My research interests are driven by the desire to be able to solve real life problems, which often leads to the need to develop a new type of a model, a new computational procedure, or a new software required for large-scale numerical experimentation. Current research includes:
The vision of this research program is to design net zero (GHG emissions) Inter-sector Integrated Energy Systems (IIES) for light industry, civic, and personal transportation sectors, without having to use CO2 capture and storage. These sectors currently emit about 40% of GHG emissions in Canada (NRCAN 2018).
Long term goals (5 to 10 yrs.) are to design net zero network of IIES clusters that are (i) resilient to subsystem failures, (ii) flexible to adapt to climate change, and (iii) maintain net zero GHG emissions as energy demands of integrated subsystems vary. Clusters will be classified based on the types of entities contained in them (e.g. one type of a cluster may contain hockey arena, residential building, bakery, and a brewery). Cross-sector integration of energy systems within clusters will make cluster GHG emissions lower than what is achievable by minimizing GHG emissions of individual sector separately.
Short term (next 5 yrs.) objectives are: (i) identify configurations of clusters (what types of civic building and of industrial plants) that lead to stable (steady) GHG emissions reduction even when individual subsystem demands vary (ii) determine integration topology so that the integrated systems are resilient & not vulnerable to cascading network failures, (iii) determine design patterns that provide flexibility with respect to changes in demand, (iv) define optimal roadmap to system configuration changes (structure and modular component capacities) to achieve GHG emissions of specified level, (v) optimal operation and integration with production scheduling, and (vi) determine the lowest possible, economically viable level of GHG emissions with presently available technologies and its dependence on carbon emissions tax.
Related recent publication: Afzali, F., et al. (2018, 2020, 2021), Li, R., et al (2021a, 2021b, 2022).
Research in this area aims to enable optimal decision making for large scale plants. Some of the solutions that we have proposed deal with:
As a part of the research, we have developed a GUI for plant model configuration in GAMS and have built a complete refinery model as an experimental workbench. This has enabled us to examine our solution methods on large problems which are much bigger than typically considered by the researchers in the area.
Recent publications: Li, F., et al. (2021a, 2021b, 2020), Mori, J., et al.(2015, 2017), Fu. G., et al (2018), Tominac, P., et al. (2017, 2018), Jalanko, M., et al. (2018a, 2018b, 2019), Castillo, C.P., et al. (2017a, 2017b).
Our work on fault identification exploits knowledge of process model structure (network topology) to construct a causal network and novel Bayesian contribution indices within the probabilistic graphical network to identify the potential root-cause variables. We demonstrate that the use of the probabilistic graphical network significantly improves fault detection capabilities when compared to methods based purely on data (Mori, J. et al. 2015).
Anomalies are events that happen very rarely and often cause significant disruptions in production. Since the number of anomalies is very small, e.g. 2 or 3 per year), one can not collect a lot of data about them, which necessitates the use of unsupervised learning models. Our recent work (Meyer, K., et al. 2021) examines detection of anomalies in resistive steel welding.
Since there are limits to the size of the nonlinear plant models which can be used for optimal planning or scheduling, industrial practice is to use linear approximations of the plant equipment. This leads to errors in the production plans and schedules. In other words, deemed optimum is not really an optimum. Some published comparisons and our own work indicate that the errors introduced by the linearized model are on the order of 2% to 4% of the optimum (even with piecewise linearization).
Our research has developed a new class of wide-boiling distillation models (hybrid models), comprised of rigorous mass balances, energy balances, and of approximate models representing the internal working of a particular piece of equipment.
This modelling approach has enabled us to model e.g. crude distillation units for use in real-time optimization (RTO) and in scheduling and planning. The accuracy of these models is within 1% of the rigorous models. The models are very small: about 150 to 200 equations (much smaller than 10,000 to 20,000 equations for the nonlinear rigorous model). Using these models in refinery planning, we have been able to demonstrate that even with the best current modelling industrial practices, there is a substantial portion of the profits which are not realized due to the using lower accuracy CDU unit models.
Recent publications: Fu, G., et al (2016, 2018), Li, F., et al (2021), Jalanko, M., et al. (2021a, 2021b)
Economic optimization of real time operation of an equipment can be approached either as an optimization of an economic model predictive controller or as a dynamic real-time optimization. Both of these approaches can be recast as solving NLP problems.
Our recent work (Wang et al, 2017a) applies normalized parametric disaggregation (NMDT) to the economic optimization of NMPC. We demonstrate that a relatively simple algorithm, which can be implemented by advanced control engineers, ensures that globally optimal performance will be achieved. If performance needs to be optimized over a longer time horizon, the size of NMPC model becomes too large to solve it in real time. D-RTO decomposes the economic real-time optimization in two layers. The top layer (D-RTO) computes optimal operating trajectories, while the lower level (MPC) guides the operation along these trajectories. State of the art research in the area uses different models in NMPC and D-RTO layers, leading to the mismatch between the models and consequently to the issues in real time performance. The work we completed recently (Wang et al, 2017b) is the first work to use the same model in both layers, which improves the real-time performance and simplifies the implementation.
Design of Software for Process Simulation, Design, Optimization and Plant Operations
Design of Methods for Plant Control and Optimization
Algorithms for production planning and scheduling
|SEP 773 Graduate||Leadership for Innovation||
Dr. Vladimir Mahalec
|SEP 700 Graduate||M.Eng. Project in Engineering Design Part II||
Dr. Vladimir Mahalec
|SEP 700 Graduate||M.Eng. Project in Engineering Design Part I||
Dr. Vladimir Mahalec