@@ -34,7 +34,7 @@ LUMEN (Liquid Upper stage demonstrator Engine) is a modular pump-fed liquid oxyg
Liquid rocket engines generate thrust by expelling mass with high velocities in accordance with Newton’s third law of motion. The liquid oxidizer and fuel are stored in tanks and pumped into a combustion chamber in which the propellants are ignited and burned. The combustion produces hot exhaust gas which is accelerated to high velocities by a nozzle. Depending on the cycle of the engine, the propellants can be transported differently to the combustion chamber. LUMEN is operated in an expander-bleed cycle which is schematically shown in Figure 1. Both propellants are pressurized by separate turbopump units. While LOX is directly injected into the combustion chamber, LNG is firstly used for the regenerative cooling of the combustion chamber in a counter-flow arrangement. The heated coolant flow is now partially remixed into the LNG max flow via a fuel mixer to actively control the fuel injection (INJ) temperature. The remaining cooling mass flow is further heated within the nozzle extension (NEM) and is then used to drive the LOX and LNG turbines. Afterwards, the turbine exhaust is vented without being combusted in the main combustion chamber (MCC). The generated thrust and therefore the operating point of LUMEN is defined by the combustion chamber pressure, mixture ratio, fuel injection temperature and cooling channel mass flow. A more detailed description of LUMEN can be found in [1, 2, 3].
This benchmark covers some challenging problems of fault diagnosis of safety-critical technical systems. During the operation of rocket engines, sensors, such as pressure transducers, thermocouples and flow meter, are subjected to high thermal and mechanical stresses and are therefore susceptible to potential failure. To prevent engine damage during testing, maximum and minimum thresholds (redlines) are defined for sensors measuring critical parameter, for example the rotational speed of the turbopump or the combustion chamber pressure. If those thresholds are exceeded, a safe shutdown of the engine is initiated. Faulty sensors can therefore cause an unnecessary test abort and thus additional costs. In addition, rocket engines are operated at physical limits and therefore susceptible to failures of components which most likely result in a catastrophic event. The diagnosis system needs to identify faulty sensors and components as fast and accurately as possible to potentially reduce the cost and prevent catastrophic failures. Additionally, misclassifications and missed faults can also result in catastrophic failures.
This benchmark covers some challenging problems of fault diagnosis of safety-critical technical systems. During the operation of rocket engines, sensors, such as pressure transducers, thermocouples and flow meters, are subjected to high thermal and mechanical stresses and are therefore susceptible to potential failure. To prevent engine damage during testing, maximum and minimum thresholds (redlines) are defined for sensors measuring critical parameter, for example the rotational speed of the turbopump or the combustion chamber pressure. If those thresholds are exceeded, a safe shutdown of the engine is initiated. Faulty sensors can therefore cause an unnecessary test abort and thus additional costs. In addition, rocket engines are operated at physical limits and therefore susceptible to failures of components which most likely result in a catastrophic event. The diagnosis system needs to identify faulty sensors and components as fast and accurately as possible to potentially reduce the cost and prevent catastrophic failures. Additionally, misclassifications and missed faults can also result in catastrophic failures.
Fault diagnosis is complicated by limited experimental data, model inaccuracies, measurement noise and inaccuracies. Furthermore, collecting experimental data in fault scenarios is not feasible due to the inherent risks and potential for catastrophic failure during testing. Additionally, the placement of sensors is constrained by factors such as extreme temperatures, high pressures, and vibrations within the engine. Development of new fault diagnosis methods is needed to address these complicating factors, for example, by combining information from multiple sources, e.g.,
* A simulation model derived by domain experts from physical insights about the system
...
...
@@ -54,7 +54,7 @@ The experimental dataset covers the transition of two stationary operating point
### Fault scenarios
In total 10 faulty test runs in the simulation domain for different fault scenarios are pre-simulated. Sensor faults are randomly introduced within the test run and are present until the end. System faults are introduced during transient operation and are also present until the end of the test run. Due to the strong coupling, a system failure influences the operation of the whole engine. As the engine is operated in open-loop, sensor faults only influence the signal of the corresponding sensor. The following faults are present in the pre-simulated dataset and can also be introduced within the simulator:
In total 10 test runs in the simulation domain for different fault scenarios are pre-simulated. Sensor faults are randomly introduced within the test run and are present until the end. System faults are introduced during transient operation and are also present until the end of the test run. Due to the strong coupling, a system fault influences the operation of the whole engine. As the engine is operated in open-loop, sensor faults only influence the signal of the corresponding sensor. The following faults are present in the pre-simulated dataset and can also be introduced within the simulator:
* Failure of the pump bearings: This fault results in an incresead torque $\tau_{\text{p}}$ absorbed by the pump of a factor $f$\
$\tau_{\text{p}} = f \tau_{\text{p}}$, $f > 1.0$
...
...
@@ -85,13 +85,11 @@ There will be three evaluation levels:
The competing diagnosis system solutions will be evaluated on each level based on the following criteria:
* False alarm rate
* Missed detection rate
* Time from fault occurrence until detection
* Fault isolation accuracy (*)
* Computation time
(*) Fault isolation accuracy is evaluated based on the ranking of the true diagnosis candidate has after a fault has been detected.
* False alarm rate - The percentage of samples does the diagnosis system state that a fault is detected when there is no fault in the system.
* Missed detection rate - The percentage of samples does the diagnosis system state that no fault is detected when there is a fault in the system.
* Fault isolation accuracy - The average probability given to the true diagnosis for all samples when a fault is correctly detected.
All performance metrics are between 0%-100%. The total score is computed as a weighted sum of these performance metrics.
For the final evaluation, a set of secret test data will be used to evaluate all participating solutions. The score of each solution per level is based on the sum of a weighted ranking in these categories. The final score is based on a weighted ranking of the score of each level.
