Biggest misuse of Control Charts & Standard Deviation

If you are an R&D associate, or a Quality Assurance manager, or a QC analyst, or a Process Technologist, or anyone from Senior Management who has a role in deciding the specification limits in a process, this article is going to be an eye-opener for you.

Before starting the discussion about Control Charts, first what are control charts?

To understand the control charts, first we need to be aware of 2 types of variation:

1. Common cause variation - which is inherent in the process and cannot be eliminated
2. Special cause variation - which is due to some assignable cause and can be eliminated after analysis

Standard deviation is used to define the UCL and LCL of the control charts. These charts are used to identify the special cause variation and check whether a process is under control.

The most common form of control charts is X Bar and R Chart. X Bar stands for the average or mean of the subgroup and R is the range of the subgroup. These both are plotted for each subgroup and the trend is analyzed to check process stability. These charts can tell when the process is out of control to give a defective product without generating a defective product in actual. When the process becomes out of control, one tries to eliminate special cause variation and thus, reducing the standard deviation.

Secondly, one must know the difference between the Control Limits and Specification Limits. Control Limits are calculated for the Control Charts and X Bar & Range on these charts are plotted. Control limits are 3 times plus and minus Standard Deviation. In control charts, there is no role of the Specification Limits. Specification Limits are set according to customer's requirement, usually by R&D or by Senior Management.

But to my greatest surprise, I have observed that some very senior people across the globe are working differently with standard deviation. They take standard deviation as deciding factor to set the specifications of a product and process. How? Let me explain.

An R&D person had to set the specifications of a product. The limits of parameters such as a, b, c etc. had to be set. He took 10 batches of the product, get those checked from QC lab to see the standard deviation, mean, range etc. for various critical parameters. Then he used the mean of these 10 batches and added & subtracted 6 times standard deviation in mean to finalize the specification limits of the concentration of the ingredient. The mistake - Control Limits are considered as Specification Limits. Standard Deviation is used to calculate specification limits.

Similarly, in batch yield calculation, rather than calculating the process losses at various stages, QA person set the yield targets considering the batch to batch yield variation - First calculating the average yield of some 5-10 batches and then adding & subtracting 3 or 6 times standard deviation to finalize upper and lower specification limits. Again, the mistake - Control Limits are considered as Specification Limits. Standard Deviation is used to calculate specification limits.

Now the question arises that why the above practice is wrong?

An R&D person, who has to launch a drug in the market, has to ensure that the ingredients of the drug must be in a certain limit to ensure the efficacy of the product. If the process is not able to manufacture the drug with those ingredients in a particular range on consistent basis batch after batch, the efficacy of the drug cannot be ensured. How can he use the standard deviation of a process to decide the ingredients' limits?

Similarly, for example if a process which is designed to give 960 Kg of product output and is bound to give a loss of 40 Kg in total at different stages, the yield limits need to be set taking care of minimum and maximum acceptable losses at different stages and total them up to define the minimum and maximum yield points. Running a process with 10 batches and using standard deviation and mean to decide the upper and lower yield limits is a completely "NO".

The basic concept of Six sigma is about reducing the variation and shifting the mean towards the center of the control limits.

But when people start using this concept in a distorted manner and use variation i.e. Standard Deviation to design specification limits of their processes, they are planning to go in a safe zone where no one can catch their inefficiencies in the process.

To conclude, we need to improve the process according to Design Specifications, not to change the specifications limits according to variation in the process.

Note: When the customer specifications limits are very broad and the company can produce the product within narrow tolerance with very low standard deviation, the usual practice is to shift the mean towards the lower side to incur cost benefits. This approach is fine till the time the process is operating beyond 6 sigma level. i.e. 8 Sigma level or 10 Sigma level or beyond that. In that case, ensuring the process compliance to 6 sigma level even after shifting the mean to lower side, product will fall within Customer Specification Limits with 3.4 defects per million.