Adaptive Information Fusion Using Evidence Theory and Uncertainty Quantification

Fault detection systems support the operator, providing insight during the decision-making while having an (unknown) fault. Data-based models are a common option for a detection system. However, systems that rely purely on data-based models are normally trained with a specific set of data, which cannot necessarily prevent data drift. Thus, an anomaly or unknown condition detection mechanism is required to handle data with new fault cases. Besides, the model’s capability to adapt to the unknown condition is equally important to anomaly detection—in other words, its capability to update itself automatically. Alternatively, expert-centered models are powered by the knowledge of operators, which provides the models with production context and expert domain knowledge. The challenge lies in combining both systems and which framework can be used to achieve this fusion. We propose a novel adaptive information fusion methodology to define fault detection systems using evidence theory and uncertainty quantification. The main contribution of this paper is providing a general framework for the fusion of
$n$
number of information sources using the evidence theory. The fusion provides a more robust prediction and an associated uncertainty that can be used to assess the prediction likeliness. Moreover, we provide a methodology for the information fusion of two primary sources: an ensemble classifier based on machine data and an expert-centered model. We demonstrate the information fusion approach using data from an industrial setup, which rounds up the application part of this research. Furthermore, we address the problem of data drift by proposing a methodology to update the data-based models using an evidence theory approach. We validate the approach using the Benchmark Tennessee Eastman while doing an ablation study of the model update parameters.