High speed limiter vs. torque prediction feedforward
Theoretically, the torque prediction feedforward is the optimum way to avoid speed changes during load changes [4] [13][14] [15]. Practically, however, there is some statistical uncertainty attached to the load predictor output. Furthermore, prediction is available only during the extreme sudden load drops. The high speed limiter contains less uncertainty, is always active and as simulation shows its performance is extremely close to the load prediction method during any kind of load fluctuation but sudden load drops. Simulation in Figure 2 assumes a perfect load prediction and should be compared to the high speed limiter simulation result [19].
Notice that load drops up to 75 seconds are handled similarly by the high speed limiter and the prediction based method. This result illustrates that the algorithms are suited for operating in parallel, because then the overspeed protection is available for a wider range of load drops.
The high speed problem during slow load variations
The overspeed protection cannot prevent the speed from exceeding the rated MCR value, even though it reduces the maximum speed reached. Therefore an additional pre-processing function of the governor setpoint was suggested.
The setpoint adaptation is an add-on to the traditional pre-processing functions like the well known barred speed ranges and load program.
A speed setpoint error is calculated by subtracting the current speed from an adjustable limit, MCR level for instance. If this error is negative (meaning that the engine is running faster than the limit) the algorithm output increases with time according to a specified ramp. If the speed error is positive the output decreases in time according to a slower ramp. In any case this output is subtracted from the current original speed setpoint. The result is that during periodic load variations the mean value of the effective speed setpoint will converge to a value which causes the engine speed not to exceed the specified limit.
Figure 3 shows a scenario which contains "large" load variations for the first 150 seconds and then smaller load variations for another 150 seconds. In both cases the effective speed setpoint fluctuates around a mean value which causes the engine to run above 100% of MCR for much less time than without the algorithm [19].
Due to its nature, this algorithm is not capable of providing protection against faster load variations, but it performs well in combination with the high speed limiter and the torque prediction based control.
The heavy running problem
During most of the time during which load variations due to weak changes are present the engine will be running "heavy". This fact makes it impossible for the engine to run at MCR speed and furthermore the governor limiters will "pull down" the fuel index in order to protect the engine. This result is undesirable. When the load decreases, the speed increases which causes the torque limiter to allow higher index. The higher index increases the speed even more. In other words the PID speed controller in the governor is no longer active and the speed variations become large.
This problem can be solved by "lifting" the index limiters by approximately 150% and furthermore implementing the original limiters in a way where they slowly affect the speed setpoint to the speed controller instead of the speed controller output. This will allow for the speed controller to stay active and still the engine speed will be kept below the limiter curves on a "long term" basis (5-10 seconds). The "lifted" curves will ensure that overloading never exceeds 15%. The control strategy is illustrated in Figure 4 [19].
An example of the above strategy is illustrated in Figure 5 for an integration time of 5 seconds [19].
The turbocharger stall problem
Previous experience of ACME project partners indicates a correlation between the exhaust gas temperature after the turbine and the distance to the surge line on the compressor map [7]. One of the first steps in the series of events which may lead to surging is a significant temperature rise and consequently there is a reason to believe that setting a limit to how fast fuel index is allowed to increase, may reduce the risk of surging. This rate limiter can be calculated with the turbine outlet temperature derivative as an independent parameter. If the temperature is stable the maximum allowed index rate is 25% per second. If the measured temperature rate is 40 degrees per second no index increase is allowed. . However, one must expect that calibration of the two parameters will be necessary during commissioning. Furthermore it is here assumed that the temperature can be measured with a time constant of 0.5 seconds. Simulation results in Figure 6 show the effect during an instantaneous speed setpoint increase from 70 to 90% of MCR [19].
Notice how the distance to the compressor surge line is kept much higher with the algorithm. Without the algorithm the compressor is practically on the surge line. The acceleration is slower with the algorithm during the initial 5 seconds but at longer terms there is no practical difference. This is satisfactory because the duration of a plant acceleration is very much longer. Furthermore higher air-to-fuel ratio will result in higher combustion efficiency and less smoke. The same phenomenon can be found during large slow load changes.
During fast load variations the algorithm does not affect the outcome significantly. During the sudden load drops the algorithm causes this acceleration to be slightly slower. The effect on the distance to the surge line is however not significantly improved in this case because the time constant in the temperature measurement is too large for this scenario. This effect is illustrated in Figure 7 [19].
The performance of the algorithm is apparently satisfactory.
2.3. The onboard TP system
The Propeller Torque Prediction Algorithm (TPA)
At this point, and prior to the presentation of the Torque Prediction System (TP System) a brief reference to the propeller torque prediction algorithm (TPA) is necessary. The main idea of the TPA is the event identification, based on pattern matching techniques. Statistical analysis of towing tank experiments with scaled ship models has allowed the determination of a correlation existing between aft ship vertical acceleration and propeller torque. The TPA has been based on this observation.
An event database containing a limited, finite number of torque profiles of interest has been constructed as the outcome of the towing tank tests and was enriched continuously and dynamically during full-scale sea trials. Based on a distance function of a set of statistically independent variables (propeller torque, propeller torque slope and vertical aft ship acceleration) known at instant to, an event from the database similar to the occurring one is determined and therefore the corresponding torque profile. Figure 8 is a conceptual illustration of the above mechanics.
The torque profile generated by the TPA is selected to match to, but not parameterized by, the occurring event. Conclusively, a finite set of predicted propeller torque profiles is available, and each and every member of this set can be assigned a unique ordinal number (index), Icurve, by simple enumeration.
