Dr. Ahmed's Research Interests

My recent research is focused on the development of novel alarm systems. Methodologies and techniques are being developed to provide process industries with a new alarm system. Significant contributions also continue to be made to the field of continuous-time identification. Although system identification is considered a mature subject, contribution to the field with new methodologies are being realized by addressing real life implementation issues. Possibilities to contribute to new fields through novel use of process models are being explored. Research efforts are dedicated to a number of different topics. Contributions are being made to model validation, valve stiction detection and alarm design.


The state of industrial alarm systems is indeed alarming. Flooding of alarms, false and nuisance alarms, along with their poor prioritization, make plant alarm systems unreliable and unmanageable. A conceptual study on the design of a novel alarm system was carried out and a risk-based alarm system design methodology was proposed. A strategy for alarm annunciation has been developed. The special features of the proposed method are tied to assignment of alarms to sets of variables and to the use of risk for alarm annunciation. Process industries are in dire need of a new approach to alarm systems. The proposed risk-based approach may be a viable alternative to the current plant alarm system. Risk-based alarm design is a shift in paradigm in alarm system design and is anticipated to change the way process alarm systems are designed.


Our proposed alarm system requires continuous estimation of the risk associated with a set of variables. This is a shift in approach in the estimation and use of risk in process industries. Typically risk estimation is done at the design stage or when there is a significant plant modification. We propose to estimate and use risk for day-to-day decision making. The impact of this approach will bring significant change in the decision making process for plant operations. The estimated risk can be used for plant optimization, performance assessment and safety management.


A list of novel methodologies for identification of continuous-time models has been developed, and a number of research articles have been published by the applicant. The methods address implementation issues in the field of identification of continuous-time models from sample measurements. Data-based modelling is a well-developed field; however, most of the modelling methods require some assumption about the quality of data. If data requirements are not met, the methods cannot be effectively applied. In real plant operations, vast amount data are collected on a regular basis. However, ‘good data’ remain precious. Contributions of the developed methodologies come from the idea of tailoring the procedure to be applicable to available data. The issues with initial unsteady conditions, raw form of data, simultaneous estimation of time delay, and irregularly sampled data are addressed in the context of continuous-time identification. To perform identification under closed-loop conditions, a new step response method is proposed. To handle the effects of disturbances, identification in the presence of disturbance has been addressed. Although step input is the most common test signal for identification, some variables may not be changed step-wise. Methods for identification from non-ideal step inputs, as well as for the sinusoidal input, have also been developed. The proposed methodologies for identification from step response have been adopted in specialized software for performance assessment and are currently in use in Syncrude Canada Ltd. Our recent development on this topic include a multi-input multi-output contnuous-time identification algorithm that allows to change all input signals simultaneously. Model validation is an integral part of model identification. However, validation of continuous-time model remains an unexplored area. A validation scheme for continuous-time models was developed that can be used for multiple-input multiple-output (MIMO) models with time delay.


Plant-wide oscillation is a common industrial phenomenon and one of the root causes of oscillation is stiction in control valves. A simple procedure to detect stiction was developed based on the shape of signals. Shape based stiction detection is not a new idea. However, it is challenging to identify the shape of a signal without a visual observation. A technique to identify the shape of a signal based on its compressibility was proposed. The two important aspects of the method are its simplicity and use of the data compression tool which is available in most plants. 


A Matlab-based graphical user interface (GUI) for identification of continuous-time models from the step response has been developed. There has been significant development in the field of step response identification over the last decade; however, none of the methods are included in the available software for system identification. The widely used Matlab SysId toolbox uses the same technique for step and other inputs. Hence, it is not designed to exploit the additional advantages that the step input offers e.g. for simultaneous estimation of the time delay, handling non-zero initial conditions, use of raw data and identification in the presence of disturbances. The unique feature of the developed Matlab GUI is tied to its capability to handle the above mentioned special conditions of data. The motivation to develop a GUI is to disseminate research work to the end-users in an appropriate and usable form. The impact of the development comes from the presentation of developed methodology in a user-friendly form so that plant personnel can use the algorithms without paying attention to theoretical details of the proposed algorithm.


Centre for Risk, Integrity and Safety Engineering (C-RISE)

230 Elizabeth Ave, St. John's, NL, CANADA, A1B 3X9

Postal Address: P.O. Box 4200, St. John's, NL, CANADA, A1C 5S7

Tel: (709) 864-8000