(584e) Multi-Class Classification of Process Faults Using Nonlinear Support Vector Machine Based Feature Selection Algorithm

The advances in information, and sensing technology has paved the way for large amount of process data collection in real time, initiating the “Big Data” era in industrial decision making processes [1]. While the modern process industry aims for smarter, safer, and more efficient operation, more operating variables are integrated under closed loop control resulting in increased process structure complexity, which obfuscates process control [2]. Thus, the industrial data outbreak immensely facilitates process monitoring by use of machine learning techniques. One of the most popular machine learning techniques is Support Vector Machines (SVMs) [3] which allows the use of high dimensional feature sets for learning problems such as classification and regression. Yet reducing the dimensionality of the feature space in data-driven modeling, known as dimensionality reduction and feature selection, is still a key task in improving model accuracy [4] as well as decreasing a priori data collection, which in turn yields enhanced efficiency in chemical processes.

In this work, we perform multi-class classification using nonlinear Support Vector Machine based feature selection algorithm for fault identification and diagnosis. Previously, we have presented a novel data-driven framework for simultaneous fault detection and diagnosis for chemical processes [5-6] that uses nonlinear Support Vector Machines (SVM) based feature selection algorithm [7]. The framework produces two-class fault-specific SVM models to detect known faults individually. In order to improve the decision making, we propose to extend the current framework by performing multiclass classification of process faults and normal operation. Here, the idea is to create a decision-tree based classification scheme where the process data is initially classified as either normal or faulty, and then examined with a series of trained models, where each defining a region of data space that separates one class from all others (one-versus-all) or from a single other class (one-versus-one) for fault detection. By incorporating multi-class SVM models, we aim to minimize the number of inspections to identify the process fault, thus enhance the current fault detection and diagnosis framework. We present results for the Tennessee Eastman process as a case study and compare our approach to existing approaches for fault detection and diagnosis.

References

[1] Reis, M. S., Gins, G. (2017). Industrial Process Monitoring in the Big Data/Industry 4.0 Era: From Detection, to Diagnosis, to Prognosis. Processes, 5(3), 35.

[2] Chiang, L. H., Russell, E. L., & Braatz, R. D. (2000). Fault detection and diagnosis in industrial systems. Springer Science & Business Media.

[3] Cortes, C., Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), 273-297.

[4] Onel, M., Beykal, B., Wang, M., Grimm, F. A., Zhou, L., Wright, F. A., Phillips, T. D., Rusyn, I., Pistikopoulos, E. N. Optimal Chemical Grouping and Sorbent Material Design by Data Analysis, Modeling and Dimensionality Reduction Techniques. 28th European Symposium on Computer-Aided Process Engineering (ESCAPE-28); Elsevier, 2018; p Accepted manuscript.

[5] Onel, M., Kieslich, C. A., Guzman, Y. A., Floudas, C. A., & Pistikopoulos, E. N. (2018). Big Data Approach to Batch Process Monitoring: Simultaneous Fault Detection and Diagnosis Using Nonlinear Support Vector Machine-based Feature Selection. Computers & Chemical Engineering.

[6] Onel, M.; Kieslich, C. A.; Guzman, Y. A.; Floudas, C. A.; Pistikopoulos, E. N. Simultaneous Fault Detection and Diagnosis of Continuous Processes via Nonlinear Support Vector Machine-based Feature Selection. 13th International Symposium on Process Systems Engineering (PSE 2018); 2018; p Accepted manuscript.

[7] Guzman, Y. A. (2016). Theoretical advances in robust optimization, feature selection, and biomarker discovery (Doctoral dissertation, Princeton University).