Case study - ar­ti­fi­cial in­tel­li­gence

DEMM Engineering & Manufacturing - - CONTENTS - DE­TAILS, VISICON­SULT.DE

AR­TI­FI­CIAL IN­TEL­LI­GENCE (AI) or ma­chine learn­ing is on the rise in al­most ev­ery in­dus­try. In ad­di­tion to prom­i­nent ex­am­ples, like au­ton­o­mous driv­ing, the in­spec­tion in­dus­try is also look­ing into this tech­nol­ogy. The Ger­man R&D driven com­pany, VisiCon­sult X-ray Sys­tems & Solutions GmbH for ex­am­ple, is cur­rently de­vel­op­ing an AI plat­form to de­tect de­fects even more re­li­ably and more au­to­mated than in the past.

X-ray tech­nol­ogy is al­ways used in the non- de­struc­tive test­ing (NDT). Un­like op­ti­cal pro­cesses, in­ter­nal de­fects like pores, in­clu­sions or ge­ome­tries can also be tested. Over the past decade of digi­ti­sa­tion, dig­i­tal de­tec­tors have been used much more heav­ily in­stead of ana­logue film. With this digi­ti­sa­tion step, the data is avail­able in real time and can be eval­u­ated by com­put­ers. This is es­pe­cially rel­e­vant in the au­to­mo­tive sec­tor with its high vol­umes and for 20 years VisiCon­sult has been de­liv­er­ing so- called Au­to­mated De­fect Recog­ni­tion (ADR) sys­tems, which de­tect er­rors au­to­mat­i­cally af­ter a hu­man ini­ti­ated com­po­nent-spe­cific pa­ram­e­ter­i­sa­tion.

Nowa­days, production cy­cles are get­ting shorter and shorter and the mar­ket is call­ing for faster ramp-up times. Fur­ther­more, the parts are go­ing to be lighter and more com­plex, while the qual­ity re­quire­ments in­crease to the same ex­tent. The clas­sic im­age pro­cess­ing meth­ods of the X-ray spe­cial­ists re­quire a high po­si­tion­ing ac­cu­racy and an ex­ten­sive com­po­nent pa­ram­e­ter­i­sa­tion.

By us­ing ar­ti­fi­cial in­tel­li­gence, this ef­fort could be mas­sively re­duced, pre­dict the spe­cial­ists from VisiCon­sult. Ini­tial tests with avail­able AI frameworks al­ready achieved de­tec­tion rates of up to 90 per­cent in a blind study com­pared to the con­ven­tional pro­ce­dure. “The tests in this area are still in their in­fancy,” says Len­nart Schu­len­burg, au­tho­rised sig­na­tory at VisiCon­sult. “Since our sys­tems are used to test safety- crit­i­cal com­po­nents, ex­ten­sive tests and qual­i­fi­ca­tions are nec­es­sary.” Au­to­ma­tion is one of the fo­cus top­ics of the fam­ily busi­ness, and in 2107 VisiCon­sult con­cluded a re­search co­op­er­a­tion with the Univer­sity of Lübeck. As part of this co­op­er­a­tion, there are now sev­eral master and doc­toral the­ses on the sub­ject of AI. “The Univer­sity of Lübeck is a leader in med­i­cal imag­ing. Since they al­ready in­ves­ti­gated heav­ily in AI, this area is al­ready far ahead of our in­dus­try. We are cur­rently trans­fer­ring this knowl­edge to our ap­pli­ca­tions,” says Schu­len­burg.

Due to qual­i­fi­ca­tion obstacles, ADR has his­tor­i­cally been ex­cluded from high- crit­i­cal ar­eas such as avi­a­tion, but this is chang­ing and there is a great in­ter­est in qual­i­fy­ing the pro­ce­dures ac­cord­ingly. There­fore, VisiCon­sult is work­ing with a lead­ing avi­a­tion group in a Euro­pean avi­a­tion re­search project to es­tab­lish ADR as a test method. The use of AI is also be­ing tested in this area, es­pe­cially in the field of Au­to­matic De­fect Recog­ni­tion within welds, where VisiCon­sult hopes for a break­through with the smart ma­chines.

Cru­cial for the suc­cess of an AI is the num­ber and qual­ity of the un­der­ly­ing im­age data. “Data is the oil of In­dus­try 4.0,” em­pha­sises Schu­len­burg. For this rea­son, VisiCon­sult has de­vel­oped a cloud so­lu­tion into which co­op­er­a­tion part­ners can up­load eval­u­ated im­age data. These are anonymised and used to pa­ram­e­terise the AI. The higher the qual­ity of the data, the bet­ter the AI can be­come. VisiCon­sults’ er­ror data­base in­cludes mul­ti­val­ued multi-user records with marked and clas­si­fied er­rors. As a spe­cial­ist for fully au­to­mated in­spec­tion sys­tems, VisiCon­sult has de­vel­oped a patent for au­to­matic ro­bot po­si­tion­ing whereby through a com­bi­na­tion of robotics, 3D sen­sors and im­age reg­is­tra­tion, a high po­si­tion­ing re­peata­bil­ity can be achieved with­out ad­di­tional com­po­nents.

This sys­tem will now be com­bined with the new AI solutions, re­duc­ing the ramp-up process from sev­eral days to a few hours. Com­po­nents are then moved in mixed op­er­a­tion on a con­veyor belt in the sys­tem and are au­to­mat­i­cally de­tected by the sys­tem. The soft­ware would recog­nise which com­po­nent is to be in­spected and guides the robotics to the ex­act in­spec­tion- po­si­tion. This would elim­i­nate set-up times com­pletely and in­crease the through­put mas­sively. As a side ef­fect, the pseudo-re­jects are re­duced and the de­tec­tion prob­a­bil­ity is in­creased. Both are re­quire­ments for an aero­space ap­proval of the sys­tem.

Newspapers in English

Newspapers from New Zealand

© PressReader. All rights reserved.