Use of Support Vector Machine to Fault Detection in Biomethanation Process

Abstract

Biomethanation processes are complex due to the multiple molecules and different microorganisms interacting on them. Hence, changes in the system inputs or operational conditions make them susceptible to failures, e.g., deviation from steady states or optimal operation points. Support Vector Machine (SVM) is a relatively simple technique that can be used to identify those deviations. In this study, SVM have been applied for fault detection in the biomethanation process. Data obtained from a model-based Dynamic Multi-Objective Optimization (DMO) have been considered as the optimal operating point. Disturbances of ±10, ±15, and ±20% in the inlet liquid flow rate from respect to the optimal were generated by simulation with an extended anaerobic digestion model (ADM1_ME). Three SVM models, quadratic, cubic, and gaussian were trained and validated with a data set of 449 and 80 points, respectively, to classify the regions of each disturbance and identify the percentage of the disturbance. Accuracies higher than 0.96 and 0.91 were achieved for all SVM models in the training and validation, respectively.