The application relates to the technical field of industrial control and process
automation, and discloses a
remote diagnosis predictive maintenance system for an industrial
instrument cluster, which comprises a topology evolution and diagnosis module, a residual error decoupling and
false alarm stripping module and a
health assessment and strategy generation module.The topology evolution and diagnosis module combines a static space topology with real-time
actuator state feedback to calculate a dynamic connection weight matrix, and outputs a suspected fault node set through a graph neural network.The residual error decoupling and
false alarm stripping module calculates an estimated controlled deviation based on a first-order
inertia transfer function, strips the estimated controlled deviation from an original residual error to obtain a decoupled residual error sequence, and performs
false alarm identification and determination.The
health assessment and strategy generation module extracts a degradation characteristic parameter based on the decoupled residual error sequence, calculates an instrument
health index, divides a health grade, and outputs a corresponding control protection strategy.The application can effectively distinguish normal process disturbances from real instrument faults, quantitatively evaluate the gradual
degradation process of the instrument, reduce the false alarm rate of monitoring, and avoid unexpected chain shutdown intervention caused by instrument degradation.