Contact

Every contact entry names its model in Type and has these keys:

YAML keyTypeRequiredDefaultRangeQuantityDescription
Contact Radiusnumberrequired≥ 0lengthDistance below which two nodes are in contact and repel each other
Contact Groupsnamed entries of sectionrequiredNamed pairs of blocks this contact model acts on
Symmetrytextoptional3DSelects the contact stiffness for "plane stress", "plane strain" or 3D (any other value)

Each entry of Contact Groups is a pair of blocks:

YAML keyTypeRequiredDefaultRangeQuantityDescription
Master Block IDintegerrequired≥ 1Block id of the master side
Slave Block IDintegerrequired≥ 1Block id of the slave side
Search Radiusnumberrequired≥ 0lengthDistance within which the global search collects candidate contact pairs; larger than the contact radius
Global Search Frequencyintegeroptional—≥ 1Steps between two global contact searches of this group; overrides the value in Contact Globals

Initialization

  • Step 1 - Identification of all surface nodes of contact blocks.
  • Step 2 - All surface nodes are stored in a list. This list exists at all cores with the current position of the surface nodes.
  • Step 3 - Connect each surface node to a geometrical surface.

The standard way for points in the parallelization is given here.

\[\begin{bmatrix}1 \\ 2 \\ 3 \\ 4 \\ 5 \\ 6 \\ 7 \end{bmatrix}_{all}\rightarrow\begin{matrix} \begin{bmatrix} 4 \\ 6 \end{bmatrix}_{C1-global}\\ \begin{bmatrix}1 \\ 2 \\ 7 \end{bmatrix}_{C2-global}\\ \begin{bmatrix}3\\5\end{bmatrix}_{C3-global}\end{matrix}\leftarrow \rightarrow\begin{matrix} \begin{bmatrix} 1 \\ 2 \end{bmatrix}_{C1-local}\\ \begin{bmatrix}1 \\ 2 \\ 3 \end{bmatrix}_{C2-local}\\ \begin{bmatrix}1\\2\end{bmatrix}_{C3-local}\end{matrix}\]

A new mapping is needed to the contact surface exchange list. The local numbering (right above) has to be mapped to the sublist and vice, versa.

\[\begin{bmatrix}1 \\ 2 \\ 6 \\ 7 \end{bmatrix}_{surface}\leftarrow\rightarrow \begin{bmatrix}1 \\ 2 \\ 3 \\ 4 \end{bmatrix}_{surface}\rightarrow \begin{matrix} \begin{bmatrix} 6 \end{bmatrix}_{C1}\\ \begin{bmatrix}1 \\ 2 \\ 7 \end{bmatrix}_{C2}\\ \begin{bmatrix}\,\,\,\,\end{bmatrix}_{C3}\end{matrix}\leftarrow \rightarrow\begin{matrix} \begin{bmatrix} 2 \end{bmatrix}_{C1-local}\\ \begin{bmatrix}1 \\ 2 \\ 3 \end{bmatrix}_{C2-local}\\ \begin{bmatrix}\,\,\,\,\end{bmatrix}_{C3-local}\end{matrix}\]

List mapping To fill the postion vector with the needed values mappings are needed.

local point ids -> global point id -> reduced contact block point id The mapping has to be done in both directions, because the contact forces have to be applied to the local core point id.


Computation

  • Step 1 - Perform a nearest neighbor search with a user defined Contact Radius. The result is a list of potential contact pairs.
  • Step 2 - A sub list nearest neighbor search is performed.
  • Step 3 - All points found are than in contact.

The quasi code is given here

# synchronize all contact nodes
synchronize_contact_points("Deformed Coordinates", "NP1",
                                               all_positions)

## loop over all contact pairs
for (cm, block_contact_params) in pairs(contact_params)
    contact_search(...)
end

function contact_search(...)
    if n < search_frequency
        potential_contact_list = global_search(...) # finds all points within the search radius
        local_search(...)
        return
    else
        synchronize_contact_points("Deformed Coordinates", "NP1",
                                               all_positions[potential_contact_list])
        local_search(...)   # finds all points within the contact radius
        return
    end
end

The structure tries to minimize the search and the communication between cores.

Synchronization All cores (except core 1) send it's displacement values to core 1. Core 1 sends them back.

Efficiency

Smaller contact areas are more efficient in numerical analysis, because less synchronisation have to occur.

Contact Models

YAML keyTypeRequiredDefaultRangeQuantityDescription
Contact Stiffnessnumberoptional100000000≥ 0Factor on the penalty stiffness that pushes nodes in contact apart
Friction Coefficientnumberoptional0≥ 0Coulomb friction coefficient between the contact surfaces

Penalty Contact

A bond-based contact formulation is computed. The contact stiffness $c\_{contact}$ is defined in YAML input file. Based on that utilzing the horizon $\delta$ the penalty stiffness shown in the table is computed. These parameter are comparable to the bond-based formulation. The algorithm was taken from Peridigm

DimensionPenalty Stiffness
plane strain:$c_{Penalty} = \frac{48 c_{contact}}{\pi 5 \delta_{slave}^3}$
plane stress:$c_{Penalty} = \frac{9 c_{contact}}{\pi \delta_{slave}^3}$
3D:$c_{Penalty} = \frac{12 c_{contact}}{\pi \delta_{slave}^4}$

Normal Contact

The distance is computed as

\[d = |\underline{\mathbf{Y}}_{master}-\underline{\mathbf{Y}}_{slave}|\]

The normal is

\[\mathbf{n} = \frac{\underline{\mathbf{Y}}_{master}-\underline{\mathbf{Y}}_{slave}}{d}\]

Both can be adopted if needed. Utilizing the contact radius $r_{contact}$ the contact force densities are

\[\mathbf{f}_{master}=c_{Penalty}\frac{(r_{contact}-d)}{\delta_{slave}}\mathbf{n}V_{slave}\]

\[\mathbf{f}_{slave} = -c_{Penalty}\frac{(r_{contact}-d)}{\delta_{slave}}\mathbf{n}V_{master}\]

Within PeriLab the functions computemasterforce__density and compute__slave__force__density are used to apply the force to the correct postions. This is needed, because if a parallel process is run, master and slave nodes not necessarily are on the same core.

Friction Contact

The model is computed if the Friction Coefficient is greater than zero. For friction the normal forces

\[\mathbf{f}_n=c_{Penalty}\frac{(r_{contact}-d)}{\delta_{slave}}\mathbf{n}\]

are used. The tangential direction must be computed. Following the Peridigm algorithm, the normal velocity of the master node $m$ and the slave node $s$ are

\[\begin{matrix}v_{m} = \mathbf{v}_m\mathbf{n}\\v_{s} = \mathbf{v}_s\mathbf{n}\end{matrix}\]

To obtain the tangential part $tan$ this value hmust be substrated from the velocity.

\[\begin{matrix}\mathbf{v}_{tan-m} = \mathbf{v}_m - v_{m}\mathbf{n}\\\mathbf{v}_{tan-s} = \mathbf{v}_s - v_{s}\mathbf{n}\end{matrix}\]

Tanking the average

\[\mathbf{v}_{avg} = 0.5(\mathbf{v}_{tan-m}+\mathbf{v}_{tan-s})\]

the relative velocity of each point can be determined.

\[\begin{matrix}\mathbf{v}_{tan-m}=\mathbf{v}_{tan-m}-\mathbf{v}_{avg}\\\mathbf{v}_{tan-s}=\mathbf{v}_{tan-s} - \mathbf{v}_{avg}\end{matrix}\]

If the length of $\mathbf{v}_{tan-m}$ and $\mathbf{v}_{tan-s}$ is greater than zero, respectively, friction occurs. The friction force is computed as using the friction coefficient $\mu$.

\[\begin{matrix}\mathbf{f}_{fric-m} = -\mu |\mathbf{f}_n|\frac{\mathbf{v}_{tan-m}}{|\mathbf{v}_{tan-m}|}\\\mathbf{f}_{fric-s} = -\mu |\mathbf{f}_n|\frac{\mathbf{v}_{tan-s}}{|\mathbf{v}_{tan-s}|}\end{matrix}\]