• Tidak ada hasil yang ditemukan

Review on Condition Monitoring of Rotating Machinery

Condition based monitoring of rotating machinery is an effective tool for the detection and diagnosis of various faults as well as maintaining production continuity in manufacturing industries. Condition monitoring is of great support to predictive maintenance and found to be much better way than corrective and preventive maintenance. The use of condition monitoring allows regular maintenance of the rotor-bearing system so that the maintenance personnel gets

alerted to avoid costly downtime and expensive emergency repairs. Thus, the condition-based maintenance (CBM) shows exactly when a machine needs replacement or is in need of maintenance and ensures optimal use of machines during their lifespan (Randall, 2011).

Condition-based maintenance (CBM) has been discretized into mainly three steps (Jardine et al., 2006) (shown in Figure 1.1), in which the first is data acquisition, the second is data processing and lastly is maintenance decision-making. Data acquisition part collects the information related to data or signals for health conditioning of the system. After that, this data is sent to the data processing domain for handling, analysing and interpreting the data. Finally, the maintenance decision part builds the steps for recommending the efficient maintenance policies. There are two prime facets in CBM program.

(a) Diagnostics: This aspect deals with the detection, isolation and identification of faults when they occur, in which fault detection is a task to indicate whether something is going wrong in the monitored system; fault isolation is a task to locate the component that is faulty; whereas fault identification is a task to determine the nature of the fault when it is detected.

(b) Prognostics: This aspect deals with the prediction of fault before its occurring in the machine. The fault prediction is an activity which decides the time for complete failure of system components due to impending fault. It also estimates the remaining useful life of the system. This is a prior event analysis and much more efficient than diagnostics to achieve zero-downtime performance.

Data Acquision Data Processing Maintenance Decision Making CBM

Figure 1.1 The three main aspects of condition-based maintenance.

Condition of a rotating equipment can be judged by collecting and visualizing the signals originating from the machine during its operation. In signal based condition monitoring technique, the vibration and current signatures are captured from the faulty machine and compared with the reference signature of healthy machine. This technique can be improved and give more as well as depth information and characteristics of various faults utilizing spectral analysis, i.e., frequency response function (FRF) such as Nyquist plot and Bode plot, fast Fourier Transform (FFT) and full spectrum techniques (to show both forward and backward whirls) as well as time-frequency analysis methods (i.e., wavelet transform, short- time Fourier transform), etc.

Model based condition monitoring strategy (Isermann and Balle, 1997; Isermann, 2005) rely on developing the explicit mathematical models of the monitored system. The models are numerically simulated to generate the vibration signals which are indicator of fault present in the machine. Usually, the different approaches, like Kalman filter, system parameters estimation, state observers and output observers as well as parity equations, etc. are used to generate residuals, which are further analysed for detecting, isolating and identifying the rotor- bearing-motor faults. Due to its capability of simultaneously identifying multiple faults in machine components, the model based approaches are usually utilized in rotor dynamic fields.

Based on natural frequency and mode shape approaches, Adams et al. (1978) and Cawley and Adams (1979) proposed that the damage location can be determined by the ratio

of change in the frequency in the two corresponding modes. The position of probable damage sites was obtained by the measurement of one pair of natural frequencies. Moreover, the actual damage site had been found out by the intersection of the curves for several pairs of modes.

This developed method was favourably applied in various laboratories, using the axial resonances of the beams, for moderate levels of damage in structure. Later, Banks et al. (1996) developed a technique using parameterized partial differential equations and Galerkin approximation approaches for the identification of damage location in a structure. They demonstrated that the natural frequencies get affected not only due to the location and severity of the damage, but also from the geometry of the damage.

As condition monitoring systems for fault identification became popular and elaborated, the analysis of vibration data yielded by these systems also becomes more-in depth and involved. Hill and Baines (1988) described the design of an expert system for the purpose of analysis of the measured vibrational data. A computer program was considered as an expert system which processes and analyses the input signals to perform the diagnosis of a fault. This requires knowledgeable and experienced traditional human experts in the field, such as rotating machinery operators, professional engineers, and maintenance managers. They came to the conclusion that an expert system approach linked with vibration monitoring would be quite beneficial, as long as care is taken in the thorough design of the rotating system.

Further, on the basis of change in natural frequency to a numerical model of a beam mounted on a simple support, cantilever support, and also on an elastic foundation (i.e., under different boundary conditions), Choy et al. (1995) developed a novel damage identification technique based on vibration theory in a beam of either uniform or non-uniform cross section.

In this technique, the damage and elastic foundation were modelled using reduction in the modulus of elasticity of a beam element and change in stiffness and damping of a Winkler

spring approach.This developed scheme was capable of identifying both location and severity of the structural damage of faulted elements for single fault and multiple faults.

Relying upon vibration based condition monitoring, Rytter (1993) has given various levels for damage identification in a structural system. These levels include (i) detection (ii) localization (iii) sizing (iv) prognosis. In the first level, the damage which is present in the structural system can be identified. Then, the geometric location of the damage can be determined in the second step. The quantification for the severity of the damage is to be done in the next level. Lastly, the remaining service life of the monitored structure can be predicted in the prognosis process. These steps for identification was observed to be excellent for researchers working in the field of structural damage.

Doebling et al. (1998) have emphasized the four levels of damage identification given by Rytter (1993). They also provided a survey on methods for detecting, locating and characterizing the fault or damage in structural as well as mechanical systems assisted with the change in the measured vibration responses. Vibration based monitoring was done with the variations in modal parameters such as modal frequencies, modal shapes, and modal damping values of the structural system. These modal parameters were obtained from the system physical parameters (i.e., mass, stiffness, and damping coefficients). The methods were explained along with their suitable applications in actual engineering systems for the recent and future scopes. They also summarized the possible difficulties that would appear while their implantation and their fidelity.

A review on condition monitoring of rotating machinery for fault detection and diagnosis methods has been given by Edwards et al. (1998). The faults, such as rotor unbalance, crack and bow in shafts were mostly studied and explained in his review paper. They mainly focussed on the modelling and diagnosis techniques for identification of those faults. They

concluded that model based monitoring methods are quite robust and effective as compared to other identification techniques. These methods can fulfil the demand of present scenario in the field of condition monitoring in rotor dynamic analysis. Later, a review paper was published in a specific manner on the crack detection and the estimation of severity of the crack in shafts (Sabnavis et al., 2004). The types and causes of rotor cracks and basics of their initiation as well as propagation were elaborately described. This review was based on primarily three methods for crack identification, i.e. vibration based methods, modal testing methods (utilize change in the mode shapes and natural frequencies) and non-traditional methods (include neural networks, fuzzy logic, and sophisticated signal processing techniques, e.g. wavelet and Wigner-Ville transforms, etc.).

Sinha (2002) presented a research concerned with health monitoring techniques for rotating machinery, such as turbogenerator sets in the power industry. His research was mainly on three main parts of rotating machine, i.e. flexible rotor, supported fluid-film bearings and flexible foundation. He provided reliable modelling for the system foundation and the mathematical modelling of the rotor (based on finite element method) and the fluid bearings.

The identification technique used measured vibration response at bearing pedestals during run- down of the machine and estimated rotor unbalance and stiffness as well as damping coefficients of the flexible foundation. The method was experimentally validated and found to be robust. It was also demonstrated that the developed condition monitoring technique can estimate reliably two major faults such as rotor unbalance and the misalignment in the rotor.

Heng et al. (2009) reviewed the detection and diagnosis techniques for monitoring the health condition of rotating machineries. In this article, the merits and demerits of present fault diagnosis techniques were elaborately discussed. To predict failures in a rotating machinery, they classified the existing methods into mainly three approaches in which the first approach

is traditional reliability approach, i.e. the event data based prediction, the second approach is prognostics approach which is based on the condition data based prediction, whereas the last way is integrated approach i.e., the prediction based on combination of both the event and condition data.

Bogue (2013) provided a detailed review on sensors and other instruments technology along with their advantageous applications in condition monitoring of rotating machinery. He concluded that the rotor system health monitoring techniques are reliant on various range of sensors based mostly on eddy current, piezoelectric, inductive, electrodynamic, magnetic and thermal technologies. Lei et al. (2013) in his review paper completely focussed on empirical mode decomposition (EMD) method for diagnosis of faults in rotating machinery. They briefly introduced the EMD method and illustrated its benefit and usefulness. The problems in applying EMD technique for diagnosis purpose were also discussed together with their optimum solutions. Then, the applications of EMD method to multiple faults diagnosis of machine in industries were overviewed in the elements of rotating system, such as rolling element bearings, gears and rotors. Later, Lu et al. (2015) combined EMD method with the modified genetic algorithm and receiver operating characteristics (ROC) to develop a hybrid kind of scheme for the purpose of fault diagnosis. They demonstrated that the proposed fault diagnosis model can achieve improvements in identification accuracy with lower feature dimensionality.

Zhang (2018) presented a review of condition monitoring techniques for gas turbines.

It was done with the motive that the efficiency and reliability of gas turbines can be monitored effectively. They described that the condition monitoring for these turbines includes the following steps:

(a) Vibration data collection and pre-processing;

(b) Sensor validation (which also included signal reconstruction in the case of sensor fault);

(c) Separation of steady-state and transient operations;

(d) Novelty detection and fault diagnostics in different operational regimes; and (e) Decision-making and fault report.

Further, the discussed methodologies were classified as knowledge-based rules, signal processing-based techniques and model-based approaches. Among them, model-based approaches were observed to be more effective and robust. The advantages and disadvantages of the methods were also discussed along with the scope of future work in the research area of condition monitoring.

For the sake of solving the problem of fault feature extraction for rotating machinery in a strong impulse noise environment, Miao et al. (2020) proposed a fault separation method based on vibration signals measured by multiple sensors. This method combined both algorithms of the median filter (MF) and an improved joint approximate diagonalization of eigenmatrices (JADE). Through the numerical simulation and experimental investigation of the vibration signal separation of a hybrid rotor, they concluded that the median filtering method can eliminate adequately the signal of impulse noise, improve the signal-to-noise ratio, and provide precondition for the accurate realization of blind separation. But, the combined method (MF-JADE) would provide a good platform for the separation of aliased signals in strong impulse noise environments.

Among studying various literature in the different types of fault detection and diagnosis techniques, it can be concluded that the quantitative model based approach is much better and preferred than qualitative signal based approach to identify the underlying fault. Model based identification method has the advantage of estimating the fault or damage parameters. These

parameters can be further used for the quantification and prognosis. The accuracy of results in mathematical model based techniques is highly sensitive to the model accuracy.