Underground » Maintenance
In this project we have developed a computer method for on-line analysis of trend monitoring data to detect and locate face equipment faults. The definition of a fault in the context of this project is any event that causes a production stoppage and downtime. The performance of the method was demonstrated using five months of Citect data collected on a longwall system.
We started the project with a mail survey of Australian longwall sites followed by extended visits to selected mines. All surveyed sites used Citect software to monitor a large number of equipment sensors and alarms. The number of variables recorded varied with the average being 6500. Most staff agreed that the potential of Citect data monitoring was not fully realised in their mines.
When we applied the tools developed in this project to the sample Citect data, we were able to correctly detect and isolate over 90% of the target faults with misclassification rates lower than 20%. When compared against similar attempts in other industries, this is a significant achievement.
To achieve this result, we have surveyed approaches used in other industries and decided to use a data-driven approach. Such approaches treat the system as a "black box". A training data set was generated to "teach" the software what fault classes were associated with what combinations of Citect values. The software self-tuned its internal coefficients using the examples provided by the training set. When provided with data of an unknown class, the software tried to classify this new reading by comparing it against the experience collected through the training set.
One outcome of the project is ready for application in the field. This is automatic processing of Citect data to generate the following statistics on a daily or weekly basis:
- uptime/downtime
- average production delay
- a log of stoppages in each shift and their durations
Such statistics can be compared against manually entered data for purposes of improving record keeping practices.
Using the techniques developed in this project we can also automatically classify the stoppages according to their causes.
Finally, the software can be installed on-line to warn operators on initiation and progress of certain fault types and identification of the cause when a shutdown occurs.
This work has been undertaken by Daniel Bongers as the major part of his PhD study. The PhD thesis is accessible through CRC Mining (www.crcmining.com.au) and covers technical details that are beyond the scope of this ACARP Report.