Explainable fault prediction using learning fuzzy cognitive maps

Abstract IoT sensors capture different aspects of the environment and generate high throughput data streams. Besides capturing these streams reporting monitoring information, there is significant potential for adopting deep learning to identify valuable insights predictive preventive maintenance. One specific class applications involves using Long Short‐Term Memory Networks (LSTMs) predict faults happening in near future. However, despite their remarkable performance, LSTMs can be very opaque. This paper deals with this issue by applying Learning Fuzzy Cognitive Maps (LFCMs) developing simplified auxiliary models that provide greater transparency. An LSTM model predicting industrial bearings based on readings from vibration developed evaluate idea. LFCM then used imitate performance baseline model. Through static dynamic analyses, we demonstrate highlight (i) which members a sequence contribute prediction result (ii) values could controlled prevent possible faults. Moreover, compare state‐of‐the‐art methods reported literature, including decision trees SHAP values. The experiments show offers some advantages over methods. LFCM, conducting what‐if analysis, more information about black‐box To best our knowledge, first time LFCMs have been simplify offer explainability.