APPLICATION OF HYBRID OBSERVATION TECHNIQUES FOR FAULT DIAGNOSIS OF ROTATING MACHINES
- The fault diagnosis of industrial facilities is one of the significant and ever-growing fields of research. Fault diagnosis can be applied to a diversity of industrial components such as rotary machines, motors, pipelines, robot manipulators, gearboxes, etc. In this research, hybrid approaches are developed for detection and classification of the rotary machine bearing faults. Rolling element bearing represents a class of nonlinear and multiple-degrees-of-freedom rotating machines that have pronounced coupling effects and can be used in various industries. Uncertain conditions in which a rolling element bearing operates, as well as nonlinearities, represent challenges for fault diagnosis that are addressed through the fault diagnosis techniques. If defects in the rolling element bearing are not identified and diagnosed in time they can lead to the failure of the whole mechanical system. The failure of the rolling element bearing results in unexpected downtimes and great economic losses. Moreover, it can be a threat to the safety of the people working in the facility. The condition monitoring of a rolling element bearing can be achieved through different techniques. This work focuses on vibration and acoustic emission analysis method because these signals are suitable for fault diagnosis in rolling element bearing. Several methods have been advised for anomaly detection and identification in rolling element bearings. These techniques can be divided into four principal divisions: model-based techniques, signal-based approaches, data-driven algorithms, and hybrid-based procedures. In this dissertation, hybrid-based techniques that uses a combination of the system modeling algorithms, observation techniques, and a machine learning-based classification are introduced for the diagnosis of bearing faults of various severities.
System modeling is the main argument in designing observation-based techniques for fault diagnosis. Numerous procedures have been used to model bearings and can be classified into two main groups: mathematical-based system modeling, and system identification techniques. The mathematical-based bearing modeling such as five-degrees-of-freedom mathematical modeling of vibration signals, and system identification techniques such as ARX-Laguerre and fuzzy ARX-Laguerre bearing vibration and acoustic emission signal modeling are prescribed in this work.
The model-based fault diagnosis techniques are reliable and robust algorithms and have been used in various applications. Observation-based algorithms are the main model-based techniques used for bearing fault diagnosis. Despite the advantages of observation-based approaches, these techniques have some limitations in the presence of uncertain and unknown conditions. Nonlinear-based observation techniques (e.g., sliding mode observer, feedback linearization observer) and linear-based observation algorithms (e.g., proportional-integral (PI) observer) are the main procedures used to develop observation to estimate the signals. The sliding mode observer is a nonlinear and high-gain observer that can improve a system's dynamic and reduce the estimator error infinite time. This technique is robust and reliable, but is prone to chattering phenomenon and limited estimation accuracy. To minimize the chattering phenomenon, the higher-order sliding mode observer is recommended in this work. This technique suffers from a somewhat reduced estimation accuracy. To improve the estimation accuracy, a higher-order super-twisting sliding mode observer was developed.
Sliding mode observer and high-order super twisting (extended-state) sliding mode observer have acceptable state estimation and works in uncertain condition; however, chattering phenomenon is the main drawback of these techniques in uncertain conditions. To minimize the effect of the chattering phenomenon, a feedback linearization observer was developed. The feedback linearization observer is a powerful technique for signal estimation. The main idea of this approach is to algebraically transform the nonlinear system dynamic parameters into a linearized system so that the feedback observation algorithm can be applied. This observer is based on the dynamics of the system's behavior, thus it works perfectly if all parameters are known. Apart from the stability and reliability of this observation technique, it suffers from a lack of robustness. To address this issue, the variable structure (extended-state) feedback linearization observer is developed in this work.
Despite the advantages of high-order super-twisting sliding mode observer and variable structure feedback linearization observer for fault diagnosis of bearing based on five-degrees-of-freedom mathematical modeling of vibration signals such as reliability and robustness, these techniques have some limitations in the presence of uncertain and unknown conditions. To decrease these limitations, the auto-regressive exogenous input (ARX) technique is advised for bearing system modeling in this work. To improve the stability and robustness of ARX modeling for vibration/acoustic emission signals, an orthonormal function technique based on the ARX-Laguerre method is developed. Moreover, The ARX-Laguerre PI observer is a linear and easy to implement technique for signals estimation but have limited robustness and accuracy. To address these issues, an extended-state technique based on a sliding mode algorithm is applied to the ARX-Laguerre PI observer to perform fault diagnosis and overcome potential problems that may appear when applying a linear observer to a nonlinear signal. Moreover, the simplicity and flexibility of the ARX-Laguerre extended-state PI observation method allow it to be applied in industrial environments for single-type and multiple-type fault diagnosis of bearing.
The ARX-Laguerre technique is robust and stable, but has some limitations when applied to nonlinear and non-stationary signal modeling. To address these problems, a fuzzy ARX-Laguerre technique for vibration and acoustic emission bearing signals is prescribed in this work. Through the high-order super-twisting (extended-state) sliding mode observer increases the robustness and reduces the chattering phenomenon, this scheme, unfortunately, suffers from the small rate chattering phenomenon and signal estimation accuracy in the presence of uncertainties and unknown conditions. Therefore, in this dissertation, the fuzzy technique is applied to the fuzzy ARX-Laguerre high-order super-twisting (extended-state) sliding mode observer to increases the signal estimation accuracy and design fuzzy ARX-Laguerre fuzzy high-order super-twisting (fuzzy extended-state) sliding mode observer.
Once the rotary machinery bearing is modeled based on a mathematical-based modeling (e.g., five-degrees-of-freedom mathematical modeling of vibration signals) or system identification techniques (e.g., ARX-Laguerre technique, and fuzzy ARX-Laguerre method), and the rotary machinery bearing signals are estimated based on the extended-state observers (e.g., high-order super-twisting sliding mode observer, variable structure feedback linearization observer, and ARX-Laguerre sliding mode PI observer) or fuzzy extended-state observer (e.g., fuzzy ARX-Laguerre fuzzy high-order super-twisting sliding mode observer), the decision regarding the bearing conditions can be made. In this work, machine learning-based classification techniques called a support vector machine (SVM) and decision tree (DT) are employed the decision-making procedure for bearing fault diagnosis to complete the proposed techniques for diagnosis the faults. Specifically, during the experiment the high-order super-twisting sliding mode observer, variable structure feedback linearization observer, ARX-Laguerre sliding mode PI observer, and fuzzy ARX-Laguerre fuzzy high-order super-twisting sliding mode observer an average fault diagnosis accuracy of 95.8%, 96.1%, 94.3%, and 99.2%, respectively.
- 필탄 파르진
- Issued Date
- Awarded Date
- Fault diagnosis; Control algorithm; Machine learning; Rotating Machine
- Authorize & License
- Files in This Item:
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.