An Evaluation of an Unhealthy Part Identification Using a 0D-1D Diesel Engine Simulation Based Digital Twin

Abstract

<div class="section abstract"><div class="htmlview paragraph">Commercial automotive diesel engine service and repair, post a diagnostic trouble code trigger, relies on standard troubleshooting steps laid down to identify or narrow down to a faulty engine component. This manual process is cumbersome, time-taking, costly, often leading to incorrect part replacement and most importantly usually associated with significant downtime of the vehicle. Current study aims to address these issues using a novel in-house simulation-based approach developed using a Digital Twin of the engine which is capable of conducting in-mission troubleshooting with real world vehicle/engine data. This cost-effective and computationally efficient solution quickly provides the cause of the trouble code without having to wait for the vehicle to reach the service bay. The simulation is performed with a one-dimensional fluid dynamics, detailed thermodynamics and heat transfer-based diesel engine model utilizing the GT-POWER engine performance tool. The prediction accuracy of the engine model is validated against a standard duty cycle data from engine testing. Multiple failure modes listed in the service troubleshooting steps for several diagnostic trouble codes are then incorporated in the engine model. Some examples of failure modes include, leakage in the plumbing of engine flow path, restricted air-filter, failed actuators, degraded heat exchanger performance etc. When a diagnostics trouble code is triggered on a vehicle, the Digital Twin simulation model initiates a design of experiment (DOE)-based analysis of associated failure modes and their different levels of failure, using a few minutes of engine transient cycle data as inputs. A set of performance parameters predicted from the model are extracted for each DOE experiment and using the same parameters from the engine data as reference, an error metric is computed for the duration of the duty cycle. This error metric is used to compare the various failure modes which are then ranked as an indicative of their closeness to the engine data. Failure modes which are responsible for the diagnostic trouble code, will tend to have a lower value of the error metric while the ones in healthy state on the real engine will be insensitive to the engine data. The Digital Twin model has been validated with real world customer truck data where the prediction of the model is compared to actual service troubleshooting results. Over multiple occasions, the Digital Twin model has been able to not only strongly indicate the actual failure, but also eliminate failure modes which are insensitive to the engine data for the trouble code. This is expected to be an assistive service tool where a robust, highly accurate, predictive and a computationally efficient simulation is run in the background whenever an engine sees an issue with performance or indicates any unhealthy characteristics. Moreover, since the simulation is capable of being run in-mission, the service technician can be alerted with indicative part failure of the vehicle even before the vehicle reaches the service bay, such that they can be prepared with appropriate maintenance actions and order parts that need replacement. This will result in a drastic reduction in vehicle downtime and cost associated with each service event.</div></div>

Improving the Accuracy of the Deep Energy Method

“…Liu et al 6 have discussed FEA's evolution, future, and transition to machine learning applications in detail. In the case of solid and fluid mechanics, machine learning models can broadly be classified into data-driven models 6,7,16,17,[8][9][10][11][12][13][14][15] and physics-informed neural networks (PINNs) [18][19][20][21][22][23][24][25] . In data-driven models, data from experimental and computational results are used to train the models.…”