Current Industrial AI research at TU Dortmund & Lamarr Institute · Prior NLP/ML work at University of Hamburg · 2022–Present
Research on intelligent systems for machining process stability — combining domain expert knowledge, acoustic signal analysis, and machine learning to support data-driven manufacturing decisions.
Evaluating whether low-cost microphones (B5, HI10, TASCAM) can capture chatter-relevant acoustic information compared with a PCB reference microphone — toward physics-informed, cost-effective industrial sensing.
Approach: Comparing PCB, B5, HI10, and TASCAM recordings using spectrogram similarity, dominant-frequency behavior, frequency components, and impact-hammer-based resonance interpretation.
Impact: Contributes toward physics-informed, cost-effective acoustic sensing for industrial machining monitoring.
Supporting thesis research on predicting tool wear from force, temperature, and acoustic machining signals — addressing shortcut learning and generalization across cutting conditions.
Approach: CNN, Random Forest, Ridge Regression, ablation studies, interpolation strategies, stride sensitivity analysis, feature engineering from multimodal signals.
Connecting physical resonance behavior of machine tools with data-driven chatter monitoring — supporting impact hammer analysis, FRF estimation, and coherence interpretation.
Transformer-based event detection and extraction from unstructured text, with geospatial mapping for situational awareness and decision support. Active GitLab contributions visible at conf.informatik.uni-hamburg.de.
Approach: Transformer-based event detection, entity normalization, web application for mapping events by location and criticality.
Time-series forecasting of per-machine and overall energy consumption using GRU/FFN models for industrial optimization and proactive monitoring.
Approach: GRU/FFN models, feature engineering, REST APIs for integration with industrial systems.
Web-based arms licensing management system for the District Kotli government office, AJK. Digitised the entirely manual paper-based licensing workflow for thousands of arms licence holders.
Problem: The existing manual system suffered from huge file volumes, security risks, no backup, and slow searching. Check posts had no online verification capability.
Solution — key features:
Architecture: Multi-tier web application, .gov domain hosting, relational database (MySQL), Object Modelling Technique (OMT), PHP backend.
Stakeholders: Govt. District Kotli (DC Office), State AJK, Licence Applicants, Check Post Investigators.
Outcome: Presented and defended at internal defence (August 2016) at Dept. CS&IT, UMSIT Kotli.
Intelligent networked monitoring and diagnostics module for condition-based real-time monitoring of tire heating presses — predicting cycle times, energy consumption, and possible errors for economic and ecological optimization in the tire industry.
Approach: LSTM, GRU, and MLP models for phase length prediction and energy forecasting in tire curing presses. Dataset preparation through REST APIs and SQLAlchemy. Unit and integration testing for robust solutions.
Goal: Increase machine performance and availability, ensure quality and reduce energy consumption in tire manufacturing.