Always same val_acc for LSTM

Hi,
I am training a Model for Timeseries Classification with 3 Labels. 
But when a LSTM model is trained, the val_acc after each epoch is always 66,14% in my case. And it wont change with different Data or batchsize/epochs. 

I Train on 1000 samples, validate on 7442 samples.

What could be the problem? Thank you