An Effective Fault Diagnosis Approach Based On Gentle AdaBoost and AdaBoost.MH

With fault causes increasingly specialized in industrial process, data-driven method is becoming an indispensable technique to handle fault diagnosis problems. As efficient and practical data-driven methods, Gentle AdaBoost and AdaBoost.MH are introduced in this paper to diagnose binary faults and multiclass faults, respectively. Moreover, F1 score is developed afterwards as the performance assessment index in order to evaluate the proposed algorithms. Tennessee Eastman Process, a benchmark dataset in chemical engineering, is then adopted as case study for methods comparison in this research. As shown in the experimental results, Gentle AdaBoost achieves promising results on binary fault diagnosis, meanwhile one against one, one against all and AdaBoost.MH are used to extend Gentle AdaBoost to multiclass fault diagnosis case, which also outperforms than supported vector machine, a popular nonlinear binary and multiclass classifier.