How we use arti­fi­cial intelligence 

Micro­vista is con­sid­ered a pio­neer in the field of auto­mat­ed analy­sis and eval­u­a­tion rou­tines of CT scan data in the indus­tri­al sec­tor. Build­ing on our many years of expe­ri­ence, we have advanced the use of AI process­es in our soft­ware solu­tions and offer you indi­vid­ual AI solu­tions for indus­tri­al use with a high degree of robust­ness against fluc­tu­a­tions in the mea­sure­ment process.

We offer

Micro­vista offers cus­tomised IO/NIO assess­ments and/or quan­tifi­ca­tion of dis­con­ti­nu­ities using robust AI techniques.

  • Object clas­si­fi­ca­tion accord­ing to IO/ NIO
Lötstellenbewertung Bauteil IO
Sol­der joint eval­u­a­tion — com­po­nent IO 
Lötstellenbewertung Bauteil NIO
Sol­der joint eval­u­a­tion — com­po­nent NIO
  • Object detec­tion and local­i­sa­tion of e.g. frac­tures, chip residues, pores

Water jack­et

  • Object seg­men­ta­tion for 2D and 3D pore analy­sis: detec­tion, local­i­sa­tion and quan­tifi­ca­tion (area, vol­ume, diameter)
Einsatz von künstlicher Intelligenz

Pore analy­sis

By com­bin­ing AI meth­ods with clas­si­cal approach­es, we are able to per­form pix­el-pre­cise mea­sure­ments of geom­e­try features.

Our ser­vice extends along the entire devel­op­ment chain, start­ing from the prob­lem def­i­n­i­tion to the imple­men­ta­tion of the ML solu­tion in exist­ing sys­tems. This includes the gen­er­a­tion of train­ing data, the selec­tion and train­ing of a suit­able ML archi­tec­ture and its com­bi­na­tion with clas­si­cal image pro­cess­ing approach­es as well as the val­i­da­tion of the ML solution.

This is how we do it

1. Solu­tion approach: In a first con­ver­sa­tion we ask you for an exact descrip­tion of the prob­lem, this can be e.g. a bound­ary pat­tern cat­a­logue. Based on this infor­ma­tion, we can quick­ly assess whether our AI approach­es are suit­able for your con­cern. We then dis­cuss how we can set up and train our AI for your use case as well as imple­ment it in your exist­ing system.

2. Learn­ing data: If you already have com­po­nent scans avail­able, you can sim­ply pro­vide them to us. If no scans are avail­able, it is best to send us sam­ple parts from which we can cre­ate suit­able scans. The learn­ing and test basis for the AI solu­tion is built from the data mate­r­i­al. If you do not have enough data, it is part of our exper­tise to deal with this prob­lem and still enable the train­ing of an AI solu­tion with the help of syn­thet­ic data sets.

3. Train­ing: The AI is then trained, i.e. the eval­u­a­tion of your images is learned, and what is learned is sub­se­quent­ly val­i­dat­ed togeth­er using ref­er­ence data. The train­ing is an iter­a­tive process that aims to achieve sen­si­tiv­i­ty and speci­fici­ty of the AI solu­tion accord­ing to the test­ing requirements.

4. Imple­men­ta­tion: The devel­oped AI solu­tion is then sup­ple­ment­ed by fur­ther soft­ware com­po­nents (e.g. result pro­cess­ing, fur­ther analy­sis tools) and imple­ment­ed direct­ly into your exist­ing sys­tem. We also offer you inte­gra­tion into our own soft­ware envi­ron­ment, which takes care of every­thing from record­ing the com­po­nent num­ber and oth­er mas­ter data to cre­at­ing a cus­tomised test report.