Teaching Machines What Engineers Already Know: QFD-Structured Neural Networks for Interpretable AI
Prof. Monica Leba and Prof. Andreea Ionica
Universitatea din Petrosani, Romania
Abstract:
Modern neural networks are powerful but opaque. Quality Function Deployment, by contrast, has spent decades encoding expert knowledge into structured, interpretable relationship matrices. This presentation asks a simple question: what happens when you build one inside the other? We introduce QFD-NN, a neural network architecture whose connectivity and initial weights are derived directly from the House of Quality. Rather than learning structure from data alone, QFD-NN uses the relationship matrix — mapping measurable inputs to technical evidence characteristics — as a structural prior that defines which connections are permitted and how strongly they should be initialized. A second matrix propagates those technical signals forward into interpretable requirement nodes aligned with the WHATs of the House of Quality. The result is a network that is not merely trained but designed: every hidden neuron has a name, every connection has an engineering justification, and every weight initialization reflects domain knowledge accumulated before a single training example is seen.
Brief Biography of the Speaker:
Professor Monica LEBA, University of Petrosani, ROMANIA
Monica Leba is Professor and Ph.D. supervisor in the field of System Control Engineering in the Computer and Electrical Engineering Department at University of Petrosani, Romania. She has a PhD (2002) in System Control Engineering from University of Petrosani, a Bachelor’s Degree in Applied Informatics (1998) and Master’s Degree in Computer Engineering (2007) from University of Petrosani. She is a member of IEEE and IFAC, Computers for Control Technical Committee. Her research interests are in applied informatics, modelling-simulation, algorithms design. She has more than 100 refereed research papers in international journals and conferences and international prizes for patents.
Professor Andreea Ionica is a Ph.D. supervisor in the field of Management and Engineering, Professor of Management in Management and Industrial Engineering Department at University of Petrosani, Romania. She has a PhD (2004) in Industrial Engineering, a Bachelor’s Degree in Engineering (1992), a Bachelor’s Degree in Economics (2002) from University of Petrosani and a Postgraduate Diploma (1998) from Institut National Polytechnique de Lorraine, France. The published papers address interdisciplinary and innovative research topics. She has more than 100 refereed research papers in international journals and conferences and international prizes for patents.