APRIL 2021 - UNCERTAINTY?
Hi! Todays newsletter contains several items. In the topic of today we discuss what uncertainty is and why it is important to understand it. Also the main idea of partial automation with the API is explained. What is the goal of automating, the advantage as well as the potential downsides? Finally some updates in on the API are discussed.
TOPIC OF TODAY
Uncertainty arises when a decision becomes non-discrete, when something cannot be classified as completely true and neither completely false. Ignoring it is in fact ignoring something is more complex then you are stating and therefore never a good idea. In general two categories of uncertainty exist:
- Aleatoric uncertainty: general randomness
- Epistemic uncertainty: randomness due to a lack of knowledge
I like baking pizza, especially the truely awesome and delicious Neopolitan version. In this process uncertainties also play an important role. There are ingredients which sometimes differ in quality (e.g. juiciness of tomatoes, strength of the gluten in the flower, amount of yiest activity). Also the recipe you select determines whether you are successfull. But in fact the recipe is not exactly alligned with the oven or the room temperatures. Is it possible to rule out all variations? No. However, the more control you have over the ingredients, the more resilient you become to the uncertainties that remain. This all to assure that a nice pizza is put on the table at dinner time!
So the better you understand the general randomness the more you can improve your recipe, or in another context: your design method. Consider the design of a pile foundation. There are many uncertainties in this process, for instance:
- Empirical design formulae used for pile design are not one-on-one applicable for each location (epistemic)
- Parameter estimation used as input for pile design is perfomed with empirical formulas (epistemic)
- Parameter estimation based on limited measurements (aleatoric/epistemic)
- General variation observed on the site considered (aleatoric/epistemic)
- Phenomena which are not yet captured in desing formulae (epistemic)
All these aspects (and more) form the total uncertainty which needs to be coped with to design a proper foundation. The more uncertainties you can understand and quantify, the better you control the remaining ones. Knowing if your parametric variability is very large should make you cautious when deciding on design methods (and the other way around).
API UPDATE #1 - PARTIAL AUTOMATION
Calculating with uncertainties can become complex, which makes it time consuming. This is a significant downside to which automation can provide aid. Automation has great upsides as the ability to standardize repetitive tasks, avoiding manual errors and coupling of processes. On the other hand it also has downsides as the limitation of creativity, limitation of accessibility and ignorance of exceptional cases. Exactly for this reason all processes in the API are only partially automated. This allows the user to continuously steer and adapt if required. The following measures were taken to assure this:
- It is impossible to combine analyses, which forces the user to think. For example, it is impossible to perform a reliability analyses and assess distribution fit directly. The user needs to look at the result of the first analysis before proceeding.
- It warns in case there is an obvious erroronous value in the data and allows you to act on it. For instance, by performing goodness of fit tests during distribution fitting (see newsletter January 2021).
- It always allows you to correct its conclusions. For example, if you do not agree with the layer identification proposed by the API you can propose and integrate a different one. Check out this example.
API UPDATE #2 - MODIFYING LAYER CHARACTERISTICS
The soil profiling API endpoint is adapted for user input. This allows for users to modify and select specific conversion formulations for specific layers. An example is visible in the Figure below. From now on two factors can be provided:
- undrained_shear_strength_factor : modification of the estimator for undrained shear strength in undrained layer
- critical_volume_angle : critical volume angle for drained layers
The values can be provided either as single value for the entire profile or per layer if the layers are already provided by the user. The documentation on the public repository has been updated likewise. Any more questions? Reach out!
FOOTNOTE
Please note that I run this service besides my job at Temporary Works Design. It is my ambition to continuously improve this project and publish corresponding newsletters on new innovations. In busy times this might be less, in quiet times this might be more. Any ideas? Let me know!