A shared reliability database for mobile mining equipment

[Truncated] Reliability data is an important input for data-driven decision making in the mining industry. Site-specific equipment reliability data in the form of historical maintenance event data is currently captured by most mining organisations however can be statistically sparse, of low quality and not fit for purpose for many modelling tasks.

A framework for a sector-specific reliability database for the mining industry has been developed. Key components of this framework include a glossary of terms and definitions, minimum and desired data sets for collection and failure taxonomies. Historical maintenance data is collected from seven organisations across four commodities and five asset classes. Data quality issues in historical maintenance data are identified and include inaccuracies, incompleteness and unstructured freeform text fields. Historical maintenance data requires data cleansing to make it fit for purpose. Interpretation of unstructured free-text fields is done to extract necessary information or cross reference other data fields.

Data cleansing is achieved by the use of rule-based data cleansing using syntactic keyword spotting. The use of a syntactic keyword spotting system is able to reduce manual cleansing effort, increase repeatability and reduce scope for human error. A data cleansing tool using syntactic keyword spotting system is developed to facilitate this. The data cleansing tool, DEST implements a rule-based cleansing system incorporating expert knowledge.

Rule development is managed by a piecewise approach where individual rules are tasked with classifying elements of any data record. Piecewise rule development is capable of sorting unstructured data with a high percentage of unique entries or missing data and enables development of rule libraries. A rule-based system comprised of 500 rules is capable of extracting information from unstructured text fields found in historical maintenance data.