Enhanced fault detection for nonlinear processes using modified kernel partial least squares and the statistical local approach

Conventional kernel partial least squares (KPLS) may not function well for detecting incipient faults in nonlinear processes. In relation to existing work, a new statistical local approach based KPLS monitoring strategy is proposed by integrating the statistical local approach into a modified KPLS framework. The advantages of the proposed technique are that (i) the new score variables constructed in the statistical local approach approximately follow Gaussian distribution, in spite of the distribution that the original data follow; (ii) after whitening of the data using KPCA, the dimension of modelling space is greatly reduced, which will definitely improve the computing speed. The new method shows more effective and sensitive performance for detecting incipient faults or slow changes of processes. This is demonstrated by a simulation numerical example and recorded data from a non-isothermal CSTR process.