Fault detection and diagnosis with statistical and soft computing methods
Book in Scopus
- Additional Document Info
- View All
Most of the research work that has been done on the field of fault detection and diagnosis has used the model based approach. This chapter presents an alternative way of carrying out a complete fault detection and diagnosis system using statistical and soft computing methods. This proposal is based only on the system's or process' history data treatment. The motivation of using process history data is basically: 1. To obtain an approach that could take into account variables correlations, that at first sight could not be so clear even for expert designers, 2. To develop a diagnostic system that learns normal operation mode directly from the process, instead of having a model based diagnostic system which depends on the expertise of the designer to manage the complexity of the system when modeling. The advantage of having a process history data based approach over a model based framework is the relatively easy way to obtain data from automated industrial processes. In most of the modern systems, can be very difficult to obtain an exact model due to the big quantity of information needed and the variables correlations, which can cause false alarms, indicating a wrong faulty component or system. Nevertheless, this kind of approaches combining statistical and soft computing methods can be supported or complemented with model based methods, in order to have a more powerful diagnosis method combining the expertise and mastery of the designer and those hidden behaviors that many systems exhibit. In this chapter it is shown how statistical methods applied in a straightforward way and combined with soft computing methods such as artificial neural networks and fuzzy logic, are ideal tools for doing diagnosis. This translates to finding the root cause of the problem using only process history data as a prior knowledge of a system, no matter if it is linear or nonlinear. This knowledge is used to give a final diagnosis, in complex scenarios whith noise presence and correlated variables, that could easily mislead to false alarms or wrong diagnosis. The organization of this chapter is as follows. First of all it is presented an introduction of why a fault diagnosis is necessary. Then a classification of fault diagnosis methods is given. After that, a presentation of the mathematical tools used is shown and then how they have been tailored in our research, in order to build complete fault detection and diagnosis systems for several applications. We present case studies that show promising results using the algorithms proposed. Finally the conclusion over this chapter is given. © 2011 by Nova Science Publishers, Inc. All rights reserved.