Simulation results
1. Introduction
Induction motors are one of the most widely used electrical machines. These motors play a very important role in the present industrial life. Therefore, the need to insure a continuous and safety operation for this motors, involves preventive maintenance programs with fault detection techniques. In general, condition monitoring schemes have concentrated on sensing specific failure modes. Electrical faults such as winding and rotor faults are responsible for an important percentage of total motor failures (Hadremi, 2000) (Bonnett et al., 2008). Preventive maintenance of three-phase induction motors also plays a very important role in the industrial life (Thorsen et al., 1995). This procedure requires monitoring their operation for detection of abnormal electrical and mechanical conditions that indicate, or may lead to, a failure of the system. In fact, in the last years monitoring induction motors becomes very important in order to reduce maintenance costs and prevent unscheduled downtimes. Therefore, there has been a substantial amount of research to provide new condition monitoring techniques for ac induction motors (Kliman et al., 1992) (Thomson, 1999). Several strategies have been proposed to access operation conditions of induction motors (Kliman et al., 1988) (Tavner et al., 1987) (Nandi et al. 1999) (Toliyat et al., 1985) (Riley et al., 1999) (Dister et al., 1998) (Henao et al. 2003) (Li 2006). However, the input stator currents represent the most widely used signal in identifying machine faults (Çolak et al., 2005) (Benbouzid, 2000) (Bayindir et al., 2007). There are several main reasons for this: most of the machine unbalances would reflect into the current waveform, the current is feasible and inexpensive to monitor and the monitoring technique is non-invasive. One of the most significant methods based on the analysis of the machine line currents is the motor current signature analysis (MCSA) (Schoen et al., 1995) (Çalis et al., 2008) (Thomson et al., 2001) (Bellini et al., 2001) (Pires et al., 2009). This method is based on the motor line current monitoring and consequent inspection of its deviations in the frequency domain. Another approach based on the analysis of the machine line currents is the Park’s vector approach (Cardoso et al., 1995). This method is based on the identification of the stator current Concordia patterns, requiring the user to have some degree of expertise in order to distinguish a normal operation condition from a potential failure. A common fact to all of these techniques is that the user must have some degree of expertise in order to distinguish a normal operation condition from a potential failure mode. Therefore, fully automatic pattern recognition methods are required to identify induction motor stator fault. Thus some artificial intelligence tools have been introduced (Nejjari et al., 2000) (Zidani et al., 2003) (Haji et al., 2001) (Ondel et al., 2006) (Martins et al., 2007). However, many of these tools require a prior identification of the system, and only then they are able to identify some faulty situation. On the other hand, the presence of noise along with the nonlinear behaviour of the machine can make this task very difficult.
Under this context, this paper proposes new procedures for the detection of a three-phase induction motor stator fault. These procedures are based on the image identification of the stator current Concordia patterns, and will allow the identification of turn faults in the stator winding as well as its correspondent severity. The identification of the faulty phase is another important feature of the proposed algorithm.
2. Pattern recognition approach
The analysis of the three-phase induction motor can be simplified using the Clark- Concordia transformation (1).
This transformation allows reducing a three-phase system into an equivalent two-phase system (with a zero-sequence component). In three phase induction motors the connection to the mains does not usually use the neutral, implying that the mains current has no homopolar component. Under this situation the three-phase induction motor
Considering ideal conditions for the motor and an unbalanced voltage supply, (3) can be obtained.
Figure 1 shows the structure of the stator currents acquisition, where they are transformed into two-phase equivalents using the Clark-Concordia transformation.
The obtained Clark-Concordia transformation results in a circular or an elliptic pattern, as can be seen in Figure 2. A healthy three-phase induction motor generates a circle (the radius of circle depends on load, being bigger for higher loads and vice versa). A fault in a stator
winding implies a distortion in the circle, assuming an elliptic pattern whose major axis orientation is associated to the faulty phase. This elliptic pattern changes according to the fault severity. For more severe faults the eccentricity of the ellipse increases.
The main idea is to implement a fully automatic pattern recognition process, which recognizes the presence of a fault, identifies the faulty phase and, furthermore, reports a severity index for that fault. The proposed algorithms deal with the diagnosis of stator winding faults by means of pattern recognition methods, based only on the acquisition of two input currents. The described algorithms will extract features of the obtained current patterns, in an automatic pattern recognition process that automatically discern the behaviour of the motor (faulty or healthy). This will allow the identification of turn faults in the stator winding and its correspondent severity.
3. Image processing based system
In order to implement a pattern recognition based fault detection, a feature-based recognition of the current stator pattern, independent of their shape, size and orientation must be obtained. Finding efficient invariants features is the key to solve this problem. Particular attention is paid to statistic moments and visual-based features obtained in the image processing system.
The proposed image-processing algorithm is divided in three stages: the image composition, the boundary representation and the feature extraction (Figure 3). The inputs for the image processing based system are the
3.1. Image composition
In the image composition stage, the
Where
Consider an object with each pixel (
3.2. Boudary representation
To implement the pattern recognition method, after the image composition it is necessary to determine the shape of the region.
To represent the boundary of the region and at the same time obtain some properties that help feature extraction the chain code algorithm is used (Ballard et al., 1982). This algorithm leads to an efficient calculation of the region area and his contour perimeter. However, the right and left upper points (tr_p, tl_p) of the region must also be obtained (Figure 5). These two points corresponds to the first and last pixels present on the first line of the region.
4. Features extraction methodologies
The last stage of the image-processing algorithm is the features extraction. There are several feactures extraction methodologies. Here are presented two effective algorithms that can be used for the induction motor fault detection: Visual features and statistic based moments methods.
4.1. Visual features
The feature extraction stage uses the area of the object and the contour perimeter to compute the index of compactness. To obtain the distance between the
The index of compactness and the distance between the
Assume that the pixels in the digital image are piecewise constant and the dimension of the bounded region image for each object is denoted by M×N, the visual features, area and perimeter, used to determine the I.O.C. can be obtained as (Amaral et al., 2001):
where
The index of compactness is then given by:
Physically, the index of compactness denotes the fraction of maximum area that can be encircling by the perimeter actually occupied by the object.
The coordinate
where the
The distance between the
where
In Figure 6 it is represented the distance between the
4.2. Statistic moments
Another possible approach to obtain efficient invariants features is based on the use of the statistic moments algorithm. This algorithm uses the regular and the central moments to obtain the features needed for the fault diagnosis (Amaral et al., 2007). The
Considering an object with each pixel (
where the sum is made over the image all M pixels.
The (p+q) order of the central moment is invariant to translation and is obtained considering the origin of the object at his gravity center (12).
The
The features used in the system recognition are the main axis angle of the
These two features can be obtained more effective using only three and four central moments in the computation of the main axis angle and eccentricity, respectively. The relation between the regular and the central moments can be obtained by:
Using these relations, the main axis angle -
A binary image can be considered as a particular case of a grey image with
5. Neuro-fuzzy classifier
Neural networks and statistical methods are particularly well adapted for data classification. However it is also possible to use a fuzzy system for the classification problem without replacing the aforementioned methods but as a different approach to achieve the same results, with the following advantages:
vague knowledge can be used;
the classifier is interpretable;
easy to implement, to use and to understand.
Since classifiers are usually derived from data, one method to obtain a fuzzy classifier is to use a neuro-fuzzy approach. The fuzzy classifier obtained by a learning procedure is called a neuro-fuzzy classifier but this “neuro-fuzzy” term strictly only applies to the creation or in the training phase. Afterwards, once the classifier is applied there only remains a fuzzy system for classification. This type of neuro-fuzzy system it is called a cooperative neurofuzzy system (Nauck et al., 1997).
The interpretation of the classifier is often illustrated by representing it in a neural network structure. Figure 8 shows the structure of the neuro-fuzzy inference system used in data classification. This figure represents the structure example of a four-layered neuro-fuzzy system with two inputs and one output. This connectionist structure effectively performs the fuzzy inferencing with some minor restrictions. In a generalized architecture for the proposed approach, there can be
Each layer has a specific functionality that contributes to the equivalence of the fuzzy inference system.
The input layer consists in the input of the data values derived from the input features. The output of each node is the degree to which the given input satisfies the linguistic label associated to the correspondent node.
The pattern layer represents the rule set of the fuzzy inference system. The number of nodes in this layer it is the same as the number of rules in the fuzzy inference system. Each node, called rule node, is a special type of radial basis function processor and represents a rule where the output of each basis function unit
The summation layer has two different types of processing units: the summation and the division units. The function of the summation unit is essentially the same as the function of the output units in the radial basis function network. The division unit only sums the weighted activations of the pattern units without using any activation function. The number of the summation units is always the same as the number of the G.R.N.N. output units.
Each node in the output layer consists in a simple division of the signal coming from the summation unit by the signal coming from the division unit, as defined by equation (22).
If the fuzzy inference system has one output then the linear function
The number of layers of the adaptive neuro-fuzzy inference system and the function of each node in the four layers are also similar to the G.R.N.N.. Thus it was constructed an adaptive network that it is functionally equivalent to a type-3 fuzzy inference system (Takagi & Sugeno fuzzy if-then rules).
In the learning of the fuzzy inference system two learning algorithms are applied. An iterative grid partition algorithm tunes the four parameters of each antecedent membership function (Amaral et al., 2002a). The first three parameters correspond to the left, the center and the right position of each membership function and the fourth parameter to its shape. The parameters of local model
6. Results
In order to obtain experimental results it was used and induction motor with the following characteristics: 1kW, 220/380V, 50Hz. The machine load is provided by means of an electromagnetic brake and the stator currents are acquired through a personal computer data acquisition board (Figure 9). After the acquisition of the stator currents, these currents are transformed into a 2D object using the
6.1. Simulation results based on visual features
In order to verify the proposed algorithm, the fault diagnostic system was first simulated in the Matlab environment. The induction motor was initially simulated without any fault. In this case the corresponding
6.2. Experimental results based on visual features
The induction motor was initially operated without any fault. In this case the corresponding
Figure 14 presents the corresponding
6.3. Simulation results based on statistic moments
In order to verify the proposed method, the system shown in Figure 9 was first simulated in the Matlab environment. The induction motor was initially simulated without any fault. In this case the corresponding
6.4. Experimental results based on statistic moments
Initially the induction motor operated with no fault and the obtained
7. Conclusions
In this chapter an image processing based classifier for detection and diagnosis of induction motor stator fault was presented. This system is based on the obtained stator currents and the correspondent Clark-Concordia transformation. This results in a circular or an elliptic pattern, of the Clark-Concordia stator currents. From the obtained current patterns it was used an image processing algorithm to identify if there is a motor fault. The image processing algorithm is divided in three stages: the image composition, the boundary representation and the feature extraction. For the the features extraction it was used two effective algorithms that can be used for the induction motor fault detection: Visual features and statistic based moments methods. The extract features of the the obtained current patterns allows to identify the turn faults in the stator winding and its correspondent severity. Several simulation and experimental results have been presented. From these results it was possible to verify that the proposed image based classifier can effectively be used for the diagnosis of induction motors.
References
- 1.
Amaral T. G. Crisóstomo M. M. Pires V. F. Almeida A. T. 2001 Visual Motion Detection with Hand-Eye Manipulator Using Statistic Moments Classifiers and Fuzzy Logic Approach- Study, Application and Comparation. Conference of the IEEE Industrial Electronics Society, IECON’01, Denver, Colorado, USA, November 29 to December 2, 2001. - 2.
Amaral T. G. Crisóstomo M. M. Pires V. F. 2002a Adaptive Neuro-Fuzzy Inference System for Modelling and Control”, Systems, Bulgária, September. - 3.
Amaral T. G. Crisóstomo M. M. Pires V. F. 2002b Helicopter Motion Control Using Adaptive Neuro-Fuzzy Inference Controller . Industrial Electronics Society, IECON’02,2090 2095 , Sevilla, Spain, November 2002. - 4.
Amaral T. G. Pires V. F. Martins J. F. Pires A. J. Crisóstomo M. M. 2007 Statistic Moment Based Method for the Detection and Diagnosis of Induction Motor Stator Fault. , POWERENG’07,106 110 , Setúbal, Portugal, April 2007. - 5.
Ballard D. H. Brown C. M. 1982 Prentice-Hall. - 6.
Bayindir R. Sefa I. 2007 Novel approach based on microcontroller to online protection of induction motors , ,48 3,850 856 , March 2007. - 7.
Bellini A. Filippetti F. Franceschini G. Tassoni C. 2001 Quantitative Evaluation of Induction Motor Broken Bars by Means of Electrical Signature Analysi s, Transactions on Industrial Applications,37 5 1248 1255 , Sep./Oct. 2001. - 8.
Benbouzid M. E. H. 2000 A review of induction motor signature analysis as a medium for faults detection, IEEE Transactions Industrial Electronics,47 5 984 993 , Oct. 2000. - 9.
Bonnett A. H. Yungb C. 2008 Increased Efficiency Versus Increased Reliability , IEEE Industrial Applications Magazine,29 36 , Jan/Feb 2008. - 10.
Cardoso A. J. Cruz M. S. M. A. Carvalho J. F. S. Saraiva E. S. 1995 Rotor Cage Fault Diagnosis in Three-Phase Induction Motors by Park’s Vector Approach, Industry Applications Conference 1995,1 642 646 , 1995. - 11.
Çalis H. Çakır A. 2008 Experimental study for sensorless broken bar detection in induction motors, Energy Conversion and Management,49 4,854 862 , April 2008. - 12.
Chen C. H. 1996 Fuzzy Logic and Neural Network Handbook , McGraw-Hill. - 13.
Çolak I. Çelik H. Sefa I. Demirbas S. 2005 On line protection systems for induction motors, Energy Conversion and Management,46 17,2773 2786 , Octuber 2005. - 14.
Driankov D. Hellendoorn H. Reinfrank M. 1993 An Introduction to Fuzzy Control , Springer-Verlag. - 15.
Dister C. J. Schiferl R. 1998 Using temperature, voltage, and/or speed measurements to improve trending of induction motor rms currents in process control and diagnostics, Proceedings of the IEEE- IAS annual meeting conference,1 312 318 , October 1998. - 16.
Hadremi B. M. El 2000 A review of induction motor signature analysis as a medium for fault detection, IEEE Transactions on Industrial Electronics,47 984 993 , October 2000. - 17.
Haji M. Toliyat H. A. 2001 Pattern recognition-a technique for induction machines rotor broken bar detection , ,16 4 312 317 , December 2001. - 18.
Henao H. Demian C. Capolino G. A. 2003 A frequency domain detection of stator winding faults in induction machines using an external flux sensor , ,39 5 1272 1279 , September/October 2003. - 19.
Kliman G. B. Stein J. 1992 Methods of motor current signature analysis , Electric Machines and Power Systems,20 nº5,463 474 , September 1992. - 20.
Kliman G. B. Koegl R. A. Stein J. Endicott R. D. Madden M. W. 1988 Noninvasive detection of broken rotor bars in operating induction motors , ,EC-3 ,4 873 879 , December 1988. - 21.
Li,W. 2006 Detection of Induction Motor Faults: A Comparison of Stator Current, Vibration and Acoustic Methods , ,12 2 165 188 , February 2006. - 22.
Martins J. F. Pires V. F. Pires A. J. 2007 Unsupervised Neural-Network-Based Algorithm for an On-Line Diagnosis of Three-Phase Induction Motor Stator Fault , ,54 259 264 , February 2007. - 23.
Nandi S. Toliyat H. A. 1999 Condition monitoring and fault diagnosis of electrical machines- A review, IEEE Industry Applications Conference 1999,1 197 204 , 1999. - 24.
Nauck D. Klawon F. Kruse R. 1997 Foundations of Neuro-Fuzzy Systems , John Wiley & Sons. - 25.
Nejjari H. Benbouzid M. 2000 Monitoring and Diagnosis of Induction Motors Electrical Faults Using a Current Park’s Vector Pattern Learning Approach , ,36 730 735 , May/June 2000. - 26.
Ondel O. Boutleux E. Clerc G. 2006 A method to detect broken bars in induction machine using pattern recognition techniques , IEEE Trans. Industry Applications,42 4 916 923 , July-August 2006. - 27.
Pires D. F. Pires V. F. Martins J. F. Pires A. J. 2009 Rotor Cage Fault Diagnosis in Three-Phase Induction Motors Based on a Current and Virtual Flux Approach , ,50 4,1026 1032 , 2009. - 28.
Riley C. M. Lin B. K. Habetler T. G. Kliman T. G. 1999 Stator Current Harmonics and their Causal Vibrations: A Preliminary Investigation of Sensorless Vibration Monitoring Applications , ,35 nº 1,94 99 ,January/February 1999. - 29.
Schoen R. S. Habetler T. G. Kamran F. Bartheld R. G. 1995 Motor Bearing Damage Detection Using Stator Current Monitoring , Applications,31 6 1274 1279 , November/December 1995. - 30.
Specht D. F. 1991 A General Regression Neural Network , ,2 nº. 6,568 576 , November 1991. - 31.
Tavner P. J. Penman J. 1987 Condition Monitoring of Electrical Machines , New York: Wiley, 1987. - 32.
Thorsen O. V. Dalva M. 1995 A survey of faults on induction motors in offshore oil industry, petrochemical industry, gas terminals, and oil refineries , Transactions on Industrial Applications,31 5 1186 1196 , September/October 1995. - 33.
Thomson W. T. 1999 A review of on-line condition monitoring techniques for three-phase squirrel cage induction motors-Past present and future, Diagnostics for Electrical Machines, Power Electronics and Drives,7 3 18 , September 1999. - 34.
Thomson W. T. Fenger M. 2001 Current signature analysis to detect induction motor faults , ,7 26 34 , July-August 2001. - 35.
Toliyat H. A. Lipo T. A. 1985 Transient analysis of cage induction machines under stator, rotor bar and end ring fault s, IEEE Transactions Energy Conversion,10 241 247 , June 1985. - 36.
Tsoukalas L. T. Uhrig R. E. 1997 Fuzzy and Neural Approaches in Engineering , John Wiley & Sons, Inc.. - 37.
Zidani F. Benbouzid M. Diallo D. Nait-Said M. 2003 Induction Motor Stator Faults Diagnosis by a Current Concordia Pattern-Based Fuzzy Decision System , Transactions Energy Conversion,18 469 475 , December 2003. - 38.
Zidani F. Benbouzid M. Diallo D. Nait-Said M. 2003 Induction Motor Stator Faults Diagnosis by a Current Concordia Pattern-Based Fuzzy Decision System , ,18 469 475 , December 2003.