Die Casting Scrap Reduction: Pareto Causes, the True Cost of a Reject, and a 30-60-90 Day Plan

Most die casting plants know their scrap rate. Very few know their scrap cost per part, and almost none know the distribution of causes by value rather than by count. That is why scrap reduction programmes stall: the team attacks the biggest count and the money is somewhere else. This article gives a realistic Pareto, maps each cause onto a variable somebody actually controls, puts a real number on a reject, and lays out a 30-60-90 day plan with owners and metrics.

Pareto of Scrap Causes in a Typical Aluminium Cell

The percentages below are bands from medium-complexity aluminium cold chamber work, measured at the point of detection. Your mix will differ, but the ranking is stable enough to plan against, and the key point is in the last column: detection point, not defect type, determines the cost.

Note that the two entries with the highest value per reject, finishing reject and handling damage, are also the two least studied. Porosity gets all the metallurgical attention and finishing gets none, yet a part that reaches final polish has absorbed the full cost of the process and then some.

Mapping Every Cause to a Controllable Variable

A scrap cause is only useful if it names a knob. The table below is the mapping we use in a first-day audit: cause, the variable that actually drives it, and the measurement that tells you the variable moved.

Two practical comments. First, several causes share one variable: die temperature stability appears in cold shut, dimensional drift and shrinkage. Stabilise it and you get three reductions for one investment, which is why it is usually the first thing we recommend. Second, pour weight is the most under-instrumented variable in the whole process. A plus or minus 3 percent pour weight band directly sets biscuit thickness, which directly sets whether intensification reaches the cavity. Most plants never weigh the pour.

What a Scrapped Part Actually Costs

Price per unit is not cost per unit. The cost of a reject is the sum of everything sunk into it, plus the value of the machine time it displaced.

Now extend the same part downstream. Add CNC machining at 90 seconds on a USD 70 per hour machine: USD 1.75. Add deburring and grinding at 40 seconds of manual labour: USD 0.24 in labour plus USD 0.10 in abrasive. Add polishing at 60 seconds: USD 0.37 labour plus USD 0.18 consumables. Add inspection and packaging: USD 0.30. The part now carries about USD 8.20 of sunk cost at the moment it reaches final inspection, and if it is rejected there the full amount is lost, plus the cost of a replacement part made from scratch.

That is the arithmetic that matters. A 3 percent scrap rate measured at the machine might cost USD 0.16 per part shipped. The same 3 percent measured at final inspection costs somewhere between USD 0.25 and USD 0.90 depending on how much value has been added, and on a high-finish part such as a faucet body or a decorative door hardware plate the number is at the top of that range.

There is a second multiplier most cost models miss: a rejected part does not just consume one slot, it consumes a slot and a replacement. If the cell is capacity constrained, each reject at final inspection effectively costs two production slots, one wasted and one re-made, and the re-made part also carries its own probability of rejection.

Why Finishing-Stage Rejects Are the Most Expensive

Four reasons, in order of financial impact.

• Value accumulation. Every operation after casting adds cost that is fully lost on rejection. By the time a part reaches the buffing wheel it carries most of its total cost.
• Detection latency. A defect found at final inspection may have been created 40 minutes and six operations earlier. If the process drifted at the polishing wheel at 09:00 and inspection catches it at 11:30, everything in between is suspect, and the correction does not tell you how much of the intervening product is affected.
• Rework rarely restores full value. A buffed-through edge can sometimes be reworked, but the reworked part has different film thickness, different edge geometry and a measurable chance of failing again. Second-pass rework yield is typically 50-70 percent, so the effective cost of a reworked part is not zero, it is the rework cost plus the probability-weighted loss of the part.
• It hides the upstream cause. A finishing reject is often a casting problem that surfaced late. Subsurface porosity that never blistered in the die opens under a polishing belt. A parting line burr that grew over 40,000 shots finally becomes unremovable at the buff. If the reject is booked to finishing, the casting process never gets fixed and the rate comes back.

The corrective pattern that works is to move detection upstream wherever the cost of detection allows. Leak test after machining rather than at the end. Burr height gauged at trim rather than judged at polish. Thermal imaging of the die once a shift rather than a scrap meeting once a week.

Where Automation Removes Rejects: Gate Removal, Deburring and Polishing

This is the section where we are directly interested, because it is also where the money is. Manual finishing generates two distinct kinds of loss: rework loss, where the finish is wrong, and handling loss, where the part is damaged moving between manual stations. DZ Machinery robotic cells address both, and the mechanism is specific.

• Consistent contact force. A force-controlled floating spindle holds contact force within a narrow band regardless of part variation, so the amount of material removed per pass is repeatable. Manual grinding varies with operator fatigue, and the variation shows up as buff-through on thin sections and incomplete edge break on others. In production we typically see edge break consistency improve from a range of 0.3-0.8 mm by hand to 0.45-0.55 mm by robot.
• Repeatable path and dwell. Dwell time at a specific point is the main driver of local heat and of geometry loss on corners. A robot dwells the same way on part 1 and part 10,000.
• Fewer handoffs. Consolidating gate removal, deburring and pre-polish into one cell with a multi-station turntable removes most of the manual pick-and-place where dents and scratches happen. Every handoff is a damage opportunity; a cell that removes three handoffs removes three of them.
• Part presentation repeatability. Dedicated fixtures present the casting to the tool in a fixed relationship. Manual finishing tolerates part wobble; automation does not, and the fixture is what makes the difference. It also means that when a defect does appear, it appears at the same place on every part, which makes it diagnosable rather than random.
• Media management. Automatic abrasive change and dressing on a defined count, rather than when the operator notices, keeps the cutting action constant across the shift. Abrasive that has run 40 percent past its useful life burns the surface and work hardens the edge.
• Data. Cycle counts, force traces and consumable life are logged. When a customer’s scrap at polish climbs, the first thing we look at is whether the abrasive change interval drifted, and the log answers it in minutes.

The measurable outcomes we normally quote to customers, and the ranges we are willing to defend:

The honest caveat is that automation amplifies the input. If the incoming castings vary by more than the cell can absorb in burr height, gate remnant size or flash thickness, the robot will reproduce that variation faithfully in the finish. The first step in any finishing automation project is to measure the variation of the incoming part, not to size the robot. We cover the interaction between casting condition and finishing in how to deburr aluminum die castings with automation, and the direct comparison between manual and robotic finishing in robotic deburring versus manual deburring.

Handling Damage: The Hidden Line Item

Handling damage is under-reported because it is attributed to whatever station discovered it. Put it on the Pareto as its own line and it usually lands between 5 and 10 percent.

• Count the handoffs. Casting to bin, bin to trim, trim to bin, bin to grinding, grinding to bin, bin to polish. Each transfer is a damage event with a probability. On a six-handoff route with 1 percent damage per transfer, over 6 percent of parts arrive damaged.
• Container design is the cheapest fix. Parts should not touch parts. Pockets, dividers and a defined pitch cost little and remove most contact damage.
• Gravity drops matter. Parts dropped 300 mm onto a steel bin bottom deform. Rubber or polymer lined bins and short drop heights change the rate immediately.
• Burrs cut the next part. Deburring early in the route reduces damage downstream, which is an argument for finishing sooner rather than later.
• Ownership has to be named. If nobody owns between-station damage, everyone books it to the last station, and the last station is usually finishing.

A 30-60-90 Day Scrap Reduction Plan

This is the plan we run with customers. It is deliberately weighted towards measurement in the first 30 days, because programmes that start by changing process parameters without a baseline cannot prove anything and usually get reversed at the first production pressure.

Metrics That Must Be on the Board

Scrap counted in pieces is a vanity metric. The board should show:

• Cost of scrap per part shipped, in currency. This is the only number that captures detection point.
• Scrap by cause, by value, weekly.
• Detection point distribution, weekly. The goal is a leftward shift.
• First-pass yield at the casting machine, and separately yield through finishing. Mixing them hides everything.
• Rework rate and rework success rate, separately.
• Pour weight Cpk and biscuit thickness Cpk.
• Die temperature band compliance, percentage of shots in band.
• Finishing edge break Cpk and abrasive change compliance.
• Handling damage per 1,000 parts.
• Top three corrective actions open, with owner and promised date.

Two behavioural points. Count rework as scrap at partial cost, not as a free pass, otherwise the board will always look good. And never let a corrective action stay open more than two review cycles; if it does, the action was wrong, not slow.

For plants sourcing rather than making castings, the same logic applies to the supply agreement: specify the defect coding you expect, require the Pareto by value, and require that finishing happens on a controlled process rather than a variable manual one. Our notes on sourcing aluminum die castings from China cover what to put in that agreement.

DZ Machinery builds robotic deburring, grinding and polishing cells with force-controlled spindles, automatic tool change and multi-station turntables, plus the fixtures, dust extraction and takt time modelling that go with them. If you are running a scrap reduction programme and have measured your incoming burr and gate variation, send us the part drawings and your current reject numbers and we will model the cell against your real yield target rather than a theoretical one.