With the development of the internet of things (IoT) and the application of 5G communication, fault diagnosis of massive multisource sensor data becomes more and more important. In this study, a compressed multisource sensor data-based fault diagnosis scheme is proposed, and its advantages include high data compression and fusion efficiency, low computational cost, and a fast online training sample updating rate. The method includes reference matrix construction, reference matrix compression and fusion, sparse vectors calculation, testing sample reconstruction, and quality evaluation. First, a reference matrix is constructed with labeled multisource sensor data, and each column in the matrix is composed of data samples collected from different sources. Then, the reference matrix is compressed using a measurement matrix, meanwhile, the multisource data samples are fused based on weighted summation during the compression. Later, sparse representation based on batch matching pursuit algorithm is conducted, in this step, the compressed testing sample is represented by the compressed reference matrix, and the output of the sparse representation is a sparse vector. After that, elements in the sparse vector corresponding to different patterns are retained exclusively while other elements are set to zero, respectively, and estimated testing samples are reconstructed with the compressed reference matrix and the processed sparse vector. Finally, based on reconstruction quality evaluation, the pattern of the testing sample is determined. Two cases are employed to validate the effectiveness of the proposed approach, including landfill gas power generator maintenance pattern recognition and multiple redundancy aileron actuator fault diagnosis, and the detection accuracy is 96.13% and 96.67%, respectively.