Intelligent fault diagnosis based on similarity analysis using generative model and multi-sensor fusion in industrial processes

Fault diagnosis is crucial in enhancing efficiency, control, reliability, and safety of modern industrial processes. In large-scale industrial systems, a fault can propagate through interconnected equipment, leading to an alarm flood and reducing the alarm management systems' performance. Therefore, a data-driven fault diagnosis method can be employed to discover the main reason for alarm floods. This research proposes a novel intelligent architecture for detecting similar patterns and diagnosing the root causes of faults in industrial systems. First, the process variables information and alarm data sequences are merged through the joint representation of the data fusion scenario. Then, pre-processed data is mapped to the secondary space by the variational auto-encoder (VAE) network, and the generative adversarial network (GAN) investigates the similarity between the patterns. Finally, similar patterns are identified with multi-sensor information fusion, and the root of the alarm flood is subsequently diagnosed. Another key innovation of the proposed algorithm is its online implementation. Therefore, this can enhance operators' decision-making abilities, enabling industrial operators to implement effective preventive measures and enhance the safety of industrial processes. A simulator, the Tennessee Eastman Process, and an experimental case study, the Saveh Rotary Cement Kiln, illustrate the efficacy of the suggested approach.