How to Combine the Seven Basic Quality Tools: Five Stages from Finding the Root Cause to Checking Process Stability

The seven basic quality tools are not a choice of one in seven; they are ordered by the questions a problem raises. First measure the problem and find the big categories with check sheets and a Pareto chart; then list possible causes with a fishbone diagram and follow them to the root cause with 5 Whys; verify with histograms and stratification; after improvement, use control charts to see whether the process is stable, and when a point goes out of control, follow a procedure written in advance. This article lines up these five stages, explains for each tool when to use it, how to draw it and how to read it, and marks which points come from ASQ (American Society for Quality) pages, which from internal documents we hold (anonymized), and which are our own additions.

The internal material is quality documents from a packaging plant, a module plant and an SMT line, with company names, personal names and part numbers removed, keeping only methods and ranges of figures; the external material is public web pages from ASQ, NIST and ISO. How far each source was read is set out in “Limitations of this article” at the end. Figures marked “assumed” or “illustrative” were made up to demonstrate calculations. This is a methods overview; no JIAHON product is required to use these tools. Sections 7 and 8 discuss measuring solder paste thickness and reflow temperature, and related product pages are given there.

1. What question each of the seven tools answers

Q: What are the seven basic quality tools? A: Check sheet, stratification, Pareto chart, cause-and-effect (fishbone) diagram, scatter diagram, histogram and control chart. This is the list commonly used in Taiwan; ASQ’s “Seven Basic Quality Tools” page lists cause-and-effect diagram, check sheet, control chart, histogram, Pareto chart, scatter diagram and stratification — the same set. ASQ also notes that newer lists often replace stratification with a flowchart, and sometimes include a run chart, so the “seven” you see may vary slightly; that isn’t a mistake, just a different version of the list.

5 Whys is not one of the seven tools; it is a questioning method, and ASQ’s 5 Whys page says it is often used together with brainstorming or a fishbone diagram, so this article places it after the fishbone. Stratification runs through every stage rather than belonging to one. The various mnemonics in circulation are not listed here; the “seven new QC tools” (affinity diagram, relations diagram, etc.) are a different set and not covered.

Figure: Five stages, each answering one question (schematic; the order of stages is our arrangement; stratification runs through all stages).

2. Stage 1: measure the problem — check sheets and Pareto charts

Q: Which fields should a check sheet include? A: Date and time, shift, line, machine and defect code, with mutually exclusive defect codes. This is our suggestion, not an ASQ requirement. Building in these fields from the start makes stratification possible later; mutually exclusive defect codes (each defect in only one category) keep the Pareto chart from double counting. An SMT example: give tombstoning, bridging, opens and solder balls each a defect code, tally one sheet per shift per line, and the weekly Pareto chart is simply the sum of these sheets.

Q: How do you draw a Pareto chart, and how does 80/20 apply to defect categories? A: Decide the categories and the measure, total each category, and arrange the bars from largest to smallest; then add percentages and a cumulative line to see how much the top few categories account for. ASQ’s Pareto page lists the steps: choose categories, choose the measure (frequency, quantity, cost or time), set the period, collect data, subtotal each category, set the left axis scale to the largest subtotal (if adding percentages and a cumulative line, use the total of all subtotals as the top of the left axis), order the bars from tallest to shortest and combine small items into “other”; the two optional steps are calculating each category’s percentage (with a percentage axis on the right) and drawing the cumulative line, which always ends at 100%.

Below is an anonymized real example: pre-shipment inspection of a module, about 30,000 units in and about 150 defects (overall yield about 99.5%, which is calculated differently from first pass yield — see First pass yield vs final yield). Counts are rounded (to the nearest 5), and we recalculated shares and cumulative values from the rounded figures.

Figure: Pareto chart drawn from the table above; numbers on bars are counts; the top three add up to about 83% (illustrative; counts rounded).

The top three account for about 83%, close to 80/20, but that is the result for this data set; no data set lands exactly on 80/20, so go by the cumulative figure you calculate yourself. This document’s response was: for the appearance items (scratches, exposed copper, polarity), incoming sampling AQL was tightened from 0.065 to 0.04, 0.05 and 0.06 respectively, and the supplier moved to 100% inspection before shipment; the electrical test item was kept under monitoring first. In other words, the items at the front of the Pareto chart were handled first, and those at the back watched first.

How SMT lines use it: add up the weekly check sheets for tombstoning, bridging, opens and solder balls into a Pareto chart, and the top two or three are what to investigate next week. Besides counts, the measure can be rework time or scrap — tombstoning may be frequent but easy to repair, opens rarer but often scrapped, so the two orderings may differ (our example, not a measurement).

A Pareto chart is only useful as a habit. An older weekly QC report from one SMT line had four fixed blocks every week: lot rejection rate and DPPM, first-article abnormalities, patrol inspection abnormalities, and the week’s top five defects. That week, appearance inspection rejected 6 of about 600 lots (about 1.0%); the top five defects totaled 49 units, with dirt the largest at 21 (about 43%), followed by solder on gold fingers at 11 and contamination at 10. “Dirt” and “contamination” are very close as names, so it is worth checking first whether the categories are mutually exclusive (our observation; the report doesn’t say). Only by using the same categories and the same chart every week can you see which category is growing.

Q: Is the top bar on the Pareto chart always the one to handle first? A: Not necessarily — first confirm it is a real process defect. In another module test report’s defect-code Pareto chart, most of the top codes were poor contact of test probes; the document notes that after sampling, fewer than 15% were SMT-related. In that case, re-judge first to separate “real soldering defects” from “test-side misjudgements”, then stratify by defect code and machine and redraw the Pareto chart. Errors in the measurement system itself also cause misjudgements — see Measurement system analysis (Gauge R&R). A Pareto chart tells you which category is larger, not why.

3. Stage 2: list possible causes for the big category — fishbone diagrams and stratification

Q: How do you draw a fishbone diagram, and how are the main bones divided? A: State the problem clearly, draw a spine and write the problem at its end as the head; then draw the main bones as categories and, under each, ask “why does this happen” layer by layer. ASQ’s fishbone page (named after its originator, Kaoru Ishikawa) lists the steps: agree on the problem statement, draw the head and spine, add the main categories, brainstorm causes, keep asking further, fill in gaps, analyze the causes and decide which to pursue (the aim is to treat the root, not the symptom), then use other tools to assign responsibility. For the main bones, ASQ mentions the “6M” proposed by Ishikawa: Materials, Machinery, Methods, Measurement, Manpower and Mother Nature (environment); category names can be changed as needed. ASQ’s steps draw a horizontal arrow pointing right as the spine, with the problem in a box at the arrowhead, so the head is on the right.

Internal practice leans toward 4M (man, machine, material, method). One packaging plant’s OCAP specification requires that for every characteristic controlled by SPC, the process engineer leads a team to list possible causes in 4M categories; another BGA line’s OCAP list draws each abnormality as a flowchart of “possible cause → possible variable → improvement action → disposition”. The first specification requires a quarterly review using a cause-and-effect diagram plus a Pareto chart, with the flowcharts updated from the review.

A fishbone diagram lists possible causes, not root causes, as ASQ’s page also says (“identify the many possible causes”). The 8D report below, on lifted shield cans, divided causes into materials, process and design, listing each cause with owner, deadline, verification result and status; all eight causes were verified, and four are shown here:

• Coplanarity (spec under 0.1 mm): 30 units measured about 0.04–0.08 mm, plus one at 115 µm; none lifted after placement and reflow, so excluded. Even the out-of-spec unit didn’t lift, so coplanarity was not the main cause of lifting; the report also estimated the cohesive height of reflowed paste at about 150 µm, enough to absorb coplanarity under 100 µm.
• Empty can through reflow 1, 2 and 3 times: shape and coplanarity didn’t worsen, so excluded.
• Solder thickness uniformity: Cpk 1.2; the report judged it not the main cause and excluded it. In the grading in section 5 of this article this Cpk is only marginal, so “excluded” means it is not the main cause of lifting, not that solder thickness has no room for improvement (this sentence is our reading).
• Laser marking too deep: 100 units each of deep and shallow marking were compared; about 3% of the deep-mark group lifted, judged to be the main cause; our notes don’t include the shallow group’s figure, so none is given here. The countermeasure was to reduce the outsourced laser marking power by about a quarter, with a one-month audit.

This report is the second half of a fishbone diagram in action: each listed cause gets a verification experiment, and only after verification is it excluded or confirmed.

On SMT, for example solder balls after reflow, the main bones could be: materials (paste storage, thawing and time in use after opening), methods (stencil aperture design, print offset, reflow profile), machinery (squeegee, placement pressure) and measurement (how the oven temperature is measured — see section 8). This division is our example, without a source; tombstoning can be divided the same way, and each small bone still needs a verification experiment to count.

Q: Which layer should stratification split first? A: Start with the sources most likely to differ: shift, machine (or line), lot (material source) and time. This order is our suggestion, not an ASQ requirement. ASQ’s histogram page gives a typical example: a bimodal histogram often means data from two shifts or two processes are mixed together, and stratification usually finds the cause. So the data sheet needs grouping fields from the start; adding them later rarely works.

4. Stage 3: question down to the root cause — 5 Whys and root cause analysis

Q: How many times do you ask “why” in 5 Whys, and where do you stop? A: Five is a guide, not a rule; stop at the layer where “if we change this, the problem won’t happen again, and we have the authority to change it”. ASQ’s 5 Whys page says the method was developed by Sakichi Toyoda, who said that by repeating “why” five times the nature of the problem and its solution become clear (our paraphrase); the page also says the root cause may be reached in fewer or more than five. The method is to write the problem on paper, draw five lines below it, write each answer to “why” on the next line, and repeat until the root cause is found. The “where to stop” sentence is our added criterion.

An anonymized report illustrates it. One lot of a module’s pre-shipment test had a yield of about 62% (about 1,300 of about 2,100 units), while about 26 other lots of the same product averaged about 99.3%; 15 units were sampled and all had gold ball shorts. The report split the causes into an “occurrence” line and an “escape” line; the layered questioning below is our arrangement of the report’s narrative — the report itself doesn’t write out each Why.

Figure: Questioning arranged from the report’s narrative (the report did not record each layer; illustrative); the occurrence line is a causal chain, the escape line parallel defenses.

The report’s process is wire bonding, not SMT, but we judge that the structure of the questioning applies to SMT (our judgement). The escape causes are what 8D calls escape points; for writing an 8D, see How to write an 8D report. The report did two more things worth learning from: it reproduced the problem with a simulated vacuum of −320 mmHg, and used design of experiments (DOE) to find that vacuum must reach −400 mmHg or below (more negative) to be safe — proving the root cause rather than just trusting the reasoning; and the countermeasures followed both lines: quarterly checks of the lines, life set by mean time between failures (MTBF), microscope magnification raised from 45× to 90×, and vacuum abnormalities added to the OCAP.

Q: If 5 Whys ends at “operator carelessness”, have we found the root cause? A: Usually you need to ask one more layer. Another internal 8D systemic cause review sheet divides root causes into four types: rules (missing or incomplete), human error, organization and system, and people (training and certification); countermeasures pair with error-proofing, including machines, systems, tools and checklists. Human error is only one type, and usually calls for asking further: was a rule written, is there error-proofing, was training formally signed off? For writing an 8D report, see How to write an 8D report; for describing a problem precisely, see 5W2H problem description and root cause for solder defects. This article covers only the 5 Whys questioning.

5. Stage 4: verify with data — histograms and scatter diagrams

Q: How does a histogram differ from a bar chart, and how do you read its shape? A: A histogram’s horizontal axis is continuous measurements, with bars touching; a bar chart’s axis is categories. The shape tells you whether the distribution mixes two things. ASQ’s histogram page recommends collecting at least 50 continuous data points, leaving no gaps between bars, and confirming before drawing conclusions that the process was running normally during sampling, otherwise the shape may mislead. The shapes the page lists are summarized below; the third column is our addition for the first thing to check.

An SMT example: if a histogram of solder paste thickness is bimodal, first look separately by shift, stencil or printer — it is often data from two sources mixed together. ASQ says bimodal data usually come from two shifts or processes; applying this to paste thickness is our use.

A question often paired with histograms is “how do specification limits differ from control limits?” Three lines are easily confused:

• Specification limits: set by design or the buyer, to judge whether a single part is usable.
• Control limits: calculated from the process’s own data, to judge whether the process has changed — not to judge whether parts conform.
• Internal limits: set within the plant, usually narrower than the specification, leaving a buffer for process drift.

A packaging plant’s process control report for memory card singulation happens to have all three: outline specification 10.900–11.100 mm, internal range 11.000–11.040 mm, measured daily at 8:00 for 30 days, actual mean about 11.03 mm; burr internal limit 15 µm or less, measured maximum about 13 µm; mold flash internal limit 1.00 mm or less, measured about 0.63–0.78 mm. The report’s outline Cpk is far above usual requirements (the exact range isn’t given here). Working back from the mean of 11.03 and the upper specification of 11.100, σ is much smaller than the specification width, so the inferred Cpk is likewise far above usual requirements — meaning it was calculated against the wide specification limits; calculated against the narrow internal range, the figure would be much smaller (our inference; the document doesn’t say). The example shows that when quoting Cpk, first state which limits it uses. For how to calculate Cpk and what is enough, see How to calculate Cp and Cpk. The internal SPC specification grades Cpk as: below 1.00 not acceptable for production, 1.00–1.33 marginal, 1.34–1.67 capable, 1.68–2.00 very good; another automatic stencil inspection report graded an aperture area Cpk of 1.406 (sample of 125) as grade B, “1.33–1.67 acceptable”. These are those two documents’ own gradings, not general standards, and the two differ slightly (1.33 falls in different groups, the starting points 0.01 apart, possibly from rounding); public sources have no single threshold, and the Cpk article linked above lists figures from NIST and textbooks.

A scatter diagram is for when “two things seem related”, for example squeegee pressure and paste thickness. ASQ describes it as used “to analyze the possible relationship between two variables”; it only shows whether there is a relationship and cannot prove which causes which — that still needs the verification experiments of stage 3.

Below is an assumed set of paired data showing what a scatter diagram needs:

These 8 pairs have a correlation coefficient of about 0.97; the points roughly form a line rising to the right, suggesting thicker paste with more pressure. The correlation coefficient only measures linear relationships, and 8 pairs are too few; if one point is an outlier, first check whether it was a recording error or a stencil change rather than deleting it. This table is our made-up example, not measurement.

6. Stage 5: is the process stable? Control charts, and what to do when out of control

Q: How do you read a control chart? A: First look for points outside the control limits, then for patterns such as several consecutive points on one side or consecutive rises or falls; if there are any, treat them as possible special causes and find the cause first, rather than adjusting parameters straight away. ASQ’s SPC page traces control charts to Walter Shewhart (early 1920s) and distinguishes two kinds of variation: common cause variation (inherent in the process and always present) and special cause variation (from outside, meaning the process is no longer in statistical control). ASQ’s control chart page lists these rules: a single point outside the control limits; two of three points on the same side of the center line beyond 2σ; four of five points on the same side beyond 1σ; eight consecutive points on the same side of the center line (or 10 of 11, 12 of 14, 16 of 20); and obvious consistent or persistent patterns. ASQ’s SPC page also warns that the more tests you apply, the higher the chance of false alarms.

Q: How are control limits calculated, and where is the constants table? A: For an Xbar-R chart, the limits use the average of subgroup means and the average range, multiplied by the constants A2, D3 and D4. The formulas in the NIST Engineering Statistics Handbook are: Xbar chart UCL = X̿ + A2·R̄, LCL = X̿ − A2·R̄; R chart UCL = D4·R̄, LCL = D3·R̄ (X̿: average of subgroup means; R̄: average subgroup range). The constants for subgroup size n = 2–7 are copied here; the NIST table is longer:

A worked example with assumed numbers: 5 measurements per subgroup (n = 5), X̿ = 10.00, R̄ = 0.30. Xbar chart UCL = 10.00 + 0.577 × 0.30 = 10.173, LCL = 10.00 − 0.577 × 0.30 = 9.827; R chart UCL = 2.115 × 0.30 = 0.635 (0.6345 rounded), LCL = 0. If the specification is 9.50–10.50, you’ll find the control limits (9.827–10.173) are narrower than the specification. That is normal, because control limits describe the variation of subgroup means, not the range of single parts — so points within the control limits don’t mean every part is in specification, and conversely, parts in specification don’t mean the process hasn’t changed. (Process capability for these assumed figures: NIST estimates σ as about R̄ ÷ d2; for n = 5, d2 = 2.326 (from general textbooks, not listed on the page), so σ is about 0.129, the specification width of 1.00 is about 7.8σ, and Cp is about 1.29.)

For subgroups larger than about 10, the standard deviation (S chart) is often used instead of the range — a general textbook point we did not check on the ASQ or NIST pages. Many SMT measurements are one value per board, such as paste thickness, with no subgroups; then an individuals chart (X-mR) is used (general knowledge, likewise not checked). For the specification, internal and control lines for paste thickness after printing, as with the singulation example above, see How to set SPI paste thickness limits.

Figure: Examples of abnormal control chart patterns (points are illustrative, not measured data).

Control chart rules differ between sources. The table compares three sources we hold; the point is to choose one set, write it into your documents, and not keep switching on the same line. If a buyer or standard specifies a particular manual, follow it; if not, choose one yourself and write it down.

“On the bad side” is the internal specification’s practice: for smaller-is-better characteristics (such as coplanarity) only the upper side counts, and for larger-is-better characteristics (such as pull strength) only the lower side, so not both sides count. The specification also sets several things outside the chart itself that are often missed:

• New processes and equipment first have a “data collection period”, accumulating 25 points before control limits are set; the rules don’t apply during collection. ASQ’s control chart page says limits calculated from the first 20 points are provisional and should be recalculated after at least 20 consecutive in-control points; another statistics software manual says 25–30 points are needed before interpretation. Starting counts differ, so set your own.
• Control limits are reviewed monthly; recalculate if Cpk changes by more than 20% from the previous month, or if the specification’s “stability index” falls below 0.95 (we did not see how it is calculated); abnormal points with known causes are excluded from the calculation.
• Machines with Cpk below 1.33 must have an improvement plan and a published monthly report.

What you do after a point goes out of control is the real value of a control chart. The OCAP (out-of-control action plan) approach prepares a flowchart in advance for each type of abnormality and posts it on the machine; when SPC goes out of control, the operator follows the chart. Review frequency is set separately in two documents: one checks the OCAP list at least every 6 months, the other reviews with a cause-and-effect diagram and Pareto chart quarterly and updates the flowcharts — two different activities. A control plan puts “what is controlled, with what gauge, how many, how often, how it is controlled, and what to do when out of specification” in one row. For example, in one packaging plant’s control plan: dicing kerf width is a special characteristic, 5 points per shift, controlled with an Xbar-R chart linked to OCAP; wire bond pull strength, 20 wires per machine per day, Xbar-S chart linked to OCAP; DI water resistivity must exceed 12 MΩ, once per shift, and the machine stops if it fails. Visual inspection has a simpler stop condition: stop when the defect rate exceeds 1% for 3 consecutive lots, and resume only after 10 consecutive normal lots following improvement.

For using control charts on paste thickness after printing, see How to set SPI paste thickness limits; for linking control plans to failure mode analysis, see How to run a PFMEA; for errors in the measurement system itself, see Measurement system analysis (Gauge R&R).

7. Control charts on SMT: trends in paste thickness and reflow temperature

Q: Can the rising and falling run rules be used for paste thickness and oven temperature? A: Yes. Both are figures measured every shift or every day anyway; charted, they can show the process drifting before defects appear. This is our use; the ASQ, statistics software and internal specification rules above are general, and the only SMT-specific source is the paste thickness one (see the second and third paragraphs below); none covers oven temperature.

How many consecutive points make a trend should follow the document you have chosen: ASQ’s control chart page only says “obvious consistent or persistent patterns” without a count; the statistics software manual (2003) in the table above lists 6 in a row rising or falling, and the packaging plant specification (2001) a 7-point trend. A solder paste printing paper (IPC conference, 2002) read for our SPI article uses only two rules: one point beyond the control limits, and 6 in a row rising or falling. The counts differ between sources (ASQ none, software manual and printing paper 6, internal specification 7); follow the document you choose.

Paste thickness: record one representative value of SPI thickness per shift or per day (for example the board average, or values at a few fixed locations; one point per board — don’t treat many measurements on the same board as separate points). Data with one value per point use an individuals chart; several boards per shift can also form a subgroup for Xbar-R (as in section 6). If the points climb steadily or fall steadily — a rising or falling run as described above — first check for paste residue on the stencil underside and apertures, changes in squeegee pressure and speed, and whether paste thawing and time after opening follow the rules. These are the check directions we list, not conclusions; confirm them with the verification experiments of stage 3. For working back from defect shapes to causes, see How to troubleshoot solder paste printing defects. Stencil wiping is one of the items; for wiper roll materials and widths see stencil wiper rolls. For setting specification and internal limits for paste thickness, see How to set SPI paste thickness limits.

Reflow temperature: run a profile every day or at every product change, and plot peak temperature and time above liquidus (TAL) each on an individuals chart. An internal reflow oven parameter control note from one line controls conveyor speed and zone setpoints; first-article checks look only at peak temperature (guarding against overheating), with a warning at ±5 °C and an alarm at ±10 °C deviation of each zone thermocouple from setpoint; the summary notes we read don’t mention control charts, so charting peak and TAL as trends is our addition. If peak climbs, or TAL steadily lengthens or shortens, again find the cause first rather than adjusting zones straight away (our summary): conveyor speed, zone settings, heater and fan condition, and whether thermocouple wires were changed are all places to look first. Two more points (our summary): peak and TAL come from the same profile, so the two charts are not independent — when the oven drifts both move together, so don’t treat both going abnormal as two problems; and control limits are not the process window: peak and TAL have their own windows from the paste data sheet and component heat ratings, while control limits are calculated from the process’s own data; they are not the same set of numbers, and in a capable process the control limits sit inside the window.

8. The data must be measured correctly before a control chart means anything

Q: Can a reflow profile be put straight onto a control chart? A: Not recommended — first confirm the measurement is right. Thermocouple locations, attachment method and the recorder all change readings; charting wrongly measured data only leads to wrong conclusions. At minimum get three things right:

• Use a real product board as the profiling board. Its heat capacity must match production; a bare board reflects only the oven, not the joint temperature.
• Put thermocouples in the right places. One each at the hottest spot, the coolest spot and heat-sensitive components; for BGAs, measure to the center of the array.
• Attach them firmly. A loose thermocouple produces a jagged profile, and neither peak nor TAL is measured correctly.

Record with a profiler, and measure each time with the same board, the same points and the same attachment, so the points on the chart are comparable; the profiling board and thermocouple wires themselves also need periodic replacement or comparison, or the trend may be the gauge drifting (our summary). How to make a profiling board and place the points is covered in a separate article. For equipment, see the reflow profiler and type K thermocouple wires. This section is our summary; sources and details are marked item by item in that article.

9. How the five stages fit together, and common misuses

Q: Which tool when, and in what order? A: Start from what you have: with only a pile of defect records, start with a Pareto chart; if you already know which category to tackle, start with a fishbone diagram; if the process is already running, start with control charts.

Figure: Common misuses (schematic; our summary).

Common misuses, also summarized by us from the documents above and the ASQ pages:

• Pareto charts counting only frequency. The most frequent is not necessarily the costliest; ASQ also lists cost and time as measures — you can draw count and cost charts from the same data.
• Treating possible causes on the fishbone as root causes. Changing things without verification easily changes nothing.
• 5 Whys stopping at “people”. Stop at operator carelessness and the only countermeasure left is “remind them”.
• Drawing control limits with too few points. ASQ: limits from the first 20 points are provisional; internal specification: 25 points in the collection period before setting limits.
• Treating control limits as specification limits. The assumed example in section 6 shows they differ.
• Histograms mixing two shifts’ data. When bimodal, stratify first rather than calculating Cpk.
• The more rules the better. ASQ notes more rules mean more false alarms.

10. Further reading on our site

This article is an overview of the toolbox; the following each cover one topic in depth: How to write an 8D report; Cp and Cpk process capability; SPI paste thickness limits; PFMEA; First pass yield vs final yield; Gauge R&R; Go / no-go gauge principles and tolerance limits; 5W2H problem description for solder defects.

External sources: ASQ: Seven Basic Quality Tools, ASQ: Statistical Process Control, ASQ: Control Chart, ASQ: Histogram, ASQ: Pareto Chart, ASQ: Fishbone Diagram, ASQ: Five Whys, NIST Engineering Statistics Handbook: Xbar and R charts, ISO 7870-2:2023 Control charts — Part 2: Shewhart control charts.

Limitations of this article

• Internal material: a 2001 SPC specification, an out-of-control action plan (OCAP) specification, control plans, weekly QC reports, a pre-shipment defect Pareto and several 8D reports, all with company names, personal names and part numbers removed. External material: ASQ pages on the seven tools, SPC, control charts, histograms, Pareto charts, fishbone diagrams and 5 Whys; the NIST Engineering Statistics Handbook’s control chart constants; and the ISO 7870-2 standard description page.
• The Chinese list of seven is the common Taiwanese usage; we found no single official source and compared it with ASQ’s English list.
• Not all internal documents are from SMT. OCAP, control plans and the wire bonding 8D come from packaging and test processes; the weekly report and visual inspection specification come from an SMT line. We took the methods; you will need to substitute the characteristics of your own process.
• Some internal figures are inconsistent, and we didn’t pick sides. The Pareto example uses only the left-hand detail (the detail total matches total defects, while the right-hand summary’s category subtotals don’t; this article rounds further); the lifted shield can report gives coplanarity ranges in two ways (0.040–0.070 and 0.037–0.077 mm), written here as “about 0.04–0.08 mm”; the defect-code report’s summary ratios don’t match recalculation from the monthly tables, so only the qualitative conclusion “fewer than 15% SMT-related” is used.
• The SMT uses in sections 7 and 8 are our own practice. Charting paste thickness and oven temperature and the listed check directions have no source (the 6-point rule comes from the paste printing paper summarized in our SPI article; we did not read the paper itself separately); the only internal material cited is one line’s reflow oven parameter control note, read only as summary notes, not the source file. The three points and sources in section 8 are marked item by item in the profiling board article.
• Internal specifications are older. The SPC specification is the 2001 edition, OCAP and control plans about 2002–2006; current AIAG or IATF SPC requirements were not checked.
• Control chart rules are not universal. ASQ, the statistics software manual and the internal specification differ, and this article does not judge which is right.
• Example figures are assumed. The control limit example (X̿ = 10.00, R̄ = 0.30) and the abnormal pattern figure were made up for demonstration.
• External sources were read as web pages only. We read the ASQ pages in full, the formulas and constants on the NIST page, and only the description page of ISO 7870-2:2023 (2nd edition, issued 14 March 2023, 48 pages), not its text.
• “Which layer to stratify first” and “where 5 Whys stops” are our suggestions, not requirements of ASQ or any standard.

FAQ

What are the seven basic quality tools? Check sheet, stratification, Pareto chart, cause-and-effect (fishbone) diagram, scatter diagram, histogram and control chart. ASQ’s Seven Basic Quality Tools lists the same set (with stratification as the seventh) and notes that newer lists often replace stratification with a flowchart and sometimes add a run chart. The various mnemonics are not listed here.

How do you draw a fishbone diagram, and should the main bones be 4M or 6M? Write the problem clearly as the head, draw a spine, then draw the main bones as categories and ask “why does this happen” layer by layer under each. Common categories are ASQ’s 6M (materials, machinery, methods, measurement, manpower, environment); internal documents often use 4M (man, machine, material, method); names can be adapted to the process. A fishbone lists possible causes, each of which must be verified before counting as a root cause.

How does 5 Whys work — how many times, and where to stop? Five is a guide; ASQ also says it may be fewer or more. Our criterion: stop at the layer where “if we change this, the problem won’t happen again, and we have the authority to change it”; if you stop at “operator carelessness”, you usually need to ask further about rules, error-proofing and training.

How do you draw a Pareto chart, and how do you read the top items? Choose categories and a measure (count, quantity, cost or time), total each category, order the bars from largest to smallest, and add percentages and a cumulative line. The cumulative line shows how much the top categories account for; 80/20 is only a guide — go by your own cumulative figure. Also confirm the top item is a real process defect, not a measurement misjudgement.

How do you read a control chart? First check for points outside the control limits, then for patterns such as consecutive points on one side or consecutive rises or falls. ASQ’s rules include a single point beyond the limits, 2 of 3 beyond 2σ, 4 of 5 beyond 1σ and 8 in a row on one side; rules differ between sources, so choose one set, write it into your documents and don’t switch often — more rules mean more false alarms.

How are UCL and LCL calculated, and where are the A2 and D4 constants? Xbar-R chart: Xbar chart UCL = X̿ + A2·R̄, LCL = X̿ − A2·R̄; R chart UCL = D4·R̄, LCL = D3·R̄. Constants A2, D3 and D4 depend on subgroup size n; the NIST Engineering Statistics Handbook has the table (this article copies n = 2–7; the NIST table is longer); for n = 5, A2 = 0.577, D3 = 0 and D4 = 2.115. ISO 7870-2 covers the full method and must be purchased.

How do control limits differ from specification limits? Specification limits are set by design or the buyer and judge whether a single part is usable; control limits are calculated from the process’s own data and judge whether the process has changed. Control limits describe the variation of subgroup means and are usually narrower than the specification; one cannot replace the other.

How do you use control charts to watch trends in paste thickness and oven temperature? Record one representative paste thickness value per shift or per day, and for reflow, one point each for peak temperature and time above liquidus (TAL) every day or at every product change; plot them on individuals charts and look for consecutive rises, falls or runs on one side. Choose one document for how many points make a trend: the statistics software manual and a paste printing paper say 6, a packaging plant specification says 7, and ASQ gives no count. The precondition is accurate measurement: profile with a real product board, put the points in the right places and attach them firmly, or the trend can’t be trusted.

How do you read a histogram, and how does it differ from a bar chart? A histogram’s horizontal axis is continuous measurements with touching bars; a bar chart’s axis is categories. ASQ recommends at least 50 continuous data points; a bimodal shape often means two shifts or processes are mixed — stratify first; a heart-cut shape often means only accepted parts’ data were used.

What is stratification, and what should be split first? Stratification groups data by shift, machine, lot or source to see whether the groups differ. Split first by the sources most likely to differ: shift, machine (line), lot (material source) and time. This is our suggestion; data sheets need grouping fields from the start, as adding them later rarely works.

In what order should the seven tools be used? With only a pile of defect records, start with a Pareto chart; if you know which category to tackle, start with a fishbone diagram and 5 Whys; to verify an improvement, compare histograms before and after; for a process in stable mass production, keep watching with control charts and write the out-of-control procedure (OCAP) in advance.

Have a similar problem? Write to JIAHON or email [email protected] to see whether there is a matching solution. We offer quick online consultations — one more person to think it through with — and will do our part to help solve the problem.