Understanding the Voice of the Process - an example on data analysis
The Voice of the Process is only one of the contributors to a Controllability Assessment. But since it might be hard to understand it, the following example shows how to listen to this contributor, and what it can bring.
About the example
The data in this example is real production data. The names of the process and its variables have been anonymized. To keep the example simple, not all variables have been used in the analysis. For the same reason, not all analysis aspects that are important will be covered. Instead, some basic aspects and the goal of the analysis will be described, to introduce the concept of the analysis.
Starting point of this example is a process with 13,6% reject over a given period of time.
Graph 1: Initial capability on product quality
Objective of the example
Investigate to what extend an analysis of the production database can explain the reasons of this reject.
The example concerns a production process where several components are assembled, and each final product is tested on a specific quality parameter once. The higher the quality value, the better the product. There is a lower specification limit, below which the product needs to be rejected.
The production process comprises some production steps and measurement steps that are needed to measure the quality.
Graph 2: Schematic representation of the proces
Many different types of data can be collected and analyzed. This usually means that the data need to be put together in a format suitable for further analysis. The table below shows columns that represent the collected variables (potential causes of quality variation). Each row in the table represents a single product, from top to bottom in the order of production. Once a table like this has been collected, different types of analysis can be performed on it.
Table 1: the data
One of the most common analyses to perform, is to graph the quality data (column Product Quality) in the order of time, in the form of a control chart:
Graph 3: Control chart of Quality
The red areas in the graph represent data that are out of control. Such data points either represent single quality values that are too far away from “normal” quality, or they represent a shift in average quality that can be considered out of control. The graph uses underlying statistics to determine whether or not areas are out of control. The graph also shows that due to the out of control situations quite some products fall below the quality specification limit. Such products need to be rejected, and there quantity is 13,6% as shown in graph 1.
The control chart in graph 3 can be used to determine the moments in time when quality went out of control. And using information like logbooks of the process, the engineer can try to find explanations (causes) of the poor product quality. Unfortunately, for complex processes this won’t work properly. Firstly because logbooks do not contain sufficient information and secondly, many events tend to occur simultaneously. So what is then the true cause of the out of control situation?
This is a good moment to jump to the conclusions in case you feel you have seen enough details for now.
Statistics can help us to continue the analysis. Analyses like regression and anova can help to identify the potential columns/variables that significantly influence quality. Though performing a correct analysis is rather complex, the results are easy to interpret.
Table 2: Main effects analysis of the variables in the data set
Table 2 shows to what extend each column/variable in the data table affects the final Product Quality. The higher the value in the column “SS-adj”, the bigger the effect of the variable on the quality. The column with the p-values can be used to assess if a variable significantly influences the Product Quality. Variables with p-values above about 0.05 can be considered to have such small influences, that their effects do not stick out above the normal noise of the quality variation. For such variables there is no clear evidence that they significantly affect Product Quality. Hence, they can’t be held responsible for the red areas in graph 3. In other words: these variables don’t affect quality in any significant way. For our example this means that variation in the process pressures and temperature doesn’t affect the quality variation. These process variables are therefore sufficiently under control. That is important information and good news.
On the other hand, several process variables show variation that does significantly affect the quality, and these are summarized in table 3.
Table 3: Variables that significantly influence Product Quality.
In fact, Table 3 represents a statistical model for the Product Quality, and the R-sq value of this table can be interpreted as follows: the variables in this table have a significant effect on the product quality, and they explain about 64% of the total amount of quality variation shown in graph 2. That means that if we would be able to fully control those variables (fully fix them), we would get the following product quality variation graph:
Graph 4: Expected quality variation after fixing the variation in the variables of table 3.
Clearly, the amount of red out of control areas has reduced in comparison to the starting situation (graph 2). There are also less data points below the specification limit. The reject level as shown in graph 5 would drop to about 5.6%.
Graph 5: Expected quality capability after fixing the variation in the variables of table 3.
In summary, the potential benefits of resolving the variability of the five significant variables looks like this:
The statistical analysis of this data set has shown that the following variables significantly influence the Product Quality:
- Process quality parameter 1
- Process quality parameter 2
- Test temperature
- Order number
If (!) the process variation of these variables could be brought fully under control, the process reject levels would drop with about 60%. Based on this, management could decide whether or not this would save enough money to allocate engineering resources to bring these specific variables further under control.
The list of significant variables includes variables of both the assembly process and the preceding component process. Hence, the focus of the engineers needs to include the component production as well.
Also, no resources need to be allocated to the variables that were found to be already sufficiently under control. Those are:
- Pressures 1, 2 and 3
- Temperature 1
- Delay time
- Test machine number
- Test fixation method
It should be noticed that the result after analysis (graph 4) still shows some red areas. This can mean that some additional variables influence the process quality, but these variables are not yet represented in the current data set. This means it might be worthwhile to investigate what other variables could be monitored in future, to resolve also this unwanted variation.
Last but not least, please remember that this analysis is a simplified example. Several more important aspects need to be taken into account. Nevertheless, the conclusions can be captured in a similar simple way.
This is what the “Voice of the Process” can tell us: where to spend our engineering resources to improve quality, and how effective this effort potentially can be.
For more information please contact us.