Auto-adaptive and Dynamical Clustering for Open-Circuit Fault Diagnosis of Power Inverters

Abstract

This paper presents a fault diagnosis approach for single open-circuit faults in inverters entirely from measurements of the stator currents. These measurements are used to extract the feature data; the feature data is then used to create clusters in an on-line, adaptive and unsupervised way. Auto-adaptive and Dynamical Clustering (AUDyC [1], [2]) is the algorithm employed for this step. Based on the derived clusters, appropriate formulations for the data labelling and fault detection and isolation are proposed. The effectiveness of the approach is validated on simulation and experiment data.

Data-driven fault detection of open circuits in multi-phase inverters based on current polarity using Auto-adaptive and Dynamical Clustering

“…i n (t) T ∈ R n , where n ∈ N is the number of phases (3 or 5). In [4], the normalized average phase currents were used as feature variables. However, in Case 2 presented in Section 2, a feature variable is possibly infinite due to the zero faulty phase current.…”

“…In this procedure, AUDyC is used for clustering the feature vectors β(t) into classes, as described in [4]. A simplified formulation of AUDyC is considered, where the distribution of class data is Gaussian.…”

“…All the removed vectors will be used in the next iterative learning loop. From the algorithm description in [4], the following condition is derived for the class creation which will be used hereafter for investigating the influence of the initial covariance σ, the membership threshold µ min and the minimum class cardinality N min on the detection time.…”

“…Hence, Fault Detection and Isolation (FDI) of OC faults in multi-phase inverters associated to a multi-phase motor is an ongoing research topic [3]. Different approaches were proposed in the literature: model-based, signal-based and data-driven [4]. The model-based approach is rather involved for real-world applications (e.g., deriving the model of a multi-phase asynchronous motor is a tedious task).…”

“…The present manuscript builds on the authors' previous work in [20,4]. Double OC faults in inverters were detected using feature variables defined in terms of the average phase currents and their average absolute values.…”