Model parameter fine-tuning and ranking methodology to improve the accuracy of threshold velocity predictions for solid particle transport

Soepyan, Frits Byron; Sarica, Cem; Subramani, Hariprasad J.; Kouba, Gene

doi:10.1016/j.petrol.2013.08.017

Cited by 9 publications

(10 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It has been shown that the prediction of these models for the same input vector may differ several orders of magnitude (Soepyan et al, 2013). An example of this behavior for arbitrarily picked 30 input vectors can be seen in Fig.…”

Section: Introductionmentioning

confidence: 87%

“…The sum of these two scores determines the best cluster set, i.e., the cluster set with the highest score is selected as the best one. Soepyan et al (2013) developed and validated a framework, Tulsa University Sand Transport -Optimization and Ranking Methodology (TUSTORM), to predict the threshold velocity for a given design/operating condition at low solid concentrations. The framework combines a data culling module, a model fine-tuning via optimization section, and a model screening and ranking protocol.…”

Section: Cluster Validationmentioning

confidence: 99%

“…Given an experimental data set, and solid transport models that predict the threshold velocity for particle transport in carrier fluid, Soepyan et al (2013) suggested that the bias of the model can be removed by fine-tuning the parameters of the models using the experimental database. Prior to fine-tuning the model parameters, the solid transport models are converted to dimensionless form, in which left-hand side of the model becomes the particle Reynolds number N Re = dpv f , and the right-hand side of the equation contains the remaining dimensionless groups.…”

Section: Model Assignmentmentioning

confidence: 99%

“…Given an extensive experimental database of threshold velocities for different inputs, the approach partitions the data into similar characteristic clusters using data mining approaches. A model evaluation technique developed by our group (Soepyan et al, 2013) is used to determine the best solids transport model to predict the threshold velocity for each cluster. The last step of the approach correctly classifies a new design/operating condition (i.e., input vectors) to one of the existing clusters, and, uses the predetermined model for that cluster to estimate the threshold velocity for the input vector.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Data clustering for model-prediction discrepancy reduction – A case study of solids transport in oil/gas pipelines

Shin

Subramani

2015

Computers & Chemical Engineering

Self Cite

View full text Add to dashboard Cite

Section: Introductionmentioning

confidence: 87%

Section: Cluster Validationmentioning

confidence: 99%

Section: Model Assignmentmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Data clustering for model-prediction discrepancy reduction – A case study of solids transport in oil/gas pipelines

Shin

Subramani

2015

Computers & Chemical Engineering

Self Cite

View full text Add to dashboard Cite

“…To remediate the above issues, in a previous work, we compiled a database that contains experimentally-observed critical velocities for different input conditions, along with the models that can be used to predict this critical velocity from the literature, and developed a systematic methodology to select the appropriate models for any given input condition (Soepyan et al, 2013a). The methodology contains three steps: (1) a data clustering component creates a reduced database, which consists of experimental data points that are representative of the input condition; (2) a model parameter fine-tuning module adjusts the parameters of the models using the reduced database to remove the bias in the models' predictions; and (3) a model screening and ranking protocol uses statistical analysis to determine the most accurate models for the input condition.…”

Section: Introductionmentioning

confidence: 99%

Confidence in Critical Velocity Predictions for Solids Transport

et al. 2014

View full text Add to dashboard Cite

Sand is often produced with oil and gas even though sand control practices are used. To successfully transport solids in pipelines, the fluid velocity must exceed the critical velocity. Solids transport models are used to predict this velocity. However, for the same input field or laboratory condition, the models' critical velocity predictions may differ by orders of magnitude. Furthermore, none of the models provide information regarding the confidence in their velocity predictions. This paper introduces a systematic methodology, which selects the appropriate solids transport models for predicting the operational critical velocity envelope to ensure solids transport in the pipe to within a predetermined confidence level. Given a field or laboratory condition as input, the approach generates: (1) velocity predictions and uncertainty bounds of all models to within a predetermined confidence level, (2) recommendation of appropriate models for calculating the critical velocity at the input condition, and (3) impact of the uncertainty of input condition, experimental data, and model on the overall uncertainty bounds. The methodology uses data clustering, model evaluation, optimization, and uncertainty propagation approaches. Based on three Case Studies, the velocity envelopes suggested by the methodology, at the 90% confidence level, are consistent with experimental observations. For input laboratory conditions, which are used for validation purposes, the contributions of model, input condition, and experimental data uncertainties to the velocity prediction uncertainty were between 50 -90%, 3 -20%, and 6 -40%, respectively. If the parameters of the selected models were fine-tuned, for the same input conditions, the above uncertainties contribute about 40 -90%, 4 -15%, and 1 -50%, respectively, to the velocity prediction uncertainty. For an input field condition (the fourth Case Study), the above values become about 35 -70%, 7 -25%, and 25 -45%, respectively; and about 25 -30%, 30 -32%, and 40 -42%, respectively, if the parameters of the models are fine-tuned prior to uncertainty estimation. The results of these Case Studies suggest that for laboratory conditions, most of the uncertainty in a model's critical velocity prediction comes from the model, while for field conditions, the uncertainties of the model, input condition, and data may have nearly equal contributions.

show abstract

Data Clustering for Uncertainty Reduction – A Case Study of Solids Transport in Oil/Gas Pipelines

Shin

2014

Proceedings of the 8th International Conference on Foundations of Computer-Aided Process Design

Self Cite

View full text Add to dashboard Cite

Model parameter fine-tuning and ranking methodology to improve the accuracy of threshold velocity predictions for solid particle transport

Cited by 9 publications

References 46 publications

Data clustering for model-prediction discrepancy reduction – A case study of solids transport in oil/gas pipelines

Data clustering for model-prediction discrepancy reduction – A case study of solids transport in oil/gas pipelines

Confidence in Critical Velocity Predictions for Solids Transport

Data Clustering for Uncertainty Reduction – A Case Study of Solids Transport in Oil/Gas Pipelines

Contact Info

Product

Resources

About