Equipment failure prediction based on neural network analysis incorporating maintainers inspection findings

The improvement of condition monitoring of critical equipment of complex systems supports the development of predictive maintenance planning. The main goal of predictive maintenance is to understand how the component degradation is affecting the performance of equipment. Based on performance degradation, the maintenance planner defines the necessity for intervention on the equipment. For complex systems, the monitored operational parameters may be used to support predictive maintenance decision-making. The combination of the parameters registered during operational campaign can be used to define the normal or abnormal equipment performance. The present paper presents a reliability-based methodology to support decision-making regarding operational performance of equipment. Based on FMEA analysis, the critical components for equipment performance are defined. The monitoring equipment is analyzed aiming at defining the registered parameters that can be used to evaluate critical components degradation associated to specific failure modes. In case of absence of monitoring parameters, an inspection route is proposed to be executed by maintainers. The main goal of the route is to use sensorial capabilities of human being to evaluate possible degradation of critical components that are not monitored by the control system. The findings of that inspection route may be used to support the evaluation of any abnormal component operational condition that may affect equipment performance. In order to combine the parameters registers and sensorial inspection results to diagnose degradation in equipment performance a Neural Network based on Radial Basis Function is developed. For the definition of normal or abnormal operational condition the decision-maker preference is modeled with utility functions that incorporate attitude toward risk. The proposed model is used to analyze the operational condition of a hydro generator. The parameters of temperature (from bearings and heat exchangers) and shaft vibration are used to characterize hydro generator operation. For other critical equipment a sensorial inspection route is proposed using mainly temperature information to support components degradation analysis. The registered and sensorial data representing one operational month are used to train the neural network regarding normal and abnormal equipment condition. After training, the Neural Network was able to detect abnormal operational condition. The use of utility functions incorporating operator attitude towards risk indicates that the definition of abnormal performance is strongly affected by operator's preference.