DHU in Garments Explained: Formula, Benchmarks and Why It's a Profit Metric (Part 3)

A report that says “defects were high today” leaves a production team with little to act on. How many garments were checked? What counted as a defect? Did the faults occur on many pieces or on a few pieces with several problems?

DHU answers one part of that question. In garment manufacturing, it means defects per hundred units. It expresses the number of faults found relative to the number of garments inspected. Used with a consistent inspection method, it helps a Bangladesh RMG factory locate rework, compare similar production periods and test whether a change has helped.

DHU is also a useful starting point for a cost discussion. It does not measure profit directly, and a lower number does not by itself prove that a shipment is acceptable. Its value comes from connecting a reliable count with the faults, causes and costs behind it.

The DHU formula counts faults, not faulty garments

DHU = total defects found ÷ total garments inspected × 100.

Suppose an inspector checks 250 garments and finds 35 defects on 20 garments. Several pieces have more than one fault.

DHU = 35 ÷ 250 × 100 = 14.

The share of garments with at least one recorded fault is different:

Faulty garment percentage = 20 ÷ 250 × 100 = 8%.

Both results describe the same inspection. DHU is fourteen defects per hundred garments. The faulty-piece result is eight affected garments per hundred inspected. Neither result alone predicts the percentage a customer will reject.

NIST's guide to attribute charts distinguishes counts of faults from proportions of nonconforming units. The distinction matters because one garment can carry an open seam, a wrong label and a stain. That is one affected garment with three recorded defects under a rule that counts each separately.

DHU can exceed 100. If 100 garments contain 140 defined faults, DHU is 140. The result is possible because defects are not limited to one per garment. Describing DHU as a percentage of defective pieces would therefore be wrong.

Decide exactly what the inspection counts

The formula is simple; the counting rules need care. An inspector checking only a sleeve attachment cannot be compared directly with someone checking the entire finished garment. The second inspection has more opportunities to find faults.

A useful data definition states the garment or component being inspected, the process stage and the required checks. It also states whether repeated faults of one type on a piece count separately. Follow the buyer's rules where they apply, and use a consistent internal definition for process monitoring.

An approved photo or physical defect sample can help inspectors interpret a rule consistently. Borderline appearance faults need an agreed reference, especially when different shifts judge them. The aim is agreement with the requirement, not agreement achieved by ignoring difficult cases.

Before starting a new report, ask two inspectors to assess the same selected pieces independently. Discuss the differences and clarify the standard. Repeat that exercise when the product, inspection method or defect catalogue changes. A dashboard cannot correct inconsistent judgement after the records have been entered.

Keep first-pass defects separate from repair checks

Imagine a garment fails at end-line, returns for repair and fails again because the repair was incomplete. The second visit is relevant evidence about repair quality. It should not quietly become another newly produced garment in the first-pass denominator.

Keep a first inspection record and link later checks to it. The system can then show defects found initially, repairs attempted, repeat repairs and final disposition. Without that link, a busy repair loop can inflate inspected quantity and make the apparent DHU improve.

The reverse error is possible too. Recording the same unresolved fault at each visit as a fresh first-pass fault inflates the numerator. Both mistakes obscure what the line originally produced.

If pieces move through inline, end-line and finishing checks, each stage can have its own rate. Adding those rates does not create a valid factory DHU. Some garments have been inspected more than once, and the scope of each check may differ. Define a common reporting boundary before combining results.

The same discipline supports right-first-time measurement. In a hypothetical group of 500 garments, suppose 50 need unplanned repair and those pieces contain 60 faults. First-pass success is 450 ÷ 500, or 90%, while DHU is 12. Subtracting DHU from 100 would give 88%, which is not the first-pass result.

Combine groups using their counts

Factory totals should reflect how many units each group contributed. Averaging displayed DHU rates gives small samples too much weight unless all sample sizes are equal.

The correct combined rate is 50 ÷ 1,000 × 100 = 5. Averaging 20 and 3.33 gives about 11.67, which misrepresents the inspected population.

This calculation is valid only when the groups use a compatible inspection basis. A weighted result still cannot make unlike products or checks comparable. Keep the combined figure for the overall view, while retaining the separate groups for diagnosis.

A zero denominator also needs explicit treatment. If no garments were inspected, DHU is unavailable. It is not zero. Showing zero would falsely suggest a defect-free inspection where no inspection occurred.

A changing style mix can move the total

Consider two hypothetical groups that keep the same individual rates. Basic styles have DHU 4, while more complex styles have DHU 20 under the stated inspection rules. This is an example of different work, not a claim that complexity excuses defects.

The combined rate more than doubles even though neither group's rate changes. The factory inspects a larger share of complex work in the second period. Blaming a general decline in workmanship would miss the reason for this movement.

The total still matters because it describes the observed fault burden. For a process comparison, also show each style group's rate. This allows management to distinguish a change in production mix from a change within a process. It also prevents a shift toward easier work from being reported as proof that a corrective action succeeded.

Keep detection time and production time distinct

A finishing inspector may find faults today on garments sewn yesterday. If the report links them only to today's sewing output, the denominator and suspected production conditions may be wrong. Preserve the inspection date, but link the finding to the relevant production batch where records permit.

When exact timing is unknown, say so. A broad batch link is more honest than an invented hourly assignment. Good traceability narrows the search for a cause without claiming detail the factory never recorded.

What is a good DHU in garments?

There is no single DHU band that establishes acceptable performance for every garment factory. A basic T-shirt and a tailored jacket differ in construction, work content and inspection opportunities. Buyer requirements and the point of inspection can differ as well.

The following bands are illustrative discussion prompts. They are not verified industry benchmarks, contractual acceptance limits or a substitute for a buyer's quality manual.

A factory can set a more useful target from its own stable, comparable history and the improvement it intends to make. State the baseline period and inspection coverage beside the target. If an external benchmark is used, establish that its garment mix and counting method match.

Low DHU does not remove the need to respond to a serious fault. A small number of safety or compliance failures may matter more than many minor appearance defects. Likewise, high DHU may be concentrated on a few garments with several repairable faults.

Keep defect severity and the observed share of faulty pieces beside the total rate. Shipment release follows the agreed inspection plan and product requirements. DHU is a process measure; it is not an AQL acceptance table.

Follow the trend without treating every change as a crisis

A rate can rise because the process changed, but it can also move through sampling variation. A new inspector, wider coverage or a harder style can change the observed number without the same change in underlying workmanship.

Plot DHU over time with sample size and relevant production events. Note a style change, machine repair, fabric-lot switch or revision to the inspection method. These notes help the team form testable explanations rather than stories built after seeing the result.

NIST's counts-chart guidance explains the statistical basis for charts of nonconformities. A u chart can monitor defects per unit when inspected quantities vary. Minitab's u-chart documentation uses the defect rate and sample size to construct the chart.

For illustration, assume a suitable stable history establishes a mean of 0.08 defects per garment, or DHU 8. For a subgroup of 500 comparable garments, the usual three-sigma limits on the DHU scale are:

Upper limit = 100 × [0.08 + 3 × square root of (0.08 ÷ 500)] ≈ 11.79.

Lower limit = 100 × [0.08 − 3 × square root of (0.08 ÷ 500)] ≈ 4.21.

A result of DHU 12 would trigger investigation under that model. These limits were not estimated from only two weekly results. They assume an appropriate historical basis and count behaviour consistent with the model.

Faults may cluster by bundle, machine or material lot. Such grouping can produce more variation than the basic model expects. Minitab's diagnostic guidance explains why this can raise false alarms in a traditional u chart. Review the grouping and chart choice before interpreting every signal as a new cause.

Control limits describe process behaviour; they do not grant permission to ship. The practical SPC guide develops that distinction further.

Worked example: trace a finishing-floor increase

Consider an illustrative weekly finishing report. Inspectors check 500 garments and record 60 faults, giving DHU 12. The previous comparable result was DHU 8. That rise merits attention, but the two numbers alone do not establish a lasting process shift.

After examining the faults, the team assigns each to one likely source stage. The location where a fault was found remains recorded separately.

The 38 sewing faults are about 63% of the total. They were found in finishing, so assigning all sixty to the finishing department would send improvement work to the wrong place.

A source assignment is still a hypothesis until supported. Measurement errors may arise from cutting, sewing, washing or the measurement method itself. Trace the affected pieces and compare intermediate checks before declaring which stage caused the problem.

Suppose skipped stitches form one major group. Coats' sewing-problem guide lists possible causes involving timing, needle selection, fabric control and thread-loop formation. The fault name alone cannot show that an operator needs retraining. A technician and operator should test the relevant setup and material conditions together.

Where the cause lies in cutting, review the RMG cutting process. Where suspect bundles continue to move, review the approval status at the cutting-to-sewing handoff. A limit on bundle quantity helps flow, but it does not replace quality release. See also Wilson confidence intervals for defect rates.

Translate the fault count into a repair-cost estimate

The economic effect depends on what must be done to recover the work. Removing a loose thread and replacing a damaged panel cannot reasonably have the same cost per defect.

For a further teaching example, assume all sixty faults above can be repaired. The following times are invented planning inputs that would need measurement in a real factory.

Assume a matching repair cost rate of BDT 5 per minute. Repair time then costs 194 × 5 = BDT 970. Add BDT 60 of extra material and 25 minutes of reinspection at the same assumed rate, or BDT 125.

Total recorded recovery cost = 970 + 60 + 125 = BDT 1,155.

Recovery cost per inspected garment = 1,155 ÷ 500 = BDT 2.31.

The BDT 5 rate is a teaching assumption, not a wage claim or factory benchmark. Its cost scope must match the work measured. ApparelDiary's cost-per-minute and CM guide explains why different minute bases produce different rates.

This observed cost should not be projected across all production if the inspected group was chosen because it looked risky. A representative sample and a consistent cost method are needed for a wider estimate. Even then, show the uncertainty around the expected repair mix.

Distinguish cost, capacity and profit

ASQ's cost-of-quality framework separates prevention, appraisal and failure costs. Routine inspection is an appraisal activity. Repair before delivery is an internal failure cost; a customer claim arises after delivery. Keeping these categories visible helps explain what an improvement actually changes.

If fewer repairs free 194 minutes, the factory gains available time. Cash savings depend on whether overtime, outside work or another expense falls. If the released time is used for extra saleable output, value that opportunity using the relevant contribution, not automatically the full sales price.

Do not count the same labour twice. A fully loaded repair rate may already include some costs that a separate lost-output estimate also assumes. Finance and production should agree the boundary before presenting the total as a profit gain.

Rank faults by count and by repair time or cost. ASQ's Pareto guidance allows frequency, money or time as the measure. A common fault may dominate the count while a less common fault consumes most recovery capacity. Serious safety and compliance risks need their own priority regardless of rank.

Turn the daily rate into a useful response

The morning review should connect a signal with a specific action. In the finishing example, the team first contains affected work, checks the suspected source and tests a correction. It then inspects later comparable output to see whether the fault has fallen.

Record the action date and what changed. If several settings change at once, it may be difficult to tell which one helped. A practical trial should be controlled enough to learn from without delaying necessary containment.

The factory KPI dashboard guide is useful for assigning ownership and review timing. Alongside DHU, display inspected quantity, faulty pieces, leading fault types, repair minutes and overdue actions. Avoid a large collection of rates that no one uses.

An honest rise in recorded DHU can follow better inspection. Do not punish the team for revealing a problem that was previously missed. Confirm the inspection basis, establish the new baseline and work on the causes. A low number achieved by undercounting leaves both the quality risk and the cost in the factory.

Part 4 examines how to reduce repeat faults through quality at the source and lean QC. DHU makes the problem visible. Sustained improvement depends on what the team learns and changes after the count.

References

• NIST: Attributes Control Charts.

• NIST: Counts Control Charts.

• Minitab: U Chart Methods and Formulas.

• Minitab: U Chart Diagnostic.

• Coats: Solutions to Sewing Problems.

• ASQ: Cost of Quality.

• ASQ: Pareto Charts.