Hi，大家好，好久不見，這篇文章是用3個層面(General to specific)講解我論文的緒論(馬達頻譜，人工智慧，專家系統壽命推斷) 。 要寫論文的同學或朋友可以參考。 然後文章可以任意複製貼上。 如果喜歡我的文章，請多多訂閱和關注我😊😊😉😉😉😉😉😉😉
Hi, everyone, long time no see, this article is to explain the introduction of my thesis (motor frequency spectrum, artificial intelligence, expert system lifetime estimation) in 3 levels (General to specific). Students or friends who want to write a paper can refer to it. Then the article can be copied and pasted at will. If you like my articles, please subscribe and follow me more😊😊😉😉😉😉😉😉😉
Hi，大家好，好久不见, 这篇文章是用3个层面(General to specific)讲解我论文的绪论(马达频谱，人工智慧，专家系统寿命推断) 。要写论文的同学或朋友可以参考。然后文章可以任意复制贴上。如果喜欢我的文章，请多多订阅和关注我😊😊😉😉😉😉😉😉😉
Explain my thesis on youtube:
https://youtu.be/gQ_hw8eUhtU
https://fb.watch/56924QCcUm/
This thesis
is mainly to explore how to really know the life of the motor by analyzing the
motor spectrum. Spectral analysis in the motor spectrogram is often the most
critical part of motor fault analysis. Among all the non-invasive analysis
detection signals (voltage, power, temperature, etc.), the signal storing the
most abundant signal is the vibrational signal. The vibrational spectrum signal
concentrates a large amount of analyzable information. For instance, the
spectrum diagram in a certain time series is rich in many vibration signals,
and any vibration signal is combined by three different factors: frequency,
phase, and amplitude. These three elements have different meanings for the
diagnosis of motor machinery: for the amplitude, engineers often look at these
spectrum change diagrams and the amplitude information received in order to
judge the seriousness of the operation. It is judged that the frequency
distribution in the motor spectrum diagram may be the location of equipment
damage or vibration source, and the frequency distribution in the motor
spectrum is an indicator of the vibration mode generated by the operation of
the equipment. Using the technology of this study, the motor spectrum can be
extracted and predicted, and the predicted spectrum can be analyzed to
determine whether the motor will malfunction within a short period of time, the
cause of the failure, and calculate how long it takes for the failure to occur.
Although
many methods of motor prognosis and health management have been proposed in the
past, almost all methods are to process the original data. The expert system is
used to judge the motor fault spectrum of the "current" motor
spectrum, which is not ideal because when the motor is judged to be
malfunctioning in the "present", most of the economic losses have
been caused. Moreover, the motor can no longer be used or the factory will stop
production. In addition, if only the "current" motor spectrum data is
processed, the compressed motor spectrum data must be first transmitted and
then analyzed and processed, which causes a time delay, leading to a decrease in
inefficiency.
Some papers
directly use the expert system to detect and judge the transformed spectral
data, but this research project explores this problem from the perspective of
artificial intelligence. That is to say, machine learning can be used to let
the program learn to predict unknown motor spectrum data from the known motor
spectrum, and make the prediction accuracy higher and higher.
This study
proposes to use five models: MLP (Multilayer Perceptron), LSTM (Long and Short
Term Memory), CNN (Convolutional Neural Network), Encoder-Decoder LSTM,
CNN-LSTM, and use these five models to predict and process the spectrum of
motor vibration. These five models each have different advantages and can also
be insufficiently complementary. Multilayer Perceptron (MLP), the model can provide
modeling for complex nonlinear systems, but MLP is prone to local extremum.
Therefore, if you want to build a model, you need to select the proper network
parameters in the MLP model (ie, the number of nodes in each layer). The
selection of layers requires a lot of effort; while long-term and short-term
memory (long and short memory, hereinafter referred to as LSTM) can model the
change of parameters in time series, LSTM has memory ability, and this memory
ability is through the LSTM area. There is a memory gate in the block that
determines whether the input value is important to be memorized and whether the
output value can be output. If the amount of
LSTM training data is insufficient, there will be a serious overfitting
problem; convolutional neural network (hereinafter referred to as CNN) is often
used for high-dimensional data, but the appropriate parameters need to be
adjusted to achieve better prediction performance; Encoder-Decoder LSTM
(EDLSTM) compresses the entire input sequence information into a fixed-length
vector, but the drawback of this approach is that this fixed-length vector may not
fully express the entire sequence information; the advantage of convolutional
neural network and long-term and short-term memory (convolution neural network
- long-term and short-term memory, hereinafter referred to as CNN-LSTM) model
is that the CNN module can be used to read the long input data as sequences or
blocks. The sequences are then pieced together by LSTM.
This study
uses the above five models to analyze the spectrum of the motor vibration and
then using the MAE, MSE as an
indicator to measure the accuracy of the performance. and use the spectrum fault diagnosis rules to
diagnose the prediction results. The research found that, according to the
above method, the life of the motor can be surely understood, and the motor can
be protected early.
論文網址: https://www.airitilibrary.com/Publication/alDetailedMesh1?DocID=U0001-3101202110161400
https://hdl.handle.net/11296/4g2mry
Explain my thesis on youtube:
https://youtu.be/gQ_hw8eUhtU
https://fb.watch/56924QCcUm/
