Simultaneous Fault Diagnosis using multi class support vector machine in a Dew Point process

Abstract There are different approaches for Process Fault Diagnosis (PFD) ranging from analytical to statistical methods, such as artificial intelligence. Support vector machine (SVM) is a relatively novel machine learning method which can be used to handle fault classification due to its good generalization ability. The PFD based on Multi Label SVM approach (MLSVM) overcomes the difficulties of the Mono Label Artificial Neural Network (MLANN) approach including the needs for a large number of data points with difficult data gathering procedure and time consuming computation. However, the existing MLSVM approach has a lower classification performance. In this paper the objective is to improve the diagnosis performance of MLSVM approach while maintaining its advantages. Therefore, a novel MLSVM approach based on multiple regulation parameters is proposed for simultaneous fault classification in a Dew Point process. The performance of the proposed MLSVM approach is compared against other classifiers approaches including MLANN and MLSVM with single regulation parameter tuning. The classification performance of the proposed approach is close to MLANN approach and superior than MLSVM with single regulation parameter. However, MLSVM has other advantages in comparison with the MLANN approach including requirement of smaller number of data, easy data gathering and lower computational burden.