Planetary gearboxes (PGs) are the core transmission link of large-scale mechanical equipment. The fault of key components occurring inside a PG usually exhibits in various locations and types, therefore, threatens the security service of mechanical equipment. To this point, an end-to-end intelligent diagnosis strategy based on Enhanced Capsule Network (ECN) is proposed to detect multiple fault types of different components in a PG. The ECN well merits the advantages of the Convolution Neural Network (CNN) and the Capsule Network (CN), which fully retains spatial information while digging in-depth fault features. The dynamic routing algorithm is applied to calculate the correlation between the different capsule layers so that accurately recognizing the fault feature. Furthermore, ECN continuously optimizes model parameters by the established margin loss functions and the input data. The experimental studies are conducted to demonstrate the effectiveness of the proposed model, which including two scenarios: 1. different faults from one component; 2. mixing the multiple faults of different components. The experimental results show that ECN featured stronger fault diagnosis capabilities than traditional CNN and CN.