The DBN-AAFPN fault diagnosis method mainly includes two parts: forward reasoning and backward reasoning.
Abstract:
The adjustable parameters of the traditional fuzzy Petri net (FPN) are single and mostly depend on expert experience. This approach lacks the adaptability to the complex ...Show MoreMetadata
Abstract:
The adjustable parameters of the traditional fuzzy Petri net (FPN) are single and mostly depend on expert experience. This approach lacks the adaptability to the complex network of sensors, which will result in insufficient accuracy of fault diagnosis. We propose a method combining the FPN with an adaptive arc and deep belief network (DBN) and improved a fast Gibbs sampling (FGS) algorithm to realize sensor fault diagnosis. First, we present the concept of adaptive arcs with label-weights based on the confidence-weights of directed arcs, which is an important component of the sensor fault model. Then, the improved FGS algorithm optimizes the model layer-by-layer, and the adjustment of the transition threshold relies on the marginal distribution of a restricted Boltzmann machine (RBM). Finally, the optimized dual-weights and dual-transition influence factors are applied to the forward and backward fuzzy reasoning of the model to achieve network adaptability. Our studies showed that this method has obvious advantages in terms of the accuracy and adaptability of complex networks compared to other FPN fault diagnosis methods. The fault reasoning confidence can provide an effective reference for maintenance personnel and improve maintenance efficiency, ensuring the reliable operation of sensors and related systems.
The DBN-AAFPN fault diagnosis method mainly includes two parts: forward reasoning and backward reasoning.
Published in: IEEE Access ( Volume: 9)