In this case study we related how the analysis of historical condition monitoring data can provide confidence, for some equipment / components, to increase the intervals between samples and reduce overall condition monitoring spend.
A key component of the process we employ to gain confidence in our decision making is the analysis of the sample comments provided by laboratory technicians. At Relialytics we employ a unique combination of text analytics and data visualisation to quickly understand the historical condition monitoring data associated with specific equipment / component types.
A good example of this is Figure 1 below. The network diagram represents a visualisation of the sample comments for the final drives from 777D off-highway trucks. Each node / dot in the diagram represents a sample. The links between the nodes provide an indication of how similar the laboratory technician’s comments are between samples. The more tightly clustered these samples are, the more similar the comments.
Figure 1 – 777D Final Drive Oil Sample Comments Network Colour Coded by Comment Type
In the case of Figure 1 it is easy to see that the vast majority of the samples are tightly clustered together indicating very similar sample comments. The associated table shows the samples are different in some respects and provide clues as to any issues experienced with the 777D final drives in the fleet. In reality the 777D final drives samples identify minimal issues.
Figure 2 shows the same 777D network colour coded to represent the evaluation code provided by the laboratory technician, i.e. whether it is an “A”, “B”, “C” or “X” sample.
Figure 2 – 777D Final Drive Oil Sample Comments Network Colour Coded by Evaluation
The “B” samples located at the edge of the large cluster represent samples that required further investigation by the maintenance crew at a reduced sample interval. From Figure 2 it can be seen that approximately 65% of the samples are normal i.e. “A” samples. 34% have the beginning of potential issues i.e. “B” samples. There were only 5 “C” samples across the 777D fleet. The “C” samples progressed from “B” samples over a period of 3,000 hours, so not rapidly and directly from an “A” to a “C” evaluation i.e. a “low rate of change” (the subject of a future blog).
Additionally the “C” samples were associated with the finals drives of 1 unit only suggesting an isolated problem and providing confidence that the 500 hour sample interval being performed by the mine site could be extended to 1,000 hours. We can illustrate the power of interpreting the network diagrams constructed from text similarity measures by comparing Figures 1 and 2 to Figures 3 and 4 (a final drive analysis for a 329D excavator).
Figure 3 – 329D Final Drive Oil Sample Comments Network Colour Coded by Evaluation
Figure 3 shows 2 large and distinct groupings of samples with different types of comments. The bottom grouping shows samples with a large number of issues requiring a number of different investigations and resampling with shorter time intervals (250 to 100 hours).
The top grouping shows more normal samples and samples requiring investigation with resampling in normal time intervals down to 250 hours.
Figure 4 – 329D Final Drive Oil Sample Comments Network Colour Coded by Evaluation
Figure 4 shows the same 777D network colour coded to represent the evaluation code. The network shows the comments progress uniformly from “A” to “X” as you work down the network demonstrating how different sample comments can be (even within an evaluation category i.e. “B” samples inhabit both distinct groupings).
We can very quickly determine from Figures 3 and 4 that the 329D excavators are not likely to be candidates for contributing to a reduction in condition monitoring spend. The above analysis demonstrates the unique text analytics and data visualisation capabilities that Relialytics can bring to your business.
If you would like to learn more about what we can do for you please contact us.