A data scientist for the avia­tion industry of tomorrow

In the en­gine man­u­fac­tur­ing busi­ness, there is noth­ing more im­por­tant than high-qual­i­ty com­po­nents. Take blisks, for ex­am­ple: in these high-tech com­po­nents that in­te­grate blade and disk, even a hun­dredth of a mil­lime­ter can make all the dif­fer­ence. In the worst case, the en­tire com­po­nent may end up be­ing worth on­ly as much as the ma­te­ri­als used to make it. Fol­low­ing cur­rent pro­ce­dures, pro­duc­tion en­gi­neers can­not be sure a blisk meets spec­i­fi­ca­tions un­til af­ter the fi­nal in­spec­tion, long af­ter the hours-long ma­chin­ing process has been com­plet­ed.

What if it were pos­si­ble to de­tect crit­i­cal de­vi­a­tions much ear­li­er on? For ex­am­ple, what if it were pos­si­ble to re­act to signs of a tol­er­ance de­vi­a­tion dur­ing the man­u­fac­tur­ing process? Then it would be pos­si­ble to avoid ex­ten­sive and cost­ly re­work­ing.

Da­ta that is pro­duced, ac­cessed or processed by man­u­fac­tur­ing ma­chin­ery dur­ing the pro­duc­tion process can play a de­ci­sive role in find­ing a so­lu­tion. But there is a huge quan­ti­ty of in­for­ma­tion. To be able to uti­lize it, some­one has to de­ter­mine which in­for­ma­tion is rel­e­vant. This per­son an­a­lyzes the da­ta, de­codes it, and finds con­nec­tions.

For blisk man­u­fac­tur­ing at MTU Aero En­gines, Dr. Oliv­er Arnold is that per­son. A 34-year-old physi­cist and da­ta re­searcher, Arnold left a po­si­tion at CERN, the Eu­ro­pean Or­ga­ni­za­tion for Nu­clear Re­search, about a year ago for his cur­rent job in Mu­nich. In the can­ton of Gene­va he had worked on high-en­er­gy ex­per­i­ments in ba­sic physics re­search. At MTU he wants to cre­ate al­go­rithms that bun­dle valu­able da­ta re­lat­ed to the ex­treme­ly com­plex blisk man­u­fac­tur­ing process.

“In plain words, we take a look at the ma­chine da­ta that is pro­duced in the process of ma­chin­ing the blisks,” Arnold ex­plains. “We try to de­tect pat­terns and in­ter­de­pen­den­cies in the da­ta that may be re­flect­ed in sub­stan­dard com­po­nent qual­i­ty, which doesn’t be­come ap­par­ent un­til lat­er in the man­u­fac­tur­ing process.” The ex­trac­tion of re­li­able trends would pro­vide pro­duc­tion en­gi­neers with a sol­id foun­da­tion for their de­ci­sion-mak­ing, en­abling them to re­act at an ear­li­er stage, im­prove process­es and cut costs.

Achiev­ing this goal is not as easy as it sounds. How can the tru­ly rel­e­vant in­for­ma­tion be ex­tract­ed from the enor­mous quan­ti­ties of da­ta sup­plied by the ma­chines? How of­ten would it be ad­vis­able to cap­ture the da­ta? What lev­el of da­ta qual­i­ty is need­ed to de­ter­mine which causal­i­ties? And what ad­di­tion­al sen­sors might be ur­gent­ly need­ed in the fu­ture—and should thus al­ready be in­clud­ed in the spec­i­fi­ca­tions for new ma­chin­ery that is about to be pro­cured? “We are still at a rel­a­tive­ly ear­ly stage,” says Arnold. “But we are mak­ing our way to­ward the so­lu­tions step by step.”

The de­vel­op­ment of a com­plete­ly da­ta-dri­ven un­der­stand­ing of process­es is still a very new sci­ence. For ex­am­ple, the Tech­ni­cal Uni­ver­si­ty of Mu­nich (TUM) was one of the very first uni­ver­si­ties to of­fer a mas­ter’s de­gree in da­ta sci­ence, in­tro­duc­ing it in 2015. Four years on, pro­fes­sion­als in this field are now in­dis­pens­able in every high-tech in­dus­try.

“The fas­ci­nat­ing thing about this for me is def­i­nite­ly the in­ter­dis­ci­pli­nary ap­proach,” says Arnold. “I my­self can­not com­pre­hend the da­ta with­out help from my col­leagues in­volved in the man­u­fac­tur­ing process. They pro­vide the process ex­per­tise and to­geth­er we trans­form it—along with the da­ta—in­to a mod­el.” One could view Arnold as an in­ter­preter: at the in­ter­face be­tween in­for­ma­tion tech­nol­o­gy and man­u­fac­tur­ing strat­e­gy, his analy­sis pro­vides sup­port for the de­ci­sion-mak­ing process.