A tool condition monitoring model combined with time-frequency domain statistics, principal component analysis (PCA) and auto associative neural network (AANN) was proposed based on motor torque signals of external motor rotation shaft and feed shaft, aiming at problems such as limited deployment of sensors in high radiation environment, noise interference of transmission chain and poor consistency of cold extrusion cutting tools. Firstly, time domain statistical features and wavelet packet energy features were extracted to form the original training set based on the torque waveform of rotary motor and feed motor. Then, the original training set was used to train the AANN model. Finally, the PCA was used to reconstruct the original training set to optimize the local structural parameters of the AANN model, and a PCA-AANN tool condition monitoring model was formed. The proposed PCA-AANN model was compared with the existing AANN model based on the torque data collected from the cutting test of the actual prototype, and experimental results showed that the introduction of PCA improved the robustness of AANN model, reduced the false alarm rate of tool operating state, and realized the accurate monitoring of tool status under radioactive environment. The proposed method provided a reference for the condition monitoring of similar transmission equipment under radioactive environment.