Applied Computational Intelligence | Research Group at Schmalkalden University of Applied Sciences

Prediction of system failures using vibroacoustic sensors

Autonomous, automatically learning system for monitoring and advance warning of degeneration of filter stages

Industrial application for
- Machine monitoring (condition based monitoring, CBM)
- Predictive maintenance (PdM)
ARD methods (automatic relevance determination): several competing approaches
- Relevance support vector method (RSVM)
- Gradient boosting method (LightGBM)
- Learning vector quantization (LVQ)
- Generalized Relevance LVQ (GRLVQ)
- Generalized Matrix Relevance LVQ (GMLVQ)
- Localized GMLVQ extension (LGMLVQ)
- Brute force approach: OLVQ1 combined with evolutionary strategies
Comprehensive test application
AI framework for process chain optimization and distributed computing
- Independent implementation
- External cloud computing versus local grid computing 23 TFLOPs, double precision, tensorflow units
- Integrated pre-processing
- Integrated feature extraction
  - Hilbert transform; estimation of the instantaneous frequency
  - Cepstrum analysis
  - Entropy estimation; approximate entropy
  - Discrete wavelet transform
  - Independent component analysis (ICA)
  - Empirical mode decomposition (EMD)
- Training
- Validation
  - Cross-validation: repeated, random samples
  - Leave-One-Subject-Out cross-validation
Funded by the Federal Ministry for Economic Affairs and Energy (BMWi)
Partner: 1 medium-sized company

Surface and overprint inspection

Automated optical assessment of the quality of overprints on extruded products

Detection of impairment of the vehicle driver

Influence of controlled amounts of alcohol on driving ability Influence of light and medium driver fatigue on driving ability

Recording of vehicle dynamic variables on a non-public test track
Research question: independent of individual driving behavior, patterns typical of mental impairment should be identified
Adaptive signal processing for feature extraction
Discriminant analysis with machine learning methods
Client: Medium-sized company

Embedded license plate recognition

Camera system with built-in pattern recognition

Smart embedded systems with geomagnetic field sensors

Recognize vehicle types in moving traffic

Signal acquisition of a geomagnetic field sensor in dowel form integrated in the road surface
Disturbance of the local geomagnetic field by ferromagnetic components in the vehicle
Detection of eight vehicle types and one residual class (further unknown types)
Determination of speed
Embedded signal analysis
Feature extraction and feature reduction based on automatic relevance determination
Embedded pattern recognition in the microcontroller system
Client: Small company

Signal analysis of correlated optical time domain reflectometry (COTDR)

Recognition of characteristic patterns in COTDR spectra

Simultaneous acquisition of COTDR spectra on hot optical fibers, i.e. active transmission lines
Objectives: Monitor quality, detect manipulations
Real-time requirements
Embedded, adaptive signal analysis
Embedded pattern recognition
Client: Subcontract in a funded project

License plate recognition

Automatic search of license plates in a multimedia database