Modification and Enhanced Testing of Data Mining-Based Algorithm to Detect Subtle Errors in Temperature Sensors in Gold Stripping Circuit Mining, Metallurgy and Exploration

Sensors are everywhere in the mining industry, with sensor information being used to monitor and operate processes. Therefore, when sensor information is wrong, economic losses can occur. Unfortunately, sensor errors are usually not detected till they become large. This is problematic as most calibrate their sensors no more than once a year (Beamex, n.d.). Using principles of data mining, where all relevant information is tapped to glean hidden information, Pothina [8] designed an algorithm to detect errors in temperature sensors in gold stripping circuit in Pogo mine, Alaska. This paper continued his work by analyzing the behavior of the algorithm on baseline data and testing the algorithm in new data and under more rigorous conditions. It also made a change to the algorithm. The modified algorithm performed very well in the new data. It also worked well under the new error conditions. Three types of errors were seeded, a fixed additive error (+ 2%), a fixed subtractive error (− 2%), and a noisy, normally distributed error, with a mean value of + 2%. Artificially seeded errors were detected within about 20 gold stripping cycles. Inherent bias in operation impacted algorithm performance by increasing the number of cycles needed to detect errors. This paper reinforced Pothina’s [8] conclusion that when data mining approach is used, sensor errors can be detected even when they are pretty low. Significant economic losses can thus be minimized.