为满足国六排放法规，搭载缸内直喷发动机的汽油车必须安装汽油机颗粒捕集器（Gasoline Particulate Filter，GPF）以限制颗粒物排放。GPF的故障诊断因标定困难、诊断频率低等问题，成为排放技术的研究重点。结合神经网络的非线性分类特性，提出了一种基于神经网络的GPF故障诊断算法。算法以GPF的诊断原理作为依据，采集稳态工况信息和对应的GPF前后压力信号作为特征输入，构成神经网络数据集；应用加噪等方法对网络结构进行优化，通过大量实验确定合适的网络参数进行泛化性能评估。评估结果表明：基于神经网络的GPF诊断算法具有良好的准确率和泛化能力。基于NI PXI系列软硬件开发了一套GPF诊断算法测试平台，对算法进行嵌入和可行性验证。测试结果表明：GPF诊断算法在发动机台架环境下能够完成故障的实时诊断和决策，满足设计要求。

To meet the requirement of China Ⅵ emission regulation, gasoline vehicles with gasoline direct injection engine must be equipped with gasoline particulate filter (GPF) to limit particulate emission. GPF fault diagnosis has become a research focus of emission control technology because of its calibration difficulty, low diagnostic frequency etc. In consideration of the nonlinear classification characteristics of neural network, a GPF fault diagnosis algorithm based on neural network was proposed. The engine steady-state condition information and corresponding pressure signals before and after GPF were collected as feature inputs to form a neural network data set according to the diagnosis principle of GPF. The system structure was optimized by applying noise and the appropriate parameters were determined through a large number of experiments to evaluate generalization capability. The evaluation results showed that the GPF diagnosis algorithm based on neural network had good accuracy and generalization performance. A GPF diagnostic algorithm test platform was developed in Ni PXI series software and hardware, and the embedding and feasibility verification of algorithm were conducted. The test results showed that the GPF diagnosis algorithm could complete the real-time diagnosis and decision-making of fault and meet the design requirements.