New supervision architecture based on on-line modelling of non-stationary data

Abstract

A new supervision system consisting of three modules is presented. The main novelty is the first module that corresponds to a modelling task. This module, which uses the auto-adaptive and dynamical clustering (AUDyC) neural network, allows us to continuously analyse and classify the functioning state of the monitored system using a dynamical modelling of all known modes (good/bad functioning modes represent different classes). The second module exploits these models of the functioning modes in order to detect ''fast'' and ''slow'' deviations. From membership degrees and from the information extracted by the monitoring module, the third module, dedicated to the diagnostics, informs the user about the functioning conditions of the system. In this paper, the main characteristics of the AUDyC and its abilities to model online non-stationary data are presented. Then, the description of the supervision system is given and some experimental results stemmed from a supervision application of a hydraulic system are discussed.

Sensor fault diagnosis of inland navigation system using physical model and pattern recognition approach

“…The update of the classes parameters is performed recursively on a sliding window of size N f en for each new observation X k = (x 1 , x 2 , ... x n ), with n the number of features. The principle of the procedure is detailed in Lecoeuche et al (2004).…”

“…normal and faulty, by classes. Some dynamical classification algorithms are proposed in the literature such as the CDL algorithm (Cluster Detection and Labeling), algorithms based on adaptive resonance theory (ART) networks (Eltoft and de Figueiredo, 1998), or AUDyC (Auto-Adaptive Dynamical Clustering) algorithm (Lecoeuche et al, 2004). The characteristics of the classes are followed online and faults can be detected using metrics between classes.…”

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

“…In this paper, we employ the normalized DC current for the feature extraction, and Auto-adaptive and Dynamical Clustering (AUDyC) algorithm [1], [2] for feature clustering. AUDyC does on-line, adaptive and unsupervised feature learning, hence the unknown modes of operation can be learnt from the measurement data.…”

On-line detection of fouling in a water circulating temperature controller (WCTC) used in injection moulding

“…The core of this supervision tool is derived from a powerful classifier: the AUDyC (AUto-adaptive and Dynamical Clustering) neural network [14,15]; the latter has been tested on other equipments such as a heat dissipater [16], and an hydraulic process [17,18]. The training of this network is unsupervised, the number of neurons on the hidden layer is not constant (each hidden neuron represents a prototype), and the number of neurons on the output layer is variable.…”