| Rotating stall and surge are two types of ?ow instabilities that have been widely recognized for axial-?ow compressors, which could cause deterioration of the air ?ow through a compressor and decrease in the pressure ratio and system e?ciency. Even worse, they could also lead to blade fracture, engine failure or shut down due to unacceptable levels of system vibration. Therefore, it is crucial to have timely and reliable prevention of the rotating stall and surge for improving the lifespan and performance of axial compressors,and ensuring the personal safety. Especially for axial-?ow compressors, as rotating stall is commonly regarded as an inception of surge, it is more important to detect rotating stall. Based on the deterministic learning theory, this thesis is mainly focused on the modeling of the aerodynamic instabilities in axial-?ow compressors and presents the approach to detecting rotating stall precursors, with the objective to extend the stable operating region of compressors, thus improving the performances of aero-engines. The main contributions of the thesis are as follows:1. We study online experiments of the modal-type stall for a low-speed axial ?ow compressor. Based on deterministic learning(DL) theory and dynamical pattern recognition, we implement an approximately accurate modeling approach and the early stall detection by using the low-speed axial ?ow compressor test rig in the key laboratory of aviation engine of Beihang University for online experimental veri?cations. Firstly, by installing high response dynamic pressure transducers arranged circumferentially around the casing of the low-speed axial ?ow compressor, the dynamic pressure data before stall and during stall inception are collected during the o?ine data processing. Then, the system dynamics underlying prestall and modal-type stall inception signal are identi?ed and stored in constant radial basis function(RBF) neural networks(NN). Secondly, since online measurement scheme design and data processing directly a?ect the reliability of the method, a good real-time operation needs to be considered. According to the studies of the problems, the online programs for the modal-type stall inception detection are implemented based on Lab VIEW. Using the method of dynamic pattern recognition in DL, it is shown that, at di?erent working speeds, the approach can successfully detect modal-type stall precursors of the compressor 0.3-1 second before the start of rotating stalls.2. The early detection of rotating stall in a low speed axial compressor under inlet distortion is studied. Inlet ?ow distortion is one of the main factors resulting in both engine stability and performance loss, which would exacerbate the instability inside the compressor ?ow ?eld and even cause the occurrence of surge. Therefore, capturing rotating stall in axial compressors with inlet distortion is very important for improving the performance and the stability of axial compressors. This thesis describes a stall detection method based on the DL theory to predict the onset of ?ow instability for axial compressors under the inlet ?ow distortion conduction. With the simulation of the inlet ?ow distortion with plugboard, a variety of experiments are conducted on a low-speed axial?ow compressor test rig of Beihang University at di?erent rotating speeds. Inlet distortion would increase the instability ?ow and make the detection schemes more di?cult to capture the weak signal of stall inception. Firstly, the thesis studies how parameter settings of the fault estimator in?uence the fault residuals and then seeks for the optimal fault estimator parameters, in order to exactly predict small oscillation faults. Secondly, by installing high response dynamic pressure transducers arranged circumferentially around the casing of the axial compressor, the dynamic data for the inlet ?ow distortion are collected. Then, using the rapid detection of small oscillation faults based on the DL method, the experimental study for the detection of stall precursors with inlet distortion is conducted. Our results show that this approach can successfully detect the inception signal of rotating stall under inlet ?ow distortion.3. The modeling and rapid detection of the spike-type stall inception in axial compressors are performed. Spike-type stall inception is characterized by a locally shortlength-scale disturbance and propagating faster than the modal-type, which is more common in axial compressor ?ow crashes. The localized nature of the spike-type stall inception and the rapid decrease of the ?ow make the detection of the indications of impending stall technically challenging. Thus, capturing the spike-type stall precursor before the rotating stall or surge is more meaningful for the active control. Based on the high-order discretization Moore-Greitzer model(Mansoux model), this thesis conducts the approximately accurate modeling and rapid detection of spike-type stall precursors.Firstly, based on the simulation study of the Mansoux-C3 modeling from MIT, the initial disturbance type of the stall is analyzed. Secondly, it studies the parameters of RBF neural network, and looks for the optimal RBF neural network structure and other methods to improve the persistent excitation level of small oscillation faults, and then further improves the performance of deterministic learning, in order to implement the approximately accurate dynamics modeling of the spike-type rotating stall. Thirdly, the locally-accurate identi?cation of the system dynamics corresponding to spike-type stall precursor is presented with deterministic learning. Finally, based on the approximately accurate modeling of the dominant system dynamics, the rapid detection for spike-type stall precursors is achieved.To sum up, this work analyzes and studies the modal-type stall, the stall with inlet?ow distortion and the spike-type stall, and conducts online experiments. Also, the proposed method to capture the stall inception in axial-?ow compressors is veri?ed in both simulations and experiments. |
PDF Full Text Request