A new approach to determining the precision and bias of on-line gauges

Abstract

On-line gauges are used in many industries to measure characteristics of raw material destined for a production process. In the coal industry, which is the focus of interest in this paper, the traditional method for determining coal characteristics has been the assay of coal samples. This method requires a great deal of work on a day-to-day basis. It has the further drawback that there is generally a time lag of at least 1 day between sampling of the coal and obtaining numerical measures of its quality characteristics. An interesting recent development is that nuclear gauges, which are capable of producing estimates of coal quality in real time, are now available commercially. The present paper demonstrates how bootstrap methods can be used to obtain a confidence interval for gauge precision using gauge output data alone. No specially constructed reference values from a sampling and assay operation are required. As such, the method is potentially an attractive alternative to the methodology set out in ISO 15239. It is also shown how gauge calibration can be checked using gauge output and one set of reference samples obtained under normal operating conditions. The method, which again uses the bootstrap, involves the construction of a simultaneous confidence band for a calibration function that relates gauge values to reference values.