Troubleshooting mainly includes three parts:
1, the fault detection signal (signal detecting stator current [stator current magnitude and current spectrum], the vibration detection signal, the temperature signal detection, magnetic flux detection method, the insulation detection, noise detection method)
2, a fault signal processing method, i.e., fault feature extraction (FFT, Hilbert transform, wavelet transform, Hilbert-Huang Transform).
3, fault recognition
- Based on analytical model method (to establish a good motor model and estimate the parameters across the state, the need for better knowledge)
- (:; Do fault classification and recognition based on neural network; optimal solution when existing samples SVM, the classification process may be established based on database comparison based expert systems) based on artificial intelligence
Asynchronous motor configuration: a stator, a rotor, bearings, machine housing, blade and the air gap
Working principle: the rotating magnetic field caused by the through current, the rotating magnetic field and the induced currents interact to form electromagnetic torque, the conversion from electrical energy to reach the mechanical energy.
Fault type:
1, the stator portion fault (12.9%)
- Looseness of the stator core and short-circuit :( stator and rotor friction between), can lead to local overheating, excessive load current, vibration and noise strengthened.
- A stator winding short circuit fault :( load overload, overvoltage, undervoltage, the insulating material in question), high heat, short-circuit current, abnormal vibration.
- Stator winding ground fault :( insulated cable is damaged, a coil connected directly to the core or base)
2, the rotor part of the fault (10%)
- Broken rotor bars (current increases and unstable, severe fever), causes:
The sudden increase in motor starting current and temperature, impact load, the rotor poor quality will have a great impact on the rotor.
3, an eccentric gap (originally should be concentric stator and rotor)
This situation will lead to vibration and noise, the worst case is the stator and rotor friction another causing excessive current, the case can be divided into:
- Static eccentric: the production of non-standard or installed in place
- Dynamic eccentricity: mechanical resonance, bearing wear or displacement
4, bearing failure
- Wear (vibration and noise increase)
- Fatigue off (an impact load, vibration and noise will also increase)
- Plastic deformation (increased vibration and noise)
- Corrosion (galvanic corrosion caused by a current, caused by moisture or chemical etching chemical liquid)
- Fault rupture (unreasonable installation work and thermal stress)
- Gluing (inadequate lubrication or high-speed heavy): the temperature will rise rapidly
- Cage damage
Fault current characteristic frequency:
1, stator winding interturn short circuit
Where f1 is the frequency of the power supply
2, broken rotor bar fault
Where f is the frequency of the power supply
3, the air gap eccentricity
4, bearing failure
Feature extraction method of fault:
1, based on Fourier transform window
Is divided into two steps, the first step is windowed in the time period to intercept the signal, the signal after the second step is taken Fourier transform analysis.
For the time domain and frequency domain, it has a central point, the window function is to add a rectangle around a central point.
Reference Links: https://blog.csdn.net/yuejiang_li/article/details/78762201
Its disadvantage is that: after the selected window function also determines the local resolution, can not change the resolution of the transformed signal with poor adaptability ; his time resolution and frequency resolution only choose one , but sometimes higher resolution will take time, sometimes need a high frequency resolution, window function is difficult to achieve.
2, wavelet transform
Reference Links: https://blog.csdn.net/cqfdcw/article/details/84995904
As can be seen from the above, a controlled displacement, b telescoping control.
When a large, time-domain observation range is widened, the frequency domain observed becomes narrow, the center is moved to low; and when a gets smaller, narrower range of observation time domain, the frequency domain is broadened observed, it is moved to the center frequency. For the steep and spikes, the rapidly changing needs good temporal resolution. For the low-band signal, the conversion is slow, requires good frequency resolution. Therefore, the wavelet transform for non-stationary signals, but its limitation is that only do break down for a low-frequency signal .
Thereby introducing a wavelet packet decomposition, wavelet packet may be low and high frequencies do decomposition while each will have to be decomposed wavelet coefficients that can be used for the final wavelet reconstruction.
There are two limitations wavelet transform, a first choice of wavelet basis influence the final result of the relatively large; the second function group is relatively fixed, can not be adjusted according to the signal after the selection.
3, Hilbert-Huang Transform
EMD-based HHT
上式可看出变换后幅值与瞬时频率均与时间有关，因此可得到频段中幅值的时频信息。
T为信号总长，边际谱反应的是每个频率的累积幅值分布，可看出有哪些频率出现，通过Hilbert变换可知道频率对应的时间点。
以HHT为基础的故障特征提取的步骤是：
- 利用EMD方法分别对原始故障信息进行分解，看信息的特点，选取n个分量为对象作为特征提取
- 例如电流选取第二、三、四、五个IMF分量作为特征量。
- 求取这四个特征量的边际谱，将边际谱作为特征向量
- 将特征向量做还原处理后作为神经网络的输入值
神经网络：
神经网络模型的不同主要集中在神经元的特性，网络的拓扑结构和学习规则上。文中主要采用RBF神经网络。
故障诊断网络结构确定的过程为：
- 分别在异步电动机正常、轴承故障、转子断条、转子断条和轴承故障并发状态下采集电流信号
- 对采集完的信号通过HHT做特征处理，提取能够体现故障信息的特征量
- 将处理好的数据作为神经网络的输入，进行训练，当误差满足要求时，神经网络结构被确定。
- 在实际过程中采集定子电流信号，经过处理后作为特征向量输入神经网络进行实时判别，神经网络输出相应状态。
用图表示为：
