To solve the difficulty in diagnosing open-circuit faults in same-arm double-switch and inner single-switch in grid-connected neutral point clamped inverters, as well as the increased computational load and diagnosis time caused by excessive detection signals, a fault diagnosis model combining symmetric-point pattern and enhanced convolutional neural network (ECA-ConvNeXt) is proposed. First, the three-phase “midpoint voltages between upper and lower bridge arms” are defined to distinguish same-arm dual-switch faults from inner single-switch open-circuit faults, while A-phase current is used to diagnose other fault types. Then, four types of signals are fused into a single “snowflake diagram” using the symmetric-point pattern, enriching data features while reducing computational complexity. Finally, dual-size convolution kernels and an efficient cross-channel attention mechanism are introduced to improve the feature acquisition and generalization ability of the ECA-ConvNeXt model. At the same time, the activation function and residual module stacking are optimized to balance diagnosis accuracy and speed. Experiments show that the proposed method achieves a diagnosis accuracy of 99.53% for 79 types of single- and double-switch open-circuit faults, including same-arm double-switch and inner single-switch faults, with an average test time of 8.82 ms, effectively balancing accuracy and speed.