Achieving manufacturing excellence is the goal pursued by every enterprise. To achieve this goal, the quality department must continuously work to identify the sources of process variation and take appropriate measures to control the goal. In the modern scientific era, quality units can already understand the production process through data analysis, which can not only increase the flexibility of production, but also ensure that products better meet customer requirements.
Three keys to realizing smart manufacturing
How to achieve manufacturing excellence based on scientific data analysis is a problem that every company faces when it requires improving product quality. Through past industrial cases of JMP cooperation, we can see that companies need to go through three key processes to achieve manufacturing excellence :
- 1. Through control charts and process capability analysis, statistical process control (SPC) is implemented to reduce process variation;
- Find sources of variation in the measurement process through measurement system analysis (MSA);
- Through design of experiments (DOE) and predictive modeling, the factors that affect measurement variation are determined , and the optimal settings of the factors are found to make the measurement process more robust.
Let’s take a case of a pharmaceutical manufacturer to see how to implement this “data analysis trilogy”. First, let's take a look at how to perform statistical process control (SPC) .
Practical case analysis
Process background
FVM Pharmaceuticals is an international pharmaceutical manufacturer specializing in the production of high-quality finished pharmaceutical products in tablets, capsules and liquids. In the pharmaceutical process, a typical manufacturing process includes: grinding the active pharmaceutical ingredient (API) into a powder of uniform particle size, and then mixing the ground material with other components to increase the volume of the API and evenly distribute it. , this blended material is then compressed into tablets and finally coated to improve shelf life, mouthfeel and other properties. At the end of this process, various quality parameters are recorded for batch acceptance.
The raw materials for the above process are concentrated emulsions containing two organic compounds, which are provided by two suppliers (A , B ) at the same time. The pharmaceutical company monitors the quality by measuring the concentration (mg/L) of the compounds through gas chromatography (GC). .
Current Issues
Recently, the quality management team discovered that there are big differences in the quality of drugs. Through troubleshooting, it is suspected that the problem is likely to come from the raw materials. To this end, the company has specially set up a quality improvement team, hoping to use statistical tools , combined with quality by design ( QbD ), and through data-based decision-making, to fundamentally find a sustainable solution and determine process changes . key driving factors to ultimately achieve the goal of optimizing process settings .
Improve process
First, the quality improvement team conducts statistical process control (SPC) of the product . Understand the stability of the process and analyze the pattern of process changes through control charts. Through process capability analysis, evaluate whether it meets customer/ external delivery deadlines. Both tools support a better understanding of processes and guide the development of improvement strategies. The team collected samples of two compound raw materials from two suppliers for 8 days and imported them into the JMP software, involving all batches. The following figure (1) is a screenshot of some data.
Figure ( 1 )
Prerequisites for analysis: Check whether the data shows normality
The premise used by many quality statistical tools is that the data comes from a normal distribution, otherwise reporting results based on mean and standard deviation will be meaningless. Use the JMP menu, click "Analyze à Distribution ", and combine the SW test and AD test results to confirm that the two compound data show a normal distribution, indicating that there is no problem. The results are shown in Figure (2) below. .
Figure II )
Of course, you can also quickly check the three parameters " day " , The relationship between " batch " and " supplier " as potential drivers of variability and the changes in the two compounds is shown in Figure ( 3 ) .
For example, batches from two suppliers are used every day, but even for batches on the same day, the results for Compound 1 and Compound 2 are very different .
Picture ( 3 )
The second step of analysis: Check whether there is any variation in the data
After completing the normality test and histogram analysis, you can use the JMP Control Chart Builder platform ( Control Chart Builder) to explore quality issues . Since variation exists in any system and process, control charts can automatically calculate control limits , which can effectively distinguish variation caused by general causes and special causes in process variation, thereby identifying and eliminating data fluctuations caused by special causes.
Another commonly used quality tool - Process Capability Analysis (Process Capability) , can effectively measure how a process performs compared with given specification limits, and helps identify underperforming processes . The specification limits used in calculations are set by external systems or according to process requirements. In this example, the lower and upper specification limits of compound 1 are 100mg/l and 500mg/l respectively, and the lower and upper specification limits of compound 2 are 300mg/l and 600mg/l .
A good process is one that is stable and consistently produces product that is well within specification limits. The above two tools, Control Chart Builder & Process Capability, are often used in combination to evaluate the stability of the process and its ability to meet customer specifications.
Through JMP 's control chart generator platform, you can quickly get the output results in Figure ( 4 ) below. The left side is the XBar&R control chart of compound 1 and compound 2 respectively , and the right side is the process capability analysis result.
Figure ( 4 ) Control chart generator results
The third step of analysis: data interpretation of analysis results
- The XBar chart is a mean control chart with batches as subgroups. Observation shows that all points of Compound 1 fall randomly within the control range (green and yellow areas). Introducing the discriminant rule, no points are circled or marked, which means that the process is " in control " or stable. After introducing the judgment rule for compound 2, it was found that batch number 11 fell into the red area and was circled or marked, which indicated that the process was out of control or unstable.
- The R chart is a range control chart with batches as subgroups and is used to monitor fluctuations in the process. Each point on the chart represents the difference between the maximum and minimum values of the batch range. Introducing the judgment rule, we can see that the 10th batch of compound 1 and the 5th batch of compound 2 exceed the control limits, indicating that there are some special reasons that cause process fluctuations.
- The right part is the process capability analysis results, used to identify poorly performing processes. Several capability indicators commonly used to evaluate include:
- Cp/Pp : The ratio of the width of the specification limit to the process fluctuation. Cp is called internal (or short-term) potential process capability (only considering fluctuations within a subgroup), and Pp is called overall (or long-term) potential process capability (considering fluctuations in all groups as a whole). They are a common indicator of how much capability can be achieved if the process mean and target mean are consistent;
- Cpl/Ppl : used to check the ability of a process to meet the lower limit of specifications;
- Cpu/Ppu : used to check the ability of a process to meet the upper specification limit;
- Cpk/Ppk : Take the minimum value of Cpl/Ppl and Cpu/Ppu.
- Cpk/Ppk equals 1.0 , which is considered a barely qualified process. Of course, some industry standards require that the minimum value of Cpk/Ppk is 1.33 or even higher.
- The current Cpk value and Ppk value of raw material compound 1 are 0.740 and 0.752 respectively , so the process is considered to be unqualified. Compound 2 is not very good either, with Cpk and Ppk values of 1.005 and 0.746 respectively .
The fourth step of analysis: explore sources of variation from the supplier’s perspective
Drag the supplier to the top of the control chart (stage area) to split the control chart by supplier, as shown in Figure ( 5 ) below.
Figure ( 5 )
As can be seen from Figure ( 5 ) above , for these two compounds, based on the existing control limits, although the two compounds provided by supplier A are very stable (the batch is within the control limits), we can find: Supply Supplier A has significantly greater batch-to-batch variation than Supplier B. Supplier B 's batch-to-batch variation was relatively small, but Supplier B 's Compound 1 was found to exhibit cause-specific variation.
In addition, we can also try to use model-driven multivariate control chart (MDMVCC ) and use principal component analysis (PCA ) to monitor the parameters of multiple processes in one control chart . In some scenarios, if you just observe each process independently, they all seem to be stable. Through multivariate control charts, you can detect the potential instability of multiple processes at the same time , and deeply investigate the contribution of individual processes to the overall process to diagnose whether the process is stable. Stablize.
Figure ( 6 )
From Figure ( 6 ) above , we can see that all batches of supplier A are within the control limits, while for supplier B , we can clearly see that three batches are out of control points and require attention.
In addition, we can also use the JMP process capability screening platform (Process Screening) to combine variation and capability into one output using target charts and process performance charts, and simultaneously check whether the compounds from two suppliers are stable and whether Conform to given specifications . This platform is especially suitable for quality inspection scenarios that require monitoring of large-volume manufacturing processes, and is highly recommended.
Figure ( 7 )
The process capability chart in Figure ( 7 ) above highlights two aspects of whether the process is healthy , namely stability index and process capability (Ppk ). A stable process will have a stability index close to 1. The higher the value of the stability index , the more unstable the process is.
Process capability Ppk is an important parameter to measure process performance. The higher the value , the lower the defect rate per million products . The ideal process (capable and stable) is shown in the green shaded area of the figure, where Ppk>1.0 and stability index <1.25 . Incompetent and unstable processes are in the red shaded area, where Ppk<1.0 and stability index >1.25 .
Goal plots display the specification-normalized mean shift on the x- axis and the specification-normalized standard deviation on the y- axis for each process and are used to quickly provide a summary view of whether a process is within specification limits.
Stable and capable processes are shown in the green (pk>2.0 ) and yellow (pk>1.0 ) shaded areas of the figure, and incompetent and unstable processes are shown in the red shaded area (Ppk<1.0 ).
For Supplier A , both processes for Compound 1 and Compound 2 are stable but have insufficient capabilities. For Supplier B , Compound 2 is stable but insufficiently capable , while Compound 1 is both insufficiently capable and unstable.
Phased conclusions and quality improvement plans
We have successively used means, range control charts, and multivariate control charts to monitor suppliers’ raw material quality and key process parameters, and require suppliers to also use these control charts to reduce process variability.
In theory, if the supply of raw materials is unstable, all supply batches should be monitored and the frequency of raw material sampling and analysis increased, but this is not currently possible because the existing gas chromatography analysis method (GC ) is very time-consuming , full inspection will not be able to meet production needs. Therefore, the rapid analysis method currently being developed, ultra-high performance liquid chromatography (UHPLC ), will be the focus of the next step. How to achieve faster raw material quality monitoring through ultra-high performance liquid chromatography measurement system? We will continue to explore this in subsequent articles.
If you are interested in the statistical methods of statistical process control (SPC) in this article, you are welcome to download the latest version of JMP and experience it for yourself. Learn more about data analysis knowledge and techniques.
Recommended reading:
