Core demands
The daily average demand for defect detection of injector valve seat is 4000-6000 pieces, and the peak demand is 12000 pieces. At present, it can only be judged by human eyes. However, there are two similar inspection processes in the fuel injector manufacturing. At present, skilled workers need to pay 4~7 people per shift to review the visual judgment process. If calculated by three shifts per day, the labor cost of visual judgment process will reach 600000/year, which is the process with the lowest input-output ratio of the whole company. Therefore, Liuzhou Yuanchuang hopes to release some manpower with the help of AI technology as soon as possible to improve the audit efficiency of quality inspection.
Solution
Step 1: filter the standard sample set according to the quality inspection target;
Step 2: train and recognize the model repeatedly through EasyDL;
Step 3: Develop automation scheme;
Step 4: Implement and deploy software and hardware to complete the automation scheme;
Step 5: The user first uploads the pictures of samples to be tested each time through the automation system, then uploads the passed recognition model in real time for judgment, and returns the corresponding processing results. Finally, the automation system classifies and flows the samples.
Relevant problems of fuel injection nozzle valve seat identified through the above automatic detection scheme: black, defect, scratch
