The elevator traction machine is a critical component of the elevator system, and its stability directly impacts the safe operation of the elevator. This study proposes a fault diagnosis method for elevator traction machines based on feature fusion and principal component analysis. Initially, the acquired signals are demodulated using principal component analysis. Subsequently, 1DCNN and 2DCNN extract and fuse the one-dimensional and two-dimensional features. Finally, a feedforward neural network classifies the fused features to detect faults. Experimental validation demonstrates that FPCA performs excellently in identifying typical faults such as encoder malfunctions, control system failures, and abnormal bearing noise, achieving a test set recognition rate of 99.58%. Four typical faults—encoder fault, control system fault, bearing damage, and vertical vibration signal under normal operating conditions—are analyzed from multiple perspectives. This study validates the value of the multi-signal fusion fault diagnosis method in practical applications.