline we apply to incoming bar stock. Define the feature that has to hold, then inspect the thing that actually controls it.
What it cost
The castings ran about USD 18 each at that quantity, which is a budget estimate for the size and alloy rather than a fixed number. Seventy-one scrapped housings is roughly USD 1,280 of casting, plus 96 hours of machine time we will never bill, plus air freight on the replacement lot, plus welding and re-machining on the 22 we saved. Call it USD 6,400 to USD 7,000 all in. The three PO clauses cost us nothing but a conversation.
The other route was to machine from 6061 plate. For this part that lands somewhere near USD 31 per piece in material and cutting time, also a budget estimate, about 70 percent more. At 300 pieces a year the casting route is still correct. At 30 pieces it would not be, and we would not be having this conversation.
If you buy castings and then machine them
Ask where the radiography was taken. If the answer is the as-cast surface, ask for it at the machined exposure depth instead. Ask how many pieces per lot were sectioned and whether the pictures travel with the parts. And ask what stock allowance sits on the machined faces, because 0.8 mm on a 4 mm wall is not a cleanup cut, it is a gamble.
A foundry that can answer those three questions has scrapped a lot before. Ours could not, the first time. It can now.
We machine cast and billet aluminium housings most weeks, and the honest comparison between the two routes lives on our casting versus machining from billet page. If you are setting up a casting supply, the first-article inspection page covers what belongs in that package, and material certification covers the paperwork that should travel with every lot. The cutting side sits under CNC milling. Send us the drawing and the casting specification and we will tell you where the machined surfaces will land relative to the porous band before you commit to a tool. The instant quote tool returns a budget estimate to start from, and anything that needs a firm number goes to our team, who confirm within one business day.