Approaches for a New Generation of Fast-Computing Catalyst Models

März, Christian; Werfel, Johannes; Kühne, Joana; Robert, Scholz

doi:10.1007/s40825-019-00153-y

Cited by 5 publications

(3 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Test cases summarized in Table 1 are fed to train the ANNs with different ratios between seen and unseen cases. For each binary reaction, the training data set is built by merging the selected number of cases in terms of γ defined by γ = number of cases for training number of entire cases (21) The cases for the training data set are not randomly chosen but always include the cases with the minimum and maximum of the inlet concentration and space velocity to bound the [1,3,5,11,15,16,18,20] 0.6 [21,23,25,26,28,30] 0.6 [31, 33,35,36,38, 40] 0.6 [1,3,5,6,8,10,11,13,15,16,18,20] 1.0 [21,22,23,24,25,26,27,28,29,30] 1.0 [31, 32,33,34,35,36,37,…”

Section: Resultsmentioning

confidence: 99%

“…de Oliveira et al 19 exploited an ANN to predict the NO x conversion efficiency of a selective catalytic reduction (SCR) device and extended their study to predict tailpipe NO x emissions from chassis dynamometer and on-road driving tests. 20 Marz et al 21 investigated several methods to simplify the high-dimensional computational model for a SCR-coated particle filter and concluded that the NN approach is the best in terms of accuracy and speed. Physics-based reactor models are designed to estimate species conversions through the reactor by solving mass, momentum, and energy conservation equations.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Feasibility Study on the Use of Artificial Neural Networks to Model Catalytic Oxidation in a Metallic Foam Reactor

Woo

Stettler

2021

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

This study investigates the feasibility of using artificial neural networks (ANNs) to predict catalytic oxidation in diesel after-treatment systems and compares their performance to that of physics-based models. Existing physics models are revisited to generate baseline data for binary reactions of major species (CO, C 3 H 6 , and NO) measured in a lab-scale microreactor comprising a metallic foam catalytic substrate. The physics model performs well to predict the measured light-off curves, which are the species conversions with ramping temperature, and the R 2 value is above 0.84 across a wide range of operating conditions. However, the model cannot perfectly capture the retarding trends observed in the CO and C 3 H 6 conversion curves after light-off. In contrast, the ANN model is capable of accurately predicting the light-off curves for operating conditions seen during the training process. This might be practically useful but is inherently limited by the availability of experimental data for training. To compensate for the drawbacks of both approaches, this study suggests a hybrid model in which a pretrained ANN is used to calculate reaction rates in the physics models. Despite the more complex data generation process for training ANNs, the hybrid model captures the light-off curves including the retarding trend and is less sensitive to the range of test conditions without renormalization as compared to the pure ANN model. This study investigates the feasibility of ANNs by comparing the pros and cons among the physics models, pure ANN, and hybrid models and suggests a step toward the most appropriate uses of ANNs in modeling exhaust after-treatment in practical applications.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Feasibility Study on the Use of Artificial Neural Networks to Model Catalytic Oxidation in a Metallic Foam Reactor

Woo

Stettler

2021

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

show abstract

“…For the time series regression problem the multivariate dataset from März et al (2020) is used. It is a result of a 1D physical simulation of an SCRF catalyst.…”

Section: Time Series Regressionmentioning

confidence: 99%

Securing Deep Learning Models with Autoencoder based Anomaly Detection

Kühne¹,

März

̈uhmann³

2021

PHME_CONF

View full text Add to dashboard Cite

Deep learning models are on the rise in many scientific fields. Their ability to solve complex and nonlinear tasks has made them very popular. However, in comparison to physical models, they struggle with extrapolation. Hence, it is important that the input data in the production stage is similar to the data seen in training. Deviations to the training often occur in real world applications due to sensor delays and drifts, aging of the system, and communication errors such as noise. Especially in safety relevant applications, securing those models against these influences, which can be seen as anomalies, is essential. In the past, Autoencoders, especially Variational Autoencoders (VAEs), have been proven useful for anomaly detection. Many researches focus on improving the Autoencoder’s separation ability for an optimal set anomaly threshold. However, the setting of the threshold is not trivial and is crucial for a good anomaly detection. Setting the threshold optimal becomes especially challenging if the anomaly is unknown in the training process, which is often the case in a real world application. The proposed method combines a deep learning model with an Autoencoder. The input data is handed to the trained Autoencoder which reconstructs the input. If the data is similar to the training data, the Autoencoder should be able to reconstruct the input data accurately. Otherwise, an anomaly is suspected. The reconstruction and the original input data are both passed through the deep learning model, generating two predictions, which are then compared. For classification and reinforcement tasks with discrete result space the prediction of non-anomalous data should lead to the same class for both samples. For those tasks this allows us to sort out samples as anomalies for which the two results are not the same, hence the threshold becomes obsolete. For regression and continuous reinforcement tasks, the difference between the two predictions can be interpreted as a safety measure and is easier to grasp than the Autoencoder’s reconstruction error, which is typically used. The advantage of this method is the distinction between samples that can or cannot be handled by the subsequent application model instead of just deciding if the input is anomalous. This leads to a higher robustness of the joint model and a better usage of resources of the deep leaning model. Moreover, the Autoencoder and the deep learning model are trained separately which makes the training a lot more stable than using coupled training methods. The proposed method is proven on both, a classification and a regression task. For the classification, the publicly available UEA multivariate time series classification dataset was used. For regression, a dataset simulating a SCR catalyst as part of an automotive exhaust gas aftertreatment system was evaluated. For both tasks common anomalies such as delay and noise were applied to the data.

show abstract