For non-stationary industrial processes, the distribution of data often deviates from historical patterns due to various factors, such as the aging of equipment and adjustments in manufacturing strategies. As a result, the conventional methods, which presume an identical and independently distributed (i.i.d.) setting, may falter in delivering satisfactory diagnostic results. This chapter focuses on the phenomenon of conditional drift, a relaxation from the stringent i.i.d. assumption, indicating that the conditional distribution of industrial data evolves over time. From a mathematical lens, we initially formulate the issue of conditional drift and propose a target mapping approach for drift adaptation, grounded in the principle of minimizing risk. Adopting this strategy, we devise an adaptive incremental identification model. This methodology enables our approach to adapt to conditional drift, enhancing its resilience in dynamic settings. The efficacy of our proposed method is substantiated through fault identification applications in two industrial processes.
Incremental Conditional Drift Adaptation for Processes Fault Identification Under Non-stationary Working Conditions
eng. Applications of Computational Methods
16.04.2025
21 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Working Group 1: Non-Stationary Processes
British Library Online Contents | 1997
|A Unified Fault Diagnosis Framework for Industrial Processes with Incremental Learning Ability
Springer Verlag | 2025
|