Art-2 and multiscale art-2 for on-line process fault detection — Validation via industrial case studies and Monte Carlo simulation

https://doi.org/10.1016/S1367-5788(02)80020-7Get rights and content

Abstract

Data from most industrial processes contain contributions at multiple scales in time and frequency. In contrast, most existing methods for fault detection are best for detecting events at only one scale. This paper provides experimental validation and insight into a new method of process fault detection based on the integration of multiscale signal representation and scale-specific clustering-based diagnosis. The multiscale ART-2 (MSART-2) algorithm models normal process operation as clusters of wavelet coefficients at different scales. It detects a process change when one or more wavelet coefficients of test data violate similarity thresholds with respect to clusters of normal data at that scale. Especially in industrial situations where the nature of the abnormal features is not known a priori, MSART provides better average performance due to its ability to adapt to the scale of the features. In contrast to most other multiresolution schemes, this framework exploits clustering behavior of wavelet coefficients of multiple variables for the purpose of scale selection and feature extraction. Detailed performance comparisons, based on rigorous Monte-Carlo simulations as well as industrial data from a large scale petrochemical process, are provided. Our results show that MSART-2 significantly improves the detection performance of the ART-2 detection algorithm over a broad range of process anomalies. Results are compared with single-scale and multiscale versions of PCA for benchmarking purposes.

References (45)

  • D. Wienke et al.

    Adaptive resonance theory based neural networks — the ‘ART’ of real-time pattern recognition in chemical process monitoring?

    Trends in analytical chemistry

    (1995)
  • D. Wienke et al.

    Adaptive resonance theory based neural network for supervised chemical pattern recognition (fuzzy ARTMAP), part 1: Theory and network properties

    Chemometrics and intelligent laboratory systems

    (1996)
  • D. Wienke et al.

    Adaptive resonance theory based neural network for supervised chemical pattern recognition (fuzzy ARTMAP), part 2: Classification of post-consumer plastics by remote NIR spectroscopy using an InGaAs diode array

    Chemometrics and intelligent laboratory systems

    (1996)
  • J.R. Williamson

    Gaussian ARTMAP: A neural network for fast incremental learning of noisy multidimensional maps

    Neural Networks

    (1996)
  • H.B. Aradhye

    Anomaly Detection Using Multiscale Methods

  • H.B. Aradhye et al.

    Clustering in wavelet domain — a multiresolution art network for anomaly detection

    IEEE Transactions on Systems, Man and Cybernetics

    (2000)
  • B.R. Bakshi

    Multiscale PCA with application to multivariate statistical process monitoring

    AIChE Journal

    (1998)
  • B.R. Bakshi et al.

    Process Monitoring by PCA, Dynamic PCA, and Multiscale PCA — Theoretical Analysis and Disturbance Detection in the Tennessee Eastman Process

  • G.A. Carpenter et al.

    Fuzzy ART: An adaptive resonance algorithm for rapid, stable classification of analog patterns

  • G.A. Carpenter et al.

    Fuzzy ARTMAP: A neural network architecture for incremental supervised learing of analog multidimensional maps

    IEEE Transactions on Neural Networks

    (1992)
  • K.C. Chou et al.

    A multiscale stochastic modeling approach to the monitoring of mechanical systems

  • M.S. Crouse et al.

    Wavelet-based statistical signal processing using hidden Markov models

    IEEE Transactions on Signal Processing

    (1998)
  • Cited by (0)

    1

    H. B. Aradhye was a graduate student at the Department of Chemical Engineering and the Department of Electrical Engineering, The Ohio State University, when this research was being done.

    View full text