How to Update Engineering Predictions with Test Data
The Bayesian procedure for updating engineering model predictions using physical test measurements — prior, likelihood, posterior and updated predictions.
When to use this guide
Use this guide when test data is available and you want to update the engineering model predictions using Bayesian methods.
Step-by-step method
- Define the prior: the current model with uncertain parameters and their distributions
- Define the likelihood: the relationship between model parameters and test measurements, including measurement uncertainty
- Collect test data: measurements with documented uncertainty
- Compute the posterior: update the parameter distributions using Bayes' theorem
- Verify the posterior: check that it is consistent with both the prior and the data
- Generate updated predictions: run the model with posterior parameter distributions
- Compare updated predictions with test data — assess the improvement
- Document: prior, likelihood, data, posterior, updated prediction and remaining uncertainty