ISSN 1060-992X, Optical Memory and Neural Networks (Information Optics), 2008, Vol. 17, No. 3, pp. 183–192. © Allerton Press, Inc., 2008.
183
Neural Clouds for Monitoring of Complex Systems
B. Lang
a
, T. Poppe
b
, A. Minin
c
, I. Mokhov
a
, Y. Kuperin
c
, A. Mekler
d
, and I. Liapakina
e
a
OOO Siemens, Fault Analysis and Prevention group
b
Siemens AG, Industry sector, Thomas.Poppe@siemens.com
c
Saint-Petersburg State University,
d
Institute of Human Brain RAS,
e
Saint Petersburg State University of Economics and Finance
e-mail: {Bernhard.Lang, Ilya.Mokhov}@siemens.com; {Alexey.Minin, Yuri.Kuperin}@gmail.com;
mekler@yandex.ru; leonikaspb@gmail.com
Received April 21, 2008; in final form June 24, 2008
Abstract—Condition monitoring is an important and challenging task actual for many areas of industry,
medicine and economics. Nowadays it is necessary to provide on-line monitoring of the complex sys-
tems status, e.g. the steel production, in order to avoid faults, breakdowns or wrong diagnostics. In the
present paper a novel machine learning method for the automated condition monitoring is presented.
Neural Clouds (NC) is a novel data encapsulation method, which provides a confidence measure regard-
ing classification of the complex system conditions. The presented adaptive algorithm requires only the
data which corresponds to the normal system conditions, which is typically available. At the same time
the fault related data acquisition is expensive and fault modeling is not always possible, especially in
case one is dealing with steel production, power stations operation, human health condition or critical
phenomena in financial markets. These real word applications are also presented in the paper.
Key words: neural clouds, one side classification, on line monitoring, early fault detection, EEG classi-
fication, fault analysis and prevention.
DOI: 10.3103/S1060992X08030016
1. INTRODUCTION
The Neural Clouds (NC) concept was successfully elaborated and applied by the Corporate Technology
Department of Siemens AG for solving the steel production optimization problem [3]. However this tech-
nique has been also transferred, with the necessary modifications, to the field of vibration analysis [4], EEG
classification and financial market analysis and prediction. The application of neuro-fuzzy methods, pre-
sented in this paper, is an attempt to make the expert condition monitoring system more intelligent and able
to face the real world problems, keeping the monitoring costs reasonable.
The concept presented in the paper is directed for the elaboration an efficient data encapsulation method
for the adaptive solving of the so called one-side classification problem. The basic idea behind the usage of
the one-side classification in the field of condition monitoring and fault analysis is that the real data, which
can be collected, usually corresponds to the normal conditions of the complex systems in question. Vice
versa data collection, corresponding to abnormal conditions, is expensive, and fault modeling is not always
available. Here one should mention that the idea of elaboration of the classifier based on a data, collected
from the system under normal operating conditions, could be extended for the case of the human health-
related measurements classification, where the data acquisition related to particular disease is even more
problematic.
The NC is applied in the following fields: analysis of vibration data [11–12], electroencephalography [1–
2, 5–9] and prediction of the American stock market declines [20–25].
THE BASIS OF THE NEURAL CLOUDS ALGORITHM
In this section the Neural Clouds method is presented. Let us assume that we are dealing with a number
of measurements from a real object. Every instance of the given data set could be considered as a point in
n-dimensional space, where n corresponds to the number of different parameters of the system under con-
sideration. First, the data set should be clustered. In most of the cases data should be normalized prior to the