A scalable algorithm for identifying multiple sensor faults using\n disentangled RNNs

Le résumé fourni par la source

The problem of detecting and identifying sensor faults is critical for\nefficient, safe, regulatory-compliant and sustainable operations of modern\nsystems. Their increasing complexity brings new challenges for the Sensor Fault\nDetection and Isolation (SFD-SFI) tasks. One of the key enablers for any\nSFD-SFI methods employed in modern complex sensor systems, is the so-called\nanalytical redundancy, which is nothing but building an analytical model of the\nsensors observations (either derived from first principles or identified from\nhistorical data in a data-driven fashion). In a nutshell, SFD amounts to\ngenerate and to monitor residuals by comparing the sensor measurements with the\nmodel predictions with the idea that the faulty sensors will result in large\nresiduals (i.e. the defective sensors generate measurement that are\ninconsistent with their expected behavior represented by the model). In this\npaper we introduce a disentangled Recurrent Neural Network (RNN) with the\nobjective to cope with the \\textit{smearing-out} effect, i.e. the propagation\nof a sensor fault to the non-faulty sensors resulting in large misleading\nresiduals. Moreover, the introduction of a probabilistic model for the residual\ngeneration allows us to develop a novel procedure for the identification of the\nfaulty sensors. The computational complexity of the proposed algorithm is\nlinear in the number of sensors as opposed to the combinatorial nature of the\nSFI problem. Finally, we empirically verify the performances of the proposed\nSFD-SFI architecture using a real data set collected at a petrochemical plant.\n

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.