Industrial big data under a large field of vision
1, the source of the data
There is a view that the source of big data is the business system, and the big data platform just collects these data by the way. This does not create additional costs for the collection of data and does not affect the operation of the original business system. However, the data in each business system is often not established for analysis purposes, and the associated relationship is lost. In this way, a lot of value in data is lost. My point is that if this view holds, data analysis should be thought of when establishing a business system. Otherwise, the value of the data will be greatly reduced. In the future, if you want to make the data play a big role, it may take a lot of time to manually process some data, combine the knowledge in the expert's mind with the data, and then store it in the database: if you wait until you analyze the data, do it again. A lot of information has long been lost. Of course, the standardization of this process itself must be done well. Otherwise it will even turn good data into junk.
2. Is there a clear business function?
Does the industrial big data platform have any specific features? Most viewpoints hold that the function of big data is excavation afterwards. For the post-mortem analysis, there are two attitudes: what data are used for what data, and data are collected for analysis. We believe that from the perspective of the future, the latter should be the direction of development. At this time, the quality and integrity of the data become very important.
In my opinion, with the improvement of ICT technology, the big data platform is likely to become a new generation of intelligent monitoring systems (GE's vision for aircraft engines should be). Different from the traditional monitoring system, the platform can memorize a large number of past cases and disposal methods. This kind of monitoring in the future is likely to serve unmanned, less human, and mobile monitoring. If this is the case, there will be extremely high demands on the data quality, transmission reliability, and implementation of big data platforms. In other words, the mutual promotion of smart manufacturing and industrial big data will greatly expand the development prospects in this field.
3, what knowledge can big data
People have long realized that the quality of data is determined by the purpose of the application. Therefore, before building a platform for big data, it is best to be able to clearly identify what kind of knowledge you want, rather than to emphasize analysis of knowledge in general. I think that one end of this knowledge is the result of the company's concern, such as quality, efficiency, energy consumption, defect rate, job rate, equipment status, and completion time, and the other end is the reason related to these factors. The knowledge we want to obtain is roughly divided into two categories: The first level of analysis is the relationship between the cause and the result, such as the relationship between the A variable and the B variable. But in reality, this relationship is often very unstable. When other factors change, this relationship will also change. Therefore, further knowledge is to know: After the factors are fixed, the relationship between variables is stable. This knowledge can be used to increase the level of production organization, review relevant personnel, find various leaks, and clearly optimize the focus. Of course, the ideal situation is to analyze the relationship of multivariate to multivariate. However, unfortunately, the results of such analysis are often inaccessible.
4. Knowledge-related human-robot problem
The process of discovering knowledge is to understand the gradual process, and it is the process of grasping the reliability of knowledge continuously. This process is often achieved by human-computer interaction. First of all, people have to put forward their own ideas about possible correlations; secondly, they can screen people's ideas through computers; then, in the screening process, there may be a large number of things that are difficult to explain or cannot be confirmed, and people need to carry out in-depth research. Comparative analysis: If necessary, even additional data and even new assumptions need to be tested and documented. I once wanted to find out the mechanism of unclear problems by finding an automated algorithm. This is difficult to achieve in reality. The best way in reality is often that the convergence of "reliability" is faster and the workload of people is relatively small. I have always believed that: The ability of humans to analyze complex problems is far behind computers. We have more material in the era of big data, but to do better, we need more intelligent people to complete. Of course, everything has counterexamples, especially the issues that all humans are very concerned about. At this time, human beings will do all sorts of preparations for machine intelligence at all costs. However, for the average enterprise, the analysis of machines instead of people is probably not worthwhile.
5, understanding of the concept
Study industrial big data, do not have to worry about how much data. Our concern is: how to make data create business value. I have always believed that the current statistical theory is mainly focused on small data sets, such as within a few dozen samples; the data mining methods of decades ago basically stopped at the scientific level, barely entered the technical scale, and rarely entered the business. Scale; the current deep learning theory, it is difficult to extensively enter the industrial field.
