This paper proposes a method for scintillation detector fault diagnosis based on the Extreme Learning Machine (ELM) algorithm. A related characteristics database is established including the falling edge time, signal amplitude of nuclear pulse signals, and energy spectrum peak channel, low channel address count of energy spectrum signal under various operating conditions. Then a diagnosis model for scintillation detector is created by improved ELM algorithm which have changed the activation function of the hidden layers and modified the number of hidden layers. An experiment system is also con...