MULTIPLE FAULTS DETECTION AND IDENTIFICATION OF THREE PHASE INDUCTION MOTOR USING ADVANCED SIGNAL PROCESSING TECHNIQUES

In this paper, we have presented the multiple fault detection and identification system for three-phase induction motor. Fast Fourier Transform (FFT) is the most used signal processing technique that offers good frequency information but failing in providing time information and handling multiple faults identification with their occurrence time. FFT also fails to detect non-stationary condition of the signal and unable to convey sudden changes, start and end of the events, drifts and trends. To obtain simultaneous time frequency information and to deal with non-stationary signals Short Time Fourier Transform (STFT) is considered optimal technique that can clearly provide time and frequency information both. In this research work, the multiple fault detection and identification system is presented by employing Short Time Fourier Transform (STFT) signal processing technique. The proposed model is designed using current signature analysis method (CSAM) for three major faults including three phase supply imbalance, single phasing condition and breakage of rotor bars. The system is simulated in MATLAB/SIMULINK and simulation is performed based on healthy and unhealthy conditions of the motor. Comparative analysis between FFT and STFT, shows STFT as a promising approach.

is considered optimal technique that can clearly provide time and frequency information both. In this research work, the multiple fault detection and identification system is presented by employing Short Time Fourier Transform (STFT) signal processing technique. The proposed model is designed using current signature analysis method (CSAM) for three major faults including three phase supply imbalance, single phasing condition and breakage of rotor bars. The system is simulated in MATLAB/SIMULINK and simulation is performed based on healthy and unhealthy conditions of the motor. Comparative analysis between FFT and STFT, shows STFT as a promising approach.

INTRODUCTION
An Induction motor is the main source of mechanical power in almost every industry including sugar, fertilizer, packing, agriculture lands, domestic and commercial water supply schemes, water filtration, RO plant, locomotives etc. Apparently, induction motors are widely accepted in industrial processes as well due to its robustness, cost effectiveness, capability to operate in rough environment and less error chance (Pandey, Zope, & Suralkar, 2012;Mortazavizadeh & Mousavi, 2014;Nandi, Toliyat, & Li, 2005;Soother & Daudpoto, 2019). However, like other motors induction motor also faces several faults due to its operating environment and usage conditions. Most of the faults are due to load variations and improper power supply arrangements (Nandi et al., 2005;Soother, Daudpoto, & Shaikh, 2018).
There are many electrical and mechanical faults related to both stator and rotor. Most described faults in the literature related to the rotor are bearing faults, broken rotor and end rings faults, and air gap eccentricity faults (Nandi et al., 2005;Mortazavizadeh & Mousavi, 2014). The faults related to the stator are imbalance in the supply phase voltages, under or over voltage, single phasing condition, reverse phase sequence and inter turn short circuit fault etc. (Nandi et al., 2005;Mortazavizadeh & Mousavi, 2014).
Presently much work is reported in this area to find, isolate and identify different types of the faults and avoid plant shutdown i.e., health of the motor is diagnosed by monitoring certain parameters. The parameter may be the vibration, torque, flux, temperature, current etc. (Mortazavizadeh & Mousavi, 2014;El Bouchikhi, Choqueuse, & Benbouzid, 2015). The condition monitoring makes it possible to detect any abnormal behavior in the motor at an early stage so that any big loss can be avoided (Gao, Cecati, & Ding, 2015). After observing any abnormal condition, the necessary preventive maintenance strategies can be applied for the removal of faults Unlike corrective maintenance strategy in which correction applied after fault has gone through motor and motor operation is disturbed (Mal et al., 2020;Ujjan et al., 2020). In this case, motor may be seriously damaged and can cause unrecoverable loss to the plant.
In vibration monitoring, faults are identified based on intensity of vibrations in healthy and unhealthy conditions. Vibration monitoring sometimes gives ambiguous result when there are fluctuations in the load so thermal monitoring is employed, in which temperature of the different sections of the machine is monitored and faults identified based on the sensors located at different sections on the motors. Thermal technique does not give good results when there are multiple faults in the motor and multiple temperature sensors requirement make it costly (Siddiqui et al., 2014).
Another technique previously used for fault detection is Air-Gap Torque monitored. In this technique motor torque is measured and non-zero frequency of the torque describes the faulty situation of the machine. Its main drawback is that there is no specific mathematical model available for fault signature (Gao et al., 2015). Stator power analyses is another useful technique used previously for unbalance fault detection in which spectral and AC components of the power signal are measured in all three phases. This technique fails to produce good results for low intensity faults (Sharma et al., 2015).
Nowadays, most used technique for condition monitoring of the motor is Motor Current Signature Analyses (MCSA). In MCSA stator current is continuously acquired and after applying a signal processing technique at current signal the frequency spectrum gives the knowledge about the health of the motor (Benbouzid, 2000;Zhongming & Bin, 2000;Gao et al., 2015). The signal processing technique to be applied depends upon the type of the fault to be detected and nature of the fault. Some types of faults are low intensity in nature.
Sometimes only information about frequency component of the signal is desired and, in some cases, both time and frequency information are required. So, it depends upon the fault which signal processing technique will be suitable for it (Nandi et al., 2005;Mortazavizadeh & Mousavi, 2014). The most common signal processing techniques employed are FFT, Short time Fourier Transform (STFT), wavelet transform (WT), Hilbert-Huang transform (HHT) and Wigner-Ville Distribution (WVD) (Gao et al., 2015).
In Mehala and Dahiya (2008), Cusidó et al. (2008) FFT does not provides better results and misses to provide useful information like start and end of the event, changes in load etc. So, a new algorithm is proposed by using STFT and WT to achieve better results. It is further reported that each technique has advantages and disadvantages that depends upon the application and constraint for exampleperformance, complexity, and desired results. It is not advisable to use STFT and WT for ordinary fault where only frequency information is needed. No doubt WT provides better results as compared to STFT but wavelet transform is complex in nature, as signal is divided into high and low frequency parts, therefore requires more calculations and computation time. It always remains an issue of selecting the basis for wavelets which matches with type of information is required. Interpreting the results of the wavelets also requires skill. STFT Phasing of supply and c) broken rotor bars. It also presents the simultaneous detection of multiple faults.
Further this paper proceeds as follows. Section 2 describes possible reasons and impacts of faults in induction motor, Section 3 presents the simulation and mathematical model of induction motor with description of STFT, Section 4 presents comparative results of the proposed system in terms of FFT and STFT analysis, and Section 5 concludes the research. https://doi.org/10.17993/3ctecno.2020.specialissue6.93-117

CAUSES AND EFFECTS OF FAULTS OVER INDUCTION MOTOR
In subsequent sections causes and effects of the faults over induction motor performance and stator current are discussed in detail that is followed by system modeling and simulations.

IMBALANCE SUPPLY VOLTAGE
There are several reasons for the imbalance in the power supply voltages. In Pakistan imbalance voltage condition is frequently faced in domestic, industrial and especially agriculture sector. The induction motors in the agriculture lands are located very far from the electrical substation. So, a very long distribution line and poor arrangement of electrical equipment causes too much voltage fluctuations and sometimes single phasing condition occurs, which can destroy the motor permanently. So, motor with applied imbalance voltage or in single phasing condition must not run for longer time to avoid any damage.
Single phasing is most serious condition that induction motor faces and it can permanently damage the motor. The reasons for single phasing may include blowing of Line fuse, supply terminal loosing, connection of motor from a distribution transformer located very far, distribution transformer phase opening, Power supply wiring conductors may face unequal impedance (Mirabbasi et al., 2009;Lee, 1999).
The imbalance and single phasing condition cause several adverse effects on the performance of the induction motor. According to National Electrical Manufacturers Association (NEMA), for a better life of induction motor it should not be operated with more than 5% unbalance in the supply (Quispe, Gonzalez, & Aguado, 2004). Due to unbalance supply motor may experience negative and pulsating torque which may produce excessive noise.
The imbalance will also increase the current imbalance in windings and temperature of the motor; this can reduce the life and efficiency of the motor.

EFFECT OF SUPPLY VOLTAGES ON STATOR CURRENT
The induction motor is operated at 3 phase supply with 50 Hz frequency. When motor is operating in normal condition the stator current spectrum will show only 50 Hz frequency.
If any type of the fault occurs, it causes sidebands near the main frequency. The frequency https://doi.org/10.17993/3ctecno.2020.specialissue6.93-117 of the generated sidebands depends upon the type of the fault. Every specific fault will have its own current signature.   Due to no. of reasons the cracks appear in the bars as well as at end rings. This may be due to thermal stress that causes overloading, magnetic stress caused by electromagnetic forces, due to electromagnetic force imbalance, vibration and noise cause stress on the Bars. Defect Under normal condition, current distribution in the rotor bars is uniform according to the load applied. Upon breakage of the bars, the resistance of the bars is increased and causes uneven distribution in current loops made by end rings and bars. So, if load is changed during induction motor operation the current distribution is greatly affected (Liang et al., 2014). This type of fault is load dependent.

EFFECT OF FAULTY BROKEN BARS
The change caused by broken rotor bars and end rings in the stator current will introduce new frequency components at the following frequencies (Messaoudi and Sbita, 2010): (2) where, k = 1, 2, 3…, N, f bb : broken rotor bar frequency, fs: electrical supply frequency, p: number of pole pairs, s: slip.
The scope of the research work is limited to the fault detection and identification. This is achieved by continuously monitoring three phase stator current as shown in Figure 3, which depicts the implementation of fault diagnoses system for three phase induction motor. This section presents the Mathematical model of the induction motor including its simulation and describes STFT analysis in the following sub-sections.

MATHEMATICAL MODEL OF THE INDUCTION MOTOR
The three-phase induction motor model is realized in Simulink using famous dq model that is explained in Simion, Livadaru, and Munteanu (2012), Robyns et al. (2012), and Batool and Ahmad (2013). According to this model the three phase quantities are converted to two phase dq model. The three phases, 120 0 electrically apart are converted into two phase voltage i.e. "d" and "q" as shown in Figure 4. The following assumptions are made while considering the two phase dq model:  The equivalent circuit of three phase induction motor is given in Figure 5 with two phases d and q. All the related parameters are shown with labeling.    (Ozpineci & Tolbert, 2003;Leedy, 2013). Three phases to two phase voltages are converted by abc-sync block and syn-abc block performs vice versa function. There are two inputs to the motor and three outputs. The inputs are three phase supply and the applied load. The outputs are three phase current, rotor speed and motor output torque.

SHORT TIME FOURIER TRANSFORM
The STFT overcomes the drawback associated with the Fast Fourier Transform. The FFT performs the function of transformation from time to frequency domain. Usually the transformation is performed to extract the additional information from the signal. FFT is converting the signal from time to frequency completely misses out the time information (Polikar, 1994). In case of induction motor faults, the time information is important because some faults are severe and are not required to persist for long time. While there is some irregular behavior or any incipient fault which usually can be tolerated for certain time and the motor does not require the urgent maintenance. FFT fails to provide time information along with frequency information. By using STFT the simultaneous time and frequency information can be obtained. So, with time-frequency results both type and time of the fault can be identified. Mathematically STFT is expressed below: where, X(τ,ω) is STFT output, x(t) is input signal ,w(τ) is the window function The window function localizes the frequency contents in time. A window function "w n " has a tapering at its end to avoid unnatural irregularities present in the signal frequency contents. The window function is the trade-off between time and frequency. The time and frequency information depend upon the size and type of the window. Larger time window will result poor time resolution and vice versa.
Although there are different types of window that are used to localize time frequency representation, but Hamming, Hann and rectangular window are the most popular. In this research work Hamming window is used which resulted the useful information.    By taking FFT of stator current, the spectral contents can be seen at 50 Hz fundamental component which is the indication of healthy motor in Figure 10, FFT can compute perfect spectral contents but it fails to provide time information in addition to frequency information. To observe frequency and time simultaneously the STFT is computed as shown in Figure 11. Figure 11 shows changes at 1 second which is due to the change in magnitude (load change) from lower to higher level which we are unable to identify using FFT. By having time domain signal, we can identify the time of any irregularity or event and in order to check that irregularity FFT can be computed but to explore simultaneous Time-Frequency information STFT can be the optimal choice.

FAULT 01 SIMULATION: POWER SUPPLY IMBALANCE
The 20 volts drop is simulated in Red Phase as power supply imbalance fault after 1 second and motor is simulated for 2 seconds which can be seen in time domain in Figure 12, having two irregularities, first at 0.5 second and second at 1 second. In order to check whether it is due to change in load or it is because of any fault, FFT is computed. The FFT shows a sideband is generated at 150 Hz in Figure 13, which indicates power supply imbalance fault as equation 1 but unable to determine when it occurred. In other words FFT fails to detect nonstationary condition of the signal. While computing STFT Figure 14, show a change in the load at 0.5 sec and a sideband is generated along with main frequency component at 1 sec. The color of the sideband generated is light because of the low magnitude of 150 Hz component.   Again, the same 20 Volt drop in Red Phase is simulated between 1.5 to 4 seconds and can be seen in Figure 15, that time domain representation is unable to convey the information about irregularities at 1 sec, 1.5 sec and at 4 second. Computing FFT of red phase for same 20 V drop shown in Fig. 16 that a sideband is generated at 150 Hz frequency which is the indication of power supply imbalance fault but again we are unaware what happened at 1sec, 1.5 sec and at 4 sec instants. By analyzing the same fault by STFT Fig. 17, gives some idea about the time of load change and type of fault by looking 150 Hz sideband between 1.5 to 4 seconds.

MULTIPLE FAULT INDUCTION
When multiple faults are present in the motor then it is necessary to find out their occurrence time in order to know the nature of the fault so that it can be decided whether motor should continue to run, or it may be stopped. Figure 27 shows FFT spectrum for broken rotor bars and power supply imbalance faults together. Three side bands along with 50 Hz frequency component are generated. Two sidebands at 36 Hz and 64 Hz are due to the result of two broken bars and one sideband at 150 Hz frequency is the result of power supply imbalance fault but the time information of these faults is missing. Figure 28 shows the computation of STFT for these two faults. It is clear from the spectrogram that broken rotor bar fault is present all the time while power supply imbalance fault has occurred at 1.5 second instant.
Similarly Figures 29 and 30 shows the six broken bars and power supply imbalance faults.
Broken bar fault is present all the time and power supply imbalance fault exist between 1 to 3 seconds.