Zero Defect Manufacturing and Lean QC in Bangladesh Garment Factories (Part 4)

A falling DHU can be welcome news without being evidence of a better process. The style may be easier. Inspectors may have checked fewer difficult operations. A repair team may have removed faults before anyone counted them. A useful improvement starts by finding out which of these stories the figures tell.

Zero defect manufacturing sets an aim: prevent faulty work from being made and passed on. It does not certify that every garment is perfect. Lean QC supports that aim by removing wasted effort. It keeps the checks that protect the order. Both ideas depend on what happens when a real fault interrupts a busy shift.

This final part of the Bangladesh RMG quality series focuses on daily practice. It builds on QA and QC, buyer audits and DHU. The examples are hypothetical. Their purpose is to show how to test a change, rather than promise a particular result.

Start with a clear definition of improvement

Suppose a line aims to reduce DHU from 15 to 5. DHU means defects per hundred units checked. If both periods cover 10,000 first-pass garments, these rates represent 1,500 and 500 defects. The reduction is 1,000 defects, or two-thirds of the starting count. The rate falls by ten defects per hundred units.

That calculation does not tell us how many garments were faulty. One garment can contain several defects. Nor does it prove that the line prevented 1,000 defects if the inspection coverage changed. The DHU guide explains this difference between defect counts and defective pieces.

Before a trial, agree on the operation, defect codes and first-pass counting point. Record the style, material and important changes in work content. Keep repaired pieces separate from new inspection entries. Do not count one fault again each time the garment returns.

A target should help the team choose work. It should never make reporting a fault feel like failing the target. The rate may fall after staff are warned about poor numbers. First ask whether the checking and records stayed sound.

Give operators a practical way to stop and signal

Toyota describes jidoka as detecting an abnormal condition and stopping. This helps prevent defects from moving through the process. Its production-system explanation includes both equipment stops and an operator calling for help. A garment factory can apply the principle without copying an automotive assembly line.

Consider an operator who finds a skipped stitch while attaching a collar. The response should be clear and local. Stop the affected task under the trained method, hold the piece and call for help. The team then checks how far back the problem might extend. A defined rule decides whether the hold must cover a bundle, machine, operation or wider area.

The signal needs a response owner. A red tag that remains unnoticed for an hour is not an effective control. The line should know who provides help and approves restart. Name a backup for an absent supervisor. Name who can release held work. Restarting the machine does not make earlier pieces acceptable.

The response can be simple, but it must cover the whole event.

Good-faith reporting must be supported when delivery pressure is high. Otherwise, operators learn that the written stop rule matters less than the hourly output target. Managers can show support by reviewing the response to a signal. Blaming the person who raised it can make the next concern harder to report.

A quality stop is not a safe maintenance procedure

An operational stop may leave a machine connected to power or stored energy. The operator must not reach into moving parts or defeat a guard. Repairs must stay within their training and approved duties.

The ILO machinery safety code covers safe use and maintenance. It also addresses isolation of energy sources. It is guidance. Local legal duties and the machine's approved instructions still apply. The factory needs trained staff and a safe procedure suited to the actual equipment.

This distinction belongs in quality training. An operator may correctly identify a seam problem while a mechanic carries out the adjustment. The quality team confirms the sewing result after safe service. Urgency does not make the service precautions optional.

Remove duplicate work before removing useful checks

Lean QC starts by watching how an inspection record moves. In some systems, one finding is written at the line, copied into a register and entered again for a report. Each transfer takes time and can change a code, quantity or order number.

A controlled entry may replace these copies if it preserves the information that users need. Link the finding to its order and stage. Record the date, quantity and follow-up owner. Corrections need a clear history. A digital form is useful only when staff can enter and retrieve reliable data during normal work.

The same review can expose travel and waiting. An inspector may walk across the floor to obtain the approved sample. A held bundle may wait because nobody knows who can release it. The sample could move to a controlled point near the task. A release deputy could reduce waiting. Either may help more than asking for faster checks.

Required checks still apply. The buyer's control plan, product risks and process history determine what must be checked. Replacing an independent inspection with an operator's self-check changes the control. It needs a sound reason and any required approval.

Measure the inspection queue

A slow inspection point can delay dispatch and encourage hurried checking. The answer depends on the workload, not on how busy the station appears.

Take an illustrative station receiving 80 garments an hour. Each garment needs 45 seconds of checking and 15 seconds of recording. With 60 usable minutes and no interruptions, its capacity is:

Capacity = 3,600 seconds ÷ 60 seconds per garment = 60 garments per hour.

The queue grows by 20 garments an hour. If a better record removes ten seconds of duplicate entry, the cycle becomes 50 seconds. Capacity rises to 72 garments an hour, but the queue still grows by eight. The paperwork change helps; it does not solve the capacity gap.

Real staffing decisions must also allow for breaks, difficult pieces, repairs and uneven arrivals. Time a reasonable mix of styles and conditions before reallocating work. Options may include moving a suitable task, adding trained cover or changing the flow of bundles. None requires silently dropping a required check.

This connects quality work with line balancing and controlled work in progress. Give the queue an owner and a response threshold. Hidden waiting should not become the normal way of meeting output targets.

Choose the first improvement with care

A count of faults can help narrow the work, but frequency is not the only reason to act. A rare safety concern may need urgent control. So may a wrong-market label, even if cosmetic faults occur more often. Use the buyer's defect rules and the product risk when setting priorities.

For a routine improvement project, choose a fault with a clear definition and enough recent evidence to study. Keep its location and operation visible in the record. “Seam defect” may combine several problems that need different fixes. Separating skipped stitches from open seams gives the team a more useful question.

Keep the first trial small enough to control. It must still cover normal changes in the chosen task. State who will run it, which work it covers and what would cause the trial to stop. Hold trial output until its required checks are complete. A promising method still needs approval before it becomes the normal method.

It also helps to record the cost of the change. A new guide may reduce repair time but need setup and maintenance. Compare those effects with the same work scope. Time freed for other work is useful capacity; it becomes a cash saving only when spending falls. This keeps a quality project from claiming a financial result that the accounts cannot support.

Teach the operation and its likely failures

Operators need more than a list of defect names. A collar-attach operator must see what an acceptable seam looks like, where to check it and what to do when it fails. Approved samples, clear limits and real practice are more useful than a lecture followed by a signature.

Training should include borderline examples where judgment is hard. Two trained people may judge a sample differently. The acceptance rule may need clearer words or a better reference. The disagreement should be resolved before a bulk order turns it into a larger dispute.

A trainer can observe the operator under normal conditions and ask them to show the response to a fault. This checks whether the skill transfers to work. A signed attendance sheet shows who attended; it does not show who can apply the method.

Self-checks place an early control near the source of a fault. They work best when the task is clear and the response is quick. Independent checks can then provide a separate view of the result, within the approved inspection system.

Use cause analysis to test explanations

Repair answers, “How do we make this piece acceptable?” Cause analysis asks, “What allowed this fault to occur, and what would stop a repeat?” Both may be needed, but they are different tasks.

The 5 Whys method helps a team look beyond the first explanation. Five is a guide, not a compulsory count. A fishbone diagram groups possible causes. These may involve material, machine, method, people, measurement or the work environment.

Neither tool turns a suggestion into a fact. If the team suspects a material change, compare the lot records. Inspect the material as well. For a suspect setting, compare the actual condition with its approved range. Use a safe method for the check. More than one cause may be active.

A 2023 Bangladesh knit garment study examined five styles at one export factory. It combined inspection data with staff discussion and cause analysis. It proposed remedies for the main defects found. That supports using both floor evidence and staff knowledge. It does not prove that the proposed fixes reduced defects. Another factory may have a different fault pattern.

Keep the observed fault separate from its proposed cause. Then assess the evidence for that cause. “Operator error” alone usually leaves too much unexplained to guide a lasting change.

Two cause reviews from receipt to production

The following cases are examples, not findings about a named factory. Each proposed cause needs evidence before it is accepted.

Wrong-market labels

Start by tracing a failed piece to the bundle, label issue and storage location. Compare the approved label with the actual item. A mixed carton is one possible cause. A wrong instruction or approved sample could produce a similar result.

If mixing is confirmed, separation should continue from receipt through packing. Clearly marked shelves may help. They fail if staff later mix both stocks in one unmarked container. The revised method needs clear identification at every transfer where the mix could recur.

A handler's check may form part of containment. Test the new storage and issue method during shift changes. Include returns and small balance quantities in the review. These less tidy conditions often reveal weaknesses that a demonstration misses.

Late trim approval

Check when the request arrived and whether it was complete. Compare its review date with the required release date. An absent approver may explain the delay. Incomplete artwork or a pending buyer decision could also play a part.

A trained deputy helps only if that person has the authority and information needed to act. Define which decisions the deputy can make and which need buyer approval. The system should reveal an ageing request before the trim becomes critical to the line.

A three-week allowance before planned output may suit an example schedule. It is not a universal lead time. Set the real allowance from supplier lead times, approval history, transport and the order's risks. Extra stock cannot compensate for an unresolved specification.

Make errors harder to create or easier to detect

ASQ's guidance on mistake proofing describes ways to prevent an error or make it immediately visible. In garments, a guide may prevent wrong placement. Other controls separate similar trims or match a label to the right order.

Choose the device for the actual failure. A guide may locate a pocket but cannot prove that the fabric shade is correct. A barcode match may confirm the codes in the system while missing a label that was printed with the wrong text. The control's limits should be explicit.

Before relying on it, challenge the method with known wrong cases. Try the wrong market, wrong size, missing label and an unreadable code where those risks apply. Use controlled samples so the challenge cannot enter saleable stock. Record which errors the check detects and what happens when it cannot decide.

An alarm also needs a response. If staff can ignore it and continue without a recorded decision, it offers less control than the display suggests. A pilot should examine normal work, awkward cases and recovery after a stop.

Prove the change worked, then watch it in use

Agree on the review measure before changing the process. For label mixing, track wrong-market labels at the same checking point. Record how many pieces were checked. Total DHU is useful context, but it may hide the target fault among unrelated seam or finishing defects.

Use a comparable mix of work where possible. Record changes in style, staff, material or checking. Reviewers can then judge other causes. Several sound shifts give more evidence than one closely supported trial. No fixed number of shifts suits every risk.

Lot acceptance remains a separate decision. NIST's acceptance-sampling guidance explains the role of lot acceptance. It differs from estimating quality or controlling the process. Passing the buyer's agreed AQL plan does not prove that a root cause has disappeared.

Maintenance belongs in the control plan where wear or drift affects the result. Use the approved checks, service limits and measurement methods for the equipment. A lower defect count alone cannot show that a maintenance action caused the improvement.

Build a daily quality habit

A short review can connect the day's faults with the work needed next. Discuss the main new risk and any held work. Review overdue actions and results from recent trials. Respond to urgent faults at once. Routine reports should help the same shift.

Follow-up matters more than the number of completed forms. An action remains open when the new method has not been tested, staff cannot use it or the target fault returns. A useful closure record explains what changed, what was checked and why the evidence supports closure.

This completes the four-part series. It covers the quality system, buyer review, defect measures and the work of reducing faults. The daily work is modest but demanding. Staff need clear authority to signal problems and a reliable response. They also need time to check whether a fix lasts.

See also Wilson confidence intervals for defect rates and buyer audit preparation.

References

• Toyota: Toyota Production System and jidoka.

• ILO: Safety and health in the use of machinery.

• ASQ: Five Whys.

• ASQ: Fishbone diagram.

• ASQ: Mistake proofing.

• Moin, Hossain and Baral: quality defect analysis in a Bangladesh garment factory, Textile & Leather Review, 2023.

• NIST: What is acceptance sampling?.