Short Shots, a Fishbone Vote and a Solution Before a Cause

Problem statement (specific, quantified, no cause, no solution): 'Short-shot scrap in the housing cell averages 4.1% on night shift versus 0.9% on day shift over six weeks (about 1,900 parts/day, target 0.5%), meaning roughly 39 defective parts per night shift; one short shot escaped to the customer in week 6. Gap to target: 3.6 percentage points on nights.' That statement names the defect, the location, the period, the magnitude and the gap. It does not name 'operator technique', which is a hypothesis, nor 'install inspection', which is a solution.

What is wrong with the fishbone:

• 31 causes is not analysis, it is brainstorming. A list of that size cannot discriminate; every cause remains equally alive.
• Dot-voting selects the most familiar or most senior-friendly hypothesis, not the most probable one. 'Operator technique' is the default blame when no data has been collected, and it conveniently locates the cause in the people with least power to object.
• The order was inverted: two fixes were made before any cause was established. The four-day dip is equally consistent with regression to the mean, with a startup effect, or with attention from being watched. It was treated as confirmation.
• No verification loop. No prediction was stated, so no result could have falsified the chosen cause.

A fishbone is a hypothesis-generation tool; the output should be a prioritized list of testable causes and the data needed to separate them.

The day/night difference is a natural experiment. Factors common to both shifts - part design, mould, tool, machine, resin specification, cycle time - are unlikely to be the cause, otherwise both shifts would fail equally. Factors that differ between shifts become the candidate set: ambient temperature and humidity, dryer dew point and resin residence/drying time (including moisture picked up after a weekend or a refill), startup conditions, staffing and experience, which machine or cavities run, and resin lot timing.

Physics supports the shortlist: short shots are a fill problem, driven by injection pressure/velocity, melt and mould temperature, venting, shot size and - critically - resin viscosity, which moisture directly affects. A dryer without dew-point monitoring is a plausible mechanism that varies by shift and season.

Data to collect before changing anything:

• Hourly dryer dew-point and temperature logs on both shifts, plotted against hourly scrap.
• Machine-recorded melt and mould temperatures and injection profiles (not operator recollection).
• Scrap by cavity, not just by cell - cavity-specific rates separate tool from process.
• Scrap versus time-into-shift, which exposes dryer saturation after a resin refill or a weekend stop.
• Resin lot numbers and arrival dates, plotted on the same run chart.

Then run a controlled trial of the leading mechanism (for example, dried-air dew point held at −20 °C on nights for two weeks) with the predicted effect stated in advance.

The inspection station is containment, not a countermeasure. It does not reduce defect generation; it adds a person, adds cycle time, and manual visual inspection of a full shift typically catches well under 100% of defects - which the customer escape already demonstrated, since inspection was in place. It also quietly legitimizes the defect as a cost of doing business.

Legitimate uses: protecting the customer while the cause is unknown, and generating data. If used, it must have an exit criterion written on day one ('remove when night shift runs ≤0.5% for four consecutive weeks with the control chart in control'), a validated detection method (challenge the inspector with known-bad parts; check agreement between inspectors), and a defined disposition for the three days of suspect stock already sitting in the bin.

Verifying the countermeasure:

• State the mechanism and predicted effect in advance - e.g. 'if dryer dew point is the cause, correcting it should bring nights to ≤0.9% within five shifts'.
• Test the prediction, don't just watch the number: run both shifts, through at least one resin lot change and one weekend startup, for three to four weeks.
• Use a control chart on hourly scrap, not weekly averages; weekly averages hid the earlier four-day dip.
• Confirm the causal link by checking the intermediate variable (dew point actually held) as well as the outcome - otherwise a coincidental seasonal change will be credited.
• Institutionalize: update the control plan and PFMEA, add dew-point monitoring to the daily checks, and only then remove the inspection station. Declaring victory on a dip is exactly what happened with the check ring.