The 5 Pillars of Statistical Process Control (SPC)

What does the term “statistical control” mean?

The Statistical Process Control is a quality control technique that was born almost a century ago through the hands-on research of Walter A. Shewhart, physicist and engineer at the time employed by the company Western Electric. The need to reduce the rate of repairs and manufacturing errors were the subject of his work and the genesis of his discoveries, later popularized by Edward Deming, another quality giant.


  1. No process, natural or industrial, is capable of supplying two exactly identical units. A good industrial process will be characterized by generating products with a high degree of homogeneity, but not exact replicas. If we measure a characteristic of a manufacturing process (or of a service), we can not expect to obtain identical readings, except for the precision limit that the measuring instrument shows us.
    Conclusion: all processes show variability and, consequently, quality is variable.
  2. We can therefore think of quality as a “controlled variable quality”, in the sense that we will obtain values ​​in each product unit that are not too far from the desired objective of the characteristic we are studying.
    Conclusion: controlling quality is admitting a variability within a predetermined range of values ​​and that this variability behaves in a “constant” way over time.
    statistical control of SPC processes. This brings us to the question of how to quantify that range of admissible variability and how to detect it. The trivial answer is to carry out a 100% control of the production, but this has problems:

    1. The inspection is not economically efficient since it is not a productive activity (that adds value), but a detector (which prevents a “negative value”). “Inject in the market).
    2. In addition, in the case that mistakes have been made, we have already spent resources on the production that we can not recover and that will have to be added to the costs of rework or loss for the bad unit detected and delayed production. Both problems are reactive, they happen once the action is done. The statistical control of processes in proactive, tries to detect as soon as possible the deviation of the quality and to avoid to the maximum bad productions.
  3. The word control implies the concept of prediction. A quality attribute is controlled when we are able to predict, within some margins, what will be its future behavior. In other words, when we are able to assign probabilities that the characteristic fits within certain limits.
    Conclusion: we only control when we are able to quantify and for this a mathematical model is necessary.
  4. In a production system there is a certain set of causes that are always present, however from time to time other “foreign” causes creep in and contaminate the statistical equilibrium of the process. These foreign causes can be, for example, lack of training of new workers, breakdowns of machines, non-compliance of raw materials, etc. These “foreign” causes are called assignable causesConclusion: there are sets of permanent or constant causes or inherent in the process, which are always acting on it, and sets of “foreign” causes or “visitors” to the process, which act intermittently and cause the process to leave its state of statistical equilibrium , understanding the latter as excursions of the measured values ​​outside the admissible limits.

Summing up the previous ideas

  1. All processes, natural or human, manifest variability.
  2. A quality attribute “in control” means that over time its variability manifests itself in a similar way, within a range of values. (Note: that a process is in statistical control does not imply that the quality is admissible)
  3. Control means being able to predict within margins.
  4. There are systems of constant causes and systems of non-constant causes that act on the processes. It is when a constant system of causes acts when we have predictability.
  5. It is possible to detect when a system of constant causes suffers a “contamination”, detect which is the new “contaminating” cause and eliminate it.

In your company, what is the variability of your quality features? Do you have a systematic record of them? Do you have predictability? Do you do proactive quality management? Are you able to detect “strange” causes?