An incipient fault diagnosis methodology using local Mahalanobis distance: Detection process based on empirical probability density estimation

•We focus on the detection of time varying incipient faults for multivariate data driven process in a noisy environment.•We propose a specifically tuned Local Mahalanobis Distance (LMD) algorithm coupled with Empirical Probability Density (EPD) estimation technique not assuming any particular data distribution.•We, first do the healthy domain estimation based on a down-sampling algorithm for anchors generation and a parameter estimation method optimally tuned and based on Generalized Extreme Value distribution (GEV) for the domain margin selection.•We secondly do the online monitoring using the approximated healthy domain information and the EPD cumulative sum technique applied to the LMD.•The performance analysis based on simulation data shows that our proposal is effective to non-Gaussian data and sensitive for incipient fault detection.•We validate our proposal using an engineering case study, the Continuous-flow Stirred Tank Reactor (CSTR), and further proves the effectiveness of our proposal We highlight the benefit of this methodology by comparing it with state-of-the-art-based solutions in terms of detection delay, detection probability, false alarm probability, and area under the receiver operating characteristic curve (AUC).