Detection of subtle sensor errors in mineral-processing circuits using data-mining techniques

The economics of a mineral-processing circuit is dependent on the numerous sensors that are critical to optimization and control systems. When a sensor goes off calibration, it results in errors that can have a severe impact on plant economics. Classical statistical approaches fail to detect when errors are “subtle,” unless they grow and become large enough to be “gross,” whereas methods based on signal processing and artificial intelligence fail in dealing with the complexity of process fluctuations. Due to the sheer volume of sensors, undetected errors are not remedied until the next calibration, which on average are a year apart across all industries [1]. This research aims to detect such subtle errors (2 percent bias) in shorter time spans of about a month — rather than wait for errors to grow — using innovative data-mining techniques and algorithms.