my learning experience
Time flies, and recalling the time when I just entered graduate school, it seems that it is still close in front of my eyes. I fell into the "pit" of fault diagnosis in a daze, and I was also at a loss at the time. I don't know what the "spectral kurtosis", "Hilbert-Huang transform", "transfer learning", "machine learning", and "LSTM" are. Alas, it seems that I am a waste who is not even familiar with the Fast Fourier Transform (FFT). I was anxious and anxious all day long, but I had a thought in my heart, that is, I can't give up. Like all my classmates, I once secretly typed in the search bar of Zhihu in the long dark night "how to get started with fault diagnosis?", "how can a novice quickly get started with fault diagnosis?" Is it?" These topics. It’s not the same as in the martial arts novel where the protagonist picks up martial arts secrets in a cave and then grows into a super master after a few days and nights of footage. The reality is even more painful. Countless days and nights of hard work are exchanged for such a little bit of improvement. and progress. But looking back and looking at all of this, I think it was indeed a very fulfilling time. Therefore, I think that these wonderful experiences and the pitfalls we have gone through should not be gradually forgotten as time goes by. We hope to sort them out bit by bit systematically, even if they can be shared among thousands of people. , inspire those few readers, and we will be very moved. ——When Brother Bearing
was studying in graduate school, the subject assigned by the tutor was to do this part of the content. The master’s degree was all electrical. At first contact with this part, it was really a bit confusing. I feel that this direction is difficult but not difficult. The main reason is that the interdisciplinary nature is relatively strong, and scattered knowledge points need to be pieced together. Each part may not need to be particularly deep in the short term. It is enough to be able to use it in the early stage, or to put it more simply, it is It can produce pictures, modify parameters, and post articles after graduation. At that time, I was not really interested in this field, and I was very confused about future employment. I didn’t know what kind of job I could find in the future, and I was very conflicted, but I thought: "Life is not a waste of time, every step counts. At that time, the cards you get are like this, although it is a bad hand, but whether you play it or not, if you don’t play it, you will be yellow.” The above are all my inner monologues back then. Although it was a bit naive, when I was doing diagnosis, I really got through the days of graduate school with such beliefs.
Later, during my postgraduate study, I took a project as the main line, and successively did the content of bearing, gear and other mechanical fault diagnosis. From the beginning, I didn’t even know what the sampling frequency was. Buy books to read, (because the books and thesis materials in this area are too scattered and there is no system, so it is very challenging for beginners, especially those who are new to this content. It’s a psychological barrier), until now I can basically transplant the algorithm I made with Matlab during my graduate study or even the content of directly calling the function library to the embedded software, and the effect of the algorithm is relatively stable. During this period, many methods have been explored, such as spectral kurtosis, Hilbert-Huang transform, various spectral analysis algorithms, and their combined optimization, preprocessing and noise reduction, and the optimization of algorithm parameters combined with intelligent algorithms such as particle swarm optimization. There's even some pattern recognition stuff, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah, blah.
Now that I am working, many things have nothing to do with the past, but I don’t want this part of the memory to fade over time. I wonder if I can spare a little time after work to study the past research content. Form a curriculum system suitable for college students to study? Of course, some content may also require payment. Work, time is really tight, but do it slowly, I hope I can stick to it.
Finally, I would like to share with you my favorite sentence:
before making a decision, any entanglement is okay. The reason for entanglement is that the choices you face are equally good or equally bad. Once you have made a decision, don't get entangled. Any entanglement is a waste of time. People always have a mentality of "what you can't get is always in turmoil" and "whether the other way is better".
Talking about Fault Diagnosis
Anyone who has a little knowledge of this field knows that fault diagnosis and life prediction include model-driven analysis methods and data-driven analysis methods. The former refers to various fault frequencies, spectrum waveforms, and various filters that you are familiar with. The latter is a fault diagnosis model based on deep learning. At present, in actual industrial applications, most of them are still model-based methods, and methods based on deep learning are more versatile, but there are also a series of problems, and there is still a long way to go. The future must be a model-based intelligent AI model algorithm, so many types of data: acceleration, current, temperature, sound, which signal source to choose? How many are the best? What is the most appropriate amount of data? How many sensors can save costs and ensure accuracy? Many of the articles in the academic world are play a trick. The author believes that no matter how deep learning develops in the future, it is not realistic to completely abandon the model. An ideal state must be based on a model-driven intelligent AI model. The advantages of the two should be fully considered and complement each other. I believe that many readers and friends are writing papers and graduating. From my experience, in summary, they need to walk on two legs, understand the mechanism, and learn deep learning. On this basis, there is a focus, focusing on in-depth research on one aspect, I believe it is still easy to produce the results of the article. For example, if you are a student majoring in computer science, in fact, the fault diagnosis and life prediction of bearings, gears, etc. are just to provide an application scenario for their models and algorithms. , Put on a hat, it looks more down-to-earth. If you are studying mechanics or electricity, you can't justify it if you don't understand the mechanism at all? To sum up, both hands must be grasped, and both hands must be hard (if time and energy are not enough, one hand should be hard). It is still not easy for students to set up a platform for the student party!
I also used my spare time to create a basic course on the introduction of bearing fault diagnosis. I believe it can help you get started with this course as quickly as possible. Interested friends can private message me to get the complete course resources.My WeChat is: ForwardTszs.
![ ](https://img-blog.csdnimg.cn/9ca4fe4fbd0241e9b8a09cf62578bfec.png