RoboTUNN Project Meeting: Update on AI-Driven Damage Analysis

At the recent Robo­TUNN pro­ject mee­ting at Fraun­ho­fer IPM and INATECH in Frei­burg, all part­ners came tog­e­ther to coor­di­na­te the next tech­ni­cal steps toward auto­ma­ted tun­nel inspec­tion. For us at Stat­Soft, one topic was cen­tral: advan­cing our models for AI-based dama­ge clas­si­fi­ca­ti­on — a key com­po­nent of the digi­tal twin and future pre­dic­ti­ve main­ten­an­ce approa­ches.

Improving Crack Classification

Seve­ral important tech­ni­cal decis­i­ons were made regar­ding dama­ge detec­tion. Going for­ward, all pre­dic­tions from the YOLO model will be trans­fer­red in full to the cen­tral data envi­ron­ment, allo­wing Stat­Soft to app­ly its own fil­te­ring logic to deter­mi­ne which detec­tions are rele­vant.
This is essen­ti­al for repro­du­ci­b­le crack clas­si­fi­ca­ti­on and for asses­sing the cri­ti­cal­i­ty of each dama­ge ins­tance.

Ano­ther step for­ward con­cerns the trai­ning of the under­ly­ing detec­tion models. Our part­ners at INATECH are eva­lua­ting, among other things, a reba­lan­cing of the trai­ning data to redu­ce class imba­lan­ce, as well as alter­na­ti­ve archi­tec­tures such as SAM3. In par­al­lel, we are asses­sing heu­ristic post-pro­ces­sing algo­rith­ms to miti­ga­te the well-known issue of over­ly wide crack anno­ta­ti­ons — a chall­enge faced across the enti­re field.

Predictive Maintenance: Initial Modeling Approaches

During work­shop ses­si­ons, seve­ral approa­ches for pre­dic­ting the future deve­lo­p­ment of struc­tu­ral dama­ge were dis­cus­sed.
Par­ti­cu­lar­ly pro­mi­sing is the use of expe­ri­men­tal data from RWTH Aachen, whe­re crack for­ma­ti­on is recor­ded under con­trol­led con­di­ti­ons. The­se data­sets could pro­vi­de valuable insights into how exis­ting dama­ge evol­ves bet­ween inspec­tions. Addi­tio­nal ide­as include mois­tu­re detec­tion, moni­to­ring joint dis­pla­ce­ments via repea­ted LiDAR scans, and inte­gra­ting wea­ther data.

All approa­ches will be eva­lua­ted by the end of Janu­ary — an important step toward the first pre­dic­ti­ve main­ten­an­ce pro­to­ty­pe plan­ned for autumn 2026.

Integration of Sensor Systems

At the same time, INATECH and RWTH are pro­gres­sing with the inte­gra­ti­on of the mul­ti­mo­dal sen­sor sys­tem onto the robots. This work is cru­cial for Stat­Soft, as image, ther­mo­gra­phy, and point cloud data will soon feed joint­ly into our clas­si­fi­ca­ti­on and assess­ment models.

Conclusion

The mee­ting in Frei­burg high­ligh­ted the strong momen­tum of the Robo­TUNN pro­ject. For Stat­Soft, the resul­ting next steps are clear: more pre­cise data pipe­lines, enhan­ced clas­si­fi­ca­ti­on models, and deepe­ned col­la­bo­ra­ti­on with our part­ners. The recent advan­ces in crack clas­si­fi­ca­ti­on, in par­ti­cu­lar, bring us clo­ser to our goal of not only auto­ma­ting tun­nel inspec­tion pro­ces­ses but also making them pre­dic­ta­ble in the long term.

Categories
Latest News
Your contact

If you have any ques­ti­ons about our pro­ducts or need advice, plea­se do not hesi­ta­te to cont­act us direct­ly.

Tel.: +49 40 22 85 900-0
E-mail: info@statsoft.de

Gui­do Band­holz (Head of Sales)