**3. Application of expert systems to training simulators**

Expert Systems (ES) are computer programs that incorporate a large amount of knowledge in a very specific field and are used to give advice or solve problems. The use of ES became a viable solution to real problems since the 1980's, since then, the use of ES has proliferating to many technological sectors, as it is demonstrated in the review of Liao (2005). Olmstadt (2000) presents a definition of ES which synthesize many of the different definitions available in the literature, this is: ES use human expertise (the result of deliberate practice on standard tasks over many years) to answer questions, pose questions, solve problems, and assist humans in solving problems. They do so by using inferences similar to those a human expert would make, to produce a justified, sound response in a brief period of time. When questioned, they should be able to produce the rules and processes that show how they arrived at the solution. The main parts of the ES are:


According to the capabilities of ES, their use in training power plant simulators has been explored as intent of minimizing the instructor role. Seifi and Seifi (2002) developed their own intelligent tutoring system for a fossil fuel power plant simulator, while Arjona et al. (2003) utilize CLIPS as foundation for their tutoring system of a part-task simulator for a steam turbine. The C Language Integrated Production System (CLIPS) is probably the most widely used expert system tool because it is fast, efficient and free (Wikipedia, 2011b). CLIPS is an inference engine initially developed to facilitate the representation of knowledge to model human expertise, it provides a cohesive tool for handling a wide variety of knowledge with support for three different programming paradigms: rule-based, object-oriented and procedural. Rule-based programming allows knowledge to be represented as heuristics, or "rules of thumb", which specify a set of actions to be performed for a given situation. Object-oriented programming allows complex systems to be modelled as modular components, which can be easily reused to model other systems or to create new components. In the procedural approach, CLIPS can be called from a procedural language, perform its function, and then return control back to the calling program (CLIPS, 2011).

#### **3.1 Knowledge acquisition and its representation**

A critical task in the development of a knowledge-based system is the knowledge acquisition, which is the process of collecting information from any source (expert knowledge, book, manuals, etc) needed to build the system. In the case of a simulator a good reference to carry out this process are the available training procedures and the operation manuals. Usually these

Besides the referred documents of the ISA, IAEA and EPRI, additional information about simulators and training programmes can be founded in EPRI (1998) and IAEA (1996, 2003). The documents published by IAEA deal about the training of operators of nuclear power

Expert Systems (ES) are computer programs that incorporate a large amount of knowledge in a very specific field and are used to give advice or solve problems. The use of ES became a viable solution to real problems since the 1980's, since then, the use of ES has proliferating to many technological sectors, as it is demonstrated in the review of Liao (2005). Olmstadt (2000) presents a definition of ES which synthesize many of the different definitions available in the literature, this is: ES use human expertise (the result of deliberate practice on standard tasks over many years) to answer questions, pose questions, solve problems, and assist humans in solving problems. They do so by using inferences similar to those a human expert would make, to produce a justified, sound response in a brief period of time. When questioned, they should be able to produce the rules and processes that show how they

 Knowledge base. It contains the knowledge of the facts and experiences of experts in a particular domain, i.e., it contains general knowledge about the expert domain. Inference engine. It is responsible of modelling the process of human reasoning. This

 User interface. This represents the method through the ES interacts with the user. This may require designing the interface using menus, dialog boxes, forms, graphics, etc. According to the capabilities of ES, their use in training power plant simulators has been explored as intent of minimizing the instructor role. Seifi and Seifi (2002) developed their own intelligent tutoring system for a fossil fuel power plant simulator, while Arjona et al. (2003) utilize CLIPS as foundation for their tutoring system of a part-task simulator for a steam turbine. The C Language Integrated Production System (CLIPS) is probably the most widely used expert system tool because it is fast, efficient and free (Wikipedia, 2011b). CLIPS is an inference engine initially developed to facilitate the representation of knowledge to model human expertise, it provides a cohesive tool for handling a wide variety of knowledge with support for three different programming paradigms: rule-based, object-oriented and procedural. Rule-based programming allows knowledge to be represented as heuristics, or "rules of thumb", which specify a set of actions to be performed for a given situation. Object-oriented programming allows complex systems to be modelled as modular components, which can be easily reused to model other systems or to create new components. In the procedural approach, CLIPS can be called from a procedural language, perform its function, and then return control back to the calling program (CLIPS, 2011).

A critical task in the development of a knowledge-based system is the knowledge acquisition, which is the process of collecting information from any source (expert knowledge, book, manuals, etc) needed to build the system. In the case of a simulator a good reference to carry out this process are the available training procedures and the operation manuals. Usually these

engine works with the information contained in the knowledge base.

plants but many of the key concepts are applicable to the fossil industry.

**3. Application of expert systems to training simulators** 

arrived at the solution. The main parts of the ES are:

**3.1 Knowledge acquisition and its representation** 

documents describe the objectives of the training session, the instructor actions (initial condition, malfunctions, etc) and the required actions to operate the unit (to turn on pumps, to open valves, etc), all this information helps to build a very complete knowledge base. These documents include normal operation (start-up, normalization and shutdown operations, for each one of the power plant systems) and in many cases abnormal operations. Additional tests can be performed in the simulator with the aim of getting supplementary information, mainly in the cases of malfunctions, where there are not enough documented records.

The acquired knowledge must be formalized and ordered with the aim of being useful to the ES; this process is named "Knowledge representation". One of the most common methods to represent knowledge is the production rules. In this method, the knowledge is divided into small fractions of knowledge or rules. A rule is a conditional structure that logically relates the information contained in the part of antecedent with other information contained in the part of the consequent. A very important feature is that the knowledge base is independent of the inference mechanism used to solve problems. Thus, when the stored knowledge become obsolete, or when new knowledge is available, it is relatively easy to add new rules, delete old ones or correct existing errors. Therefore, there is no need of reprogramming all the expert system. The rules are stored in hierarchical sequence logic, but this is not strictly necessary. It may be in any sequence and the inference engine will use them in the right order to solve a problem. This approach is also called IF-THEN rules and some of its main benefits are their modularity and that each rule defines a relatively small and independent piece of knowledge. However, the process of coding the rules can be a cumbersome chore for personnel little familiar with this kind of responsibilities. Tavira-Mondragón et al. (2010b) describes a graphic tool which serves to build training exercises for a combined cycle power plant simulator, with no guidance of a human instructor. This editor contains a group of blocks where each block represents a rule (or a group of rules), and each block is customized by their characteristic parameters. Figure 10 shows in a schematic way, the graphic representation of the malfunction insertion during a training

Fig. 10. Knowledge representation in CLIPS.

Fossil Fuel Power Plant Simulators for Operator Training 107

can use any one of his consoles to supervise and control any process of the power plant. The instructor console is provided with two monitors, hence, besides of using the instructor functions described previously, he can display any screen of the operator consoles with the purpose of watching any operative action carried out by the trainee. In many architectures of simulators is common to find an additional PC, the maintenance station (not shown in Figure 11), which serves as a backup if the instructor console fails or as a test station, this means that any software modification is tested and validated in this station before any

The simulation software is designed with the purpose that the response of simulator is comparable with the results observed in the reference plant under similar conditions. As expected, besides the mathematical models, it is required the execution software or simulation environment. Tavira-Mondragón et al. (2010a) describes the software architecture for a simulation environment. The software architecture of the simulation environment has four main parts: the real-time executive, the operator module, the instructor console module, and mathematical models. Each one of these modules can be hosted in the same or in different PC, and they are connected through the TCP/IP protocol under Windows operating system. A brief description of each module is shown in the

 Real-Time Executive. The real-time executive module coordinates all simulation functions, so it includes the mathematical model launcher, the managers for: interactive process diagrams, global area of mathematical models and instructor console. Additionally it includes data base drivers and the main sequencer, which sequences all

 Operator Module. The operator module is in charge of the operator HMI and manages the information flow with the executive system. The HMI consists of interactive process diagrams, which are animations with static and dynamic parts. The static part is constituted by a drawing of a particular flow diagram whereas the dynamic part is configured with graphic components stored in a library which are related to each one of the plant's equipment, e.g., pumps, valves, motors, etc. These components have their

 Instructor Console Module. This module carries out all the tasks related to the graphical interface of the instructor and a module to dynamically update the instructor console

The better option for the operator console is to emulate via software the consoles of the actual plant, this represent the less cost option compared with the acquisition of such consoles, in this way a graphic imitation of the actual HMI provides a suitable operation interface. This HMI is a graphical application based on a multi-window environment with interactive process diagrams, these diagrams are organized in hierarchical levels following the organization of the power plant systems, i.e. boiler, turbine, etc. There are two main types of diagrams: information diagrams and operation diagrams. The first type shows values of selected variables. The values are presented as bar or trend graphs. The trainee

following paragraphs (the mathematical models are discussed in the next section).

own properties and they are established during the simulation.

**4.3 Main features of the human machine interface for trainees** 

change is carried out in the simulator.

the simulator functions in real-time.

with the simulation information.

**4.2 Software architecture** 

session and the corresponding rules generated by the editor, which are used during the execution of CLIPS as tutoring system. In such way, the ES is responsible of tracking the status of the simulation to determine the group of rules that should be fired. Due to its inference engine, and according to the configuration of the simulation exercise, the ES is able to modify the simulation process, because it can insert malfunctions, modify values of selected process variables, and change the status of the simulation without the intervention of a human instructor.

The use of ES is especially suitable for training standalone systems, because these systems incorporate: a simulator of a power plant, an intelligent tutor to guide the training session, and besides it can include the trainee evaluation and study material in some multimedia format as theoretical support of the training objectives. Naturally, the HMI for the trainee must be designed bearing in mind that the user, in addition of its operation interfaces, will need "a window" to observe the tutor messages.
