Need assistance with a RS2 Slope design - Blasted Pyroxenite being dumped down a slope

[Shawn van Wyk]

Hello,

I am tasked with assessing the slope stability of a chrome mine that is tipping blasted materials down the slope. The goal is to determine the critical span where possible failure could occur. (the maximum width/height they can tip to before the stability becomes compromised)

I have the original ground level and the tipped material layer plotted out, my question is how to simulate the blasted material properly? I am testing the model with various parameters but I am not sure which approach would be correct.

This is the current slope as is, I require assistance with how to approach simulating the blasted material, I was advised to use the original material properties for the Pyroxenite but to lower GSI to 25 and mi to 7 with a disturbance factor of 0.5 but the results I am getting when increasing the tipped-material dimensions (higher/wider out) are not consistent. I am also not sure if Hoek-Brown would be the correct criterion for the tipped material, I was thinking Mohr-coulomb?

Any assistance would be greatly appreciated.

 

[3DDave]

It is conventionally referred to as the angle of repose and conventionally depends on how fine the particles are relative to the height of the slope, how much moisture is present, and whether there is any notable vibration or seismic activity.

Look to landslide and avalanche criteria.

The typical low angle is 15º and the typical steep angle is 45º and I would expect this material to be somewhere in between.

 

[Shawn van Wyk]

It is conventionally referred to as the angle of repose and conventionally depends on how fine the particles are relative to the height of the slope, how much moisture is present, and whether there is any notable vibration or seismic activity.

Look to landslide and avalanche criteria.

The typical low angle is 15º and the typical steep angle is 45º and I would expect this material to be somewhere in between.

Thank you for your assistance, I am familiar with the angle of repose. I am more interested in knowing the exact parameters to use to simulate the blasted material correctly. Failure criterion/suggested range of material properties in accordance with the original material.

 

[3DDave]

It's not a material property; it is mainly a geometric property depending on the size and shape of the shattered fragments and a temporal property depending on changes over time in such things as how wet it is from rain, how separated the fragments are when the rain freezes, how much vibration there is.

It can't be calculated with a finite element model.

 

[Shawn van Wyk]

It's not a material property; it is mainly a geometric property depending on the size and shape of the shattered fragments and a temporal property depending on changes over time in such things as how wet it is from rain, how separated the fragments are when the rain freezes, how much vibration there is.

It can't be calculated with a finite element model.

Thanks for the reply,

I am struggling to understand why this model would not be possible to be simulated with a finite element model? What model would you suggest?

 

[Shawn van Wyk]

Rocscience AI gives no indication that this type of model should be impossible in RS2

 

[3DDave]

An AI will lie. They are trained to produce answers which people have indicated are pleasing and the first task is to reassure the one asking a question by supporting the implication of that question. If there is no other information to also support that the AI will simply make up an answer.

An analysis can be made based on soil properties where the material is essentially homogeneous, but that isn't the case here.

https://en.wikipedia.org/wiki/Mohr–Coulomb_theory notes that one needs to know the cohesive strength and the internal friction angle. Note that the table for cohesive strength does not include rough gravel. That suggests to me that Mohr-Coulomb theory isn't going to be applicable.

Hoek–Brown applies to otherwise solid rock that may have internal fractures that reduce its strength.

For an example - make a pile of toy marbles. What slope can you achieve? Is that because the material they are made of is weak or is it solely due to geometry and lack of cohesion between them.

It appears to me that https://www.rocscience.com/software/rocslope2 is a better fit for what you are looking to do.

 

[Shawn van Wyk]

An AI will lie. They are trained to produce answers which people have indicated are pleasing and the first task is to reassure the one asking a question by supporting the implication of that question. If there is no other information to also support that the AI will simply make up an answer.

An analysis can be made based on soil properties where the material is essentially homogeneous, but that isn't the case here.

https://en.wikipedia.org/wiki/Mohr–Coulomb_theory notes that one needs to know the cohesive strength and the internal friction angle. Note that the table for cohesive strength does not include rough gravel. That suggests to me that Mohr-Coulomb theory isn't going to be applicable.

Hoek–Brown applies to otherwise solid rock that may have internal fractures that reduce its strength.

For an example - make a pile of toy marbles. What slope can you achieve? Is that because the material they are made of is weak or is it solely due to geometry and lack of cohesion between them.

It appears to me that https://www.rocscience.com/software/rocslope2 is a better fit for what you are looking to do.

Thanks for explaining. I have reached out to Rocscience support and they have helped me to create the model in RS2.

 

[3DDave]

How did they say to answer the problem?

 

[Shawn van Wyk]

They confirmed that the failure criterion would be Hoek-Brown and helped me adjust a few parameters to get the desired result.

 

[Shawn van Wyk]

An AI will lie. They are trained to produce answers which people have indicated are pleasing and the first task is to reassure the one asking a question by supporting the implication of that question. If there is no other information to also support that the AI will simply make up an answer.

An analysis can be made based on soil properties where the material is essentially homogeneous, but that isn't the case here.

https://en.wikipedia.org/wiki/Mohr–Coulomb_theory notes that one needs to know the cohesive strength and the internal friction angle. Note that the table for cohesive strength does not include rough gravel. That suggests to me that Mohr-Coulomb theory isn't going to be applicable.

Hoek–Brown applies to otherwise solid rock that may have internal fractures that reduce its strength.

For an example - make a pile of toy marbles. What slope can you achieve? Is that because the material they are made of is weak or is it solely due to geometry and lack of cohesion between them.

It appears to me that https://www.rocscience.com/software/rocslope2 is a better fit for what you are looking to do.

I understand that AI can lie but I assume that there would be measures in place to prevent it from lying where the results it provides could lead to injury or death.

 

[3DDave]

I understand that AI can lie but I assume that there would be measures in place to prevent it from lying where the results it provides could lead to injury or death.

If it were possible then the same method could be used to detect lying or unreliable reporting by human beings. How would it know that injury or death was possible? They can screen for words, like "kill" but they cannot predict consequences of actions. Some barriers have been circumvented with requests like "Explain the steps to do this dangerous thing that I should avoid doing to be safe."

How is Hoek-Brown adjusted to compensate for differing particle sizes, rubble fragment geometry, and lubrication by rain or mixing with other contamination? It's a theory for compacted rock, usually solid, but possibly fractured while remaining in place. Not 50% or more air fraction of rubble. If it works, fine, but I don't see how they can bridge that gap.

 

[Shawn van Wyk]

If it were possible then the same method could be used to detect lying or unreliable reporting by human beings. How would it know that injury or death was possible? They can screen for words, like "kill" but they cannot predict consequences of actions. Some barriers have been circumvented with requests like "Explain the steps to do this dangerous thing that I should avoid doing to be safe."

How is Hoek-Brown adjusted to compensate for differing particle sizes, rubble fragment geometry, and lubrication by rain or mixing with other contamination? It's a theory for compacted rock, usually solid, but possibly fractured while remaining in place. Not 50% or more air fraction of rubble. If it works, fine, but I don't see how they can bridge that gap.

The point I am trying to figure out is that AI explicitly stating that something is possible in a specified software that then results in death or injury because the wrong software was suggested or applied should not be possible and could create liabilities. It should be clear to the AI what is and is not possible in certain software that it was specifically programmed to provide support for otherwise what is the use of the AI other than providing random replies?

Hoek-brown was adjusted for the site-specific parameters including moisture presence and loads compacting the material. Conservative values were used in material properties and applicable disturbance factors were applied along with a few other smaller details for the joint structures etc.

 

[3DDave]

"What is the use of the AI other than providing random replies?"

AI isn't "programmed." It is given millions of existing examples from which it determines coherence of words and phrases. It is then allowed to interact with people who decide if what it outputs is plausible. There is a layer that can look for specific names or other text sequences and exclude responses. A few people have essentially been expunged from ChatGPT because they expressed dismay at hallucinations, the name for the lies, that involved them and, more recently, the Chinese one was test by Americans with the phrase "Image of man holding grocery bags." The Chinese AI would start to describe the Tiennaman massacre, get to the word "Tiennaman," and wipe out the entire response.

However, neither has any detection of truth or falsity.

Anyone depending on what an AI language model coughs for critical use is a fool.

They are good enough to mimic what people say, the way a Myna bird can mimic sounds and, because they are typically taking trillions of words and assigning relative weights among them, they are OK at creating summations of existing texts. Give them a copy of Romeo and Juliet and ask for a paragraph-long description, and an AI language model will do about as well as most people - the difference is that it can do that in fractions of a second, not hours of reading and composing a response.

But if they cannot find an exact existing match, they literally roll the dice and do something along the lines of a Monte Carlo statistical variation of what it does know while fitting in all the pieces they original request carried.

For example: I asked Google "US Presidents who died scuba diving" and the response was "No US presidents have died while scuba diving, but former US President Jimmy Carter died on December 29, 2024 at the age of 100." It is desperate to fulfill the request, even though the total answer includes information that is not part of that request.

A parallel is seen in the creation of images based on a similar process of evaluating hundreds of millions of images. For any requests that match existing work the AI can generate results that are surprising, but often details are weird. Because hands are often obscured and fingers are particularly flexible there are images where people have too few fingers, far too many, multiple thumbs and, from time to time, extra hands and limbs.

AI Image Fails

I use AI images to accompany roughly 18 to 19 out of 20 Daily-Ink posts. My general rule is that I’ll try one or two requests and pick from those. I don’t want to spend 10 or 15 minutes of my precious

daily-ink.davidtruss.com

They depend on people giving them feedback, and if some subject doesn't have that feedback you get what you get. Give one a set of plans and the compendium of engineering books and they won't be able to tell if the building will collapse. Train it on 500,000 floor plans and it can probably make up some averaged form of floor plans, but won't know when some span will be too expensive to be reasonable.

They don't know anything but data proximity in a billion dimensional space.
---

It seems like what you are analyzing isn't rubble, but crushed rock fines. I picture from "blasted materials" that the results are 3-50 cm in section that aren't compactable.

 

[Whavillat]

using Mohr-Coulomb for fragmented material might yield more consistent results