Core Concepts and Implementation Logic of Statistical Process Control (SPC)
I. Starting from the analysis of control charts: Why is it necessary to reinforce the foundation?
Recently, when the team was discussing a certain control chart, many people had a vague understanding of concepts such as "sources of variation" and "steady-state judgment" - this is actually the entry threshold for statistical process control (SPC). Coincidentally, Chapter V (Analysis of Process Variation) and Chapter VIII (Application of Control Charts) of the Red Book of "Management" systematically organize these contents. Take this opportunity to string together the scattered basic concepts into a logical chain to facilitate everyone's reference in practice.
II. The essence of fluctuations: The boundary between accidental factors and special factors
The outputs of all processes (such as product dimensions and non-conformity rates) will fluctuate. The root lies in two types of influencing factors:
Random factors: The inherent "background noise" of the process itself, such as the slight vibration during machine operation, the minor compositional differences in raw materials, and the small action deviations of operators. The characteristics of this type of factors are as follows: The individual impact is extremely weak, their occurrence is irregular, and they cannot be completely eliminated. However, the overall fluctuation range is small, and they are considered "acceptable normal fluctuations" in engineering by default.
Special factors (abnormal factors): "Interference sources" introduced from the outside, such as precision deviations caused by damaged machine parts, unqualified raw material batches, and misoperations by operators. The characteristics of this type of factors are: large influence amplitude, clear causal relationship, and being identifiable and eliminable - they are the key "targets to be eliminated" by SPC.
Simply put: Random factors are unalterable, while special factors are must - be - altered. Distinguishing between the two is the core goal of SPC.
III. The inevitability of fluctuations: It is "abnormal" when there are no fluctuations
Many people mistakenly think that "stable quality = no fluctuations", which is a typical misunderstanding. Fluctuation is an inherent attribute of the process
1. Randomness of factors: There are countless tiny influencing factors in the process (such as a 0.1℃ change in temperature and slight wear of the cutting tool). The occurrence time and influence degree of each factor are unpredictable, but they will jointly act on the output.
2. Falsehood of no fluctuation: If there is absolutely no fluctuation in the quality characteristics of a certain process, it is either due to data falsification (the results have been manually modified) or the resolution of the measuring instrument is too low (for example, using a ruler to measure micron - level dimensions, which simply cannot capture the fluctuations).
3. Controllability of fluctuations: It is impossible to completely eliminate fluctuations. However, by eliminating special factors, the amplitude of fluctuations can be compressed to the "level of random factors" — this is the value of SPC: to reduce fluctuations rather than eliminate them.
IV. Sources of fluctuations: Full coverage of 5M1E
The specific factors affecting quality fluctuations can be summarized using the "5M1E" framework (man, machine, material, method, measurement, environment):
Man: Differences in the skills and fatigue levels of operators.
Machine: The precision and wear condition of the machine.
Material: Differences in the composition and batches of raw materials;
Method: Setting of process parameters and execution of operation procedures;
Measurement: The accuracy and calibration status of measuring instruments;
Environment: Temperature, humidity, and dust concentration in the workshop.
Among these six categories of factors, there are both accidental factors (such as slight fatigue of operators) and special factors (such as damage to machine parts). The key to judgment is: Is the factor inherent to the process? Does the impact amplitude exceed the acceptable range?
V. The core of the control chart: Use the "3σ line" to distinguish between normal and abnormal
The first control chart (P control chart, used to monitor the nonconforming product rate) invented by Shewhart in 1924 essentially separates "random fluctuations" from "abnormal fluctuations" using statistical boundaries (control lines)
- The control line is not determined by subjective judgment but calculated using the historical data of the process. Usually, it takes mean ± 3 times the standard deviation (3σ) (under the normal distribution, 99.73% of the data will fall within this range).
- When the data points fall within the control limits and there are no abnormal patterns (such as 7 consecutive points on the same side or consecutive upward trends), it indicates that the process is only affected by random factors and is in a statistical process state (stable state) – this is the ideal state pursued by SPC: the process output is stable and predictable, and the future product characteristics will remain within the current fluctuation range.
VI. How to judge abnormalities? The Eight Patterns in GB/T 4091
How to judge whether the process is out of control? GB/T 4091 - 2001 General Control Charts presents eight abnormal patterns. The core logic is the low - probability event (probability of occurrence)(<5%):
1. One data point falls outside the 3σ control limit (probability: 0.27%).
2. Seven consecutive points are on the same side of the center line (probability 1.56%).
3. Seven consecutive points rising or falling (probability 0.78%);
4. Among 11 consecutive points, at least 10 are on the same side.
5. Among 14 consecutive points, at least 12 are on the same side.
6. Among 17 consecutive points, at least 14 are on the same side.
7. Among 20 consecutive points, at least 16 are on the same side.
8. At least 2 out of 3 consecutive points are between 2σ and 3σ (warning zone).
The design of these patterns balances "false alarms" (misjudging normal fluctuations as abnormal) and "missed alarms" (misjudging abnormal fluctuations as normal). For example, the probability of "one point out of 3σ" is extremely low (0.27%), so it can be firmly determined as abnormal; while the probability of "seven consecutive points on the same side" is about 1.56%, which belongs to "potential abnormalities that need attention".
VII. "Acceptance Criteria" for Process Control Status
Whether the process is in a state of statistical control (stable state) depends on four conditions:
1. Random data dispersion: The points are evenly distributed around the center line (mean) without clustering.
2. Points within the control limits: All data do not exceed the boundaries of "mean ± 3σ".
3. No abnormal pattern: There are no special fluctuations such as chains, trends, or cycles (e.g., 5 consecutive points increasing).
4. Predictable output: Future product characteristics will remain within the current range of fluctuations (for example, the non-conformity rate tomorrow will not suddenly double).
Conversely, as long as any of the following conditions is met, it is considered "out of control":
- There are special factors (such as machine breakdown).
- The point exceeds the control limit;
- Abnormal patterns such as chains, trends, and cycles appear.
VIII. The "two types of uses" of control charts: analysis and monitoring
Control charts are not "one-time tools". Instead, they are divided into two types: those for analysis and those for control, corresponding to the entire "diagnosis - monitoring" process of the process.
Analytical control chart: Used for "finding problems" - when SPC is first applied to a process, it is almost impossible for the process to be in a stable state directly (there must be special factors). At this time, use the analytical control chart to collect data, identify abnormalities (such as points outside 3σ), eliminate or correct special factors (such as replacing machine parts) until the process reaches a stable state.
Control chart for control: Used for "maintaining achievements" - after the process is stable, "fix" the control limits of the control chart for analysis as the standard for daily monitoring. For example, measure 5 products every hour and plot the data points on the control chart for control. If the points are within the limits and there is no abnormal pattern, it indicates that the process is normal. If a point goes out of the limits or there is an abnormal pattern, immediately stop the machine for inspection (such as whether the raw materials have changed batches).
IX. "Standard Process" for SPC Implementation
In summary, the core steps of process statistical control are:
1. Select characteristics: Determine the quality characteristics to be monitored (such as the critical dimensions of the product, the non - conformance rate of assembly) - select the characteristics that "have a significant impact on product quality and are measurable".
2. Data collection: Collect data according to the specified sample size (e.g., 5 samples per hour) and frequency (e.g., once an hour). The data should be real and continuous without omission or modification.
3. Diagnose with the analytical control chart:
In control: If the process meets the steady-state conditions, extend the control limits of the analysis control chart to serve as the control lines of the control chart for control purposes. At the same time, evaluate the process capability (for example, calculate CPK, i.e., the process capability index). Note: The process capability is only meaningful when the process is in control (the output of an out-of-control process is unpredictable, and the calculated CPK is inaccurate).
Uncontrolled: Immediately check for special factors (conduct a one-by-one investigation using 5M1E). After correction, collect data again until it becomes controlled.
4. Monitor with control charts: In daily production, continuously plot data points on the control charts. Once an abnormality is detected, handle it immediately - the aim is to "nip special factors in the bud" and maintain the process in a stable state.
Finally: The essence of SPC is "prevention"
Many people regard SPC as a "post - event inspection tool", which is actually wrong. The core of SPC is prevention. By using control charts to identify "early signals of abnormality" (such as three consecutive rising points), problems can be resolved before they become serious (such as a large number of defective products being produced). It's like a doctor using a thermometer to measure body temperature. A temperature of 37℃ is normal (in a stable state), and 38℃ is abnormal (out of control). At this time, taking medicine costs much less than waiting until the fever reaches 40℃ before starting treatment.
Remember: SPC is not a "statistical game", but a process management tool that "lets data speak". Only by understanding the basic concepts can you truly use control charts correctly.