Efficient combustion control has increasingly become a quality requirement for automobile manufacturers because of its impact on pollutant and greenhouse gas emissions. In view of this, the management system development of modern internal combustion engines is mainly aimed at combustion control. The real-time detection of in-cylinder pressure characteristic parameters has a considerable significance on the closed-loop combustion control of the internal combustion engine. This paper presents a detection method in which the start of combustion, peak pressure, maximum pressure rise rate, and phase of maximum pressure rise rate are identified through vibration acceleration signal. In order to analyze the relationship between vibration and in-cylinder pressure signal, experimental data are acquired in a diesel engine by implementing various injection strategies and engine operating conditions (speed and load). The results show that the start of combustion can be detected by analyzing its relationship with the peak position of the filtered vibration signal, and the phase of the maximum pressure rise rate can be identified by examining its relationship with the zero-cross position that is adjacent to the right of the peak position. Moreover, the filtered vibration signals are also truncated in the same length and utilized as inputs for algorithms to detect the peak pressure and the maximum pressure rise rate. The algorithms are mainly performed on data compression (or feature extraction) and target regression. Major algorithms, such as one-dimensional convolutional neural network, compression sensing, wavelet decomposition, multilayer perceptron, and support vector machine, are tested. Various experimental results verify that for the test engine the phase detection accuracy of the start of combustion and maximum pressure rise rate is less than 1.7°CA for a 95% prediction interval width. For the detection of the peak pressure and maximum pressure rise rate, the normalized error threshold is set as 0.05, then the accuracies can be not less than 95%.

由于对污染物和温室气体排放的影响，有效的燃烧控制越来越成为汽车制造商的质量要求。有鉴于此，现代内燃机的管理系统开发主要针对燃烧控制。缸内压力特性参数的实时检测对内燃机的闭环燃烧控制具有相当重要的意义。本文提出了一种通过振动加速度信号识别燃烧开始、峰值压力、最大压力上升率和最大压力上升率相位的检测方法。为了分析振动与缸内压力信号的关系，通过实施各种喷射策略和发动机运行条件（速度和负载）在柴油发动机中获取实验数据。结果表明，通过分析其与滤波后的振动信号峰值位置的关系可以检测到燃烧的开始，通过检查其与相邻过零位置的关系可以识别最大压力上升率的相位。在峰值位置的右侧。此外，过滤后的振动信号也被截断为相同的长度，并用作算法的输入，以检测峰值压力和最大压力上升率。这些算法主要在数据压缩（或特征提取）和目标回归上执行。主要算法，如一维卷积神经网络、压缩感知、测试了小波分解、多层感知器和支持向量机。各种实验结果验证，对于95%的预测区间宽度，试验发动机的燃烧开始和最大压力上升率的相位检测精度小于1.7°CA。对于峰值压力和最大压力上升率的检测，将归一化误差阈值设为0.05，则准确度不低于95%。