Research and application of improved auto-associative hierarchical neural network with time delay estimation for fault diagnosis of process

For the fault diagnosis of penicillin fermentation process, large amount of sample data and severe time delay are the critical problems, which have not been fully addressed. Aiming at solving these problems, an improved auto-associative hierarchical neural network with time delay estimation is proposed. There are three steps to develop the fault diagnosis method. Firstly mutual information estimation and time delay estimation are separately applied for selecting key variables and calculating delay time among variables. Then auto-associative neural network (ANN) is used because it can compress the time delay series by a certain ratio. It is noted that the input and output of ANN at different sample interval are regarded as the same while ignoring their time-series relationship. Hence, it is developed with forgetting factor and expecting factor to reveal the time-series relationship. Finally, the improved ANN is combined with hierarchical neural network (HNN) as a whole to construct a new auto-associative hierarchical neural network (AHNN) to identify the process faults. Its application to penicillin fermentation process showed the proposed approach can effectively deal with large amount of sample data and identify the fault at a high accuracy rate.