This chapter is divided into four subchapters. The first subchapter gives generalized information about the modeling of internal combustion engines; the second describes the concrete implementation of the GrinSH WTE model. The third subchapter gives an introduction to the subject of signal adaptation and provides the necessary distinction between the PMRS developed here and other methods. The fourth subchapter describes the mathematical background and the implementation of the GrinSH PMRS.
2.2. Modeling of the GrinSH WTE
The basic components of the WTE model developed in this work were the engine model with a speed controller and the generator and frequency converter model.
The entire process of the WTE model operation could be described as a balance between driving power from the engine and extracted power from the generator. This balance could be described as an interaction of the torque increasing processes, torque decreasing processes, and torque oscillations.
The torque-increasing processes were represented by the thermodynamic processes due to the combustion of gas in the cylinders. The torque decreasing processes were represented by the inertia of the crankshaft and of the generator rotor. The torque oscillations were represented by the torque due to the mass oscillation of the crankshaft and due to compression and expansion of the gases in the cylinders [
13].
The mechanical structure of the WTE model was built as shown in
Figure 3. Using this structure, the entire drivetrain can be described by the differential Equations (1)–(7). These equations describe the spring-mass interaction between all cylinders to connect the torques in the single cylinders with the torques in the flywheels and the generator wheel.
The constants k and D and the value of the inertia of all components used in the differential equations are provided by MTU. The angular speed ω and the angle of rotation θ can be calculated by the WTE model using the geometry data of mechanical components, also provided by MTU.
To solve the differential Equations (1)–(7), the torques of all components must be calculated.
The torque of each cylinder pair is calculated as a superposition of the torques of two cylinders connected in this block, minus the torque, which arises from the spring-mass interaction with other sections of the drivetrain. The torque of each cylinder is calculated using Equation (8). Each cylinder contributes to the entire torque with its unique time delay, which is calculated from the angular speed, according to the firing order.
In Equation (8), the cylinder torque is calculated as a superposition of the three components, leading to the entire torque. These are the combustion torque, TCombustion(θ), compression and expansion torque, Tcompr/exhaust(θ), and the mass torque due to the oscillation of moving masses in the cylinder, TOscillation(θ).
The contribution of the combustion is calculated using the empiric Equation (9), which was implemented by Isermann et al. [
13]:
The input parameters for calculating the combustion are θmax, which is the crankshaft angle of the maximal torque in one cycle. In this application, it is set to 370° according to the data provided by MTU. The parameter θ describes the position of the rotational mass during the initialization. The parameter Tstatic is the static torque, given by the average value over one full cycle. This value is calculated by the speed controller from the gas amount due to Equation (10):
where, H (kJ/kg) is the low calorific value of the fuel, g (kg) is the amount of fuel injected into the cylinder during one cycle, and µ (g/kWh) is the specific gas consumption, which can be taken from a lookup table, derived from the specification given in
Figure 4.
Figure 4 also provides information about the operating range of the rotational speed, the torque, and the mechanical power of the MTU engine. The left y-axis shows the mechanical torque of the engine in Nm, corresponding to the dashed line in the graph. The right y-axis shows the power of the engine in kW, corresponding to the solid lines in the graph. The x-axis shows the rotational speed of the engine in RPM. Additionally, the lines of constant specific gas consumption are drawn in the map as the thin lines with the values written on them.
The contribution of the oscillation torque is calculated according to Equation (11):
The constants , , and are calculated from the geometrical data of the engine.
The variables , , and are computed in the simulation for each cylinder, based on the firing order and the rotational speed of the crankshaft.
The contribution of the compression and expansion in the cylinder is calculated according to Equation (12):
where
is the compression-expansion pressure, which can be calculated from the geometry of cylinder and from amount of inserted fuel.
The crankshaft of the engine is loaded with a squirrel cage induction generator, which is connected to the grid via the frequency converter. Hence, the generator can run with variable speeds in a speed range, as shown in
Figure 4. The generator and frequency converter model is implemented as a PT1 (first-order low pass filter) module with the gain factor set to 0.5 and the time constant set to 0.001 s.
The speed controller is built as a PI (proportional-integral) controller managing the gas insert value of the gas according to the operating point of the engine.
The controller is adjusted for different operating points. Around this operating point, the behavior of the controlled model can be considered linear.
The MTU provided the test results, generated with the confidential MTU model of the engine. This data is used here to tune the controller for the angular speed of 1500 RPM and for three load steps, which are represented by the red line in the subplot (b) of
Figure 5.
The adjustment of the GrinSH speed controller was achieved by a stepwise approach to the results of the MTU Model.
Figure 5 shows the adaptation of the PI controller of the GrinSH WTE model speed controller to the results of the MTU brake power test. The subplot (a) shows the reference result of MTU (blue line) compared with the GrinSH WTE model result (red line). Subplot (b) of
Figure 5 shows the load step input used for both models in order to generate the response seen in the upper subplot. The adaptation was carried out for three power steps. The results of the adaptation are analyzed in the following for each power step.
Power step from 9 kW to 250 kW at 10 s. The speed control with the GrinSH model started with a 0.4 s delay. The magnitude of the speed change was the same for both models. The falling gradient of the speed signal was the same for both models. The rising gradient of the GrinSH model, beginning in the inflection point at 15 s was flatter than the one of the MTU Model. Final adjustment of the rotational speed to the reference value of 1500 RPM took place for the MTU Model after 12.1 s and for the GrinSH Model after 14.2 s. This delay was mainly due to the flatter rising gradient of the GrinSH model.
Power step from 250 kW to 490 kW by 30 s. The speed control with the GrinSH model started with a 0.5 s delay. The magnitude of the speed change was the same for both models. The falling gradient of the speed signal was flatter than in the MTU Model. Setting the speed to the reference value of 1500 RPM took place in the MTU Model 10.2 s after the beginning of the control process. The entire control process with the GrinSH Model lasted 14 s. Here, the reference speed was overregulated up to a maximum of 0.012%.
Power step from 290 kW to 670 kW by 50 s. The speed control with the GrinSH model started with a 1.8 s delay. The magnitude of the speed change was the same for both models. The falling gradient of the speed signal was flatter than by MTU Model. Setting the speed to the reference value of 1500 RPM took place in the MTU Model approximately 7 s after the beginning of the control process. The entire control process with the GrinSH Model lasted 14 s. The result of the MTU simulation was considered insufficient by the MTU itself because the speed drop and recovery time exceeded the operating point.
The results show that the GrinSH model represented the MTU model of the gas engine only in proximity. The adjustment of the controller allowed the GrinSH WTE model to achieve the same variation of rotational speed as at the reference model of the MTU in terms of magnitude and slope of the reference rotational speed. However, the time delay of up to 1.8 s for the worst reproduced operating points occurred, and the duration of the transients was up to 50% longer for the GrinSH model for one operating point. However, it must be considered that the worst results of the GrinSH Model simulation corresponded to the operating point where the validation model had also not achieved a steady operating point.
In the context of this work, the GrinSH gas engine model will be used to find out if control of the gas motor with the aid of a PMRS module represents the desired behavior. This means that the focus was the ability of a PMRS module to compensate for the temporal and magnificent disparity of the rotational speed. The GrinSh model achieved the same magnitude of speed reduction with the same gradient as the MTU Model. These are two very important criteria for the planned application. The GrinSh Model shows a more inert control behavior compared to the MTU model; however, the control time remains in the acceptable range for the planned application. The temporal delay and the longer signal setting process are additional factors that the PMRS module has to compensate for.
For the better adjustment of the GrinSH model, the measurement results of the real engine tests are necessary.
2.3. Preventive Signal Adaptation
It is a common case that the control system, represented by a real or modeled process, needs to be controlled in the way that it reproduces a certain behavior.
In the case of simple control systems, which can be considered linear, the correct parameterization of the PID controller is often sufficient to achieve this goal [
15].
However, often it is the case that the system dynamics of the processes to be controlled are so complex that the settings of the PID controller are no longer sufficient to achieve the desired behavior of the system. A compensator is often used here.
There are many different types of compensators. In most cases, the compensators adjust the phase shift of the systems to be controlled. These compensators are based on certain transfer functions, which advance or retard the phase, i.e., they are lead or lag compensators, respectively. However, there are also other applications besides phase control, which the compensator can work with, for example, the correction of offset and gain errors.
Another way to reach the goal of reproducibility of desired behavior by the given system is the implementation of “model predictive control” (MPC) [
16].
In general, the MPC application is based on advanced controller hardware and requires high computational effort. To keep them applicable for real time applications, high computing power is required [
17].
For specific applications such as the control of the engine within the operating scenario given by a WT, solutions with less computational afford than used for MPC applications are required. Looking for such solutions, the authors of this work decided to use inversed transfer function of the model for pre-calculation of the control input signal. This approach can be used in applications, where the main objective is set to the definition of the signal, which leads to the predefined behavior of the controlled model. The challenge of this application is to find the transfer function, which describes the modeled process and the inversion of it [
18].
In this work, the inversion of the transfer function, which represents the model of the WTE was used. Further, the output of the reference WT was preprocessed using the inverted transfer function. Finally, this preprocessed signal is used as an input for the WTE simulation. The method, as well as the mathematical background, are discussed in detail in the following section.
2.4. Inversion of the Input Signal
The aim of the method described below is to retrieve the reference signal of the WT model as precisely as possible by means of the WTE model. The crucial parameters to be emulated are the power and the rotational speed on the generator of the WT. The frequency converter sets the power of the WTE generator. This is implemented with almost no time lag and without mechanical influences on the overall system. The rotational speed of the generator results from the dynamics of the engine. The MTU gas engine is controlled via rotational speed only. For this reason, only the set point of the speed controller of the WTE model was modified by the method described below.
The idea of the method is to find the transfer function
G, which represents the model, and then to use its inverse function
G*, on the input signal
u, coming from the reference WT model. This mathematical operation creates a new input signal
u*, which, being used as the input to the WTE model, generates the output signal equal to the signal of the reference WT model. Reversing the order of the transfer function and the inverse transfer function would not change the result [
19] (see
Figure 6).
The mathematical background of this method is presented by Equations (13) and (14), which show that if a transfer function
G* is an inverse
G, the overall input and output of a serial connection of both systems remain identical.
However, to apply these principles to the GrinSH application, the WTE model needs to be expressed as a transfer function, which is not possible because of the non-linearity of the WTE model. The approximation of the WTE model by PT1 and PT2 transfer functions is a compromise, which allows the use of the principle shown in
Figure 6 by replacing
G* with inverted PT1 and PT
2 followed by the WTE model itself instead of
G (
Figure 7).
The PT1 and PT2 transfer functions used to approximate the WTE model are the following:
Figure 8 shows the PT1 and the PT2 functions plotted together with the WTE model response to a stepwise increase in the angular speed (blue line).
For a better understanding of the results given in
Figure 8, a block diagram was added in the left upper corner of
Figure 8. The colors chosen in the block diagrams correspond to the colors of the simulation results shown in
Figure 8. The blue curve gives the reference signal for the simulation (step function). The red signal is the output of the WTE simulation without the use of any PMRS module. The yellow line is the result of the processing of the step function with the PT1 (15) transfer function. The purple curve is the result of processing the step function with the PT2 (16) function.
The inversion of PT1 and PT2 functions is performed using the approach introduced and tested by Joerg J. Buchholz and Wolfgang v. Gruenhagen [
18] as a proper inversion of transfer functions of dynamic systems. This method solves the problem that the Simulink Matlab is not able to simulate the inversion of strictly proper transfer functions. The strictly proper transfer function is given if the degree of its numerator is smaller than the degree of its denominator, which applies to the transfer function (15) and (16) used in this work.
The method of proper inversion is based on the idea that the transfer function, which is to be inverted, is used in the feedback branch of a control loop with a very high controller gain
K. The transfer function of the inverter is then given by Equation (17).
This application can be implemented in the Simulink environment as a block diagram, shown in
Figure 9.
To test the method introduced above, a step function was processed with a PT2 inverted function and with the PT2 function connected in series. The subplot (a) shows the input u according to the nomenclature introduced in
Figure 9. The subplot (b) shows the u* and the subplot (c) shows the y*. The Gain Factor
K was set to 150,000.
As can be seen from
Figure 10 that the proper inversion approach managed to generate the inverted function from the proper transfer function (16), which can then be used to reproduce the reference signal y* = u in good approximation.
Due to the results shown in
Figure 10, the method introduced here and tested on the step function using the PT2 inversed transfer function followed by the application of the transfer function lead to a sufficient result.
The method illustrated in
Figure 7 was applied to the real scenario generated by the WT model and tested on the WTE model with a PMRS module, consisting of a PT1 or PT2 inverse function, connected in series before the WTE model. The results and discussion of this application will be presented in the next chapter.