Parity space-based fault isolation using minimum error minimax probability machine

摘要

In this paper, a new parity space-based fault isolation (FI) approach is proposed for linear discrete time-invariant systems. The main contribution lies in the integrated design of FI system and fault isolability properties without any assumptions on the probability distribution of unknown inputs. First, a fault isolability evaluation index is defined as the worst-case probability of correct FI in different faulty scenarios, and a new objective function is proposed for the integrated design of parity space-based residual generator, evaluation logic and fault isolability properties. Based on this, the FI problem is formulated as a bank of optimal multiclass classification problems. Next, using the technique of minimum error minimax probability machine (MEMPM), a group of feasible parity space matrices for residual generation are obtained. Furthermore, the FI method by MEMPM is presented with a desirable trade-off between the fault isolability properties in different faulty scenarios. To show the effectiveness of the proposed method, a satellite attitude control system subject to six faults is considered.