Research Of Theory And Method For Fault Diagnosis And Fault-tolerant Control Based On Rough Sets

Fault is sometimes ineluctable in the actual system of industry production. For maintaining the control performance of system after it went wrong, it is very necessary to study the theories and methods of fault diagnosis and fault-tolerant control of system. Rough set theory (RST) is a kind of new mathematical tool of soft computing. Some useful knowledge can be extracted from a data sample applying RST. The purposes in this paper are that building firstly the rough set models of systems, making fault detection with the models built, extracting control rules from the data sets of control behaviors of human-operators using RST, and designing the apery rough set intelligent controllers applying the control rule sets extracted under normal and faulty states of systems;while the system is at normal, it is controlled by the controller designed under normal state, if detecting out the system went wrong, the controller can be automatically switched to fault's one by the fault-tolerant control system, and then the system will be controlled by it. For carrying out above goals, the modeling method and designing method of the apery rough set intelligent controllers based on RST are proposed. The key of two kinds of methods is how to extract the decision tables (rule sets) from data records of the input-output of system and the control behaviors of operators. The concrete method is that building the primordial decision information table, discretizating, retecting according to the significances of condition attributes relative to decision attribute, taking the original decision table, completizating, and verifying the rules in the decision table. In the process studying discretization techniques, a kind of perfect algorithm is presented for the limitation of greedy algorithm improved, making it have practicality;aiming at the discretization result changes necessarily as the values of attributes of a decision information table are changed, two theorems are advanced and proved, the theorems show that the discretization result obtained in terms of a sample can not be as a result of variables in an actual system. Completeness definition and measurement, the necessity of completization andspace describing method are given about a continuous decision table;in addition, the completization method of linear subsection interpolation of continuous decision table is proposed. Finally, modeling, fault diagnosis, controller design and fault-tolerant control of non-linear systems consisted of one-tank, two-tank and three-tank based on RST are respectively realized. The experiment results state that the strategy of fault diagnosis and fault-control based on RST proposed is valid.