LAUSR

Abstract Abnormity identification plays a significant role in keeping a process healthy and safe. The rule-based expert system (RB-ES) is an efficient approach in this field. However, the relationship between propositions is fuzzy and uncertain when the process knowledge is summarized into rules. Moreover, the process data utilized as observed evidence to activate the uncertain rules face the imprecise measurement problem. The imprecise measurement problem comes from noises, heavy dust, grease pollution, and other circumstance factors that influence sensors. To address the above issues, an abnormity identification framework based on fuzzy logic rules is proposed in this work. Two kinds of uncertain rules with fuzzy linguistic information and expert estimated threshold in premises are specially considered. First, a deviation estimation method based on probability distribution is proposed to extend the imprecise data into the interval data to solve the uncertain problem caused by imprecise measurement. Second, the fuzzy piecewise membership function is utilized to calculate the certainty factor (CF) of the premise in rules, which describes the occurrence risk of an event or abnormity. Third, the inference steps based on the CF expert system with uncertain rules and imprecise activated variables are presented to identify the abnormity in the complex industrial processes. The proposed method is applied to the thickening process in hydrometallurgy and performs better than the traditional rule-based method and the PLS method on the actual process data. The accuracy of the proposed method still maintains higher when more noises are added, generally 8%–10% higher than the other two methods. The proposed method also shows advantages in identifying multi-abnormity and reducing model establishing workload.