SPC - Passive Design of Experiments

In his book, "Out of the Crisis," Dr. W. Edwards Deming made a statement that, "No process is truly stable."  He also stated that if you treat every observation as a special cause and you react, you will increase variation rather than solve problems.  I wonder if this is similar to the words that are spoken...some are important and others simply help to pass the time.  It is important to be aware and listen to the words.  It is also important to listen to the process.

Placing an SPC chart on a process is the "end game" of performing statistical process control.  Design of Experiments requires that potentially important controlling factors are identified, rational ranges of control/changes are determined, the study is planned, executed, and the results are analyzed.  Statistical Process Control is no different.

First study and document every action and energy transfer of the process.  For each action and energy transfer, what are their conditions for success?  When a condition is not good enough it becomes a cause.  Is it possible to change the process such that these conditions will tell the operator or process engineer when the condition might be changing?  If so, you have an EVENT signal and you can use Event Base Sampling rather than random time or count based sampling (e.g. once an hour or after every X parts).  Standard events are new batch of material, new box of parts, tool change, at startup, first part, etcetera.  Sample just before the event and plot the point.  Sample just after the event and look for change.  Pay special attention for the next six points to see if the process shifts.  If the first point is outside the control limits...the process was hit by a truck...huge impact.  If you do not see the change until after the seventh point, the change was more subtle but the event is statistically significant.  Your passive designed experiment found a factor that is significant.  You have a reason for the chart and you know how to interpret cause and effect.

There are many types of statistically significant signals.  Trends are gradual changes from tool wear, gradual loosening of fixtures or building up of debris, addition of new material into a batch of material (% change), clogging of vents, etc.  Cycles are periodic changes in energy (air pressure, voltage, power factor, hydraulic pressure, etc.) and temperatures.  Shifts are conditions that control a function which change suddenly.  Study the sequence of functions (actions and energy transfers).  Document the factors that control each function and stay at the function level.  No...every idea is not a good idea.  All ideas should be respected as should team members.  The good ideas have rational connections to the functions.  Everything that directly "touches" the function is a potential first level cause.  Understand these conditions and select those which are known to change suddenly or gradually.  Develop a strategy to observe these factors and their changes.  Use the SPC chart to confirm the cause and effect relationships.

Often events occur and there is nothing that can be done.  The process must run and the change must be accepted.  The shift occurs and the important question becomes, "Is the product still okay?"  Perform tests or develop rules at that step and if the shift does not violate the rules, keep running the process and ship the parts.  If the product fails the test, the control of nonconforming product process must be initiated.

The SPC chart is the last step in performing passive designed experimentation.  Companies who begin with the SPC charts are doomed to be data loggers.  Changes occur...why?  ...let's go find something!  All that a special cause signal means is that the process is statistically different as compared to the time when the control limits/parameters were set.

First, determine the controlling factors.  Second, monitor those factors for change.  Third, confirm the changes with the appropriate SPC Chart.  Fourth, develop strategies to control the dominant conditions/controlling factors.  When those factors cannot be controlled (e.g. new batch of material that is in specification but causes a shift), develop rules to determine if the product is okay to process and ship.  This will limit and control all nonconforming product based on the signal called the Event.  It will also give you a rational business reason to have a chart.  If you study the process and cannot justify the expense for the actions, the chart is a bad business decision.  Anytime you have a high Cost of Poor Quality, you have a valid business reason for passive Design of Experiments or SPC.