Deep feature learning of in-cylinder flow fields to analyze cycle-to-cycle variations in an SI engine

Abstract

Machine learning (ML) models based on a large data set of in-cylinder flow fields of an IC engine obtained by high-speed particle image velocimetry allow the identification of relevant flow structures underlying cycle-to-cycle variations of engine performance. To this end, deep feature learning is employed to train ML models that predict cycles of high and low in-cylinder maximum pressure. Deep convolutional autoencoders are self-supervised-trained to encode flow field features in low dimensional latent space. Without the limitations ascribable to manual feature engineering, ML models based on these learned features are able to classify high energy cycles already from the flow field during late intake and the compression stroke as early as 290 crank angle degrees before top dead center ([Formula: see text]) with a mean accuracy above chance level. The prediction accuracy from [Formula: see text] to [Formula: see text] is comparable to baseline ML approaches utilizing an extensive set of engineered features. Relevant flow structures in the compression stroke are revealed by feature analysis of ML models and are interpreted using conditional averaged flow quantities. This analysis unveils the importance of the horizontal velocity component of in-cylinder flows in predicting engine performance. Combining deep learning and conventional flow analysis techniques promises to be a powerful tool for ultimately revealing high-level flow features relevant to the prediction of cycle-to-cycle variations and further engine optimization.

Towards CRISP-ML(Q): A Machine Learning Process Model with Quality Assurance Methodology

“…Using unlabeled data and pre-trained models: Labeling data could be very expensive and might limit the available data set size. Unlabeled data might be exploited in the training process, e.g., by performing unsupervised pre-training [107] and semi-supervised learning algorithms [108][109][110].Complementary transfer learning could be used to pre-train the network on a proxy data set (e.g., from simulations) that resembles the original data to extract common features [111].…”

Towards CRISP-ML(Q): A Machine Learning Process Model with Quality Assurance Methodology

“…Unlabeled data might be exploited in the training process, e.g. by performing unsupervised pre-training [87,88] and semi-supervised learning algorithms [89,90]. Complementary, transfer learning could be used to pre-train the network on a proxy data set (e.g.…”

The Influence of Flow on Cycle-to-Cycle Variations in a Spark-Ignition Engine: A Parametric Investigation of Increasing Exhaust Gas Recirculation Levels

“…With the dawn of significant improvements in advanced imaging, laser, and computing technologies over the last decade, engine CCV has been studied with detailed spatial and temporal scales. Recent experimental studies have related the in-cylinder velocity to fired engine CCV (Buschbeck et al 2012;Zeng et al 2015Zeng et al , 2019Bode et al 2017Bode et al , 2019Schiffmann et al 2017;Fach et al 2022;Dreher et al 2021;Hanuschkin et al 2020) using high-speed particle image velocimetry (PIV) combined with other techniques such as in-cylinder pressure measurements as well as flame and spark imaging. In particular, Buschbeck et al and Zeng et al studied homogenous mixtures with varying equivalence ratios to examine the influence of the flame speed on cyclic performance; it was shown that large-scale flow structures can have a significant effect on the flame development and subsequent combustion speed, a finding which is amplified when the flame speed is slower (Buschbeck et al 2012;Zeng et al 2019).…”