Application of a Neural-Network-Based Algorithm for the Real-Time Correction of the In-Cylinder Pressure Signal Sensed with a Piezoelectric Washer

The objective of this work is to analyze the signal of a piezoelectric washer installed under the spark plug and to compare the combustion metrics evaluated with such signal to the indexes from a standard piezoelectric sensor for the in-cylinder pressure measurement, considered as the reference. In the first part of the article, the spectrum analysis of the piezoelectric washer pressure trace is proposed. It is demonstrated how such a signal can be used to measure the main combustion and knock indexes. Nevertheless, due to the intrinsic characteristics of the system, the knock index evaluated from the raw pressure trace cannot be directly used to estimate the instantaneous knock intensity. For this reason, a model-based algorithm for Real-Time (RT) application is developed to calculate a corrective factor of the high-frequency content of the signal. With such an algorithm, the logarithmic mean value of the Maximum Amplitude of Pressure Oscillation (MAPO) can be accurately evaluated through an Artificial Neural Network (ANN) to properly scale the indexes calculated from the washer signal. The algorithm is further developed with respect to a previous work of the authors by introducing a new function to account for the intake air temperature and the fuel quality effects on the knock indexes, and it needs both some variables provided by the Engine Control Unit (ECU) and the raw washer combustion indexes as inputs. In the last part of the work, the algorithm is validated at the engine test bench under steady-state and transient conditions by reproducing dynamic speed and load profiles.