Abstract： This paper considered the fault detection and reconstruction problem for multiple types of sensor failures in a unified framework. The multiple failure modes pose a challenge to performance monitoring and fault reconstruction. First, a fault model was proposed to represent the existence of additive and multiplicative sensor faults in a single framework. From a data-driven perspective, both necessary and sufficient conditions for fault detectability were derived using Rayleigh-Ritz lemma. Additionally, a data-driven fault reconstruction approach based on threshold fault subspace extraction algorithm was proposed. The proposed fault reconstruction method demonstrates superiority over the traditional method in terms of reconstruction errors. Numerical simulations fully verify the main results.