Three-dimension data of batch process unfolded two-dimension data can cause some information loss, and the global and local structures of data may change in the process. Therefove, a temporal extension global-local neighborhood preserving embedding based on tensor factorization(TTGNPE) algorithm is proposed, which can deal directly with three-dimension data to avoid folding information loss. Neighborhood preserving embedding is introduced in the global and local structure preserving of the data space to fully extract global and local feature information of data. By combining moving the data window to handle process dynamic time-variance, the contribution plot method is applied to diagnose fault variables after a failure being detected. The superiority of the proposed algorithm on the penicillin fermentation process is verified for the fault detection and diagnosis of the batch process.