A Neural Network Observer for Injection Rate Estimation in Common Rail Injectors with Nozzle Wear

摘要

The objective of this study is to present a neural observer that estimates changing injection behavior due to wear and aging effects within the nozzle of a common rail diesel injector. Using a dynamic identification system in combination with a modified learning rule, the neural observer is applicable to a wide range of problem sets. A multilayer perceptron (MLP) network with three layers and few neurons in the hidden layer ensures fast computing and high efficiency; network learning is based on quasi-Newton optimization and an additional line search algorithm. Modeling the bottom part of the injector introduces a simulation model, which is validated with experimental data from a solenoid common rail diesel injector. Estimation results conform well with the altered plant and therefore demonstrate the significant benefit of using the proposed neural network observer concept.