NAKAYA Makoto1 NAKABAYASHI Akio1 OHTANI Tetsuya1
We have developed the world's first on-line tracking simulator which matches the physical world with the virtual world. This paper describes this tracking simulator and its application to plant operations in the future. As examples of its use, we describe the prediction operation which ensures optimum plant operation by knowing the future plant behavior, and a method to create more accurate models by adjusting model parameters as variables. Regarding its applications, we explain the importance of various uses of plant models during the plant life-cycle, and then describe an application to production control which is achieved by direct product quality control with frequent re-scheduling while maintaining quality.
- Instrument & Control Research Center, Corporate R&D Headquarters
INTRODUCTION
Since the beginning of this century, resource prices have been escalating because of economic growth in developing countries such as China and India, and the chemical plants of multinational European and American companies producing basic materials have grown in scale with their huge capital investment. In contrast, chemical manufacturers in Japan are focusing on global niches, and are rapidly shifting to functional materials for competitive industries such as cars and electronics.
Regarding future plant operations, large overseas plants require even safer operations to ensure the stability of product supply, and also more efficient operations to reduce costs. Meanwhile, functional materials plants must handle irregular production to flexibly meet the unique specifications and demands of customers. Highly accurate plant models are needed to decide how to operate actual plants, ranging from basic material chemical plants to functional material plants.
Yokogawa, which is introducing technological innovation into plant operations by integrating the real world with the virtual world, has developed the world's first tracking simulator which runs a dynamic simulator on-line.1 2 3 4 This tracking simulator can be used not only for training plant operators, but also for new applications in the field of operation support. This paper describes examples of using the tracking simulator and its application to plant operations in the future.
TRACKING SIMULATOR DEVELOPED BY YOKOGAWA
 |
| Figure 1 Periodical prediction screen |
We have developed two model parameter identification methods in order to precisely simulate the behavior of the act ual plant. One method is a sequential parameter identification method, where a few model parameters are updated periodically and made to conform to the process values which are taken into the simulator. With this method, the small number of updating parameters reduces the modeling error caused by mismatches between the actual plant and the model. However, the values of these parameters could exceed the expected physical range in some cases.
The other method uses a data reconciliation technology with dynamic compensation to solve the above problem, where multiple parameters over a broad range of the plant model are updated periodically.5 Although data reconciliation usually assumes a steady state, our method allows parameters to be identified even in an unsteady state.
With these two parameter identification techniques, we have succeeded in developing a tracking simulator which precisely simulates the target.
EXAMPLES OF USING THE TRACKING SIMULATOR
 |
| Figure 2 Case study screen |
We introduce two examples of using the tracking simulator. In the first example, the simulator is used for predicting future plant operations.6 In the second, it is used for configuring high-accuracy models by estimating model parameters exactly.7
Operation considering future prediction
Even though automation of plant operations is making progress by introducing advanced control, there are still many cases where series of operations are entrusted to operators. In this case, usually, the operators conduct operations, check the results by trends, and then make sure that the settings have been reflected to the plant.
However, when plant operations have large time constants, it takes a long time until the response can be confirmed. For example, when conducting a series of operations for changing grades, the loss in products increases if there are mistakes during the operations, because it takes a long time until the result of the operations can be checked. Also, since operations to a plant have a big impact on the plant, there is heavy psychological pressure on the operators. To solve these issues, we suggest predicting plant operations.
 |
| Figure 3 Parameter estimation as a function of current density |
The future prediction has two functions. One is a "periodic prediction function" which shows how a plant behaves if the current operating condition is kept. The other is a "case study function" which shows the result of a virtual operation on the current state of the plant.
With the periodic prediction function, the system predicts the plant behavior periodically and shows the predicted result on the screen. Figure 1 indicates that the result of the latest prediction is different from that of the last prediction because of a disturbance. In such a situation, the operator can figure out how to deal with the plant by using the case study function. Figure 2 shows the result of each analysis using the case study function when three different conditions are input. The operator can take the most appropriate action considering several conditions.
Though application to future prediction is described by taking a disturbance as an example, it can also be used to optimize plant operation. With these functions, a plant can reach a required condition even faster, and plant operations can be improved. At the same time, it is possible to examine the validity of operations in advance, and this can reduce the psychological pressure. Moreover, the automatic simulation of a plant upon an input operation and the automatic verification of the result of the simulation can be a foolproof measure ensuring input operations to the control system.
Configuration of high-accuracy model by parameter mapping
 |
| Figure 4 Result of the simulation by model modification |
Generally, simulation parameters are considered as constants. But with the tracking simulator, model parameters can be identified as a function of operating conditions. The result of applying this to a fuel-cell power generating system is described below.
A fuel cell generates electricity with the reverse reaction of elect rolysis of water by conducting hyd rogen gas to the anode and air to the cathode. This is expressed by the following reaction formula.
Anode side: H2 → 2H+ + 2e-
Cathode side: ½ O2 + 2H+ + 2e- → H2O
We have used ion conductivity, which expresses the condition of hydrogen ion conductivity in a moist polymer membrane between a node and cathode, as a tracking parameter, and then conducted the tracking simulation to make the ion conductivity correspond to the output voltage of the fuel cell. After the tracking starts, we make the output voltage of the fuel cell correspond to the measured value by adjusting the ion conductivity parameter every time the load current changes. Since the volume of water in the polymer membrane, through which hydrogen ions are conducted, varies with the load current, it is logical to express the ion conductivity as a function of the load current. The result of the ion conductivity obtained as a function of the load current is shown in Figure 3, which was estimated when the load current of the fuel cell became steady after it had been changed. Figure 4 shows the result with only tracking simulation and the result of another simulation using the ion conductivity model as a function of the load current. Because an adjusting parameter cannot respond to sharp changes, a transient response cannot be expressed accurately with the tracking simulator alone. However, it was verified that satisfactory simulation results can be obtained when applying the approximation formula derived from Figure 3.
When configuring a simulation model, it was verified that the model parameter which has been treated as a constant can be embedded into the model as a function of some variables. With this method, more accurate simulations become possible.
USE OF MODELS IN PLOANT MODEL LIFE-CYCLE
 |
| Figure 5 Using a model in the plant life-cycle |
The plant model using the tracking simulator which we have developed is a physical model based on the material balance and energy balance. Because investing in this type of plant model is generally enormously expensive, it is important to effectively utilize the plant model throughout the plant life- cycle from master plan to decommissioning which is indicated in Figure 5. It is assumed that the model is used not only for designing, start up, running and operating a plant and extending plant facilities, but also for analysis, prediction and diagnosis, and optimization.
As shown in the center part of Figure 5, there are various types of plant models, such as a rigorous model which is defined based on physical principles like the tracking simulator, and a simple model where the relationship between inputs and outputs is defined. At present, a model is created individually for each intended purpose such as designing and training. From the operational standpoint of handling a model as shown in the bottom part of Figure 5, the current usual procedure is to create a model, modify it to fit an actual plant, and reuse it as many times as possible corresponding to intended usages.
The use of a model in the future when handling a model, namely when configuring, modifying, and reusing a model, is described below.
- Model configuration
When configuring a model, even for a rigorous model, commercial simulators are equipped with a tool to configure the model, thus users generally have an environment where they can easily configure the model using a GUI (Graphic User Interface). However, in order to create an accurate model, it is necessary to be well acquainted with the physical and chemical phenomena of an intended process. If a satisfactory degree of approximation cannot be obtained with the plant device model provided by a simulator vendor, users often have to analyze phenomena and define a model equation by themselves. Thus, model configuration requires further technological development for efficient application of models. - Model modification
As for model modification including the tuning process to adjust the model to match the measured values, the time for modification can be reduced by applying the function of our tracking simulator, which identifies model parameters and modifies a model automatically, and by using this function online. - Reusing a model
With model transformation technologies, it is possible to reuse a plant model by converting it to a required application. The example of this usage is explained in the next section.
APPLICATION PROPERTY CONTROL FOR FLEXIBLE PRODUCTION
 |
| Figure 6 Quality control using tracking simulator |
As mentioned in the Introduction, when producing high value-added functional materials (which are also called fine chemicals), generally, the approach is to produce small quantities of many products. It is anticipated that the chemical plants in Japan, which are aiming for global niches, will focus on fine chemicals where small quantities of many products are manufactured. In this area, even though customers of chemical manufacturers are in the same type of business, products must be manufactured to meet the individual specifications given by each customer, and it is important to meet each user's needs.
For manufacturing highly-functional materials with small quantities and many kinds, it is necessary to establish a style of production control which can reschedule in order to keep the stock as small as possible, switch grades frequently and effectively while maintaining the quality of the product, and respond to sudden changes of order from a customer.
Currently, in order to realize this ty pe of flexible production system, we are investigating using plant models. Especially for quality control, we intend to create a production system where an operator can directly specify the quality of products, while the required product quality used to be achieved by controlling variables such as temperature, pressure and flow rate. With the tracking simulator capable of simulating the actual plant's behavior, sufficient and necessary information to control quality can be obtained in real-time. Thus, information such as physical information which used to be measured by inserting a sensor and compositional value which required a lot of time to analyze, can be accurately and instantly obtained from a virtual space.
Figure 6 shows the structure of the system which controls the quality of a product using the tracking simulator. When the properties which determine the product quality cannot easily be expressed with physical and chemical models, the properties are estimated by connecting the tracking simulator with soft-sensors. Although a conventional advanced control combining a multi-variable control package and chemical soft-sensors has been restricted to controlling a plant locally, with the tracking simulator, the virtual information of the entire plant is available. For example, it is possible to directly control the quality and significant indicators under constraint conditions, such as usable energy quantity and CO2 emissions.
CONCLUSION
We have described our tracking simulator and its application to plant operation in the future. Regarding the latter, we described that it is important to modify the model on a daily basis to maintain accuracy and transform it depending on each purpose in order to utilize the plant model, based on physical and chemical principles, effectively throughout the plant life-cycle. In the area of flexible production of diverse products expected in the future, our goal is to realize a production control supporting tool which can guarantee the amount of production, time required for production and product quality by using the plant model technology.
REFERENCES
- M. Nakaya, G. Fukano, et al., "On-line Simulator for Plant Operation," SICE Annual Conference 2005, 2005, pp. 3811-3815
- M. Nakaya, G. Fukano, et al., "On-line Simulator for Plant Operation," Proceedings of the 6th World Congress on Control and Automation, 2006, pp. 7882-7885
- A. Nakabayashi, G. Fukano, et al., "Application of Tracking Simulator to Reforming Process," SICE-ICASE International Joint Conference 2006, 2006, pp. 1871-1875
- Gentarou Fukano, Yasushi Onoe, et al., "Application of Tracking Simulator to Steam Reforming Process," Yokogawa Technical Report English Edition, No. 43, 2007, pp. 13-16
- Tatenobu Seki, Gentarou Fukano, et al., "Innovative Plant Operation by Using Tracking Simulator," Yokogawa Technical Report Vol. 52, No. 1, 2008, pp. 35-38 in Japanese
- Tatenobu Seki, Gentarou Fukano, et al., "Transient Response Prediction with Tracking Simulator," SICE the 7th Control Division Conference, 72-2-3, 2007, in Japanese
- M. Nakaya, K. Kawaguchi, et al., "MODEL PARAMETER ESTIMATION BY TRACKING SIMULATOR FOR THE INNOVATION OF PLANT OPERATION," Proceedings of the 17th World Congress The International Federation of Automatic Control, 2008, pp. 2168-2173