Modeling the emissions of a gasoline engine during high-transient operation using machine learning approaches

To perform a suitable optimization method in terms of emission and efficiency for an internal combustion engine, first highly accurate and possible real-time capable modeling for the transient operations should be provided. In this work, the modeling of NO x and HC raw emission (before exhaust aftertreatment systems) in a six-cylinder gasoline engine under highly transient operation was performed using machine learning approaches. Three different machine learning methods, namely Artificial Neural Network, Long Short-Term Memory, and Random Forest were used and the results of these models were compared with each other. In general, the results show a significant improvement in accuracy compared to other studies that have modeled transient operations. Furthermore, the shortcoming of Artificial Neural Network for the prediction of the HC emission by the transient operation is observed. The coefficient of determination ( R2) for the best model for NO x prediction is 0.98 and 0.97 for the training data and test data, respectively. This value is 0.9 and 0.89 for the best HC prediction model.

Incipient Surge Detection in Large Volume Energy Systems Based On Wigner-Ville Distribution Evaluated On Vibration Signals

Compressor response investigation in nearly unstable operating conditions, like rotating stall and incipient surge, is a challenging topic nowadays in the turbomachinery research field. Indeed, turbines connected with large-size volumes are affected by critical issues related to surge prevention, particularly during transient operations. Advanced signal-processing operations conducted on vibrational responses provide an insight into possible diagnostic and predictive solutions which can be derived from accelerometer measurements. Indeed, vibrational investigation is largely employed in rotating-machine diagnostics together with time-frequency analysis such as smoothed pseudo-Wigner Ville (SPWVD) time-frequency distribution (TFD) considered in this paper. It is characterized by excellent time and frequency resolutions and thus it is effectively employed in numerous applications in the condition monitoring of machinery. The aim and the innovation of this work regards SPWVD utilization to study turbomachinery behavior in detail in order to identify incipient surge conditions in the centrifugal compressor starting from operational vibrational responses measured at significant plant locations. To this aim, an experimental campaign has been conducted on a T100 microturbine connected with different volume sizes to collect significant data to be analyzed. The results show that SPWVD is able to successfully identify system evolution towards an unstable condition, by recognizing different levels and features of the particular kind of instability that is going to take place within the plant. Instability phenomena regarding rolling bearings have also been identified and their interaction with surge onset has been investigated for diagnostic purposes.

Study on the Transient Characteristics of the Centrifugal Pump during the Startup Period with Assisted Valve

The startup period, one of several transient operations in a centrifugal pump, takes note of some problems with these devices. Sometimes a transient high pressure and high flow rate over a very short period of time are required at the startup process. The pump’s dynamic response is delayed because of the rotational inertia of the pump and motor. Our research focuses on how to get a large flow in a short time when the pump cannot meet the requirements alone without a large power driver. To achieve a strong response in the startup process, a ball valve is installed downstream of the pump. The pump’s transient behavior during such transient operations is important and requires investigation. In this study, the external transient hydrodynamic performance and the internal flow of the pump during the transient startup period are studied by experiments and simulations. In order to find an appropriate matching method, different experiments were designed. The content and results of this paper are meaningful for performance prediction during the transient pump-valve startup period.