[yhddd]

I am performing a transient water and air multiphase flow simulation using CEL to initialize the phase volume fraction as follows:
Air volume fraction:step(z/1[m]-0.98-3.005)
Water phase volume fraction: step(0.98-z/1[m]+3.005)

However, floating-point overflow occurs during the first iteration, using the K-e model. If I switch to the SST model, this problem will not occur

If I don't use CEL，I also won't have this problem

However, I think k-e model can be applied to this problem. How can I solve it

[ghorrocks]

Does the error occur during definition of the initial condition or when it starts to iterate?

Either way, I think the problem is likely to be your initial condition. Make sure you define a sensible initial value for k and epsilon, noting that epsilon cannot be zero (it makes turbulent viscosity undefined), and k cannot be zero (it makes the epsilon equation undefined). In fact the k-e model is not very good for very low values of k or e, which is why it is not used for low Re number flows. The SST model is much better at low Re number.

What Re number is the flow you are modelling? If it is a low Re then the k-e model is not suitable and will not work well (as you have found).