434732 A Method of Proactive Model-Based Alarm System Design

Thursday, November 12, 2015: 10:43 AM
Salon G (Salt Lake Marriott Downtown at City Creek)
Taha Mohseni Ahooyi1, Jeffrey E. Arbogast2, Warren D. Seider3, Ulku G. Oktem4 and Masoud Soroush1, (1)Department of Chemical and Biological Engineering, Drexel University, Philadelphia, PA, (2)Applied Mathematics R&D, American Air Liquide, Newark, DE, (3)Department of Chemical and Biomolecular Engineering, University of Pennsylvania, Philadelphia, PA, (4)Risk Management and Decision Center, Wharton School,University of Pennsylvania, Philadelphia, PA

Process safety has been an active field of research for many decades. Along with the development of more sophisticated control systems to improve product quality and plant efficiency, modern safety systems have evolved to comply with tighter safety regulations, criteria, and standards required by governments, regulatory agencies, and the process industries.  

Alarm systems provide a direct sensible feedback from the current status of an operating plant. This is usually done through monitoring the signal transmitted by a physical sensor connected to a process variable of interest and converting the signal to a visual or annunciator alarm whenever the signal crosses a limit. This is to alert the plant personnel to take proper safety action to prevent an incident which is about to occur as a result of the observed deviation. However, multiple problems have emerged with the advancement of the alarm systems. First, as implementation of alarm systems is becoming cheaper and more feasible, an unnecessarily large number of alarms are employed, leading to many false alarms. Second, complex interactions among variables of a process plant can cause many alarms to trigger redundantly or unnecessarily in response to a single event. Different approaches have been introduced to address these problems, including alarm signal processing [1], multivariate analysis [2], alarm threshold optimization [3], and model-based monitoring [4, 5].

In this work, we combine plant-wide operability analysis and alarm systems, and present a method of soft (model-based) alarm system design. The method is model-based as it makes use of a deterministic first-principles model to simulate the plant dynamic behavior over a receding time-horizon into the future. Such simulations provide a basis to project the plant operability status in the future and to generate soft alarms. The proposed method offers the following advantages. First, the method systematically accounts for interactions among plant variables, as it utilizes a first-principles model with an adequate level of complexity. Second, the method can provide alarms tied to unmeasurable state variables. This feature is particularly interesting as conventional alarms respond to measurable plant variables only. Third, since the soft alarm system can provide alarm signals before an incident occurs, operators can take the proper precautionary and mitigating actions proactively. Techniques that ensure the computational efficiency of the method will be presented. The application and performance of the method will be shown using a polymerization plant example.

References

[1]. Chen, T. (2010). On reducing false alarms in multivariate statistical process control. Chemical Engineering Research and Design, 88(4), 430-436.

[2]. Stefatos, G., & Hamza, A. B. (2010). Dynamic independent component analysis approach for fault detection and diagnosis. Expert Systems with Applications, 37(12), 8606-8617.

[3]. Jiang, R. Y. (2010). Optimization of alarm threshold and sequential inspection scheme. Reliability Engineering & System Safety, 95(3), 208-215.

[4]. Mohseni Ahooyi, T., Soroush, M., Arbogast, J. E., Seider, W. D., & Oktem, U. G. (2014). Maximum‐likelihood maximum‐entropy constrained probability density function estimation for prediction of rare events. AIChE Journal, 60(3), 1013-1026.

[5]. Mohseni Ahooyi, T., Arbogast, J. E., & Soroush, M. (2015). Applications of the Rolling Pin Method. 1. An Efficient Alternative to Bayesian Network Modeling and Inference. Industrial & Engineering Chemistry  Research. 54 (16), 4316–4325.


Extended Abstract: File Not Uploaded
See more of this Session: Process Monitoring and Fault Detection I
See more of this Group/Topical: Computing and Systems Technology Division