Design log Dams N.º 06
Koyna Dam, 1967: why a concrete dam cracks at the "neck"
Abstract. In 1967 an earthquake cracked Koyna Dam, in India, right where its downstream face changes slope. I modeled it in OpenSees and the video shows how that damage appears, second by second.
In December 1967 an earthquake shook the area around Koyna Dam, in western India. Koyna is a concrete gravity dam about 100 m tall, and it did not collapse. But when engineers inspected it, they found horizontal cracks on both faces, all at roughly the same height: right where the downstream face changes slope.
Since then, Koyna has been the example almost everyone who models dams uses to test their model. If your model does not crack the dam in that spot, something is off.
In this note we will cover three things: why the dam cracks right there, how I modeled it, and how to read the video that comes with the note.
Why it cracks at the “neck”
Look at the dam’s shape in the video. At the bottom it is wide, a big triangle. Higher up, at about two thirds of the height, the downstream face turns almost vertical and the dam narrows suddenly. That zone is often called the neck.
When the earthquake moves the base, the wide part moves almost as a block. The top part, slender and heavy, lags behind and then overshoots, like the tip of a whip. All that back-and-forth concentrates at the neck, where the section is smallest.
Add something you already know about concrete: it resists compression very well, but in tension it holds barely a tenth of that. When the crest leans one way, one face of the neck stretches. If it stretches more than the concrete can take, a crack opens.
How I modeled it
I used OpenSees in Python, with a plane model of the dam’s tallest section:
- Concrete with a material that loses stiffness in tension as it cracks (tensile damage, from 0 intact to 1 fully open).
- Reservoir full, on the left side.
- Foundation rock in three layers. Its properties are assumed, not measured at Koyna, and the video says so.
- Ground motion from Koyna 1967, horizontal and vertical components.
Before running the earthquake I checked the basics: the vibration period. My model gives about 19 rad/s (close to 3 Hz, so s), practically the same as the Abaqus reference model and Chopra’s classic solution.
How to read the video
The video has four panels moving together:
- The dam (left). The deformed shape is exaggerated 150 times so you can see it. Color is damage: yellow is sound concrete, purple and black is cracked.
- The record (top right). The red line marks which second of the earthquake you are at. The horizontal peak reaches about 0.5 g.
- Crest displacement (middle). How far the crest moves relative to the base. The largest peaks are around 6 to 8 cm.
- Cracked area (bottom). What percentage of the dam has nearly full damage.
Watch the bottom panel as the video runs. Damage does not grow smoothly: it climbs in steps. Each step matches a crest displacement peak. The crack moves forward only when the whip strikes hard, and between strikes it stays put. By the end, 4 to 5% of the dam is cracked. Damage concentrates at the neck, where the real Koyna cracked, and at the heel, where the dam meets the rock.
What I take away
- Validate with something simple first. The period is cheap to compute and tells you whether mass and stiffness are right before you spend hours on nonlinear analysis.
- Tensile damage is very sensitive. The energy concrete dissipates as it cracks changes how far the crack travels. Declare it and justify it; do not leave the default.
- Separate what is measured from what is assumed. The rock here is hypothetical. The crack pattern looks like the real one, but that does not make this a “true” Koyna model.
If your dam model does not crack Koyna at the neck, do not use it on another dam yet.
Academic model built with public data.
Educational, reference-only content. Opinions are my own and do not represent any employer. On a real project, the engineer of record and the governing code decide.
← Back to notes