This paper addresses the problem of the detection of new failures modes in industrial plants. Since its associated optimization problem is NP-hard, an efficient method based on taboo search and Bezdeck criterion is proposed, intensification and diversification strategies are also included because of the non connectivity of the solutions space. The proposed approach exercised on a simulated industrial system was shown to exhibit good performances in dealing with the occurrence of new faults in industrial processes, despite that such search algorithm is perfectly general, it can be easily extended to more complicated schemes.