| NSF Org: |
CBET Division of Chemical, Bioengineering, Environmental, and Transport Systems |
| Recipient: |
|
| Initial Amendment Date: | August 22, 2017 |
| Latest Amendment Date: | August 22, 2017 |
| Award Number: | 1704833 |
| Award Instrument: | Standard Grant |
| Program Manager: |
Raymond Adomaitis
CBET Division of Chemical, Bioengineering, Environmental, and Transport Systems ENG Directorate for Engineering |
| Start Date: | September 1, 2017 |
| End Date: | August 31, 2021 (Estimated) |
| Total Intended Award Amount: | $107,675.00 |
| Total Awarded Amount to Date: | $107,675.00 |
| Funds Obligated to Date: |
|
| History of Investigator: |
|
| Recipient Sponsored Research Office: |
3451 WALNUT ST STE 440A PHILADELPHIA PA US 19104-6205 (215)898-7293 |
| Sponsor Congressional District: |
|
| Primary Place of Performance: |
220 S. 33rd Street Philadelphia PA US 19104-6393 |
| Primary Place of
Performance Congressional District: |
|
| Unique Entity Identifier (UEI): |
|
| Parent UEI: |
|
| NSF Program(s): |
Proc Sys, Reac Eng & Mol Therm, GOALI-Grnt Opp Acad Lia wIndus |
| Primary Program Source: |
|
| Program Reference Code(s): |
|
| Program Element Code(s): |
|
| Award Agency Code: | 4900 |
| Fund Agency Code: | 4900 |
| Assistance Listing Number(s): | 47.041 |
ABSTRACT
Model predictive control is widely being implemented in many industries, such as chemical plants and oil refineries, leading to substantial improvement in operations. The use of process monitoring through model-based sensors has enabled industries to predict and improve processes. Prior research has introduced novel safety systems using models, which generate alarm signals that can provide warnings of pending problems. This research project involves developing a process improvement model will not only prove useful for the chemical and petrochemical industries, but will also benefit the food, nuclear, aircraft, and petroleum industries by identifying potential hazards. Deployment of this model would result in saving lives, reducing workplace injuries, and economic benefits. The researchers are collaborating with the Air Liquide Corporation, which will ensure the industrial relevance and practicality of the results of this research and will enhance the dissemination of research results. The data resulting from this research project will also provide improved security of industrial operations. Additionally, the researchers are developing educational modules and projects based on the outcomes of this research for use in graduate and undergraduate engineering courses at Drexel University and the University of Pennsylvania.
The objectives of this research project are to study: (1) robust large-scale state-estimate prediction (robust to process-model mismatch and unmeasured inputs), (2) offline optimization-based calculation of the worst-case combinations of process-model parameter values and the most extreme control actions, (3) efficient implementation of the model-predictive safety system for large-scale plants, and (4) implementation and testing of the model-predictive safety system first on the steam-drum system of an integrated steam-methane reformer/pressure-swing adsorber unit through simulations, and then on a steam-drum system in a real integrated steam-methane reformer/pressure-swing adsorber system in real time at Air Liquide. The research team also is developing industrial guidelines for adding and maintaining model-predictive safety systems as a complement for existing functional (safety-instrumented) systems. The involvement of the industrial collaborator enriches the training of graduate and undergraduate students involved in the project. The research project also is being integrated with the Drexel Co-op Program, and undergraduate students, preferably from underrepresented groups, are being recruited for six-month long research internships.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
Process safety is the first objective in every process industry. Despite continuous efforts to improve the safety of processes, the level of human and financial losses due to incidents in the U.S. process industries is still significant. Furthermore, due to the reactive and generally univariate nature of existing functional safety systems (Figure 1), a more-intense operation may not be realistic without innovation in functional safety. Thus, further improvement in process functional safety methods and procedures is needed. Among these efforts are the development and deployment of software packages that predict frequencies and consequences of incidents based upon historical data. However, these packages are usually unable to predict the probabilities of incidents that have never happened before. This inability has motivated the study of methods capable of predicting hazardous conditions in processes.
We introduced and developed the concept of model-predictive safety (MPS), which represents a new paradigm in functional safety; that is, the use of process model (digital twin) predictions to detect operation hazards before they lead to safety risks. MPS uses a dynamic model of a process to predict future process behavior and forecast the potential consequences of future incidents with reasonable accuracy. Such forecasts are then used to determine and implement optimal proactive actions. MPS allows for a systematic utilization of dynamic process models to generate alarm signals (alerts) for the predictive detection and proactive prevention of operation hazards (OHs) in real time. It uses a process model (digital twin) to predict the process safety status over a moving prediction horizon and to generate alarm signal(s) indicating the presence of a present or future OH with reasonable accuracy (Figure 2); it generates alarm signals that alert the process personnel to imminent and potential future OHs before the actual OHs occur. This combined predictive and proactive (prescriptive), real-time use of process models in process safety is an attractive unique feature of MPS. Unlike conventional safety systems that are individually reactive to current conditions through specifically designed logic, MPS systematically accounts for process nonlinearities and interactions among process variables and generates predictive alarm signals. Therefore, this new paradigm in functional safety is analogous to the evolution in process control from only single-loop control (e.g., proportional-integral-derivative control) toward multivariable model-predictive control.
The implementation of an MPS system requires off-line calculations of (a) the most aggressive action that MPS should take to prevent each operation hazard when uncertain model parameters take their nominal values, and (b) the most aggressive action that MPS should take to prevent each operation hazard when uncertain model parameters take their worst-case values. We developed novel min-max optimization methods that can be used offline to perform the calculations systematically. The performance and ease-of-use of the methods were shown using simulated chemical process examples.
To enable the distributed implementation of MPS on a large-scale plant, we used the concept of community structure to optimally decompose large-scale systems into multiple sub-systems. We formulated community detection as a multi-objective optimization problem and employed the concept of weighted modularity to conduct partitioning and identify the subsystems. Based on the non-sorting genetic algorithm, we developed and deployed a computer algorithm that generates a list of non-dominated solutions. The algorithm was tested on the Tennessee Eastman process to show its application and performance. We also used the concept of community detection to decompose a large-scale system into a set of observable subsystems, allowing for the distributed implementation of state estimators and MPS on large-scale systems.
3 Ph.D. (1 female) and 6 (2 female) undergraduate students were trained in this project. The Ph.D. and undergraduate students all successfully graduated. Ahmad Arabi Shamsabadi and Hossein Riazi, two of the Ph.D. graduates, joined Pall Corporation and Evonik, respectively. The third Ph.D. graduate is on internship at Honeywell.
Last Modified: 01/18/2022
Modified by: Warren D Seider
Please report errors in award information by writing to: awardsearch@nsf.gov.
