Sensors occupy a pivotal role in indoor fire detection, and any malfunction within their operation can potentially escalate into severe accidents. Multi-sensor networks utilized in indoor fire early warning systems are inherently characterized by their high dimensionality and intricate linear interdependencies, presenting substantial operational complexities. To address these complexities, the paper introduces a fault diagnosis method based on Kernel Principal Component Analysis combined with Autoencoders (KPCA-AE). This methodology harmoniously integrates KPCA and AE, where KPCA is utilized to reduce the dimensionality of the data, thereby facilitating the training of the AE model. The proposed method adeptly and precisely identifies faults based on reconstruction errors, introducing a novel and effective paradigm for fault diagnosis in indoor fire sensors.