As was pointed out in the literature review, the history of condition monitoring and fault diagnosis is as old as the induction motor itself. The induction motors have been initially relied on simple protection such as over-current and over-voltage to ensure safe operation. In spite of these tools, many companies are still faced unexpected system failures and reduced motor lifetime. Redundancy and conservative design techniques have been adopted for improving the reliability of induction motor drive systems against a variety of faults that could occur. However, these techniques are expensive to realize.
Condition monitoring is leading to incipient fault detection and prediction of induction motors, which has attracted many researchers in the past few years owing to its considerable influence on the safe operation of many industrial processes. Early detection, prediction and correct diagnosis of incipient faults could allow preventive maintenance to be performed and provide sufficient time for controlling the shutdown of product line. It could reduce the financial losses and avoid catastrophic consequences. As discussed above this topic could be treated under three headings: thermal, current and vibration monitoring. These methods have its own advantages and disadvantages, this is the reason of why that the thermal monitoring and vibration monitoring have been paid less attention than current monitoring. Previous studies of condition monitoring have not dealt with thermal monitoring and have paid less attention because of the thermal sensors, which are needed to access the motor performance such as thermocouples, resistance temperature detectors (RTD), winding thermostat and thermistor. It has been reported that the thermal monitoring was
insulation life, which have effects on the motor resistance to either environmental or mechanical effects [147], [148].
Such approaches, however, have failed to address the induction motor faults without thermal monitoring technique. In one hand, recently, the researchers have used thermal cameras to monitor the rotating machinery and read the device temperature in healthy and faulty conditions without any access to the motor (contactless) based on the image processing technique to detect the motor faults. On the other hand, up to now, the research has tended to focus on the current monitoring (electrical monitoring) because it does not need to any additional sensors, as the current and voltage transformers are connected to the protection system at all times. Thus, the MCSA was very popular for monitoring the induction motor since it is non-intrusive detection (does not disconnect the electrical circuit), safe to operate (no contact between the motor and the current transformer) and remote sensing (current transformer could be place anywhere for monitoring) [9], [69], [149]–[152].
All the studies reviewed so far, however, suffer from the fact that MCSA is not appropriate for analysing the non-stationary signals. Another problem with this approach is that it fails to take the low signal to noise ratio into account, which makes the MCSA non-sensitive under certain conditions such as in inverter-fed motor as stated in [4], [147], [153], [154]. There would be therefore a definite need for vibration monitoring for mechanical faults detection because it allows different locations for sensors to be mounted on the motor, while MCSA relying on the radial rotor movement. Consequently, in case of bearing fault, the MCSA has difficulty in distinguishing non-drive-end or drive-end if two bearings have similar physical
characteristics. Furthermore, the vibration signal has higher signal to noise ratio than the MCSA as shown in table 3-3 [155].
Table 3-3: Common differences between the vibration and current signals [152].
Fault type VIB MCSA
Electrical faults detection × √
Radial rotor movement analysis × √
Cheap installation × √
Able to apply in rough environment × √
Mechanical faults detection at early stage √ × Easy to distinguish between different bearings √ ×
Mean Time To Failure (MTTF) √ ×
Higher signal to noise ratio √ ×
Moving on to consider the AI techniques for induction motor faults detection based on data mining. As indicated previously, most of AI techniques such as (ANN, GA, NN and SVM) have been applied and validated successfully for diagnosing the motor faults with different classification accuracy. Although extensive research has been carried out on the use of AI for induction motor fault detection and prediction, no single study exist shows that there is one best technique for all kind of motors to diagnose the faults. This is because, the bigger the dataset the more complex task for classifiation, the noise in the data may lead to insufficient and irrelevant to orginal class, and the overfitting problem is difficult to overcome because it affects the classification system. Therefore, several studies have revealed that the development of induction motor fault detection based on AI techniques is still in its early stages. Consequently, despite that the considerable work have been done in this field, much more work are required to bring such techniques into the mainstream of induction motor fault diagnosis. Due to the limitations and strengths of these techniques, the
could have an effect on the developing on the rotating machinery condition monitoring for fault diagnosis scheme.
The previous sections have shown that many researches have been performed for IM fault detection by relying on the traditional methods and some AI techniques. Furthermore, the problem of IM fault still exists in many manufacturing applications. For that reason, the need for data mining algorithms are very important to detect and predict the fault before it happens based on the motor previous behaviour (data) in order to reduce the breakdowns of the electric machines.
Considering the aforementioned shortcomings of the methods that were used in motor condition monitoring, this research is aimed to address these disadvantages by presenting new classification technique. This technique is based on data mining rule discovery that are simple in algorithm design, and easy to apply for three kinds of condition monitoring technique, which are thermal, current and vibration monitoring, based on simple digital image and signal processing.
The next chapter describes the proposed hybrid approach by combing the Bees Algorithm and Data Mining methods that are used for condition monitoring in order to detect, classify and diagnose the induction motor faults at an early stage.
4 CHAPTER
4
PROPOSED BEE FOR MINING (B4M)
“In this chapter, the combination of the proposed methods Bees Algorithm and Data Mining for induction motor faults detection have been described and explained in details, so called Bee for Mining (B4M). The proposed method (B4M) has been tested and validated based on the UCI dataset and its performance has been compared with other well-known classifiers. The proposed method has been translated as a software code or toolbox package using MATLAB software version “R2015a””.