Measurement data: The microscope of the process in continuous numerical values
Measurement data refers to continuous numerical values obtained through precise measuring tools. Its core characteristic is "infinitely divisible" - for example, the diameter of a shaft in machining (10.02mm, 10.03mm), the weight of an injection-molded part (25.1g, 25.2g), and the temperature of an electronic product (45.5°C, 45.6°C). These data are like an "electrocardiogram of the process" and can capture fluctuations as small as 0.01mm. For instance, when the shaft diameter gradually changes from 10.00mm to 10.05mm due to tool wear, measurement data can directly reflect this trend, which is precisely the core requirement of SPC (Statistical Process Control) - to provide early warnings of process abnormalities through the analysis of fluctuations in continuous data.
For machined products in continuous batch production (such as 1000 shafts per day), measurement data is the golden raw material for SPC. For example, use the X - R control chart (mean - range chart): Take 5 shafts from each batch, measure the shaft diameters, then calculate the mean (X) and range (R), and draw the control chart. If the points fall within the control limits and there is no abnormal trend (such as 7 consecutive rising points), it indicates that the process is stable. If the points exceed the control limits, immediately conduct investigations (such as tool wear, changes in raw material hardness). This ability to prevent problems before they occur can only be achieved with measurement data, because it can reflect the gradual change of the process, rather than waiting until non - conforming products appear to take action.
Counting data: The "efficient judge" in discrete scenarios
Counting data refers to discrete and non - continuous results. The core is to "count the quantity" or "make a binary judgment" — it cannot be described by continuous numerical values and can only be obtained through "comparing with standards" or "checking for defects". Common scenarios are divided into three categories:
1. Qualified/unqualified judgment: Measure the holes with go - no go gauges (it is qualified if the go gauge can enter and the no - go gauge cannot enter), and conduct tooling positioning comparison (it is qualified if it is consistent with the standard part).
2. Defect number statistics: The number of scratches and burrs in the visual inspection (for example, a casting has 3 scratches).
3. Defects per unit: The number of defective points per unit area/length (such as the number of rust spots on each square meter of steel plate, the number of insulation damages on each meter of cable).
The "application areas" of count data are scenarios where measurement is difficult or the cost is too high. For example, internal defects in large castings (ultrasonic flaw detection can only determine "crack presence/absence" and cannot measure the crack length), destructive experiments (testing the material strength will damage the sample, and one can only count "qualified/unqualified"), and appearance defects (the length of scratches is not important; what matters is "how many places there are").
For this type of scenario, SPC uses discrete data tools: such as the P chart (proportion of non - conforming products, which analyzes the proportion of non - conforming products per day) and the C chart (number of defects, which analyzes the total number of defects in each batch of products). These tools do not require continuous numerical values and can be used as long as the "quantity" can be counted, thus solving the problems of "unmeasurable" or "too expensive to measure". For example, use the P chart to analyze the non - conforming rate of go - no - go gauges. If the upper limit rises from 2% to 5% on a certain day, it indicates that the tooling may be worn and needs to be calibrated immediately.
The dilemma of small-batch production: When SPC "fails"
A friend's company produces large-scale wind power equipment, with a monthly output of only 2 - 5 units. Each wind turbine consists of large components such as the tower, blades, and generator, and has a high degree of customization (for example, different customers require different tower heights), resulting in extremely high measurement costs (for example, the aerodynamic performance test of the blades requires a dedicated site, and each unit can only be tested once). At this time, the conventional SPC completely "fails": SPC requires at least 20 - 30 consecutive samples to calculate the control limits, and a monthly output of 5 units simply cannot meet the sample size requirement. The control chart will turn into a "random dot plot", making it impossible to judge the process stability.
What should be done? The answer is product review - the "individual evaluation method" for small - batch, high - value products. Use the scoring system to convert "qualitative defects" into "quantitative scores" to replace the batch data analysis of SPC.
Product review: The "quality check-up form" for small-batch scenarios
The core of product review is to assign a quality score to each product. By deducting points weighted by the "severity of defects", the question of "whether there are defects" is transformed into "how serious the defects are". The specific rules are as follows:
A total score of 100 points: represents "perfect without defects";
Severe non-conformities: Defects that affect the function or safety (such as cracks in the blade structure and insulation failure of the generator), with 10 points deducted for each case. This type of defect will cause the wind turbine to fail to operate or lead to safety accidents, and "zero tolerance must be exercised".
General non-conformities: Defects that affect performance or lifespan but do not endanger safety (such as scratches on the tower coating, insufficient bolt torque), with a deduction of 5 points for each occurrence. This type of defect will reduce the product lifespan (for example, scratches lead to coating peeling and tower corrosion), but will not cause immediate failure.
Minor non-conformities: Appearance or label defects that do not affect the function (such as misaligned logo stickers or missing instruction manuals), deduct 1 point for each case - such defects only affect the aesthetics or user experience.
Calculation logic: The score of each product = 100 - (the number of serious defects × 10 + the number of general defects × 5 + the number of minor defects × 1). For example, if a certain wind turbine has 1 serious defect (blade crack), 2 general defects (coating scratches), and 3 minor defects (misaligned logo stickers), the score is 100 - (1 × 10 + 2 × 5 + 3 × 1) = 77 points. The lower the score, the more serious the quality problem.
From scoring to improvement: Grasp the vital few with the Pareto chart
For small-batch products, "scoring each unit" is the key. If the monthly production is 3 units, 3 scores can be obtained; if the monthly production is 5 units, 5 scores can be obtained. Organize these scores and the corresponding defect records into a monthly report, and conduct an analysis using the Pareto chart:
- The horizontal axis represents "defect types" (serious defects: blade cracks, generator insulation failure; general defects: coating scratches, insufficient bolt torque; minor defects: misaligned trademark stickers, missing instruction manuals).
- The vertical axis represents the "deduction percentage" (for example, 20 points are deducted for blade cracks, accounting for 40% of the total deduction; 15 points are deducted for coating scratches, accounting for 30%).
The core of Plato's principle is the "20/80 principle" - 20% of the defects lead to 80% of the deduction of points. For example, the monthly report of a friend's company shows that blade cracks (serious defects) account for 45% of the total deduction of points, and coating scratches (ordinary defects) account for 30%. Together, they account for 75% - these are the "vital few" defects and should be resolved as a priority.
- For blade cracks: Check whether the mold is worn (which may cause cracks during blade forming) and whether there are impurities in the raw materials.
- For coating scratches: Optimize the tower barrel handling process (replace steel ropes with soft slings to reduce scratches).
Logic for selecting data types
Measurement and attribute types are not an "either-or" choice but are selected based on process characteristics and measurement capabilities.
- Capable of measuring and in need of measuring subtle changes → Measurement type (e.g., machined shaft diameter);
- Unable to measure, too costly to measure, or only need to judge pass/fail → Attribute type (e.g., appearance defects);
- Small batches, high value → Product review (Quantify the quality of each unit using a scoring system and focus on improvement points using a Pareto chart).
In essence, the core of all quality tools is to "replace experience with data" – whether it is continuous numerical values of the measurement type, discrete quantities of the counting type, or scores from product audits. Ultimately, it is all about "making quality visible" and providing a direction for improvement. For small - batch production, product audit is not a "second - best option" but the optimal solution for "precisely matching the scenario". After all, quality control that can cover each product is more practical than SPC which "accumulates sample size".