In application, it is often given some specific defect patterns, and machine vision is used to identify whether they have occurred or not. However, it is often encountered that many obvious defects are missed because they have not occurred before or the patterns are too diverse.
With the rapid development of science and technology, machine vision is becoming more and more familiar. Can the machine vision inspection system, which is mainly composed of industrial cameras and analysis software, completely replace the manual visual inspection? If yes, what is the applicable scope? If not, what are the conditions and difficulties?
Point 1:
replace? To be exact, it can only replace some. At present, the living space of small and medium-sized enterprises is limited, and automation upgrading requires funds and talents. Due to various limitations, they are not willing to invest a lot of money, so machine vision detection will only appear in large enterprises with strong strength. At present, the labor cost is the largest, the enterprise is also operating at high cost, and the capital pressure is great. Unless it has to buy, it will make do with it.
Point 2:
In the application of various defect detection, polishing is a difficult point. The algorithm is too difficult to do if the obtained image is to be seen and the result can only be given after careful consideration. On the contrary, if we do a good job in lighting in the early stage and highlight the features to be detected, the algorithm is not difficult.
However, there are also problems. It does not mean that the algorithm is good enough. There is still a huge gap between machine vision and human beings! The biggest gap is intelligence. Yes, the smart camera is far from the word "smart", which is mainly reflected in the recognition of unexpected defects.
In application, it is often given some specific defect patterns, and machine vision is used to identify whether they have occurred or not. However, it is often encountered that many obvious defects are missed because they have not occurred before or the patterns are too diverse. If it is a person, although he is not asked to detect the defect in the operation process document, he will notice it and have a greater chance of catching it.
At present, there are many smart camera suppliers and many analysis software suppliers. Although their software algorithms have their own characteristics, they are similar in use and have very similar functions. However, they all process the pictures obtained by the camera according to fixed patterns and steps, analyze an expected feature from the pictures, and then give the discrimination results. There is no revolutionary intelligent algorithm.
In general, we may not see image analysis algorithms with human intelligence in a short time (10 years), because the relevant theories are not ready.
However, machine vision inspection still has a market, because it solves a very serious problem: instability. No matter what kind of reward and punishment system you design, the operator of manual visual inspection will have a relatively high rate of missed inspection. But there is no problem with the machine vision inspection equipment. As long as it is something you have written in the algorithm, it will be carefully implemented every time. For the quality control of the factory, we are more willing to give up the benefits of wisdom brought by manual visual inspection and choose machine vision, which is silly but meticulous. For the factory, the most important thing is "controlled". We can accept that some special defects have been leaked, but we must know that some frequently occurring unacceptable defects can be found.
In terms of cost, machine vision is cheaper than manpower. The average factory has four shifts, and the cost of four operators is more than 200000 per year. And 200000 yuan is basically enough to develop a set of machine vision. Therefore, it can be concluded that the visual inspection system can not completely replace the artificial visual inspection in the short term. The main reason is that it is not intelligent enough. But for the quality control of the factory, machine vision is more reliable than people.
2018-03-09
7265