Roberto Melli

A prototype expert system for the conceptual synthesis of thermal processes

Abstract An expert system capable of assisting the process engineer in thermal processes design selection can improve both the quality and the variety of the solutions proposed. Conceptual design is part of the engineering design process which converts design requirements into an acceptable design solution. Nowadays, engineering design problems have become more and more complex and the increasing complexity has made synthesis a very difficult task: it is during this stage of the design process that the creativity and experience of an engineer are most needed. Inappropriate component requirements and/or coupling can significantly reduce the capability of a process to meet design demands. The prototype presented in this paper (COLOMBO) is an attempt to show that an expert system can be of valid help in finding answers to conceptual process selection. An exhaustive search across the data base allows the user to evaluate plant configuration constituted by virtually every technically feasible combination of components. Generally, the problems under exam result in a goal-driven search process and, therefore, a backward chaining approach is used here. The data base has been subdivided into two main classes: equipment and resources. Each “equipment” object has in-flows and out-flows (energy, matter,…) which define its operational capabilities. Interconnection of these objects is possible when there is compatibility between the out-flows of one and the corresponding in-flows of the next object. Of course, connectivity must satisfy all the constraints imposed to the problem domain. Presently only a few components have been included in the data base and, therefore, only few conventional configurations are produced by the expert system. Nevertheless COLOMBO produced four configurations for a power generation plant and three process configurations for a cogeneration plant which resolve to be equivalent to the human expert choices.

Optimal operation and sensitivity analysis of a large district heating network through pod modeling

District heating networks (DHNs) are crucial infrastructures for the implementation of energy efficiency and CO2 reduction plans, especially in countries with continental climate. DHNs are often complex systems and their energy performance may be largely affected by the operating conditions.Optimal operation of DHNs involves the optimization of the pumping system. This is particularly important for large networks and for low temperature networks. A common practice to perform optimization consists in using a phyical model. Nevertheless, simulation and optimization of DHNs may involve large computational resources, because of thire possible large extension and the number of scenarios to be examined. A reduced model, obtained from the physical model, can be effectively applied to multiple simulations of a network, with significant reduction of the computational time and resources.In this paper, a large district heating system, which supplies heating to a total volume of buildings of about 50 million of cubic meters, is considered. The use of a reduced model based on proper orthogonal decomposition (POD) is investigated. Various operating conditions corresponding to partial load operation are analyzed using a fluid-dynamic model of the network. Results show that optimal settings are not particularly regular with respect to load variation. This means that any variation in the thermal load generally involves changes in the set points of all groups. For this reason, a sensitivity analysis is performed using the POD model in order to check the opportunities to limit the number of variations in the pumping settings without significant penalization of the total pumping power.The proposed approach is shown to be very effective for the optimal management of complex district heating systems.Copyright

An Example of Thermo-Economic Optimization of a CCGT by Means of the Proper Orthogonal Decomposition Method

This paper presents an application of the Proper Orthogonal Decomposition (POD) technique (also called Karhunen-Loeve Decomposition, Principal Component Analysis, or Singular Value Decomposition) to the thermo-economic optimization of a realistic Combined-Cycle Gas Turbine (CCGT) process. The novel inverse-design approach proposed here employs the thermo-economic cost of the two products as objective function.The proposed procedure does not require the generation of a complete simulated set of results at each iteration step of the optimization, because POD constructs a very accurate approximation to the function described by a certain number of initial simulations, and thus “new” points in design space can be extrapolated without recurring to repeated process simulations. Thus, the often taxing computational effort needed to iteratively generate numerical process simulations of incrementally different configurations is substantially reduced by replacing much of it by easy-to-perform matrix operations: a non-negligible but quite small number N of initial process simulations is used to calculate the basis of the POD interpolation and to validate (i.e., extend) the results.Since the accuracy of a POD expansion depends of course on the number N of initial simulations (the “snapshots”), the computational intensity of the method is certainly not negligible: but, as successfully demonstrated in the paper for a realistic CCGT inverse process design problem, the idea that additional full simulations are performed only in the “right direction” indicated by the gradient of the objective function in the solution space leads to a successful strategy at a substantially reduced computational intensity. This “economy” with respect to other classical “optimization” methods is basically due to the capability of the POD procedure to identify the most important “modes” in the functional expansion of the vector basis consisting of a subset of the design parameters used in the evaluation of the objective function.Copyright

Multidisciplinary DSS as Preventive Tools in Case of CBRNe Dispersion and Diffusion: Part 1: A Brief Overview of the State of the Art and an Example – Review

The paper addresses some important issues related to the need for a timely, reliable and accurate tool for the early warning in case of CBRNe events. The state-of-the-art of the currently available tools is briefly presented in the first part of the two-papers set. While the accurate calculation of the dispersion of both lighter- and heavier-than-air contaminants in complex three-dimensional domains is definitely possible with commercially available CFD packages, the time needed to obtain a reliable numerical solution, under the pertinent atmospheric conditions prevailing at the time of the attack, exceeds the requirements of a first-aid intervention. Therefore, it would be advisable to combine these CFD packages with some sort of “intelligent” Decision Support System that makes use of multidisciplinary knowledge base and of some kind of detection-diagnostic-prognostic Expert System. The DSS could be interfaced with some standard early detection tools and ought to include an enhanced diagnostic/prognostic utility based on a specific series of local CFD simulations of dispersion events. Its use ought to be relatively easy for trained personnel. Since the database for the CFD dispersion calculation is by definition “local”, detailed maps of the presumable target areas must be included in the database. The second part of this paper presents a detailed description and one example of application of such an Expert Assisted CFD dispersion calculation, named FAST-HELPS (Fast Hazard estimate of low-level particles spread).

A Thermoeconomic-Artificial Intelligence Combined Approach to the Diagnosis of Energy Systems

In this paper, the combined application of two different approaches to the diagnosis of energy systems is proposed. The first approach is based on thermoeconomic analysis. It consists on the filtration of effects due the dependence of the efficiencies of components on their operating condition. This is obtained through productive models which relate resources and products. With respect to physical models, these are generally less accurate but more compact and thus more suitable to deal with the available measurements in real plants. The second approach is an artificial intelligence (AI) technique. This is based on the calculation of appropriate indicators that experience shows to be affected by possible anomalies. Malfunctions are detected and recognized through the analysis of deviations registered by the indicators during plant operation. The diagnosis methods are applied to a gas turbine plant with real anomalies. These are investigated in order to highlight possible advantages and disadvantages of the two methods and the benefits that can be reached through their combined application.Copyright