“…A potential solution is to consider adaptive thresholds [19,24], and use apriori knowledge, either qualitative [33,62,31,23] or quantitative [53,25,47], to derive non-constant thresholds to take the time-varying uncertainties and disturbances into account. Furthermore since the residual depicted in Figure 1 is non-periodic, diagnosis approaches for machines working in non-stationary operating conditions [16,52,43,51] are not applicable. This paper instead proposes an adaptive statistical residual evaluation method, which exploits quantitative a-priori knowledge in the form of data.…”