AI Models Trained in Labs Fail in Real Factories—Until Now
A new study shows that machine learning models trained in controlled laboratory settings often break down when deployed in actual industrial environments. Transfer learning techniques can close this gap by up to 30%, potentially unlocking AI-powered predictive maintenance across manufacturing and critical infrastructure industries where equipment failures are costly.
Originaltitel: Lab-to-Field Generalization Gap: Assessment of Transfer Learning for Bearing Fault Detection
<p>The integration of Artificial Intelligence into industrial maintenance remains challenging due to the scarcity of high-quality data representing faulty conditions. Machine Learning models trained on laboratory testbed data often fail to generalize effectively in real workshop environments. This study evaluated the effectiveness of Transfer Learning models in handling this domain shift challenge compared with Machine Learning models. Their potential to address the generalization gap was assessed by analyzing the model adaptability from lab-recorded data to data from emulated workshop conditions, where real-world variability was replicated by embedding synthetic noise into the lab-recorded data. The case study focuses on detecting rotor unbalance through bearing vibration signals at varying speeds. A Support Vector Classifier was trained on the transformed features for both models for binary classification. Model performance was assessed under varying data availability and noise conditions to evaluate the impact of these factors on classification accuracy, sensitivity, and specificity. The results show that Transfer Learning outperforms Machine Learning, achieving up to 30% higher accuracy under high-noise conditions. Although the Machine Learning model exhibits greater sensitivity, it misclassifies balanced cases and reduces specificity. In contrast, the Transfer Learning model maintains high specificity but has difficulty detecting mild unbalance levels, particularly when data availability is limited.</p>