A new hybrid modelling strategy for fault detection and diagnosis of the air handling unit by combining slow and kernel slow features analyses

The operation of air handling unit (AHU) always exhibits hybrid linear, nonlinear and time-varying dynamic characteristics, however the traditional single modelling strategy based methods are not effective in tackling these hybrid characteristics. To resolve this issue, a new hybrid modelling strategy based fault detection and diagnosis (FDD) method has been developed by this study. The method has integrated slow features analyses (SFA) with kernel SFA (KSFA), and we called this the hybrid SFA (HSFA). Specifically, to effectively handle the AHU's linear and time-varying dynamic properties, the SFA was first employed to mine linear slow features (SFs) of input data. Subsequently, to deal with the AHU's nonlinear and time-varying dynamic characteristics, the KSFA was applied in the residual subspace of the input data to capture the nonlinear SFs. To detect the fault, the T 2 statistic was constructed using the linear and nonlinear SFs while the SPE statistic was established using the residuals of nonlinear SFs. To further identify the fault variable, a novel reconstruction based contribution plot was developed based on the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:msubsup> <mml:mi>T</mml:mi> <mml:mi>L</mml:mi> <mml:mn>2</mml:mn> </mml:msubsup> </mml:math> statistic in the HSFA's principal subspace. The experiments and comparisons with the ASHRAE RP-1312 dataset have demonstrated the superiority of the proposed HSFA based approach in terms of the FDD.