Use of machine learning on SCADA data for asset’s prognostics health management
Alexandre Cesa A B and Elliot Press AA KPMG, 235 St George’s Terrace, Perth WA6000, Australia.
B Corresponding author. Email: acesa@kpmg.com.au
The APPEA Journal 60(2) 602-605 https://doi.org/10.1071/AJ19054
Accepted: 4 March 2020 Published: 15 May 2020
Abstract
The timely detection of anomalies in the process industry is paramount to ensure effective and safe operation of plant. There typically exists an abundance of historical data recorded in Supervisory Control and Data Acquisition (SCADA) systems, which is most often used for understanding past events through, for example, root cause analysis. It is envisaged that higher levels of insight could be achieved from the same datasets by utilising more advanced analytical techniques such as machine learning frameworks. This would enable moving from a ‘diagnosis–mitigation’ (i.e. a root cause analysis) paradigm to a more desirable ‘detection–prediction–prognosis–prevention’ paradigm. Machine learning techniques can be used on SCADA data to support the detection of plant anomaly conditions that do not necessary manifest as process alarms for example. We used a Bayesian network framework on the Tennessee Eastman Plant benchmark problem to demonstrate the technique’s capability. Our model proved to be effective in detecting anomalous plant conditions in most situations.
Keywords: anomaly detection, artificial intelligence, asset health, Bayesian network, PHM, remaining useful life.
Alexandre Cesa is an Associate Director at KPMG Management Consulting. He has 15+ years of experience in Engineering Asset Management in the resources industry in Australia. Alexandre is a CPEng and his background includes BEng and MSc (Mechanical), GDBEAM and MBA. He is currently pursuing a PhD in Data Science and Chemical Engineering with Curtin University. |
Elliot Press is a Data Science and AI consultant at KPMG Management Consulting. He has a background in biological sciences and cognitive evolution. Utilising a strong background in understanding evolutionary equilibriums, selection and agent behaviour, he has an academic interest in the modelling and simulation of complex systems, agent based modelling and Bayesian probability. Elliot is currently pursuing a MSc in Data Science, and has a BSc (Honours) in Cognitive Biology. |
References
Al Dhafeeri, M. A. (2007). Identifying sources key to detailed troubleshooting of amine foaming. Oil & Gas Journal 105, 56–67.Aldrich, C., and Auret, L. (2013) ‘Unsupervised Process Monitoring and Fault Diagnosis with Machine Learning Methods.’ (Springer: London.)
Chandola, V., Banerjee, A., and Kimar, V. (2009). Anomaly detection: a survey ACM Computing Surveys 41, 15.
| Anomaly detection: a surveyCrossref | GoogleScholarGoogle Scholar |
Downs, J., and Vogel, E. (1993). A plant-wide industrial process control problem. Computers & Chemical Engineering 17, 245–255.
| A plant-wide industrial process control problem.Crossref | GoogleScholarGoogle Scholar |
Kwon, D., Hodkiewicz, M. R., Fan, J., Shibutani, T., and Pecht, M. G. (2016). IoT-based prognostics and systems health management for industrial applications. IEEE Access : Practical Innovations, Open Solutions 4, 3659–3670.
| IoT-based prognostics and systems health management for industrial applications.Crossref | GoogleScholarGoogle Scholar |
Lei, Y., Li, N., Guo, L., Li, N., Yan, T., and Lin, J. (2018). Machinery health prognostics: a systematic review from data acquisition to RUL prediction. Mechanical Systems and Signal Processing 104, 799–834.
| Machinery health prognostics: a systematic review from data acquisition to RUL prediction.Crossref | GoogleScholarGoogle Scholar |
Russell, E. L., Chiang, L. H., and Braatz, R. D. (2000). ‘Data-driven techniques for fault detection and diagnosis in chemical processes.’ (Springer: London/New York.)
Santos, A., Figueiredo, E., Silva, M. F. M., Sales, C. S., and Costa, J. C. W. A. (2016). Machine learning algorithms for damage detection: kernel-based approaches. Journal of Sound and Vibration 363, 584–599.
| Machine learning algorithms for damage detection: kernel-based approaches.Crossref | GoogleScholarGoogle Scholar |
