Development of Generic Vibration-Based Rotor Fault Diagnosis for Rotating Machine

Abstract

Vibration-based condition monitoring techniques have been widely adopted by different industries to assess the health status of their critical equipment and machinery. Rotating machines are at the heart of most of industrial processes, therefore their reliable and safe operation is a topic of high interest in the field. To carry on the signal based vibration analysis, it is required an individual with an appropriate level of knowledge and expertise on several subjects, such as signal processing techniques, vibration analysis, as well on the specific dynamics of the studied machine and its working conditions. The need of improved mechanism and techniques, which can accurately perform regardless the human/operator expertise, has led to an extensive development of intelligent systems proposals for the fault detection and diagnosis in rotating machinery. Such techniques appears as a possibility since their capability to identify certain patterns of failure and correlate them to certain type of defects. In this study, it is developed a smart rotor fault detection model, based on vibration signals and artificial neural networks (ANN). The model is intended to be useful for industrial applications, contributing as an alternative to traditional vibration-based condition monitoring. The methodology adopted to develop this model consists on the utilisation of available experimental vibration data to define the architecture of the ANN. Followed by the development of a finite element (FE) model of the experimental rig, which provides an understanding of the dynamics of the machine as well allows the generation of mathematically simulated data. The features optimisation and validation of results using FE data generated are also carried out. Finally, a proposal for integration of a smart fault detection model into the frame of industrial internet of things (IIoT) is presented. The available experimental vibration data include the baseline condition, also referred as healthy in further sections of this work, and four typical rotor related faults, namely misalignment, shaft bent, looseness in bearing pedestal and rotor rub. Two different rotational speeds are considered in the study. The experimental conditions are simulated into the FE model in order to emulate the defects and generate additional vibration signals with the same characteristics. This contributes to a better understanding of the dynamics of the machine and experimental observations. The selected features are optimised along the study, starting with their extraction only from acceleration in time domain, to follow with the addition of velocity in frequency domain and further normalisation. Different features are analysed and compared in order to determine their relevance into provide an effective characterisation and classification of the studied machine conditions. A 2-steps scheme of the developed model is further proposed. The first step performs the fault detection by indicating whether the machine is healthy or faulty, while the second step provides further information on the nature of the defect when it has been detected on the previous step. The delivered results show a 100% of accuracy in the fault diagnosis when analysing samples collected at a unique speed. Same results are observed when the model is blindly tested with unknown data obtained at a different speed to the used for training the ANN.

Date of Award	31 Dec 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Parthasarathi Mandal (Supervisor) & Jyoti Sinha (Supervisor)