As one of the most common and effective quality-relevant fault monitoring techniques, projection to latent structures(PLS) and its improved algorithms have been wildly used in many industries to provide assurance for high-quality products. In this paper, a new enhanced sparse projection to latent structures(ESPLS) algorithm is proposed to achieve quality-relevant fault monitoring with better sensitivity. The algorithm implements sparse orthogonal decomposition on input process variable space. Two indices based on quality-relevant subspace and quality-irrelevant subspace with major variation are developed for fault detection and analysis. Experiments on Tennessee Eastman Process (TEP) chemical benchmark reveal its outstanding performance in fault detection and superior accuracy in differentiating the quality-relevant and irrelevant impact of the given fault.