Baidu PaddlePaddle easydl: when AI enters the factory, "small bearing" can also turn "big industry"

“ There are places to turn , There are bearings .” Electric toys in daily life 、 Bicycle 、 Electric cars and even car wheels can't rotate without it . Bearings are hidden inside various parts , Not only ensure their normal operation , It even affects the reform of the industry .

0.001mm

Is the distance from safety

Bearing , Talk small but not small , Speak big . General bearings consist of two rings （ Inner and outer rings ） Composed of balls , Fixed on a shaft , Ensure normal rolling . The accuracy of bearings is based on 0.001mm To measure , The manufacturing deviation of ordinary mechanical parts is 0.01mm.

Take the car for example , A car usually has about 30 Kind of 50 The remaining sets of bearings , Distributed on the car chassis 、 transmission 、 Wheel hub 、 brake disc 、 Engine, etc , The bearing rotating at high speed during driving has an absolute connecting role , There is no room for a slip , It is also related to the life and quality of the car .

The data released by the State Administration of market supervision shows ,2021 In, China implemented automobile recall 232 Time , involve 873 Thousands of cars . among , The number of vehicles recalled due to defects in engines and electronic and electrical components accounts for 84%. From the cause of defect recall , Vehicles recalled for manufacturing reasons account for 15%.

Although the automobile bearing is a small part , But it can be seen everywhere in cars , And the power of the car 、 reliability 、 Safety and comfort are closely related . therefore , How to ensure the yield of factory bearings 100% crucial .

「 Grinding burns 」 problem

  Both brain and money

In bearing production , Quality inspection is an important part . Usually , The quality inspector uses artificial naked eye observation 、 Pick out the greasy dirt on the surface by means of quality inspection such as finger rotation bearing 、 Bearing with defects such as scratches . But man is not a machine , We can never achieve the yield 100%.

“ Visual fatigue is easy to appear in manual quality inspection , Employees can't stay focused for a long time , There will be missed detection 、 False detection , Affect the yield of products . Even if employees work every day 10 Hours , It is also difficult to cope with the continuous production demand of the production line . meanwhile , Factories also face high labor costs .” Wesken （ Xiamen ） Intelligent Technology Co., Ltd （ Hereinafter referred to as wesken ） Lishuihan, head of the instrument business department, said . For him and his team , Slave guest The production workshop of the household takes back the demand , And solving business problems through technology research and development is their important task .

At first , Facing the needs of customers, Li shuihan and his team will rely on their own business experience , Analyze the needs 、 test , Through the previous eddy current testing to solve . Eddy current testing can detect the length of the bearing 、 Thickness and other parameters , But the oil floating on the bearing surface 、 defects , In especial “ Grinding burns ” Always difficult to solve .

Grinding burn is an undesirable phenomenon of workpiece after grinding , Due to improper processing and other reasons , Under the action of higher grinding temperature, the metallographic structure of the surface layer of the workpiece is changed , In essence, it belongs to a hidden defect on the surface of the workpiece .

Many parts with grinding burns are often used in the working environment of alternating load , Such as steering knuckle in automobile powertrain 、 transmission shaft 、 Bearings and other parts , The surface quality of the workpiece is highly required . This pair has 20 For Li shuihan, who has years of research and development experience , Finding the technical method to solve this kind of problem is also a difficult problem .

later , Based on the understanding of deep learning , Lishuihan built his own algorithm team internally , Try to pass through AI Vision algorithm solves “ Grinding burns ” This problem . But after more than a year , Nearly one million R & D costs have been invested , Subject to the effect of technology and algorithm development , Finally, it failed to land on the customer's production line .

Baidu Flying propeller EasyDL help AI Enter the factory

Win the super big order , Cost reduction 90%

Solve grinding burn through visual inspection , Is there a low-cost development method ？

On trial OpenCV、Halcon Wait for a series of products , Li shuihan finally locked the key to solve the problem in the market with mature algorithm ability AI Development platform —— Baidu Flying propeller EasyDL.

Flying propeller EasyDL It is a zero threshold for enterprises AI Development platform , Zero algorithm basis can customize high precision AI Model , Based on the industrial level deep learning platform Flying propeller ,EasyDL Built in a rich Baidu large-scale pre training model and automatic model optimization mechanism , It can realize one-stop intelligent annotation 、 model training 、 Service deployment and other whole process operations .

Different from eddy current testing previously applied , Wesken based on Baidu Flying propeller EasyDL A visual inspection system for finished bearings is built based on the deep learning algorithm of . adopt AI Visual inspection , It can deal with complex and diverse bearing defects , Such as end face 、 Outer diameter scratch , what's more , The grinding burn problem that has been troubled before has been well solved .

With Baidu Flying propeller EasyDL Blessing , Wesken's finished bearing visual inspection system has been put into operation in just three months . It can not only improve the speed of automatic troubleshooting ,100ms One test can be completed within , It can also maintain the false positive rate at 3% about , Improve the accuracy of detection .

“ Compared with traditional development methods , Our project uses EasyDL Model development , The overall development cost is reduced 90%.” Lishuihan mentioned , Based on Baidu Flying propeller EasyDL Developed finished bearing visual inspection system , also Help wesken get the first place in the domestic bearing field 、 Seventh in the world “ Our group ” Super big order , Help the group quickly improve the efficiency of quality inspection of production lines .

“ Small bearing ” turn “ Big industry ”

AI The upgrade is expected in the future

In the field of industry , In order to get rid of manpower, quality inspection must carry out technological upgrading and transformation . Every enterprise will experience the pain of transformation , It depends on who is fast and efficient . Take wesken for example , From traditional eddy current testing to machine vision 、 Deep learning , Wesken not only through Baidu Flying propeller EasyDL Realize the upgrading of its technology , It also helps its customers achieve the purpose of reducing costs and increasing efficiency .

At present , All walks of life are also facing standardization 、 Intelligent 、 The need for Automation . Guangxi Liuzhou Yuanchuang EFI, a national high-tech enterprise, passed Baidu Flying propeller EasyDL In depth cooperation , Developed a quality inspection system for the valve seat of the fuel injection nozzle , Help enterprises save 60 Labor cost of more than 10000 yuan . Beijing discrimination technology , Baidu is preset in its edge quality inspection camera Flying propeller EasyDL platform , So that quality inspectors who do not understand the code can easily use , Let industry AI Landing is easier .

Baidu Flying propeller EasyDL Provide... For the industry AI The basis of transformation “ base ”, Provide powerful AI power , Improve the success rate of enterprise digital intelligence transformation .

Actually , Every enterprise is a “ Bearing ”, Let's rotate together , Jointly drive the industry AI to ground .

notes ： Flying propeller EasyDL Model capability is provided by Wenxin big model