Cluster analysis is a module in PANalytical’s HighScore Plus software that allows decision making on a strongly reduced data set that still represents the original incoming flow of data. As a result not every individual measurement has to be analyzed.
X-ray diffraction is a powerful tool for quality control and quality assurance in cement. It allows quantification of all clinker phases as well as additives such as slag, fly ash and pozzolan. Moreover, modern X-ray diffraction equipment allows rapid measurement enabling process control with a high frequency. However, the combination of rapid measurements and the large amount of parameters yield a large flow of information. This may complicate data evaluation and decision making.
X-ray diffraction is a powerful tool for quality control and quality assurance in cement. It allows quantification of all clinker phases as well as additives such as slag, fly ash and pozzolan. Moreover, modern X-ray diffraction equipment allows rapid measurement enabling process control with a high frequency. However, the combination of rapid measurements and the large amount of parameters yield a large flow of information. This may complicate data evaluation and decision making.
Cluster analysis is a module in PANalytical’s HighScore Plus software that allows decision making on a strongly reduced data set that still represents the original incoming flow of data. As a result not every individual measurement has to be analyzed. Cluster analysis simply compares the XRD measurements and indicates which scans are similar or different. It is a very easy-to-use technique and important information can be obtained without the knowledge of crystalline structures or crystallographic background.
In this application note various examples of the application of cluster analysis in different stages of the cement plant are discussed.
Application of cluster analysis in cement processing can be performed as:
Cluster analysis (1) (2) is a method that greatly simplifies the analysis of large amounts of data by:
This drastically reduces the amount of data which has to be processed, because only representative scans, outliers and sometimes the most different scans must be analyzed in more detail. Further cluster analysis can be used to discover hidden features/structures in the data. The cluster analysis implemented in HighScore Plus is basically an automatic 3-step process, but additional visualization tools are present to judge and influence the clustering.
Clustering of raw materials and production samples is based on a master file with characteristic relevant measurements for each type of material and an automatic decision if a sample belongs to an existing cluster or not. In this example a cluster analysis of 44 different fly ash raw material samples was performed. The results are given in Figure 1. In the 3D graph, each data point represents one scan. The scans that are close to each other are very similar, indicating that the phases are also very similar. The cluster analysis clearly shows three types of fly ashes. According to this database every further investigated sample can be classified to a certain cluster. As mentioned before, the cluster analysis can be performed completely automated. Also a pass/ fail decision can be made whether a raw material is suitable or not. If a scan does not belong to a cluster an error message will appear. This error indicates that it is a completely different material; it may be a badly prepared, damaged or wrongly labeled sample. The difference in quantitative Rietveld results of the fly ashes in Table 1 confirms that three different types can be distinguished.
Table 1. Quantitative Rietveld phase analysis of fly ash raw materials coming from different sources
Figure 1. Cluster analysis of fly ash raw materials coming from different sources
Clustering of final products, meaning blended cements in the building materials industry, is based again on a master file with characteristic relevant measurements for each type of material and an automatic decision whether a sample belongs to an existing cluster or not. In the second example a cluster analysis of 268 different fly ash cement samples was performed. The fly ash cements contained different amounts of the different types of fly ashes described in the first example. The results are given in Figure 2 and Table 2. The cluster analysis enables a clear distinction between two types of fly ash cement with different fly ash contents. Therefore it is possible to classify different types of cement and also to check if cements fulfill a certain norm or not. Furthermore different types of control files can be adapted and automatically chosen for each material in order to perform an accurate Rietveld analysis.
Table 2. Quantitative Rietveld phase analysis of fly ash cements with different amounts of raw materials coming from different sources
Figure 2. Cluster analysis of fly ash cements with different amounts of raw materials coming from different sources
Cluster analysis also allows to distinguish between clinkers produced at different kilns. In this example a cluster analysis was performed for clinkers produced at 6 different kilns. From Figure 3 it is obvious that the clinkers produced in these kilns are very different from eachother. By performing detailed phase analysis on the most representative scan of each cluster, one can correlate the mineralogical characteristics with the kiln characteristics. This information is easily correlated with the properties of the finished cement to identify which kilns produce more or less similar material. This allows the manufacturer to identify ways to optimize the process by using clinkers from different sources. Comparisons of this kind are an excellent way to ensure best practice across a group. It is even possible to identify the plant where a clinker was produced. This can help for example in the case of customer complaints.
Figure 3. Cluster analysis of Portland cement clinkers produced at 6 different kilns
In order to control the clinker production process cluster analysis can be used as well. A typical example of material from two different kilns is shown in Figure 4. This time the variation of the blue scans shows that one of the kilns is quite unstable while the other (green scans) runs much more stable. Again, this information sheds light on the impact of different operating practices, valuable for plant optimization.
Via the cluster analysis in Figure 4 it can be concluded that the production has a higher degree of variation than the other. Subsequent full phase quantification to find sources of variation. Further analysis can be done by detailed study of the scans involved and corrective actions can be taken. Properties of the clinker and cement are directly related to phase information, that can be used to optimize variables such as kiln temperature, mill conditions, and blend proportions.
Figure 4. Cluster analysis of Portland cement clinkers produced at 2 different kilns with stable and unstable process conditions
Cluster analysis is a powerful, yet easy-to-use early- warning tool for process and quality control in cement production. A pass/fail analysis enables to decide about the quality of incoming raw materials. The quality of different clinkers and all types of cement can be determined as well. Furthermore it is possible to stabilize the clinker production.
