| الملخص: | The traditional equipment for tailpipe emission monitoring basically adopts the static fixed-point detection mode. Not only does this detection mode fail to ensure accurate analysis of vehicle exhaust emissions during dynamic operation, but its measurement results are also susceptible to interference from human intervention or the detection environment. Therefore, a remote vehicle exhaust gas detection system based on an integrated deep learning model is proposed. Firstly, the TLZ7x-EasyEVM development board designed based on the SOM-TLZ7x core board is used as the on-board hardware platform, and the FS704UM network module is interconnected with the development board through the GPIO interface to realise wireless remote data transmission. The NHA-502 exhaust gas analyser was used to detect the components of CO, CO2, O2 and NOx in the exhaust gas, and the performance parameters of the vehicle were monitored in real time through the OBD-II interface. Then, in order to further improve the efficiency of feature information extraction by sliding window, an incremental computation-based feature information calculation method is proposed to extract feature information from preprocessed vehicle exhaust data. Secondly, using the extracted seven types of feature information as inputs in the tailpipe monitoring centre, the LSTM neural network, which has the ability to remember long-term temporal information, is used as the weak predictor of integrated learning, and the weak predictor is weighted and combined with the strong predictor using the AdaBoost integrated learning algorithm. The effectiveness of the proposed system is verified by experiments on an Audi A4 vehicle. The results show that the proposed system achieves an accuracy of 91.33% for the detection of abnormal exhaust conditions. © 2025, Taiwan Ubiquitous Information CO LTD. All rights reserved.
|
|---|